Shorten component readme

- At the end there were not that many unique facts in there. It now only lists the relevant bits without much text around it.
Merge pull request #4428 from element-hq/toger5/component-assets-as-files
2026-11-13 01:12:29 +00:00 · 2026-11-10 14:31:12 +01:00 · 2026-11-10 14:18:30 +01:00 · 2026-11-10 07:46:51 +01:00 · 2026-11-10 07:38:23 +01:00 · 2026-11-10 07:36:15 +01:00
49 changed files with 1424 additions and 456 deletions
@@ -71,10 +71,27 @@ jobs:
          flags: unittests
          fail_ci_if_error: true
  playwright:
-    name: Run end-to-end tests
+    name: Run end-to-end tests (${{ matrix.part }})
    if: ${{ !inputs.disable_e2e_tests }}
    timeout-minutes: 60
    runs-on: ubuntu-latest
+    # We manually shard the tests across runners. Playwright's own shards cut
+    # the list of tests into equal counts in file order, which would leave the
+    # widget suite, the slowest tests by far, together at the end.
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - part: chromium-1
+            args: --project=chromium --shard=1/2
+          - part: chromium-2
+            args: --project=chromium --shard=2/2
+          - part: widget-1
+            args: --project=widget --shard=1/2
+          - part: widget-2
+            args: --project=widget --shard=2/2
+          - part: firefox-mobile
+            args: --project=firefox --project=mobile
    steps:
      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
        with:
@@ -107,17 +124,64 @@ jobs:
        run: pnpm playwright install --with-deps
      - name: Run backend components
        run: |
+          # Make sure the directory for the embedded build exists as Docker might
+          # create it with wrong access rights otherwise.
+          mkdir -p embedded/web/dist
          docker compose -f docker-compose-dev.yml -f docker-compose-playwright.yml pull
          docker compose -f docker-compose-dev.yml -f docker-compose-playwright.yml up -d
          docker ps
      - name: Run Playwright tests
        env:
          USE_DOCKER: 1
-        run: pnpm exec playwright test
+          ARGS: ${{ matrix.args }}
+        # Unquoted on purpose: the part's arguments are several words
+        run: pnpm exec playwright test $ARGS
      - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
        if: ${{ !cancelled() }}
+        with:
+          name: blob-report-${{ matrix.part }}
+          path: blob-report
+          if-no-files-found: error
+          # Only read by the merge below.
+          retention-days: 1
+
+  e2e_report:
+    # Combines the blobs into a single HTML report and blocks PRs on failure.
+    name: Run end-to-end tests
+    needs: playwright
+    if: ${{ always() && !inputs.disable_e2e_tests }}
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        with:
+          # Inside a called workflow, github.repository is the caller so we need to make sure to still check out the EC repo here.
+          repository: ${{ inputs.matrix-js-sdk-sha && 'element-hq/element-call' || github.repository }}
+          persist-credentials: false
+      - name: Enable Corepack
+        run: corepack enable
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
+        with:
+          cache: "pnpm"
+          node-version-file: ".node-version"
+      - name: Install dependencies
+        run: scripts/layered.sh --ignore-pnpmfile
+        env:
+          PR_HEAD_REPO_OWNER: ${{ github.event.pull_request.head.repo.owner.login }}
+          JS_SDK_GITHUB_BASE_REF: ${{ inputs.matrix-js-sdk-sha }}
+      - name: Download the parts' reports
+        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
+        with:
+          pattern: blob-report-*
+          path: all-blob-reports
+          merge-multiple: true
+      - name: Merge into one HTML report
+        run: pnpm exec playwright merge-reports --reporter html ./all-blob-reports
+      - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
        with:
          name: html-report
          path: playwright-report
          if-no-files-found: error
          retention-days: 4
+      - name: Fail on a failed part
+        if: ${{ needs.playwright.result != 'success' }}
+        run: exit 1
@@ -221,119 +221,68 @@ See also:

 - [Developing with linked packages](./docs/linking.md)

-#### Element Call as a component (experimental)
+#### Element Call as a component

-Element Call can also be embedded directly into another React application
-rather than being loaded in an iframe as a widget. `pnpm build:component`
-builds it as a library into `component/dist` (the bundle, its stylesheet and
-type declarations), and
+Element Call can be rendered inside another React app instead of an iframe.
+The host supplies its own Matrix client, and the call stays inside the element
+it is mounted in.

-```sh
-pnpm dev:component
-```
-
-serves a harness on port 3001 that stands in for such an application: it signs
-in twice against the development backend and shows two calls side by side, in
-resizable boxes, with page furniture of its own around them. Use it to see how
-Element Call behaves when it does not own the page — the size it is given,
-whether it stays inside its container, and what it says to its host, which is
-logged along the bottom. The harness is served with the same development
-certificate as the app, so unless the development CA is trusted, the browser
-needs a certificate exception for `https://localhost:3001` as well (see the
-note under [Backend](#backend)). It reads the same `public/config.json` as
-`pnpm dev` if one exists, and runs with Element Call's defaults otherwise.
-
-The call lays itself out for the size of the element it is mounted in, not the
-window: a host that shrinks the container to a corner of its page gets the
-picture-in-picture layout, just as a host that shrank the whole iframe used to.
-The breakpoints in the stylesheets the component uses are
-`@container element-call` queries against its root element for the same reason;
-for the standalone app the root is the page, so they mean what the media queries
-they replaced did. (The standalone-only views, such as the home and login pages,
-still use plain media queries, since the component never shows them.)
-
-The component's stylesheet is confined to the element it is mounted in: the
-build rewrites every selector so that it matches only Element Call's root or
-what is inside it, with `html`, `body` and `:root` standing for that root (see
-`component/build/scopeStylesToRoot.ts`). A host's own page keeps its styles,
-and Element Call brings its own fonts and design tokens along.
-
-The component speaks every language the app does. English is bundled in; the
-other locales are split into chunks the host's bundler loads the first time
-they are needed. It starts in the browser's language, and follows the host's
-own language setting through the `language` prop (`supportedLanguages` lists
-the tags it accepts, and anything else falls back to its base language or to
-English). The `theme` prop works the same way and takes the same values as the
-widget's `theme` URL parameter: `light`, `dark`, `light-high-contrast` or
-`dark-high-contrast`. Both can change while a call is running without
-disturbing it.
-
-A host must call and await `initializeElementCall(config)` once before
-rendering the component: it loads the `Intl` polyfills, applies the
-deployment-wide `config.json`-style configuration and sets up translations.
-The component itself takes the host's `client` and the `roomId` to call in, an
-`intent` saying what the user asked for (which decides whether to show the
-lobby, ring, and so on), an optional `config` overriding what the intent
-implies, and an optional `hostBridge` through which Element Call tells the host
-that the user has joined or hung up, that it wants to stay on screen, and so
-on. The host makes its own requests (`join`, `hangUp`, `setDeviceMute`) through
-the handle exposed on `ref`. The full API is documented in the type declarations
-(`component/index.tsx` and `component/host.ts`).
-
-A few things differ from the widget on purpose: the component draws a solid
-background rather than a gradient unless told otherwise, never offers to edit
-the user's profile (the account is the host's), scopes its keyboard shortcuts to
-its own root element so that several instances can share a page, and only shows
-its own post-call and error screens when the host has not supplied a `close()`
-callback; with one, it asks the host to unmount it instead. The
-[global JS controls](./docs/controls.md) on `window` are unchanged and remain
-page-wide, so with several instances on one page they apply to all of them.
-
-Each release publishes the package to npm, with the same version as the release:
+##### Install

 ```sh
 npm install @element-hq/element-call-component
 ```

-To use an unreleased commit, a host can instead install it as a git dependency
-on the `component` directory of this repository,
+Releases publish the package with the same version as Element Call. For an
+unreleased commit, depend on this repository directly:

 ```json
 "@element-hq/element-call-component": "github:element-hq/element-call#main&path:/component"
 ```

-whose `prepare` script runs the build on install. That build needs pnpm (via
-Corepack) on the host's machine, runs a full `pnpm install` of this repository,
-since it inherits the `--max-old-space-size` setting of
-the app build; the host's pnpm also has to allow it to run at all
-(`allowBuilds` in its `pnpm-workspace.yaml`). Note that `component/` is a pnpm
-project of its own for this reason, so pnpm commands run from inside that
-directory target it rather than the repository; run them from the repository
-root. The host imports the component from
-`@element-hq/element-call-component` and the stylesheet from
-`@element-hq/element-call-component/style.css`, and has to provide `react`,
-`react-dom`, `matrix-js-sdk`, `@vector-im/compound-web` and
-`@vector-im/compound-design-tokens` itself, since the bundle leaves them
-external.
-Compound's stylesheet are the host's to load too: the component shares
-the host's copy of Compound, (required for example for tooltips)
+This builds on install, so the host needs pnpm (via Corepack) and must allow
+the build (`allowBuilds` in its `pnpm-workspace.yaml`).

-The component is large, and a host will usually load it lazily, only once a
-call is shown. Everything a host needs in order to talk about a call before
-then is also exported
-from `@element-hq/element-call-component/api` (a small entry point that does not
-load the component).
+##### Use

- props and the types they are made of
- the handle
- the host bridge
- `UserIntent`
- `HeaderStyle`
- `BackgroundStyle` enums
- `configurationForIntent`, the defaults each intent implies
-  Import from it wherever a value such as
-  `BackgroundStyle.Solid` is needed on a path that must stay light; the main
-  entry point re-exports all of it too.
+1. Provide the externals yourself: `react`, `react-dom`, `matrix-js-sdk`,
+   `@vector-im/compound-web`, `@vector-im/compound-design-tokens`, and
+   Compound's stylesheet.
+2. Import `@element-hq/element-call-component/style.css`.
+3. Await `initializeElementCall(config)` once before rendering.
+4. Render the component with `client`, `roomId` and an `intent`; optionally
+   pass `config`, `hostBridge`, `language` and `theme` (both can change
+   mid-call). Control it through the `ref` handle (`join`, `hangUp`,
+   `setDeviceMute`).
+
+Load the component lazily. Types, enums and `configurationForIntent` are
+available without it from `@element-hq/element-call-component/api`. The full
+API is in `component/index.tsx`, `component/api.ts` and `component/host.ts`.
+
+##### How it differs from the widget
+
+- It lays out for its container, not the window: a small container gets the
+  picture-in-picture layout.
+- Its styles and keyboard shortcuts are scoped to its root, so the host page is
+  untouched and several instances can share a page. The
+  [global JS controls](./docs/controls.md) stay page-wide.
+- It draws a solid background by default and never offers profile editing.
+- If the host bridge implements `close()`, it asks to be unmounted instead of
+  showing its own post-call and error screens.
+- Sounds, images and the MediaPipe runtime are separate files the host's
+  bundler copies (webpack 5, Vite, Rollup); MediaPipe is fetched only when a
+  background effect is first used.
+
+##### Develop
+
+- `pnpm build:component` builds the library into `component/dist`.
+- `pnpm dev:component` serves a test host on `https://localhost:3001`: two
+  calls side by side in resizable boxes, with host messages logged at the
+  bottom. It uses the development backend, `public/config.json` if present,
+  and needs the same certificate exception as `pnpm dev` (see
+  [Backend](#backend)).
+- Run pnpm commands from the repository root; `component/` is its own pnpm
+  project.

 ### Backend

@@ -26,8 +26,7 @@
  "features": {
    "feature_element_call_video_rooms": true,
    "feature_video_rooms": true,
-    "feature_group_calls": true,
-    "feature_release_announcement": false
+    "feature_group_calls": true
  },
  "default_federate": true,
  "default_theme": "light",
@@ -40,7 +39,10 @@
  },
  "setting_defaults": {
    "breadcrumbs": true,
-    "feature_group_calls": true
+    "feature_group_calls": true,
+    "releaseAnnouncementData": {
+      "room_list_section": true
+    }
  },
  "jitsi": {
    "preferred_domain": "meet.element.io"
@@ -26,8 +26,7 @@
  "features": {
    "feature_element_call_video_rooms": true,
    "feature_video_rooms": true,
-    "feature_group_calls": true,
-    "feature_release_announcement": false
+    "feature_group_calls": true
  },
  "default_federate": true,
  "default_theme": "light",
@@ -40,7 +39,10 @@
  },
  "setting_defaults": {
    "breadcrumbs": true,
-    "feature_group_calls": true
+    "feature_group_calls": true,
+    "releaseAnnouncementData": {
+      "room_list_section": true
+    }
  },
  "jitsi": {
    "preferred_domain": "meet.element.io"
@@ -19,7 +19,12 @@ services:
      - ./backend/playwright_homeserver-othersite.yaml:/data/cfg/homeserver.yaml:Z
  element-web:
    # Pin to a SHA so that upstream cannot break our tests. Renovate handles regular updates.
-    image: ghcr.io/element-hq/element-web:develop@sha256:e063472b6bdf6d0bfa1d43dfa4ef7effee1efd2c9684daee665d1bedfe746a47
+    image: ghcr.io/element-hq/element-web:develop@sha256:37d88b4426ff93a884b31c63d5319a7bcee32580f189c724e6b09f793a6de390
  element-web-1:
    # Pin to a SHA so that upstream cannot break our tests. Renovate handles regular updates.
-    image: ghcr.io/element-hq/element-web:develop@sha256:e063472b6bdf6d0bfa1d43dfa4ef7effee1efd2c9684daee665d1bedfe746a47
+    image: ghcr.io/element-hq/element-web:develop@sha256:37d88b4426ff93a884b31c63d5319a7bcee32580f189c724e6b09f793a6de390
+    volumes:
+      # The other site opens the embedded package under test in place of the
+      # copy its image bundles (the first site opens the dev server's app.
+      # Built by the Playwright run (see build.setup.ts).
+      - ./embedded/web/dist:/app/widgets/element-call:ro,Z
@@ -34,6 +34,10 @@ translations. The browser is Playwright's, so a fresh clone needs

 ## End-to-end

+The Element Web instance `HOST1` opens the app the dev server serves. The instance
+`HOST2` (the other site) opens the embedded package as a host ships it. This package
+is built by the `build-embedded` setup project and mounted in `docker-compose-playwright.yml`.
+
 - `playwright/*.spec.ts` — standalone, on Chromium and Firefox.
 - `playwright/widget/` — widget mode, where cross-context bugs surface.
 - `playwright/component/` — the component in a host page, via the harness on port
@@ -48,6 +52,8 @@ Test what a user observes: the peer sees the change, it survives a reconnect, it
 right after a reload. Reuse `playwright/spa-helpers.ts`,
 `playwright/widget/test-helpers.ts`, `playwright/fixtures/`.

+To mark new release announcements as viewed, add them in `backend/ew.test*.config.json`.
+
 ```sh
 pnpm test                   # unit + storybook
 pnpm backend                # Synapse + LiveKit, required for e2e
@@ -26,6 +26,19 @@ process.env.NODE_EXTRA_CA_CERTS = join(
  "backend/dev_tls_local-ca.crt",
 );

+const desktopChrome = {
+  ...devices["Desktop Chrome"],
+  permissions: ["clipboard-write", "clipboard-read", "microphone", "camera"],
+  ignoreHTTPSErrors: true,
+  launchOptions: {
+    args: [
+      "--use-fake-ui-for-media-stream",
+      "--use-fake-device-for-media-stream",
+      "--mute-audio",
+    ],
+  },
+};
+
 /**
 * See https://playwright.dev/docs/test-configuration.
 */
@@ -43,11 +56,11 @@ export default defineConfig({
   * Reporter to use. See https://playwright.dev/docs/test-reporters
   *
   * On CI, `list` prints a line per test as it finishes, which keeps the job
-   * log moving; `github` annotates failures on the PR; `html` is uploaded as
-   * an artifact and must never try to open a browser there.
+   * log moving; `github` annotates failures on the PR; the `blob` format is
+   * uploaded by sharded CI runs and combined into a HTML report by the workflow.
   */
  reporter: process.env.CI
-    ? [["list"], ["github"], ["html", { open: "never" }]]
+    ? [["list"], ["github"], ["blob"]]
    : [["list"], ["html"]],
  /* Stop a broken run early on CI instead of waiting out every retry. */
  maxFailures: process.env.CI ? 5 : undefined,
@@ -63,29 +76,28 @@ export default defineConfig({
  },
  /* Configure projects for major browsers */
  projects: [
+    {
+      // Builds the embedded package into the directory the other site's
+      // Element Web container serves it from (see docker-compose-playwright.yml).
+      name: "build-embedded",
+      testMatch: "**/widget/embedded/build.setup.ts",
+    },
    {
      name: "chromium",
-      testIgnore: "**/mobile/**",
-      use: {
-        ...devices["Desktop Chrome"],
-        permissions: [
-          "clipboard-write",
-          "clipboard-read",
-          "microphone",
-          "camera",
-        ],
-        ignoreHTTPSErrors: true,
-        launchOptions: {
-          args: [
-            "--use-fake-ui-for-media-stream",
-            "--use-fake-device-for-media-stream",
-            "--mute-audio",
-          ],
-        },
-      },
+      testIgnore: ["**/mobile/**", "**/widget/**"],
+      use: desktopChrome,
+    },
+    {
+      // Element Call as a widget in Element Web, on Chromium. This uses its
+      // own project so that it can be sharded on CI.
+      name: "widget",
+      dependencies: ["build-embedded"],
+      testMatch: "**/widget/**/*.@(spec|test).ts",
+      use: desktopChrome,
    },
    {
      name: "firefox",
+      dependencies: ["build-embedded"],
      testIgnore: "**/mobile/**",
      use: {
        ...devices["Desktop Firefox"],
@@ -106,6 +106,66 @@ async function drawsPerSecond(page: Page): Promise<number> {
  });
 }

+test.describe("the camera menu's preview", () => {
+  test("shows the camera with the effect in force", async ({ page }) => {
+    test.setTimeout(120_000);
+    await page.goto("/");
+    await SpaHelpers.createCall(page, "Preview user", "Preview in the menu");
+    const camera = await cameraColour(page.locator("video").first());
+
+    await page.getByRole("button", { name: "Camera", exact: true }).click();
+    const preview = page.getByRole("menu").locator("video");
+    await expect(preview).toBeVisible();
+    await expect
+      .poll(async () => distance(await averageColour(preview), camera), {
+        timeout: 20_000,
+      })
+      .toBeLessThan(30);
+
+    const tile = page.getByRole("menuitemradio", { name: "Background 1" });
+    await tile.click();
+    const picture = await averageColour(tile.locator("img"));
+    await expect
+      .poll(async () => distance(await averageColour(preview), picture), {
+        timeout: 60_000,
+      })
+      .toBeLessThan(distance(camera, picture) / 2);
+  });
+
+  test("shows the call's own camera once joined", async ({ page }) => {
+    test.setTimeout(120_000);
+    await page.goto("/");
+    await SpaHelpers.createCall(
+      page,
+      "Call preview user",
+      "Preview in the call's menu",
+      true,
+    );
+    await expect(page.getByTestId("incall_videomute")).toBeEnabled({
+      timeout: 10_000,
+    });
+    const camera = await cameraColour(page.locator("video").first());
+
+    await page.getByRole("button", { name: "Camera", exact: true }).click();
+    const preview = page.getByRole("menu").locator("video");
+    await expect(preview).toBeVisible();
+    await expect
+      .poll(async () => distance(await averageColour(preview), camera), {
+        timeout: 20_000,
+      })
+      .toBeLessThan(30);
+
+    const tile = page.getByRole("menuitemradio", { name: "Background 1" });
+    await tile.click();
+    const picture = await averageColour(tile.locator("img"));
+    await expect
+      .poll(async () => distance(await averageColour(preview), picture), {
+        timeout: 60_000,
+      })
+      .toBeLessThan(distance(camera, picture) / 2);
+  });
+});
+
 test.describe("background effects section", () => {
  test("section stays within the call area at every size", async ({ page }) => {
    test.setTimeout(120_000);
@@ -151,6 +211,46 @@ test.describe("background effects section", () => {
  });
 });

+test.describe("a pipeline that fails to build", () => {
+  test("offers no effects for the session, and keeps the choice", async ({
+    page,
+  }) => {
+    test.setTimeout(120_000);
+    await page.goto("/");
+    await SpaHelpers.createCall(page, "Failing user", "Pipeline fails");
+    const preview = page.locator("video").first();
+    const camera = await cameraColour(preview);
+    await page.route("**/*.tflite*", async (route) => route.abort());
+
+    await page.getByRole("button", { name: "Camera", exact: true }).click();
+    const tile = page.getByRole("menuitemradio", { name: "Background 1" });
+    await tile.click();
+    await expect(
+      page.getByText("Background effects are not supported on this platform."),
+    ).toBeVisible({ timeout: 30_000 });
+    await expect(tile).toBeDisabled();
+    await page.keyboard.press("Escape");
+    // The camera and microphone carry on without it.
+    expect(distance(await cameraColour(preview), camera)).toBeLessThan(30);
+    await expect(page.getByTestId("incall_mute")).toHaveAttribute(
+      "aria-checked",
+      "true",
+    );
+
+    await page.unroute("**/*.tflite*");
+    await page.reload();
+    await expect(page.getByTestId("lobby_joinCall")).toBeVisible();
+    await page.getByRole("button", { name: "Camera", exact: true }).click();
+    await expect(tile).toHaveAttribute("aria-checked", "true");
+    const picture = await averageColour(tile.locator("img"));
+    await expect
+      .poll(async () => distance(await averageColour(preview), picture), {
+        timeout: 60_000,
+      })
+      .toBeLessThan(distance(camera, picture) / 2);
+  });
+});
+
 test.describe("the chosen effect", () => {
  test("chosen effect survives leaving and rejoining", async ({ page }) => {
    test.setTimeout(120_000);
@@ -29,47 +29,10 @@ export interface MyFixtures {
  addUser: (username: string, host: string) => Promise<UserBaseFixture>;
 }

-// Minimal config.json for the local element-web instance
-const CONFIG_JSON = {
-  default_server_config: {
-    "m.homeserver": {
-      base_url: "https://synapse.m.localhost",
-      server_name: "synapse.m.localhost",
-    },
-  },
-
-  element_call: {
-    participant_limit: 8,
-    brand: "Element Call",
-  },
-
-  // The default language is set here for test consistency
-  setting_defaults: {
-    language: "en-GB",
-    feature_group_calls: true,
-  },
-
-  // the location tests want a map style url.
-  map_style_url:
-    "https://api.maptiler.com/maps/streets/style.json?key=fU3vlMsMn4Jb6dnEIFsx",
-
-  features: {
-    // We don't want to go through the feature announcement during the e2e test
-    feature_release_announcement: false,
-    feature_element_call_video_rooms: true,
-    feature_video_rooms: true,
-    feature_group_calls: true,
-  },
-};
-
 export const widgetTest = test.extend<MyFixtures>({
  // allow per-test override: `widgetTest.use({ callType: "dm" })`
  callType: ["room", { option: true }],
-  asWidget: async ({ browser, context, callType }, pUse) => {
-    await context.route(`http://localhost:8081/config.json*`, async (route) => {
-      await route.fulfill({ json: CONFIG_JSON });
-    });
-
+  asWidget: async ({ browser, callType }, pUse) => {
    const brooksDisplayName = `brooks_${Date.now()}`;
    const whistlerDisplayName = `whistler_${Date.now()}`;

@@ -164,13 +127,14 @@ export const widgetTest = test.extend<MyFixtures>({
   * Provide a way to add additional users within a test.
   * The returned user will be registered on the default homeserver, the name will be made unique by appending a timestamp.
   */
-  addUser: async ({ browser }, use) => {
+  addUser: async ({ browser }, use, testInfo) => {
    await use(
      async (
        username: string,
        host: string = HOST1,
      ): Promise<UserBaseFixture> => {
-        const uniqueSuffix = Date.now();
+        // Tests in parallel workers can reach this in the same millisecond.
+        const uniqueSuffix = `${Date.now()}_${testInfo.parallelIndex}`;
        const { page, clientHandle, mxId } = await TestHelpers.registerUser(
          browser,
          `${username.toLowerCase()}_${uniqueSuffix}`,
@@ -0,0 +1,20 @@
+/*
+Copyright 2026 Element Creations Ltd.
+
+SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-Element-Commercial
+Please see LICENSE in the repository root for full details.
+*/
+
+import { test as setup } from "@playwright/test";
+import { execFileSync } from "node:child_process";
+import { mkdirSync } from "node:fs";
+
+const EMBEDDED_DIST = "embedded/web/dist";
+
+setup("build the embedded package", () => {
+  setup.setTimeout(10 * 60 * 1000);
+  mkdirSync(EMBEDDED_DIST, { recursive: true });
+  execFileSync("pnpm", ["build:embedded", "--outDir", EMBEDDED_DIST], {
+    stdio: "inherit",
+  });
+});
@@ -0,0 +1,56 @@
+/*
+Copyright 2026 Element Creations Ltd.
+
+SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-Element-Commercial
+Please see LICENSE in the repository root for full details.
+*/
+
+import { expect, type Route } from "@playwright/test";
+
+import { widgetTest } from "../../fixtures/widget-user.ts";
+import { HOST2, TestHelpers } from "../test-helpers.ts";
+
+/** Where the other site's Element Web serves the embedded package from */
+const CONFIG_JSON = "**/widgets/element-call/config.json";
+
+const HOST_ANSWERS: Record<string, (route: Route) => Promise<void>> = {
+  "a 404": async (route) => route.fulfill({ status: 404 }),
+  "a failed fetch": async (route) => route.abort("failed"),
+  "an HTML page": async (route) =>
+    route.fulfill({
+      status: 200,
+      contentType: "text/html",
+      body: "<!doctype html><title>Element</title>",
+    }),
+};
+
+widgetTest.skip(
+  ({ browserName }) => browserName === "firefox",
+  "The widget suite does not run on Firefox, see the other specs here",
+);
+
+for (const [answer, respond] of Object.entries(HOST_ANSWERS)) {
+  widgetTest(
+    `reaches the lobby when config.json gets ${answer}`,
+    async ({ addUser }) => {
+      const { page } = await addUser("Brooks", HOST2);
+      await TestHelpers.createRoom("Config Room", page);
+
+      let answered = 0;
+      await page.route(CONFIG_JSON, async (route) => {
+        answered++;
+        await respond(route);
+      });
+
+      await TestHelpers.startCallInCurrentRoom(page);
+
+      await expect(
+        page
+          .locator('iframe[title="Element Call"]')
+          .contentFrame()
+          .getByTestId("lobby_joinCall"),
+      ).toBeVisible({ timeout: 15000 });
+      expect(answered).toBeGreaterThan(0);
+    },
+  );
+}
@@ -144,7 +144,11 @@ export class TestHelpers {

    await this.dismissStartupToasts(page);

-    await TestHelpers.setDevToolElementCallDevUrl(page);
+    // The other site's Element Web opens the embedded package it serves itself
+    // (see docker-compose-playwright.yml), not the dev server's app.
+    if (host !== HOST2) {
+      await TestHelpers.setDevToolElementCallDevUrl(page);
+    }

    const clientHandle = await page.evaluateHandle(() =>
      window.mxMatrixClientPeg.get(),
@@ -153,9 +157,10 @@ export class TestHelpers {
    return { page, clientHandle, mxId };
  }

-  // Dismisses any toasts that appear on startup, such as "Failed to load service worker" or "Back up your chats".
-  // Toast can be stacked, and only the top one can be dismiss, so just look at what is on top and
-  // dismiss (if part of expected toats)
+  // Dismisses the toasts a fresh device gets after sign-in.
+  //
+  // Toasts stack and only the top one can be dismissed, so look
+  // at what is on top each time.
  public static async dismissStartupToasts(page: Page): Promise<void> {
    const expectedToasts: {
      title: string;
@@ -163,6 +168,7 @@ export class TestHelpers {
      confirm?: string;
    }[] = [
      { title: "Failed to load service worker", button: "OK" },
+      { title: "Verify this device", button: "Later" },
      { title: "Back up your chats", button: "Not now" },
      {
        title: "Turn on key storage",
@@ -213,30 +219,6 @@ export class TestHelpers {
    }
  }

-  /**
-   * Clears what Element Web stacks over the room list after sign-in. The
-   * release announcement is held back while any toast is up, so it is closed
-   * after the toasts, not before.
-   */
-  public static async clearRoomListOverlays(page: Page): Promise<void> {
-    await TestHelpers.dismissStartupToasts(page);
-    await TestHelpers.closeReleaseAnnouncement(page, "Introducing Sections");
-  }
-
-  public static async closeReleaseAnnouncement(
-    page: Page,
-    name: string,
-  ): Promise<void> {
-    try {
-      await page
-        .getByRole("dialog", { name })
-        .getByRole("button", { name: "OK" })
-        .click({ timeout: 2000 });
-    } catch {
-      // Announcement not shown; nothing to do
-    }
-  }
-
  public static async expandAllSections(page: Page): Promise<void> {
    try {
      await page
@@ -282,15 +264,14 @@ export class TestHelpers {
  /**
   * Opens the room list's "New" menu and picks the given item.
   *
-   * While the sections announcement is anchored to the "New" button, the
-   * button has no menu at all, so the overlays go first. Driven by keyboard,
-   * as a toast arriving afterwards sits over the button and swallows clicks.
+   * Driven by keyboard, as a toast arriving late sits over the button and
+   * swallows clicks.
   */
  public static async chooseFromRoomListNewMenu(
    page: Page,
    itemName: string,
  ): Promise<void> {
-    await TestHelpers.clearRoomListOverlays(page);
+    await TestHelpers.dismissStartupToasts(page);
    await page
      .getByRole("navigation", { name: "Room list" })
      .getByRole("button", { name: "New", exact: true })
@@ -427,9 +408,8 @@ export class TestHelpers {
      name: "Toggle Invites section",
    });
    await expect(invitesSection).toBeVisible({ timeout: 10000 });
-    // Toasts and the announcement sit over the top of the room list, where
-    // this header is
-    await TestHelpers.clearRoomListOverlays(page);
+    // Toasts sit over the top of the room list, where this header is.
+    await TestHelpers.dismissStartupToasts(page);
    if ((await invitesSection.getAttribute("aria-expanded")) !== "true") {
      await invitesSection.click();
    }
@@ -8,7 +8,7 @@ overrides:
  '@livekit/components-core>rxjs': ^7.8.1
  '@livekit/track-processors>@mediapipe/tasks-vision': ^0.10.18
  tar: ^7.5.11
-  minimatch: ^10.2.3
+  minimatch: ^11.0.0
  glob: ^10.5.0
  qs: ^6.15.2
  js-yaml: ^4.2.0
@@ -1145,6 +1145,9 @@ packages:
    resolution: {integrity: sha512-O8jcjabXaleOG9DQ0+ARXWZBTfnP4WNAqzuiJK7ll44AmxGKv/J2M4TPjxjY3znBCfvBXFzucm1twdyFybFqEA==}
    engines: {node: '>=12'}

+  '@isaacs/nonrecursive-glob-match@1.0.3':
+    resolution: {integrity: sha512-SI7kyKHq4XXDRozw+r9GlRZkia1ZkFrpLW98r2IC2Bpr8YkFGE3fvjzmxCG0x78A++69HmDNrDvSMYBZmHGpKA==}
+
  '@joshwooding/vite-plugin-react-docgen-typescript@0.7.0':
    resolution: {integrity: sha512-qvsTEwEFefhdirGOPnu9Wp6ChfIwy2dBCRuETU3uE+4cC+PFoxMSiiEhxk4lOluA34eARHA0OxqsEUYDqRMgeQ==}
    peerDependencies:
@@ -3530,8 +3533,8 @@ packages:
  boolbase@1.0.0:
    resolution: {integrity: sha512-JZOSA7Mo9sNGB8+UjSgzdLtokWAky1zbztM3WRLCbZ70/3cTANmQmOdR7y2g+J0e2WXywy1yS468tY+IruqEww==}

-  brace-expansion@5.0.9:
-    resolution: {integrity: sha512-ScQ4IuvIEF1TMlP7Zt+vjJ//9zlPb2SDcxWxM3bk8s6t6GGdJ7KO1dCcTidOPJKePW30LE/2cT7wCyPho9/Wxg==}
+  brace-expansion@5.0.12:
+    resolution: {integrity: sha512-YovQ3rzhaLMIrDjNDMkNS01tea93qhEhG5xy8f6+R0l+dw3Ki+5sCoIoI942iuLZTHWogWktgwVDhU09iNEimQ==}
    engines: {node: 20 || >=22}

  braces@3.0.3:
@@ -4768,9 +4771,9 @@ packages:
  minimalistic-crypto-utils@1.0.1:
    resolution: {integrity: sha512-JIYlbt6g8i5jKfJ3xz7rF0LXmv2TkDxBLUkiBeZ7bAx4GnnNMr8xFpGnOxn6GhTEHx3SjRrZEoU+j04prX1ktg==}

-  minimatch@10.2.6:
-    resolution: {integrity: sha512-vpLQEs+VLCr1nU0BXS07maYoFwlDAH0gngQuuttxIwutDFEMHq2blX+8vpgxDdK3J1PwjCJiep77OitTZ4Ll1A==}
-    engines: {node: 18 || 20 || >=22}
+  minimatch@11.0.4:
+    resolution: {integrity: sha512-/UJ1dEoVQTjPNDh6SGlUo+ueLMmoHTas4q+8yaAR/5t6nLAIiBDJtoYUh20IYleix7ehiQkgczb84PE3uzHkxg==}
+    engines: {node: 20 || >=22}

  minimist@1.2.8:
    resolution: {integrity: sha512-2yyAR8qBkN3YuheJanUpWC5U3bb5osDywNB8RzDVlDwDHbocAJveqqj1u8+SVD7jkWT4yvsHCpWqqWqAxb0zCA==}
@@ -6902,7 +6905,7 @@ snapshots:
      ignore: 5.3.2
      import-fresh: 3.3.1
      js-yaml: 4.3.1
-      minimatch: 10.2.6
+      minimatch: 11.0.4
      strip-json-comments: 3.1.1
    transitivePeerDependencies:
      - supports-color
@@ -6981,7 +6984,7 @@ snapshots:
    dependencies:
      '@humanwhocodes/object-schema': 2.0.3
      debug: 4.4.3(supports-color@7.2.0)
-      minimatch: 10.2.6
+      minimatch: 11.0.4
    transitivePeerDependencies:
      - supports-color

@@ -7117,6 +7120,8 @@ snapshots:
      wrap-ansi: 8.1.0
      wrap-ansi-cjs: wrap-ansi@7.0.0

+  '@isaacs/nonrecursive-glob-match@1.0.3': {}
+
  '@joshwooding/vite-plugin-react-docgen-typescript@0.7.0(typescript@5.9.3)(vite@8.2.4(@types/node@24.13.3)(esbuild@0.28.2)(jiti@2.7.0)(sass@1.103.1)(terser@5.46.1)(yaml@2.9.0))':
    dependencies:
      glob: 10.5.0
@@ -8776,7 +8781,7 @@ snapshots:
      '@typescript-eslint/types': 8.69.0
      '@typescript-eslint/visitor-keys': 8.69.0
      debug: 4.4.3(supports-color@7.2.0)
-      minimatch: 10.2.6
+      minimatch: 11.0.4
      semver: 7.8.5
      tinyglobby: 0.2.17
      ts-api-utils: 2.5.0(typescript@5.9.3)
@@ -8791,7 +8796,7 @@ snapshots:
      '@typescript-eslint/types': 8.72.0
      '@typescript-eslint/visitor-keys': 8.72.0
      debug: 4.4.3(supports-color@7.2.0)
-      minimatch: 10.2.6
+      minimatch: 11.0.4
      semver: 7.8.5
      tinyglobby: 0.2.17
      ts-api-utils: 2.5.0(typescript@5.9.3)
@@ -9116,7 +9121,7 @@ snapshots:

  boolbase@1.0.0: {}

-  brace-expansion@5.0.9:
+  brace-expansion@5.0.12:
    dependencies:
      balanced-match: 4.0.4

@@ -9661,7 +9666,7 @@ snapshots:
      json-stable-stringify-without-jsonify: 1.0.1
      levn: 0.4.1
      lodash.merge: 4.6.2
-      minimatch: 10.2.6
+      minimatch: 11.0.4
      natural-compare: 1.4.0
      optionator: 0.9.4
      strip-ansi: 6.0.1
@@ -9781,7 +9786,7 @@ snapshots:

  filelist@1.0.6:
    dependencies:
-      minimatch: 10.2.6
+      minimatch: 11.0.4

  fill-range@7.1.1:
    dependencies:
@@ -9881,7 +9886,7 @@ snapshots:
    dependencies:
      foreground-child: 3.3.1
      jackspeak: 3.4.3
-      minimatch: 10.2.6
+      minimatch: 11.0.4
      minipass: 7.1.3
      package-json-from-dist: 1.0.1
      path-scurry: 1.11.1
@@ -10006,7 +10011,7 @@ snapshots:
      jiti: 2.7.0
      jsonc-parser: 3.3.1
      magic-string: 1.2.3
-      minimatch: 10.2.6
+      minimatch: 11.0.4
      ora: 9.4.1
      react: 19.2.8
      react-i18next: 17.0.13(i18next@26.4.1(typescript@5.9.3))(react-dom@19.2.8(react@19.2.8))(react@19.2.8)(typescript@5.9.3)
@@ -10449,9 +10454,10 @@ snapshots:

  minimalistic-crypto-utils@1.0.1: {}

-  minimatch@10.2.6:
+  minimatch@11.0.4:
    dependencies:
-      brace-expansion: 5.0.9
+      '@isaacs/nonrecursive-glob-match': 1.0.3
+      brace-expansion: 5.0.12

  minimist@1.2.8: {}

@@ -27,7 +27,7 @@ overrides:
  # Security fixes:
  # - https://github.com/advisories/GHSA-7r86-cg39-jmmj
  # - https://github.com/advisories/GHSA-23c5-xmqv-rm74
-  "minimatch": "^10.2.3"
+  "minimatch": "^11.0.0"
  # Security fix: https://github.com/element-hq/element-call/security/dependabot/109
  "glob": "^10.5.0"
  # Security fixes:
@@ -225,6 +225,51 @@ describe("createLivekitMediaTrack", () => {
    );
  });

+  // LiveKit's media stream track is the processed one while a processor runs,
+  // and that one names no device; the camera behind it does
+  it.each([
+    { deviceId: "cam-1", switches: false },
+    { deviceId: "cam-2", switches: true },
+  ])(
+    "leaves a processed camera alone unless its device changes: $deviceId",
+    async ({ deviceId, switches }) => {
+      vi.stubGlobal("MediaStreamTrack", class {});
+      onTestFinished(() => {
+        vi.unstubAllGlobals();
+      });
+      const camera = Object.assign(Object.create(LocalVideoTrack.prototype), {
+        source: Track.Source.Camera,
+        _mediaStreamTrack: {
+          readyState: "live",
+          label: "",
+          getSettings: () => ({ deviceId: "cam-1" }),
+        },
+        // What LiveKit shows as the track, with no device of its own
+        processor: {
+          processedTrack: { label: "", getSettings: () => ({}) },
+        },
+      }) as LocalVideoTrack;
+      const switchActiveDevice = vi.fn().mockResolvedValue(true);
+      const { participant, publication, room } = fakes({
+        track: camera as unknown as FakeTrack,
+      });
+      const track = createLocalLivekitMediaTrack(
+        scope,
+        Object.assign(participant, {
+          isLocal: true,
+        }) as unknown as LocalParticipant,
+        publication as unknown as TrackPublication,
+        Object.assign(room, { switchActiveDevice }),
+        vi.fn(),
+      );
+
+      await track.setDevice(deviceId);
+      expect(switchActiveDevice.mock.calls).toEqual(
+        switches ? [["videoinput", deviceId, false]] : [],
+      );
+    },
+  );
+
  // LiveKit stops a camera that is turned off; the processor is built when it
  // is turned on again
  it.each([
@@ -199,11 +199,13 @@ export function createLocalLivekitMediaTrack(
        throw new MatrixRTCError(`A ${base.source} track has no device`);
      // LiveKit decides whether to restart by the constraints it keeps, which
      // a browser that does not echo the device id back (Firefox) leaves
-      // without one; a track already on the device is left alone here
+      // without one; a track already on the device is left alone here. Asked
+      // of the camera track itself: with a processor on, LiveKit's media
+      // stream track is the processed one, which names no device
      const current = publication.track;
      if (
        current instanceof LocalTrack &&
-        current.mediaStreamTrack.getSettings().deviceId === deviceId
+        (await current.getDeviceId(false)) === deviceId
      )
        return;
      await room.switchActiveDevice(
@@ -141,6 +141,7 @@ export const Default: Story = {
    backgroundImageRefusal: undefined,
    addBackgroundImage: fn(),
    removeBackgroundEffect: fn(),
+    cameraTrack: null,
    hangup: fn(),
    buttonSize: "lg",
    showFooter: true,
@@ -33,7 +33,9 @@ import { useBehavior } from "../useBehavior";
 import { type LayoutSwitchViewModel } from "../state/LayoutSwitchViewModel";
 import { LayoutSwitch } from "../room/LayoutSwitch";
 import { type BackgroundEffectGridItem } from "./BackgroundEffectGrid";
+import { SelfPreview } from "./SelfPreview";
 import { type UnusableReason } from "../tracks/backgroundImages";
+import { type LocalVideoMediaTrack } from "@element-hq/matrixrtc-sdk";

 export interface BackgroundImageRefusal {
  reason: UnusableReason | "not-kept";
@@ -92,6 +94,8 @@ export interface FooterState {
  backgroundEffectSettling: boolean;
  /** Why the last file offered couldn't be kept, if it couldn't. */
  backgroundImageRefusal: BackgroundImageRefusal | undefined;
+  /** The camera to preview, with the effect in force; null while it is off. */
+  cameraTrack: LocalVideoMediaTrack | null;
  showFooter: boolean;

  /* This is needed for WindowMode = "flat" */
@@ -181,6 +185,7 @@ export const CallFooter: FC<FooterProps> = ({
  const backgroundEffectSettling = useBehavior(vm.backgroundEffectSettling$);
  const addBackgroundImage = useBehavior(vm.addBackgroundImage$);
  const removeBackgroundEffect = useBehavior(vm.removeBackgroundEffect$);
+  const cameraTrack = useBehavior(vm.cameraTrack$);
  const refusal = useBehavior(vm.backgroundImageRefusal$);
  const refusalMessage = useMemo(() => {
    switch (refusal?.reason) {
@@ -264,6 +269,7 @@ export const CallFooter: FC<FooterProps> = ({
        backgroundEffectSettling={backgroundEffectSettling}
        onAddBackgroundImage={addBackgroundImage}
        onRemoveBackgroundEffect={removeBackgroundEffect}
+        selfPreview={<SelfPreview track={cameraTrack} />}
        backgroundImageRefusal={refusalMessage}
        backgroundEffectNotice={
          backgroundEffectNotice === "unavailable"
@@ -7,14 +7,17 @@ Please see LICENSE in the repository root for full details.

 import { afterEach, describe, expect, it, vi } from "vitest";
 import { BehaviorSubject } from "rxjs";
+import { type LocalVideoMediaTrack } from "@element-hq/matrixrtc-sdk";

 import {
  flushPromises,
  testScope,
  mockMuteStates,
  mockMediaDevices,
+  mockVideoTrack,
 } from "../utils/test";
-import { constant } from "../state/Behavior";
+import { type Behavior, constant } from "../state/Behavior";
+import { type BackgroundEffectsState } from "../state/BackgroundEffects";
 import type { CallViewModel } from "../state/CallViewModel/CallViewModel";
 import type { Layout } from "../state/layout-types";
 import type { SpotlightTileViewModel } from "../state/TileViewModel";
@@ -78,6 +81,7 @@ vi.mock("livekit-client", () => ({
 function buildMinimalCallViewModel(layout: Layout): CallViewModel {
  return {
    layout$: constant(layout),
+    camera$: constant(undefined),
    edgeToEdge$: constant(false),
    showHeader$: constant(false),
    hangup: (): void => {},
@@ -109,6 +113,16 @@ const pipLayout: Layout = {
  spotlight: {} as SpotlightTileViewModel,
 };

+const noPipeline = constant<BackgroundEffectsState>({
+  supported: false,
+  processor: undefined,
+});
+
+/** The provider's verdict, which the SDK's check stands for here. */
+function pipelineAsTheSdkSays(): Behavior<BackgroundEffectsState> {
+  return constant({ supported: sdkSupportMock(), processor: undefined });
+}
+
 const twoMicsAndOneCamMediaDevices = mockMediaDevices({
  audioInput: {
    available$: constant(
@@ -140,7 +154,7 @@ describe("createCallFooterViewModel", () => {
        buildMinimalCallViewModel(gridLayout),
        mockMuteStates(),
        twoMicsAndOneCamMediaDevices,
-        constant(false),
+        noPipeline,
        /* reactionIdentifier */ undefined,
        { showControls: true, header: HeaderStyle.Standard },
      );
@@ -183,7 +197,7 @@ describe("createCallFooterViewModel", () => {
            select: vi.fn(),
          },
        }),
-        constant(false),
+        noPipeline,
        /* reactionIdentifier */ undefined,
        { showControls: true, header: HeaderStyle.Standard },
      );
@@ -202,7 +216,7 @@ describe("createCallFooterViewModel", () => {
        buildMinimalCallViewModel(layout),
        mockMuteStates(),
        twoMicsAndOneCamMediaDevices,
-        constant(false),
+        noPipeline,
        /* reactionIdentifier */ undefined,
        { showControls: true, header: HeaderStyle.Standard },
      );
@@ -225,7 +239,7 @@ describe("createCallFooterViewModel", () => {
        buildMinimalCallViewModel(gridLayout),
        mockMuteStates(),
        twoMicsAndOneCamMediaDevices,
-        constant(false),
+        noPipeline,
        /* reactionIdentifier */ undefined,
        { showControls: true, header: HeaderStyle.Standard },
      );
@@ -271,7 +285,8 @@ describe("createCallFooterViewModel", () => {
        testScope(),
        mockMuteStates(),
        twoMicsAndOneCamMediaDevices,
-        constant(false),
+        pipelineAsTheSdkSays(),
+        /* camera$ */ constant(null),
        /* openSettings */ undefined,
        /* hangup */ undefined,
        /* showLogo */ false,
@@ -356,21 +371,52 @@ describe("createCallFooterViewModel", () => {

    it("passes on the wait for the first effect", () => {
      platformMock.mockReturnValue("desktop");
-      const settling$ = new BehaviorSubject(true);
+      const pipeline$ = new BehaviorSubject<BackgroundEffectsState>({
+        supported: true,
+        processor: undefined,
+        wait: "settling",
+      });
      const vm = createLobbyFooterViewModel(
        testScope(),
        mockMuteStates(),
        twoMicsAndOneCamMediaDevices,
-        settling$,
+        pipeline$,
+        /* camera$ */ constant(null),
        /* openSettings */ undefined,
        /* hangup */ undefined,
        /* showLogo */ false,
      );
      expect(vm.backgroundEffectSettling$.value).toBe(true);
-      settling$.next(false);
+      pipeline$.next({ ...pipeline$.value, wait: undefined });
      expect(vm.backgroundEffectSettling$.value).toBe(false);
    });

+    it("offers nothing once the pipeline fails to build, keeping the choice", () => {
+      platformMock.mockReturnValue("desktop");
+      backgroundEffectSetting.setValue("blur");
+      const pipeline$ = new BehaviorSubject<BackgroundEffectsState>({
+        supported: true,
+        processor: undefined,
+      });
+      const vm = createLobbyFooterViewModel(
+        testScope(),
+        mockMuteStates(),
+        twoMicsAndOneCamMediaDevices,
+        pipeline$,
+        /* camera$ */ constant(null),
+        /* openSettings */ undefined,
+        /* hangup */ undefined,
+        /* showLogo */ false,
+      );
+      expect(vm.backgroundEffect$.value).toBe("blur");
+
+      pipeline$.next({ supported: false, processor: undefined });
+      expect(vm.selectBackgroundEffect$.value).toBeUndefined();
+      expect(vm.addBackgroundImage$.value).toBeUndefined();
+      expect(vm.backgroundEffect$.value).toBe("none");
+      expect(backgroundEffectSetting.getValue()).toBe("blur");
+    });
+
    it("adding is unavailable at the limit", () => {
      sdkSupportMock.mockReturnValue(true);
      const added$ = addedBackgrounds.added$ as BehaviorSubject<
@@ -427,6 +473,44 @@ describe("createCallFooterViewModel", () => {
      expect(store.remove).toHaveBeenCalledWith("spare");
    });

+    it("previews the lobby's camera while there is one", () => {
+      const track = mockVideoTrack({});
+      const camera$ = new BehaviorSubject<LocalVideoMediaTrack | null>(track);
+      const vm = createLobbyFooterViewModel(
+        testScope(),
+        mockMuteStates(),
+        twoMicsAndOneCamMediaDevices,
+        noPipeline,
+        camera$,
+        /* openSettings */ undefined,
+        /* hangup */ undefined,
+        /* showLogo */ false,
+      );
+      expect(vm.cameraTrack$.value).toBe(track);
+      // The lobby releases the camera when it is turned off
+      camera$.next(null);
+      expect(vm.cameraTrack$.value).toBeNull();
+    });
+
+    it("previews the call's camera only while it is not muted", () => {
+      const muted$ = new BehaviorSubject(false);
+      const track = mockVideoTrack({ muted$ });
+      const vm = createCallFooterViewModel(
+        testScope(),
+        { ...buildMinimalCallViewModel(gridLayout), camera$: constant(track) },
+        mockMuteStates(),
+        twoMicsAndOneCamMediaDevices,
+        noPipeline,
+        /* reactionIdentifier */ undefined,
+        { showControls: true, header: HeaderStyle.Standard },
+      );
+      expect(vm.cameraTrack$.value).toBe(track);
+      muted$.next(true);
+      expect(vm.cameraTrack$.value).toBeNull();
+      muted$.next(false);
+      expect(vm.cameraTrack$.value).toBe(track);
+    });
+
    it("availability is the same before and during a call", () => {
      for (const supported of [true, false]) {
        sdkSupportMock.mockReturnValue(supported);
@@ -436,7 +520,7 @@ describe("createCallFooterViewModel", () => {
          buildMinimalCallViewModel(gridLayout),
          mockMuteStates(),
          twoMicsAndOneCamMediaDevices,
-          constant(false),
+          pipelineAsTheSdkSays(),
          /* reactionIdentifier */ undefined,
          { showControls: true, header: HeaderStyle.Standard },
        );
@@ -11,13 +11,12 @@ import {
  distinctUntilChanged,
  map,
  type Observable,
+  of,
  switchMap,
 } from "rxjs";
 import { supportsAudioOutputSelection } from "livekit-client";
-import {
-  supportsBackgroundProcessors,
-  supportsModernBackgroundProcessors,
-} from "@livekit/track-processors";
+import { type LocalVideoMediaTrack } from "@element-hq/matrixrtc-sdk";
+import { supportsModernBackgroundProcessors } from "@livekit/track-processors";
 import { logger } from "matrix-js-sdk/lib/logger";

 import {
@@ -33,6 +32,7 @@ import {
 } from "../tracks/backgroundImages";

 import { type CallViewModel } from "../state/CallViewModel/CallViewModel";
+import { type BackgroundEffectsState } from "../state/BackgroundEffects";
 import { type MenuOptions } from "./MediaMuteAndSwitchButton";
 import { type MediaDevices } from "../state/MediaDevices";
 import {
@@ -87,7 +87,9 @@ function buildDeviceBehaviors(
  mediaDevices: MediaDevices,
  /** return empty arrays for  audioOptions and videoOptions*/
  disableSwitcher$: Behavior<boolean>,
-  backgroundEffectSettling$: Observable<boolean>,
+  backgroundEffectsState$: Observable<BackgroundEffectsState>,
+  /** The camera as the pipeline has it, for the menu's preview. */
+  camera$: Observable<LocalVideoMediaTrack | null | undefined>,
 ): Pick<
  ViewModel<FooterSnapshot>,
  | "audioOptions$"
@@ -107,6 +109,7 @@ function buildDeviceBehaviors(
  | "addBackgroundImage$"
  | "backgroundImageRefusal$"
  | "removeBackgroundEffect$"
+  | "cameraTrack$"
 > {
  const options$ = (
    available$: Behavior<Map<string, MenuOptions["label"]>>,
@@ -123,10 +126,14 @@ function buildDeviceBehaviors(
      ),
    );

-  const supported = supportsBackgroundProcessors();
+  // Also false once the pipeline has failed to build this session.
+  const supported$ = backgroundEffectsState$.pipe(
+    map(({ supported }) => supported === true),
+    distinctUntilChanged(),
+  );
  const slow = !supportsModernBackgroundProcessors();
-  const offered$ = disableSwitcher$.pipe(
-    map((switcherDisabled) => !switcherDisabled && supported),
+  const offered$ = combineLatest([disableSwitcher$, supported$]).pipe(
+    map(([switcherDisabled, supported]) => !switcherDisabled && supported),
  );
  // A new object for each refusal, so the same one twice is shown twice.
  const refusal$ = new BehaviorSubject<BackgroundImageRefusal | undefined>(
@@ -157,8 +164,10 @@ function buildDeviceBehaviors(
    ),
    selectVideoButtonOption$: constant(mediaDevices.videoInput.select),
    backgroundEffect$: scope.behavior(
-      backgroundEffectSetting.value$.pipe(
-        map((raw) => (supported ? serializeEffect(parseEffect(raw)) : "none")),
+      combineLatest([backgroundEffectSetting.value$, supported$]).pipe(
+        map(([raw, supported]) =>
+          supported ? serializeEffect(parseEffect(raw)) : "none",
+        ),
      ),
    ),
    selectBackgroundEffect$: scope.behavior(
@@ -186,7 +195,10 @@ function buildDeviceBehaviors(
      ),
    ),
    backgroundEffectSettling$: scope.behavior(
-      backgroundEffectSettling$.pipe(distinctUntilChanged()),
+      backgroundEffectsState$.pipe(
+        map(({ wait }) => wait !== undefined),
+        distinctUntilChanged(),
+      ),
      false,
    ),
    // Kept and offered, not put on: that waits for the user to choose it.
@@ -208,6 +220,17 @@ function buildDeviceBehaviors(
      ),
    ),
    backgroundImageRefusal$: refusal$,
+    // None while muted: the call mutes its track rather than releasing it.
+    cameraTrack$: scope.behavior(
+      camera$.pipe(
+        switchMap((camera) =>
+          camera
+            ? camera.muted$.pipe(map((muted) => (muted ? null : camera)))
+            : of(null),
+        ),
+      ),
+      null,
+    ),
    // Never the one in force: taking it away would leave nothing chosen.
    removeBackgroundEffect$: scope.behavior(
      offered$.pipe(
@@ -237,8 +260,8 @@ function buildDeviceBehaviors(
 * @param callModel - The root CallViewModel; provides layout, grid mode, reactions, etc.
 * @param muteStates - Audio and video mute state + toggles.
 * @param mediaDevices - Available and selected input devices.
- * @param backgroundEffectSettling$ - Whether the first effect chosen is still
- *   being prepared.
+ * @param backgroundEffectsState$ - The background pipeline's state: whether
+ *   effects run here, and the wait for the first.
 * @param reactionIdentifier - The local user's reaction identifier string, or
 *   undefined when reactions are not supported (hides the reaction button).
 * @param options - `showControls`: whether the call controls should be shown.
@@ -249,7 +272,7 @@ export function createCallFooterViewModel(
  callModel: CallViewModel,
  muteStates: MuteStates,
  mediaDevices: MediaDevices,
-  backgroundEffectSettling$: Observable<boolean>,
+  backgroundEffectsState$: Observable<BackgroundEffectsState>,
  reactionIdentifier: string | undefined,
  options: { showControls: boolean; header: HeaderStyle },
 ): ViewModel<FooterSnapshot> {
@@ -268,7 +291,8 @@ export function createCallFooterViewModel(
      scope,
      mediaDevices,
      disableDeviceSwitcher$,
-      backgroundEffectSettling$,
+      backgroundEffectsState$,
+      callModel.camera$,
    ),
    // candidat to move into the FooterViewModel
    showFooter$: callModel.showFooter$,
@@ -331,8 +355,9 @@ export function createCallFooterViewModel(
 * @param scope - ObservableScope that bounds the lifetime of derived behaviors.
 * @param muteStates - Audio and video mute state + toggles.
 * @param mediaDevices - Available and selected input devices.
- * @param backgroundEffectSettling$ - Whether the first effect chosen is still
- *   being prepared.
+ * @param backgroundEffectsState$ - The background pipeline's state: whether
+ *   effects run here, and the wait for the first.
+ * @param camera$ - The lobby's camera, for the camera menu's preview.
 * @param openSettings - Callback to open the settings modal, or undefined.
 * @param hangup - Callback to leave/cancel, or undefined (hides the button).
 * @param showLogo - Whether to show the Element Call logo.
@@ -341,7 +366,8 @@ export function createLobbyFooterViewModel(
  scope: ObservableScope,
  muteStates: MuteStates,
  mediaDevices: MediaDevices,
-  backgroundEffectSettling$: Observable<boolean>,
+  backgroundEffectsState$: Observable<BackgroundEffectsState>,
+  camera$: Behavior<LocalVideoMediaTrack | null>,
  openSettings: (() => void) | undefined,
  hangup: (() => void) | undefined,
  showLogo: boolean,
@@ -389,7 +415,8 @@ export function createLobbyFooterViewModel(
      scope,
      mediaDevices,
      constant(false),
-      backgroundEffectSettling$,
+      backgroundEffectsState$,
+      camera$,
    ),
  };
 }
@@ -97,3 +97,35 @@ Please see LICENSE in the repository root for full details.
 .refusal {
  margin: var(--cpd-space-2x) var(--cpd-space-4x) 0;
 }
+
+.selfPreview {
+  position: relative;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  flex-shrink: 0;
+  aspect-ratio: 16 / 9;
+  margin-block-end: var(--cpd-space-2x);
+  overflow: clip;
+  border-start-start-radius: var(--cpd-space-3x);
+  border-start-end-radius: var(--cpd-space-3x);
+  background: var(--cpd-color-bg-subtle-secondary);
+  color: var(--cpd-color-icon-secondary);
+}
+
+.selfPreview > video {
+  position: absolute;
+  inset: 0;
+  inline-size: 100%;
+  block-size: 100%;
+  object-fit: cover;
+  /* Rounded itself too: a video can be drawn on a layer of its own, which the
+     frame's rounded clip doesn't always reach. */
+  border-radius: inherit;
+}
+
+/* Up to the menu's frame, over the padding at its top. */
+.selfPreviewPinned,
+.deviceListLeadsWithPreview {
+  margin-block-start: calc(-1 * var(--cpd-space-5x));
+}
@@ -12,6 +12,7 @@ import type { Meta, StoryObj } from "@storybook/react-vite";
 import { MediaMuteAndSwitchButton } from "./MediaMuteAndSwitchButton";
 import styles from "./MediaMuteAndSwitchButton.module.css";
 import { type BackgroundEffectGridItem } from "./BackgroundEffectGrid";
+import { SelfPreview } from "./SelfPreview";
 import { shippedBackgrounds } from "../tracks/backgroundEffectCatalog";
 import meterStyles from "./MicrophoneLevelMeter.module.css";
 import { MediaDevicesContext } from "../MediaDevicesContext";
@@ -60,6 +61,50 @@ const WithAMicrophone: FC<{ children: ReactNode }> = ({ children }) => {
  return <>{children}</>;
 };

+/** A camera for the preview: a figure against a room, drawn to a stream. */
+const FakeCamera: FC = () => {
+  const [stream, setStream] = useState<MediaStream>();
+  useEffect(() => {
+    const canvas = document.createElement("canvas");
+    canvas.width = 640;
+    canvas.height = 360;
+    const context = canvas.getContext("2d")!;
+    const draw = (): void => {
+      context.fillStyle = "#6b7f8e";
+      context.fillRect(0, 0, 640, 360);
+      context.fillStyle = "#c9a27e";
+      context.beginPath();
+      context.arc(320, 150, 62, 0, 2 * Math.PI);
+      context.fill();
+      context.fillStyle = "#34495e";
+      context.beginPath();
+      context.ellipse(320, 380, 150, 140, 0, Math.PI, 2 * Math.PI);
+      context.fill();
+    };
+    draw();
+    // Redrawn, so the stream keeps delivering frames.
+    const timer = setInterval(draw, 200);
+    const captured = canvas.captureStream(10);
+    setStream(captured);
+    return (): void => {
+      clearInterval(timer);
+      captured.getTracks().forEach((track) => track.stop());
+    };
+  }, []);
+  return (
+    <video
+      data-testid="preview-camera"
+      autoPlay
+      muted
+      playsInline
+      ref={(video): void => {
+        if (video && stream && video.srcObject !== stream)
+          video.srcObject = stream;
+      }}
+    />
+  );
+};
+
 /** Supplies a call-sized root. Without one the list is bounded by the whole Storybook frame. */
 const WithACallArea: FC<{ children: ReactNode; blockSize?: number }> = ({
  children,
@@ -171,8 +216,8 @@ export const VideoMute: Story = {
      { label: { type: "name", name: "Camera 1" }, id: "1" },
      { label: { type: "name", name: "Camera 2" }, id: "2" },
    ],
-
    selectedOption: "1",
+    selfPreview: <SelfPreview track={null} />,
  },
 };

@@ -199,6 +244,8 @@ export const VideoUnmute: Story = {
    selectedBackgroundEffect: "none",
    onSelectBackgroundEffect: fn(),
    onAddBackgroundImage: fn(),
+    // As the footer always gives the camera menu one.
+    selfPreview: <FakeCamera />,
  },
 };

@@ -344,6 +391,9 @@ export const BackgroundEffectsSettling: Story = {
      checked: false,
    }))
      await expect(other).toHaveAttribute("aria-busy", "false");
+    await expect(menu.queryByTestId("preview-camera")).toBeNull();
+    const frames = document.body.querySelectorAll(`.${styles.selfPreview} svg`);
+    await expect(frames).toHaveLength(1);
  },
 };

@@ -501,6 +551,113 @@ export const RemovingWithLiveSelection: Story = {
  },
 };

+/** Where the call leaves room, the camera holds the top while the options scroll. */
+export const BackgroundEffectsWithPreview: Story = {
+  args: {
+    ...VideoUnmute.args,
+    // Enough that the list fills its share, so the preview's cost counts.
+    options: Array.from({ length: 8 }, (_, i) => ({
+      label: { type: "name" as const, name: `Camera ${i + 1}` },
+      id: `${i + 1}`,
+    })),
+  },
+  decorators: [
+    (Story): JSX.Element => (
+      <WithACallArea blockSize={640}>
+        <Story />
+      </WithACallArea>
+    ),
+  ],
+  play: async ({ canvasElement }) => {
+    const canvas = within(canvasElement);
+    await userEvent.click(canvas.getByRole("button", { name: "Camera" }));
+    const picture = await within(document.body).findByTestId("preview-camera");
+    const preview = picture.parentElement!;
+    const list = document.body.querySelector<HTMLElement>(
+      `.${styles.deviceList}`,
+    )!;
+    const frame = document.body
+      .querySelector("[role='menu']")!
+      .getBoundingClientRect();
+
+    const box = preview.getBoundingClientRect();
+    await expect(list.contains(preview)).toBe(false);
+    await expect(box.top - frame.top).toBeLessThanOrEqual(2);
+    await expect(frame.width - box.width).toBeLessThanOrEqual(2);
+    await expect(box.width / box.height).toBeCloseTo(16 / 9, 1);
+    await expect(box.bottom).toBeLessThanOrEqual(
+      list.getBoundingClientRect().top,
+    );
+
+    const listShare = Math.round(640 * 0.6);
+    // The list runs on through the menu's padding at its foot, which isn't
+    // part of its share.
+    const listEnd =
+      list.getBoundingClientRect().bottom -
+      Number.parseFloat(getComputedStyle(list).paddingBlockEnd);
+    await expect(listEnd - box.top).toBeLessThanOrEqual(listShare + 1);
+
+    await expect(list.scrollHeight).toBeGreaterThan(list.clientHeight);
+    list.scrollTop = list.scrollHeight;
+    await expect(preview.getBoundingClientRect().top).toBe(box.top);
+  },
+};
+
+/** With the camera off the preview says so, and effects can still be chosen. */
+export const BackgroundEffectsPreviewWithTheCameraOff: Story = {
+  args: {
+    ...VideoUnmute.args,
+    enabled: false,
+    selfPreview: <SelfPreview track={null} />,
+  },
+  play: async ({ args, canvasElement }) => {
+    const canvas = within(canvasElement);
+    await userEvent.click(canvas.getByRole("button", { name: "Camera" }));
+    const menu = within(document.body);
+    await menu.findByRole("group", { name: "Background effects" });
+    const preview = document.body.querySelector(`.${styles.selfPreview}`)!;
+    await expect(preview.querySelector("video")).toBeNull();
+    await expect(preview.querySelector("svg")).not.toBeNull();
+
+    await userEvent.click(menu.getByRole("menuitemradio", { name: "Blur" }));
+    await expect(args.onSelectBackgroundEffect).toHaveBeenCalledWith("blur");
+  },
+};
+
+/** In a call too short to hold it, the camera leads the options and scrolls away. */
+export const BackgroundEffectsPreviewScrollsWithTheList: Story = {
+  args: VideoUnmute.args,
+  decorators: [
+    (Story): JSX.Element => (
+      <WithACallArea blockSize={300}>
+        <Story />
+      </WithACallArea>
+    ),
+  ],
+  play: async ({ canvasElement }) => {
+    const canvas = within(canvasElement);
+    await userEvent.click(canvas.getByRole("button", { name: "Camera" }));
+    const picture = await within(document.body).findByTestId("preview-camera");
+    const preview = picture.parentElement!;
+    const list = document.body.querySelector<HTMLElement>(
+      `.${styles.deviceList}`,
+    )!;
+    await expect(list.contains(preview)).toBe(true);
+    const cameras = within(document.body).getByRole("group", {
+      name: "Camera",
+    });
+    await expect(
+      preview.compareDocumentPosition(cameras) &
+        Node.DOCUMENT_POSITION_FOLLOWING,
+    ).toBeTruthy();
+
+    list.scrollTop = list.scrollHeight;
+    await expect(preview.getBoundingClientRect().bottom).toBeLessThan(
+      list.getBoundingClientRect().top,
+    );
+  },
+};
+
 /** In a short call the effects scroll into view with the list. */
 export const BackgroundEffectsScrollWhenTheyDoNotFit: Story = {
  args: {
@@ -14,10 +14,12 @@ import {
  type FC,
  useEffect,
  type ReactElement,
+  type ReactNode,
 } from "react";
 import {
  Alert,
  Button,
+  InlineSpinner,
  Menu,
  MenuItem,
  MenuScrollArea,
@@ -87,6 +89,8 @@ export interface MediaMuteAndSwitchButtonProps {
  onRemoveBackgroundEffect?: (id: string) => void;
  /** Why the last file offered couldn't be used; a new object each time. */
  backgroundImageRefusal?: { text: string };
+  /** The user's own camera, shown at the top of the camera menu. */
+  selfPreview?: ReactNode;
  /**
   * For any toggle and option this method will be called.
   * So toggles need to be implemented by listening here and setting the right toggle item to `enabled`
@@ -103,9 +107,16 @@ const LIST_SHARE_OF_CALL = 0.6;
 /** Smallest device list height in px, so a short call still shows more than one device. */
 const MIN_LIST_HEIGHT = 160;

+/** The menu's width as its stylesheet sets it, to place the preview before
+    the menu is drawn. */
+const MENU_WIDTH = 296;
+
 /** Space kept between the menu and the call area's sides. */
 const MENU_MARGIN = 16;

+/** Below the preview, as its stylesheet draws it. */
+const PREVIEW_GAP = 12;
+
 export const MediaMuteAndSwitchButton: FC<MediaMuteAndSwitchButtonProps> = ({
  enabled,
  busy,
@@ -124,6 +135,7 @@ export const MediaMuteAndSwitchButton: FC<MediaMuteAndSwitchButtonProps> = ({
  onAddBackgroundImage,
  onRemoveBackgroundEffect,
  backgroundImageRefusal,
+  selfPreview,
  onSelect,
 }) => {
  // Requested but not yet selected. Keyed by kind too, since Chrome uses
@@ -202,21 +214,53 @@ export const MediaMuteAndSwitchButton: FC<MediaMuteAndSwitchButtonProps> = ({

  // Measured on the call area: CSS can't size the portalled menu against it.
  const rootElement = useRootElement();
-  const [listMaxHeight, setListMaxHeight] = useState<number>();
+  const [listShare, setListShare] = useState<number>();
+  const [previewBlock, setPreviewBlock] = useState(0);
  useEffect(() => {
    if (!menuOpen) return;
    // Followed, since a host can resize the call while the menu is open.
    const subscription = observeElementSize$(rootElement)
      .pipe(
-        map(({ height }) =>
-          Math.max(MIN_LIST_HEIGHT, Math.round(height * LIST_SHARE_OF_CALL)),
+        map(({ width, height }) => ({
+          height: Math.round(height * LIST_SHARE_OF_CALL),
+          width: Math.min(MENU_WIDTH, Math.round(width - 2 * MENU_MARGIN)),
+        })),
+        distinctUntilChanged(
+          (a, b) => a.height === b.height && a.width === b.width,
        ),
-        distinctUntilChanged(),
      )
-      .subscribe(setListMaxHeight);
+      .subscribe(({ height, width }) => {
+        setListShare(height);
+        setPreviewBlock(Math.round((width * 9) / 16) + PREVIEW_GAP);
+      });
    return (): void => subscription.unsubscribe();
  }, [menuOpen, rootElement]);

+  // Paid for out of the list's share, so a preview makes the menu no taller.
+  const hasPreview =
+    iconsAndLabels === "video" &&
+    selfPreview !== undefined &&
+    listShare !== undefined;
+  const previewPinned =
+    hasPreview && listShare - previewBlock >= MIN_LIST_HEIGHT;
+  const listMaxHeight =
+    listShare === undefined
+      ? undefined
+      : Math.max(
+          MIN_LIST_HEIGHT,
+          listShare - (previewPinned ? previewBlock : 0),
+        );
+  const preview = hasPreview && (
+    <div
+      aria-hidden
+      className={classNames(styles.selfPreview, {
+        [styles.selfPreviewPinned]: previewPinned,
+      })}
+    >
+      {backgroundEffectSettling ? <InlineSpinner size={32} /> : selfPreview}
+    </div>
+  );
+
  useEffect(() => {
    if (menuOpen) devices.requestDeviceNames(); // No-op after the first call
  }, [menuOpen, devices]);
@@ -407,9 +451,12 @@ export const MediaMuteAndSwitchButton: FC<MediaMuteAndSwitchButtonProps> = ({
          />
        }
      >
+        {previewPinned && preview}
        <MenuScrollArea
          ref={trackFocusSource}
-          className={styles.deviceList}
+          className={classNames(styles.deviceList, {
+            [styles.deviceListLeadsWithPreview]: hasPreview && !previewPinned,
+          })}
          data-testid="device-list"
          style={
            {
@@ -418,6 +465,7 @@ export const MediaMuteAndSwitchButton: FC<MediaMuteAndSwitchButtonProps> = ({
            } as CSSProperties
          }
        >
+          {hasPreview && !previewPinned && preview}
          {audioMeter && speakerOptions && (
            <MenuScrollSection title={t("settings.devices.speaker")}>
              {deviceItems(
@@ -0,0 +1,10 @@
+/*
+Copyright 2026 Element Creations Ltd.
+
+SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-Element-Commercial
+Please see LICENSE in the repository root for full details.
+*/
+
+.mirrored {
+  transform: scaleX(-1);
+}
@@ -0,0 +1,47 @@
+/*
+Copyright 2026 Element Creations Ltd.
+
+SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-Element-Commercial
+Please see LICENSE in the repository root for full details.
+*/
+
+import { act, render } from "@testing-library/react";
+import { describe, expect, it } from "vitest";
+import { BehaviorSubject } from "rxjs";
+
+import { SelfPreview } from "./SelfPreview";
+import styles from "./SelfPreview.module.css";
+import { mockVideoTrack } from "../utils/test";
+
+describe("SelfPreview", () => {
+  it("detaches only its own element from a camera shown twice", () => {
+    const track = mockVideoTrack({});
+    const first = render(<SelfPreview track={track} />);
+    const second = render(<SelfPreview track={track} />);
+    const [a, b] = [first, second].map((r) =>
+      r.container.querySelector("video"),
+    );
+    expect(track.attach).toHaveBeenCalledWith(a);
+    expect(track.attach).toHaveBeenCalledWith(b);
+
+    first.unmount();
+    expect(track.detach).toHaveBeenCalledExactlyOnceWith(a);
+  });
+
+  it("mirrors the camera while it faces the user, and only then", () => {
+    const facingMode$ = new BehaviorSubject<"user" | "environment">("user");
+    const { container } = render(
+      <SelfPreview track={mockVideoTrack({ facingMode$ })} />,
+    );
+    const video = (): HTMLVideoElement => container.querySelector("video")!;
+    expect(video().classList.contains(styles.mirrored)).toBe(true);
+    act(() => facingMode$.next("environment"));
+    expect(video().classList.contains(styles.mirrored)).toBe(false);
+  });
+
+  it("says the camera is off while there is none", () => {
+    const { container } = render(<SelfPreview track={null} />);
+    expect(container.querySelector("video")).toBeNull();
+    expect(container.querySelector("svg")).not.toBeNull();
+  });
+});
@@ -0,0 +1,48 @@
+/*
+Copyright 2026 Element Creations Ltd.
+
+SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-Element-Commercial
+Please see LICENSE in the repository root for full details.
+*/
+
+import { type FC, useEffect, useRef } from "react";
+import { constant, type LocalVideoMediaTrack } from "@element-hq/matrixrtc-sdk";
+import { VideoCallOffIcon } from "@vector-im/compound-design-tokens/assets/web/icons";
+
+import styles from "./SelfPreview.module.css";
+import { useBehavior } from "../useBehavior";
+
+const noFacingMode = constant(undefined);
+
+interface Props {
+  /** The camera as the pipeline has it, or none while the camera is off. */
+  track: LocalVideoMediaTrack | null;
+}
+
+/**
+ * The user's own camera with the effect in force: the pipeline's own track,
+ * shown a second time rather than captured again.
+ */
+export const SelfPreview: FC<Props> = ({ track }) => {
+  const videoRef = useRef<HTMLVideoElement | null>(null);
+  useEffect(() => {
+    const element = videoRef.current;
+    if (!track || !element) return;
+    track.attach(element);
+    // This element only: the pre-join preview may be showing the same camera
+    return (): void => {
+      track.detach(element);
+    };
+  }, [track]);
+  const facingMode = useBehavior(track?.facingMode$ ?? noFacingMode);
+
+  if (!track) return <VideoCallOffIcon width={32} height={32} />;
+  return (
+    <video
+      ref={videoRef}
+      muted
+      playsInline
+      className={facingMode === "user" ? styles.mirrored : undefined}
+    />
+  );
+};
@@ -108,6 +108,22 @@ describe("Config.initWith", () => {
    expect(Config.get().media_quality).toEqual(DEFAULT_CONFIG.media_quality);
  });

+  it("falls back to the defaults when config.json is not JSON", async () => {
+    vi.spyOn(globalThis, "fetch").mockResolvedValue(
+      new Response("<!doctype html><title>Element</title>", {
+        status: 200,
+        headers: { "content-type": "text/html" },
+      }),
+    );
+    const warnSpy = vi.spyOn(logger, "warn").mockImplementation(() => {});
+    Reflect.set(Config, "internalInstance", undefined);
+
+    await expect(Config.init()).resolves.toBeUndefined();
+
+    expect(Config.get().media_quality).toEqual(DEFAULT_CONFIG.media_quality);
+    expect(warnSpy).toHaveBeenCalledTimes(1);
+  });
+
  it("replaces a config initialized earlier", () => {
    Config.initWith({ ssla: "https://first.invalid/ssla" });
    Config.initWith({ ssla: "https://second.invalid/ssla" });
@@ -156,5 +156,10 @@ async function downloadConfig(fetchTarget: string): Promise<ConfigOptions> {
    return {};
  }

-  return response.json();
+  try {
+    return await response.json();
+  } catch (e) {
+    logger.warn(`Config at ${fetchTarget} is not JSON, using defaults`, e);
+    return {};
+  }
 }
@@ -7,30 +7,30 @@ Please see LICENSE in the repository root for full details.

 import { type RelationType } from "matrix-js-sdk";

-import catSoundOgg from "../sound/reactions/cat.ogg?url";
-import catSoundMp3 from "../sound/reactions/cat.mp3?url";
-import clapSoundOgg from "../sound/reactions/clap.ogg?url";
-import clapSoundMp3 from "../sound/reactions/clap.mp3?url";
-import cricketsSoundOgg from "../sound/reactions/crickets.ogg?url";
-import cricketsSoundMp3 from "../sound/reactions/crickets.mp3?url";
-import dogSoundOgg from "../sound/reactions/dog.ogg?url";
-import dogSoundMp3 from "../sound/reactions/dog.mp3?url";
-import genericSoundOgg from "../sound/reactions/generic.ogg?url";
-import genericSoundMp3 from "../sound/reactions/generic.mp3?url";
-import lightbulbSoundOgg from "../sound/reactions/lightbulb.ogg?url";
-import lightbulbSoundMp3 from "../sound/reactions/lightbulb.mp3?url";
-import partySoundOgg from "../sound/reactions/party.ogg?url";
-import partySoundMp3 from "../sound/reactions/party.mp3?url";
-import deerSoundOgg from "../sound/reactions/deer.ogg?url";
-import deerSoundMp3 from "../sound/reactions/deer.mp3?url";
-import rockSoundOgg from "../sound/reactions/rock.ogg?url";
-import rockSoundMp3 from "../sound/reactions/rock.mp3?url";
-import waveSoundOgg from "../sound/reactions/wave.ogg?url";
-import waveSoundMp3 from "../sound/reactions/wave.mp3?url";
-import baduntssSoundOgg from "../sound/reactions/baduntss.ogg?url";
-import baduntssSoundMp3 from "../sound/reactions/baduntss.mp3?url";
-import bellSoundOgg from "../sound/reactions/bell.ogg?url";
-import bellSoundMp3 from "../sound/reactions/bell.mp3?url";
+import catSoundOgg from "../sound/reactions/cat.ogg?url&no-inline";
+import catSoundMp3 from "../sound/reactions/cat.mp3?url&no-inline";
+import clapSoundOgg from "../sound/reactions/clap.ogg?url&no-inline";
+import clapSoundMp3 from "../sound/reactions/clap.mp3?url&no-inline";
+import cricketsSoundOgg from "../sound/reactions/crickets.ogg?url&no-inline";
+import cricketsSoundMp3 from "../sound/reactions/crickets.mp3?url&no-inline";
+import dogSoundOgg from "../sound/reactions/dog.ogg?url&no-inline";
+import dogSoundMp3 from "../sound/reactions/dog.mp3?url&no-inline";
+import genericSoundOgg from "../sound/reactions/generic.ogg?url&no-inline";
+import genericSoundMp3 from "../sound/reactions/generic.mp3?url&no-inline";
+import lightbulbSoundOgg from "../sound/reactions/lightbulb.ogg?url&no-inline";
+import lightbulbSoundMp3 from "../sound/reactions/lightbulb.mp3?url&no-inline";
+import partySoundOgg from "../sound/reactions/party.ogg?url&no-inline";
+import partySoundMp3 from "../sound/reactions/party.mp3?url&no-inline";
+import deerSoundOgg from "../sound/reactions/deer.ogg?url&no-inline";
+import deerSoundMp3 from "../sound/reactions/deer.mp3?url&no-inline";
+import rockSoundOgg from "../sound/reactions/rock.ogg?url&no-inline";
+import rockSoundMp3 from "../sound/reactions/rock.mp3?url&no-inline";
+import waveSoundOgg from "../sound/reactions/wave.ogg?url&no-inline";
+import waveSoundMp3 from "../sound/reactions/wave.mp3?url&no-inline";
+import baduntssSoundOgg from "../sound/reactions/baduntss.ogg?url&no-inline";
+import baduntssSoundMp3 from "../sound/reactions/baduntss.mp3?url&no-inline";
+import bellSoundOgg from "../sound/reactions/bell.ogg?url&no-inline";
+import bellSoundMp3 from "../sound/reactions/bell.mp3?url&no-inline";

 export const ElementCallReactionEventType = "io.element.call.reaction";

@@ -8,18 +8,18 @@ Please see LICENSE in the repository root for full details.
 import { type ReactNode, useEffect } from "react";

 import { type CallViewModel } from "../state/CallViewModel/CallViewModel";
-import joinCallSoundMp3 from "../sound/join_call.mp3";
-import joinCallSoundOgg from "../sound/join_call.ogg";
-import leftCallSoundMp3 from "../sound/left_call.mp3";
-import leftCallSoundOgg from "../sound/left_call.ogg";
-import handSoundOgg from "../sound/raise_hand.ogg";
-import handSoundMp3 from "../sound/raise_hand.mp3";
-import screenShareStartedOgg from "../sound/screen_share_started.ogg";
-import screenShareStartedMp3 from "../sound/screen_share_started.mp3";
-import declineMp3 from "../sound/call_declined.mp3?url";
-import declineOgg from "../sound/call_declined.ogg?url";
-import timeoutMp3 from "../sound/call_timeout.mp3?url";
-import timeoutOgg from "../sound/call_timeout.ogg?url";
+import joinCallSoundMp3 from "../sound/join_call.mp3?url&no-inline";
+import joinCallSoundOgg from "../sound/join_call.ogg?url&no-inline";
+import leftCallSoundMp3 from "../sound/left_call.mp3?url&no-inline";
+import leftCallSoundOgg from "../sound/left_call.ogg?url&no-inline";
+import handSoundOgg from "../sound/raise_hand.ogg?url&no-inline";
+import handSoundMp3 from "../sound/raise_hand.mp3?url&no-inline";
+import screenShareStartedOgg from "../sound/screen_share_started.ogg?url&no-inline";
+import screenShareStartedMp3 from "../sound/screen_share_started.mp3?url&no-inline";
+import declineMp3 from "../sound/call_declined.mp3?url&no-inline";
+import declineOgg from "../sound/call_declined.ogg?url&no-inline";
+import timeoutMp3 from "../sound/call_timeout.mp3?url&no-inline";
+import timeoutOgg from "../sound/call_timeout.ogg?url&no-inline";
 import { useAudioContext } from "../useAudioContext";
 import { prefetchSounds } from "../soundUtils";
 import { useLatest } from "../useLatest";
@@ -75,7 +75,6 @@ import {
  type RTCSlotOptions,
  constant,
 } from "@element-hq/matrixrtc-sdk";
-import { map } from "rxjs";
 import {
  advancedCamera,
  advancedScreenShare,
@@ -181,7 +180,7 @@ export const ActiveCall: FC<ActiveCallProps> = (props) => {
              props.muteStates,
              mediaDevices,
              capture,
-              backgroundEffectsState$.value.processor,
+              backgroundEffectsState$.value,
            ),
    });

@@ -195,9 +194,7 @@ export const ActiveCall: FC<ActiveCallProps> = (props) => {
        ...callViewModelOptionsFromParams(urlParams),
        encryptionSystem: props.e2eeSystem,
        capture,
-        videoProcessor$: scope.behavior(
-          backgroundEffectsState$.pipe(map((state) => state.processor)),
-        ),
+        backgroundEffectsState$,
        hostBridge,
        autoLeaveWhenOthersLeft,
        waitForCallPickup: waitForCallPickup && sendNotificationType === "ring",
@@ -241,7 +238,7 @@ export const ActiveCall: FC<ActiveCallProps> = (props) => {
      vm,
      props.muteStates,
      mediaDevices,
-      backgroundEffectsState$.pipe(map(({ settling }) => settling ?? false)),
+      backgroundEffectsState$,
      `${props.client.getUserId()}:${props.client.getDeviceId()}`,
      { showControls: urlParams.showControls, header: urlParams.header },
    );
@@ -16,7 +16,6 @@ import {
  type LocalMedia,
  type LocalVideoMediaTrack,
 } from "@element-hq/matrixrtc-sdk";
-import { map } from "rxjs";

 import inCallStyles from "./InCallView.module.css";
 import styles from "./LobbyView.module.css";
@@ -134,9 +133,7 @@ export const LobbyView: FC<Props> = ({
        muteStates,
        devices,
        captureSettings(),
-        mediaScope.behavior(
-          backgroundEffectsState$.pipe(map((state) => state.processor)),
-        ),
+        backgroundEffectsState$,
      ),
    );
    return (): void => {
@@ -158,13 +155,15 @@ export const LobbyView: FC<Props> = ({
    null,
  );
  useEffect(() => {
+    if (mediaVm === null) return;
    const footerScope = new ObservableScope();
    setFooterVm(
      createLobbyFooterViewModel(
        footerScope,
        muteStates,
        devices,
-        backgroundEffectsState$.pipe(map(({ settling }) => settling ?? false)),
+        backgroundEffectsState$,
+        mediaVm.videoTrack$,
        openSettings,
        hangup,
        // Logo and header are connected: only show the logo in SPA with header.
@@ -179,6 +178,7 @@ export const LobbyView: FC<Props> = ({
    backgroundEffectsState$,
    hangup,
    hideHeader,
+    mediaVm,
    muteStates,
    openSettings,
  ]);
@@ -12,8 +12,8 @@ import { type RingingMediaViewModel } from "../state/media/RingingMediaViewModel
 import { useBehavior } from "../useBehavior";
 import { useInitial } from "../useInitial";
 import { prefetchSounds } from "../soundUtils";
-import ringtoneMp3 from "../sound/ringtone.mp3?url";
-import ringtoneOgg from "../sound/ringtone.ogg?url";
+import ringtoneMp3 from "../sound/ringtone.mp3?url&no-inline";
+import ringtoneOgg from "../sound/ringtone.ogg?url&no-inline";
 import { type UseAudioContext, useAudioContext } from "../useAudioContext";
 import { useLatest } from "../useLatest";

@@ -34,23 +34,6 @@ describe("prefetchSounds", () => {
    expect(bytesOf(sounds.beep)).toEqual([1, 2, 3]);
  });

-  it("decodes base64 data: URLs without fetching them", async () => {
-    // "OggS" — the bytes 0x4f 0x67 0x67 0x53
-    const sounds = await prefetchSounds({
-      beep: { ogg: "data:audio/ogg;base64,T2dnUw==" },
-    });
-    expect(fetchSpy).not.toHaveBeenCalled();
-    expect(bytesOf(sounds.beep)).toEqual([0x4f, 0x67, 0x67, 0x53]);
-  });
-
-  it("leaves out a sound with a percent-encoded data: URL", async () => {
-    const sounds = await prefetchSounds({
-      beep: { ogg: "data:audio/ogg,Ogg%53" },
-    });
-    expect(fetchSpy).not.toHaveBeenCalled();
-    expect(sounds).toEqual({});
-  });
-
  it("leaves out a sound whose response is not okay", async () => {
    fetchSpy.mockResolvedValue(new Response(null, { status: 404 }));
    const sounds = await prefetchSounds({
@@ -60,18 +43,16 @@ describe("prefetchSounds", () => {
  });

  it("leaves out a sound that fails to load, keeping the others", async () => {
-    fetchSpy.mockRejectedValue(new TypeError("Failed to fetch"));
+    // Fetched in the order given
+    fetchSpy
+      .mockRejectedValueOnce(new TypeError("Failed to fetch"))
+      .mockResolvedValueOnce(
+        new Response(new Uint8Array([1, 2, 3]), { status: 200 }),
+      );
    const sounds = await prefetchSounds({
      beep: { ogg: "https://example.org/beep.ogg" },
-      boop: { ogg: "data:audio/ogg;base64,T2dnUw==" },
+      boop: { ogg: "https://example.org/boop.ogg" },
    });
    expect(Object.keys(sounds)).toEqual(["boop"]);
  });
-
-  it("leaves out a sound with a malformed data: URL", async () => {
-    const sounds = await prefetchSounds({
-      beep: { ogg: "data:audio/ogg;base64" },
-    });
-    expect(sounds).toEqual({});
-  });
 });
@@ -32,38 +32,12 @@ function getPreferredAudioFormat(): "ogg" | "mp3" {

 const preferredFormat = getPreferredAudioFormat();

-/**
- * Decode a `data:` URL into the bytes it carries.
- *
- * The component build inlines the sound files as `data:` URLs. Those could be
- * `fetch`ed like any other URL, but that fetch is subject to the host page's
- * Content Security Policy, and `connect-src` lists rarely allow `data:` (a `*`
- * source does not cover it). Reading the bytes out of the URL directly needs no
- * network access, so no policy can stop it.
- * @param url The `data:` URL to decode.
- * @returns The decoded bytes.
- */
-function decodeDataUrl(url: string): ArrayBuffer {
-  const separator = url.indexOf(",");
-  if (!url.startsWith("data:") || separator === -1)
-    throw new Error("Not a valid data: URL");
-  const header = url.slice("data:".length, separator);
-  // Vite only percent-encodes SVGs; every other inlined asset is base64.
-  if (!header.split(";").includes("base64"))
-    throw new Error("Only base64 data: URLs are supported");
-  // `decoded` is a binary string: one char per byte
-  const decoded = atob(url.slice(separator + 1));
-  const bytes = Uint8Array.from(decoded, (c) => c.charCodeAt(0));
-  return bytes.buffer;
-}
-
 /**
 * Load the bytes of a single sound file.
- * @param url Where the sound is: an ordinary URL to fetch, or a `data:` URL.
+ * @param url Where the sound is.
 * @returns The bytes, or null if the sound could not be loaded.
 */
 async function loadSound(url: string): Promise<ArrayBuffer | null> {
-  if (url.startsWith("data:")) return decodeDataUrl(url);
  const response = await fetch(url);
  if (isFailure(response)) return null;
  return await response.arrayBuffer();
@@ -6,7 +6,7 @@ Please see LICENSE in the repository root for full details.
 */

 import { beforeEach, describe, expect, it, vi } from "vitest";
-import { BehaviorSubject, distinctUntilChanged, map } from "rxjs";
+import { BehaviorSubject, distinctUntilChanged, map, of } from "rxjs";
 import { type BackgroundProcessorWrapper } from "@livekit/track-processors";

 import {
@@ -47,23 +47,28 @@ function fakePipeline(): {
  };
 }

-/** One letter per state: idle, waiting for a frame, attached. */
+/** One letter per state: idle, preparing, waiting for a frame, attached. */
 function letter(state: BackgroundEffectsState): string {
+  if (state.supported === false) return "x";
+  if (state.wait === "preparing") return "p";
  if (state.processor === undefined) return "i";
-  return state.settling ? "w" : "a";
+  return state.wait === "settling" ? "w" : "a";
 }

 describe("the pipeline's state", () => {
  function testState({
    effect,
+    answer = "-y",
    firstFrame = "",
    expected,
  }: {
    effect: string;
+    /** When the trial build answers, from being asked. */
+    answer?: string;
    firstFrame?: string;
    expected: string;
  }): void {
-    withTestScheduler(({ behavior, schedule, expectObservable }) => {
+    withTestScheduler(({ behavior, cold, schedule, expectObservable }) => {
      const transformer = {
        onFirstFrame: undefined as (() => void) | undefined,
      };
@@ -72,6 +77,7 @@ describe("the pipeline's state", () => {
        effect$: behavior(effect, { n: "none", b: "blur" }),
        setEffect: vi.fn(),
        added$: new BehaviorSubject<AddedBackground[] | undefined>([]),
+        canSegment: () => cold(answer, { y: true, n: false }),
        pipeline: fakePipeline().pipeline,
        transformer,
      });
@@ -84,15 +90,21 @@ describe("the pipeline's state", () => {

  it("defaults to no effect", () => testState({ effect: "n", expected: "i" }));

+  it("attaches only once it is known the pipeline builds", () =>
+    testState({ effect: "nb", answer: "--y", expected: "ip-w" }));
+
+  it("offers none where the pipeline fails to build", () =>
+    testState({ effect: "nb", answer: "-n", expected: "ipx" }));
+
  it("waits from the first effect until a frame carries it", () =>
    testState({
-      effect: "    nb-n-b",
-      firstFrame: "  --f",
-      expected: "  iwa",
+      effect: "    nb---n-b",
+      firstFrame: "---f",
+      expected: "  ipwa",
    }));

  it("attaches on first use and stays attached", () =>
-    testState({ effect: "nb-n", firstFrame: "--f", expected: "iwa" }));
+    testState({ effect: "nb--n", firstFrame: "---f", expected: "ipwa" }));
 });

 describe("background effects", () => {
@@ -109,6 +121,7 @@ describe("background effects", () => {
      effect$,
      setEffect,
      added$,
+      canSegment: () => of(true),
      pipeline: fake.pipeline,
      transformer: { onFirstFrame: undefined },
      ...options,
@@ -118,8 +131,6 @@ describe("background effects", () => {
    effect$.next(raw);
    await flushPromises();
  };
-  const blur = async (on: boolean): Promise<void> =>
-    choose(on ? "blur" : "none");

  beforeEach(() => {
    effect$ = new BehaviorSubject("none");
@@ -128,6 +139,17 @@ describe("background effects", () => {
    fake = fakePipeline();
  });

+  it("offers none for the session where the pipeline fails to build", async () => {
+    const effects = build({ canSegment: () => of(false) });
+    await choose("blur");
+
+    expect(effects.state$.value.supported).toBe(false);
+    expect(effects.state$.value.processor).toBeUndefined();
+    expect(effects.state$.value.wait).toBeUndefined();
+    // A browser update may fix it, so the next session asks again.
+    expect(effect$.value).toBe("blur");
+  });
+
  it("puts a shipped background on as that picture", async () => {
    build();
    await choose(`image:${shippedBackgrounds[0].id}`);
@@ -141,10 +163,10 @@ describe("background effects", () => {

  it("switches in place rather than reattaching", async () => {
    const effects = build();
-    await blur(true);
+    await choose("blur");
    const pipeline = effects.state$.value.processor;
-    await blur(false);
-    await blur(true);
+    await choose("none");
+    await choose("blur");

    expect(effects.state$.value.processor).toBe(pipeline);
    expect(fake.switches).toEqual([
@@ -157,18 +179,18 @@ describe("background effects", () => {
  it("keeps switching after a switch fails", async () => {
    vi.mocked(fake.pipeline.switchTo).mockRejectedValueOnce(new Error("lost"));
    build();
-    await blur(true);
-    await blur(false);
+    await choose("blur");
+    await choose("none");
    expect(fake.switches).toEqual([{ mode: "disabled" }]);
  });

  it("switches one at a time, skipping those overtaken", async () => {
    const finished = fake.holdNext();
    build();
-    await blur(true);
-    await blur(false);
-    await blur(true);
-    await blur(false);
+    await choose("blur");
+    await choose("none");
+    await choose("blur");
+    await choose("none");
    expect(fake.switches).toHaveLength(1);

    finished();
@@ -8,9 +8,15 @@ Please see LICENSE in the repository root for full details.
 import {
  combineLatest,
  distinctUntilChanged,
+  first,
+  from,
  map,
  Observable,
+  type ObservableInput,
+  of,
  scan,
+  startWith,
+  switchMap,
 } from "rxjs";
 import {
  type BackgroundProcessorWrapper,
@@ -38,8 +44,12 @@ const blurRadius = 15;
 export interface BackgroundEffectsState {
  supported: boolean | undefined;
  processor: undefined | VideoProcessor;
-  /** From the first effect chosen until a frame carrying it is drawn. */
-  settling?: boolean;
+  /**
+   * The effect is not ready:
+   * preparing, asking whether the pipeline can be built.
+   * settling, once it is attached until a frame carrying the effect
+   */
+  wait?: "preparing" | "settling";
 }

 export interface BackgroundEffectsOptions {
@@ -51,6 +61,8 @@ export interface BackgroundEffectsOptions {
  setEffect: (id: EffectId) => void;
  /** The backgrounds the device keeps, undefined until it has said. */
  added$: Behavior<AddedBackground[] | undefined>;
+  /** Builds a segmenter on trial, and answers whether it could. */
+  canSegment: () => ObservableInput<boolean>;
  /** The background processor pipeline to be switched between effects. */
  pipeline: BackgroundProcessorWrapper;
  /** Tells, once, that a frame carrying an effect has been drawn. */
@@ -70,6 +82,7 @@ export function createBackgroundEffects(
    effect$,
    setEffect,
    added$,
+    canSegment,
    pipeline,
    transformer,
  }: BackgroundEffectsOptions,
@@ -77,9 +90,11 @@ export function createBackgroundEffects(
  // Wrapped once, so the call and the preview see one processor throughout.
  const processor = fromLivekitProcessor(pipeline);
  const choice$ = effect$.pipe(map(parseEffect));
-  const wanted$ = choice$.pipe(
-    map((effect) => effect.kind !== "none"),
-    distinctUntilChanged(),
+  const wanted$ = scope.behavior(
+    choice$.pipe(
+      map((effect) => effect.kind !== "none"),
+      distinctUntilChanged(),
+    ),
  );

  combineLatest([choice$, added$])
@@ -95,6 +110,18 @@ export function createBackgroundEffects(
        setEffect("none");
    });

+  // Asked once an effect is first wanted. A browser update may fix what
+  // fails, so the choice is kept for the next session rather than cleared.
+  const buildable$ = scope.behavior<boolean | undefined>(
+    supported
+      ? wanted$.pipe(
+          first(Boolean),
+          switchMap(() => from(canSegment())),
+          startWith(undefined),
+        )
+      : of(undefined),
+  );
+
  const drewAFrame$ = scope.behavior(
    new Observable<boolean>((subscriber) => {
      subscriber.next(false);
@@ -105,18 +132,25 @@ export function createBackgroundEffects(
    }),
  );

+  // In one step, so that nothing sees the pipeline neither preparing nor
+  // attached.
  const state$ = scope.behavior(
-    combineLatest([wanted$, drewAFrame$]).pipe(
-      scan<[boolean, boolean], BackgroundEffectsState>(
-        (previous, [wanted, drewAFrame]) => {
+    combineLatest([wanted$, buildable$, drewAFrame$]).pipe(
+      scan<[boolean, boolean | undefined, boolean], BackgroundEffectsState>(
+        (previous, [wanted, buildable, drewAFrame]) => {
          // Attached the first time an effect is wanted and never detached
          // after, so someone who never turns one on pays for none of it.
          const enable =
-            previous.processor !== undefined || (supported && wanted);
+            previous.processor !== undefined || (wanted && buildable === true);
+          const preparing = supported && wanted && buildable === undefined;
          return {
-            supported,
+            supported: supported && buildable !== false,
            processor: enable ? processor : undefined,
-            settling: enable && !drewAFrame,
+            wait: preparing
+              ? "preparing"
+              : enable && !drewAFrame
+                ? "settling"
+                : undefined,
          };
        },
        { supported, processor: undefined },
@@ -70,9 +70,11 @@ import {
 } from "./CallViewModel.ts";
 import { type Handler, MuteState } from "../MuteStates.ts";
 import {
+  type LocalVideoMediaTrack,
  type RTCParticipation,
  type VideoProcessor,
 } from "@element-hq/matrixrtc-sdk";
+import { type BackgroundEffectsState } from "../BackgroundEffects";

 initializeWidget();

@@ -1692,28 +1694,113 @@ describe("a camera with a video processor", () => {
        muteStates,
        mockMediaDevices({}),
        {},
-        processor,
+        { supported: true, processor },
      ).find((request) => request.source === "camera"),
    ).toMatchObject({ processor });
  });

+  it("leaves the camera out of the join while the processor is prepared", () => {
+    const muteStates = mockMuteStates();
+    vi.spyOn(muteStates.video.enabled$, "value", "get").mockReturnValue(true);
+    const sources = (wait?: "preparing"): string[] =>
+      initialPublishRequests(
+        muteStates,
+        mockMediaDevices({}),
+        {},
+        {
+          supported: true,
+          processor: undefined,
+          wait,
+        },
+      ).map((request) => request.source);
+    expect(sources("preparing")).not.toContain("camera");
+    expect(sources()).toContain("camera");
+  });
+
+  it("waits to make a camera track until it is known the processor builds", async () => {
+    const setHandler = vi.spyOn(MuteState.prototype, "setHandler");
+    onTestFinished(() => setHandler.mockRestore());
+    const effects$ = new BehaviorSubject<BackgroundEffectsState>({
+      supported: true,
+      processor: undefined,
+      wait: "preparing",
+    });
+    let participation: RTCParticipation | undefined;
+    withCallViewModel(
+      {},
+      (_vm, _session, { rtcParticipation }) => {
+        participation = rtcParticipation;
+      },
+      { backgroundEffectsState$: effects$ },
+    );
+    const camera = participation!.localMember$.value?.tracks$.value?.find(
+      (track) => track.source === "camera",
+    ) as LocalVideoMediaTrack;
+    const turnedOn = cameraHandler(setHandler)(true);
+    await flushPromises();
+    expect(camera.setEnabled).not.toHaveBeenCalledWith(true);
+
+    effects$.next({ supported: true, processor });
+    await turnedOn;
+    expect(camera.setEnabled).toHaveBeenCalledWith(true);
+    expect(participation!.publish).not.toHaveBeenCalled();
+  });
+
+  it("turns the camera on only once its processor is on the track", async () => {
+    const setHandler = vi.spyOn(MuteState.prototype, "setHandler");
+    onTestFinished(() => setHandler.mockRestore());
+    const effects$ = new BehaviorSubject<BackgroundEffectsState>({
+      supported: true,
+      processor: undefined,
+      wait: "preparing",
+    });
+    let participation: RTCParticipation | undefined;
+    withCallViewModel(
+      {},
+      (_vm, _session, { rtcParticipation }) => {
+        participation = rtcParticipation;
+      },
+      { backgroundEffectsState$: effects$ },
+    );
+    const camera = participation!.localMember$.value?.tracks$.value?.find(
+      (track) => track.source === "camera",
+    ) as LocalVideoMediaTrack;
+    // Attaching takes its time, as the pipeline is built on the track
+    let attached!: () => void;
+    vi.mocked(camera.setProcessor).mockReturnValueOnce(
+      new Promise<void>((resolve) => {
+        attached = resolve;
+      }),
+    );
+    const turnedOn = cameraHandler(setHandler)(true);
+    effects$.next({ supported: true, processor });
+    await flushPromises();
+    expect(camera.setProcessor).toHaveBeenCalledWith(processor);
+    expect(camera.setEnabled).not.toHaveBeenCalledWith(true);
+
+    attached();
+    await turnedOn;
+    expect(camera.setEnabled).toHaveBeenCalledWith(true);
+  });
+
  // The SDK builds the processor on the camera's restart, before the room
  // sees a frame, so the track is unmuted rather than published anew
  it("turns on a camera turned off before the processor was chosen", async () => {
    const setHandler = vi.spyOn(MuteState.prototype, "setHandler");
    onTestFinished(() => setHandler.mockRestore());
-    const videoProcessor$ = new BehaviorSubject<VideoProcessor | undefined>(
-      undefined,
-    );
+    const effects$ = new BehaviorSubject<BackgroundEffectsState>({
+      supported: true,
+      processor: undefined,
+    });
    let participation: RTCParticipation | undefined;
    withCallViewModel(
      {},
      (_vm, _session, { rtcParticipation }) => {
        participation = rtcParticipation;
      },
-      { videoProcessor$ },
+      { backgroundEffectsState$: effects$ },
    );
-    videoProcessor$.next(processor);
+    effects$.next({ supported: true, processor });
    await cameraHandler(setHandler)(true);
    await flushPromises();

@@ -37,6 +37,7 @@ import {
  combineLatest,
  distinctUntilChanged,
  filter,
+  firstValueFrom,
  fromEvent,
  map,
  merge,
@@ -77,6 +78,7 @@ import {
  type ReactionInfo,
  type ReactionOption,
 } from "../../reactions";
+import { type BackgroundEffectsState } from "../BackgroundEffects";
 import { shallowEquals as arrayShallowEquals } from "../../utils/array";
 import { type MediaDevices } from "../MediaDevices";
 import { type MuteStates } from "../MuteStates";
@@ -170,13 +172,16 @@ export function initialPublishRequests(
  muteStates: MuteStates,
  mediaDevices: MediaDevices,
  capture: CaptureSettings,
-  processor?: VideoProcessor,
+  effects?: BackgroundEffectsState,
 ): PublishRequest[] {
  const requests: PublishRequest[] = [];
  if (muteStates.audio.enabled$.value)
    requests.push(publishRequest("microphone", mediaDevices, capture));
-  if (muteStates.video.enabled$.value)
-    requests.push(publishRequest("camera", mediaDevices, capture, processor));
+  // Left out while the processor is prepared; the call publishes it after
+  if (muteStates.video.enabled$.value && effects?.wait !== "preparing")
+    requests.push(
+      publishRequest("camera", mediaDevices, capture, effects?.processor),
+    );
  return requests;
 }

@@ -187,8 +192,8 @@ export interface CallViewModelOptions {
   * defaults otherwise.
   */
  capture?: CaptureSettings;
-  /** Background blur and the like, applied to the camera as it changes. */
-  videoProcessor$?: Behavior<VideoProcessor | undefined>;
+  /** The background effects state (preparing, supported...) including the processor itself. */
+  backgroundEffectsState$?: Behavior<BackgroundEffectsState>;
  /**
   * The application hosting Element Call, which can ask it to hang up and wants
   * to know when the user joins or leaves. Defaults to no host.
@@ -363,6 +368,8 @@ export interface CallViewModel {
  keyRotationSuppressed$: Behavior<boolean>;
  /** The local member, once our membership has been seen in the room. */
  localMember$: Behavior<LocalRTCMember | null>;
+  /** Our own camera track, while one is published. */
+  camera$: Behavior<LocalVideoMediaTrack | undefined>;
  remoteMembers$: Behavior<RemoteRTCMember[]>;
  /** The transports the client holds a connection to, for the developer panel. */
  connectedTransports$: Behavior<TransportMetadata[]>;
@@ -488,7 +495,21 @@ export function createCallViewModel$(
    sentCallNotification$ = of(null),
  } = options;

-  const { capture = {}, videoProcessor$ = constant(undefined) } = options;
+  const {
+    capture = {},
+    backgroundEffectsState$ = constant<BackgroundEffectsState>({
+      supported: false,
+      processor: undefined,
+    }),
+  } = options;
+  const videoProcessor$ = scope.behavior(
+    backgroundEffectsState$.pipe(map(({ processor }) => processor)),
+  );
+  // Resolves once it is known whether the processor builds
+  const processorKnown = async (): Promise<unknown> =>
+    firstValueFrom(
+      backgroundEffectsState$.pipe(filter(({ wait }) => wait !== "preparing")),
+    );
  const localTracks$ = scope.behavior<
    (LocalAudioMediaTrack | LocalVideoMediaTrack)[]
  >(
@@ -512,6 +533,19 @@ export function createCallViewModel$(
    "screenShare",
  ) as Behavior<LocalVideoMediaTrack | undefined>;

+  const processorAttached = followProcessor(
+    scope,
+    camera$,
+    videoProcessor$,
+    logger,
+  );
+
+  // A camera turned on waits for its processor to be known, then to be on
+  // the track: LiveKit restarts the track with whatever processor it has then.
+  const cameraReady = async (): Promise<void> => {
+    await processorKnown();
+    await processorAttached();
+  };
  const supply = (source: "microphone" | "camera"): TrackSupply => ({
    publish: async () =>
      rtcParticipation.publish(
@@ -522,6 +556,7 @@ export function createCallViewModel$(
          source === "camera" ? videoProcessor$.value : undefined,
        ),
      ),
+    ready: source === "camera" ? cameraReady : undefined,
  });
  followMuteState(
    scope,
@@ -548,6 +583,29 @@ export function createCallViewModel$(
  );
  followDevice(scope, camera$, mediaDevices.videoInput, "camera", logger);

+  // A camera wanted at the join was left out of it while the processor was
+  // prepared (see initialPublishRequests), and is published once that is known
+  if (
+    backgroundEffectsState$.value.wait === "preparing" &&
+    muteStates.video.enabled$.value
+  )
+    processorKnown()
+      .then(async () => {
+        if (camera$.value !== undefined || !muteStates.video.enabled$.value)
+          return;
+        await rtcParticipation.publish(
+          publishRequest(
+            "camera",
+            mediaDevices,
+            capture,
+            videoProcessor$.value,
+          ),
+        );
+      })
+      .catch((e) => {
+        logger.error("Failed to publish the camera", e);
+      });
+
  // The output device follows the selection onto the client the same way. A
  // host that routes audio itself picks the output outside the browser, so the
  // client is left alone
@@ -561,8 +619,6 @@ export function createCallViewModel$(
        });
      });

-  followProcessor(scope, camera$, videoProcessor$, logger);
-
  const members$ = scope.behavior<(LocalRTCMember | RemoteRTCMember)[]>(
    combineLatest(
      [rtcParticipation.localMember$, rtcParticipation.remoteMembers$],
@@ -1508,6 +1564,7 @@ export function createCallViewModel$(
    leave,
    toggleScreenSharing: toggleScreenSharing,
    sharingScreen$: sharingScreen$,
+    camera$,

    tapScreen: (): void => screenTap$.next(),
    tapControls: (): void => controlsTap$.next(),
@@ -21,6 +21,7 @@ import { type MediaDevices } from "./MediaDevices";
 import { MuteStates } from "./MuteStates";
 import { nullHostBridge } from "../HostBridge";
 import { flushPromises, mockMediaDevices, testScope } from "../utils/test";
+import { type BackgroundEffectsState } from "./BackgroundEffects";

 describe("LobbyMediaViewModel", () => {
  it("opens what starts enabled in one capture and follows the devices", async () => {
@@ -135,7 +136,7 @@ function setup(
    muteStates,
    mediaDevices,
    {},
-    constant(processor),
+    constant<BackgroundEffectsState>({ supported: true, processor }),
  );
  return { media, vm, muteStates, mediaDevices };
 }
@@ -13,7 +13,6 @@ import {
  type ObservableScope,
  type PublishRequest,
  trackBySource$,
-  type VideoProcessor,
 } from "@element-hq/matrixrtc-sdk";
 import { logger as rootLogger } from "matrix-js-sdk/lib/logger";
 import { combineLatest, debounceTime, map } from "rxjs";
@@ -25,6 +24,7 @@ import {
 import { type MediaDevices } from "./MediaDevices";
 import { type MuteStates } from "./MuteStates";
 import { followDevice, followProcessor } from "./localMediaControls";
+import { type BackgroundEffectsState } from "./BackgroundEffects";

 export interface LobbyMediaViewModel {
  /** The camera, once open, for the preview. */
@@ -42,9 +42,13 @@ export function createLobbyMediaViewModel(
  muteStates: MuteStates,
  mediaDevices: MediaDevices,
  capture: CaptureSettings,
-  videoProcessor$: Behavior<VideoProcessor | undefined>,
+  /** No camera is opened while its pipeline is prepared: it would show the room without the processor. */
+  backgroundEffectsState$: Behavior<BackgroundEffectsState>,
 ): LobbyMediaViewModel {
  const logger = rootLogger.getChild("[LobbyMedia]");
+  const videoProcessor$ = scope.behavior(
+    backgroundEffectsState$.pipe(map(({ processor }) => processor)),
+  );
  const microphone$ = trackBySource$(
    scope,
    media.tracks$,
@@ -64,13 +68,19 @@ export function createLobbyMediaViewModel(
    muteStates.video.enabled$,
    microphone$,
    camera$,
+    backgroundEffectsState$,
  ])
    .pipe(debounceTime(0), scope.bind())
-    .subscribe(([audio, video, microphone, camera]) => {
+    .subscribe(([audio, video, microphone, camera, { wait }]) => {
      const open: PublishRequest[] = [];
      if (audio && microphone === undefined && !opening.has("microphone"))
        open.push(publishRequest("microphone", mediaDevices, capture));
-      if (video && camera === undefined && !opening.has("camera"))
+      if (
+        video &&
+        camera === undefined &&
+        !opening.has("camera") &&
+        wait !== "preparing"
+      )
        open.push(
          publishRequest(
            "camera",
@@ -21,6 +21,8 @@ import { type Handler } from "./MuteStates";
 /** How a mute switch gets a track where there is none. */
 export interface TrackSupply {
  publish: () => Promise<unknown>;
+  /** Awaited before a switch turned on acts: a camera waits for its processor to be known. */
+  ready?: () => Promise<unknown>;
 }

 /**
@@ -45,6 +47,7 @@ export function followMuteState(
  let desired = muteState.enabled$.value;
  muteState.setHandler(async (enabled) => {
    desired = enabled;
+    if (enabled) await supply.ready?.();
    const track = track$.value;
    if (track !== undefined) {
      desired = await track.setEnabled(enabled);
@@ -94,18 +97,23 @@ export function followDevice(
 /**
 * Attached once the camera track exists rather than carried in the publish
 * request, so that a new track and a change of processor take the same path.
+ * Answers with a wait for the latest attachment to have gone through: a
+ * camera turned on before then would restart without its processor.
 */
 export function followProcessor(
  scope: ObservableScope,
  camera$: Behavior<LocalVideoMediaTrack | undefined>,
  videoProcessor$: Behavior<VideoProcessor | undefined>,
  logger: Logger,
-): void {
+): () => Promise<void> {
+  let attached = Promise.resolve();
  combineLatest([camera$, videoProcessor$])
    .pipe(scope.bind())
    .subscribe(([camera, processor]) => {
-      camera?.setProcessor(processor).catch((e) => {
-        logger.error("Failed to set the video processor", e);
-      });
+      attached =
+        camera?.setProcessor(processor).catch((e) => {
+          logger.error("Failed to set the video processor", e);
+        }) ?? Promise.resolve();
    });
+  return async () => attached;
 }
@@ -13,13 +13,19 @@ import {
 } from "@livekit/track-processors";
 import { ImageSegmenter } from "@mediapipe/tasks-vision";

-import { BackgroundEffectTransformer } from "./BackgroundEffectTransformer";
+import {
+  BackgroundEffectTransformer,
+  canSegment,
+} from "./BackgroundEffectTransformer";

 const frame = {} as VideoFrame;
 const controller = {} as TransformStreamDefaultController<VideoFrame>;

 describe("BackgroundEffectTransformer", () => {
-  afterEach(() => vi.restoreAllMocks());
+  afterEach(() => {
+    vi.restoreAllMocks();
+    vi.unstubAllGlobals();
+  });

  it("loads a background chosen before it was built", async () => {
    vi.spyOn(VideoTransformer.prototype, "init").mockResolvedValue();
@@ -41,6 +47,24 @@ describe("BackgroundEffectTransformer", () => {
    expect(load).toHaveBeenCalledWith("/background.jpg");
  });

+  it("runs MediaPipe from its own bundle", async () => {
+    vi.spyOn(VideoTransformer.prototype, "init").mockResolvedValue();
+    const create = vi
+      .spyOn(ImageSegmenter, "createFromOptions")
+      .mockResolvedValue({} as ImageSegmenter);
+
+    await new BackgroundEffectTransformer({ backgroundDisabled: true }).init(
+      {} as VideoTransformerInitOptions,
+    );
+
+    const [fileset, options] = create.mock.calls[0];
+    expect(fileset.wasmLoaderPath).toMatch(/vision_wasm_internal\.js/);
+    expect(fileset.wasmBinaryPath).toMatch(/vision_wasm_internal\.wasm/);
+    expect(options.baseOptions?.modelAssetPath).toMatch(
+      /selfie_segmenter\.tflite/,
+    );
+  });
+
  it("reports the first frame carrying an effect, once", async () => {
    vi.spyOn(BackgroundTransformer.prototype, "transform").mockImplementation(
      async function (this: BackgroundTransformer) {
@@ -78,4 +102,33 @@ describe("BackgroundEffectTransformer", () => {
    expect(transformer.options.blurRadius).toBeUndefined();
    expect(transformer.options.imagePath).toBeUndefined();
  });
+
+  it("segments on the CPU where the GPU refuses", async () => {
+    vi.spyOn(VideoTransformer.prototype, "init").mockResolvedValue();
+    const build = vi
+      .spyOn(ImageSegmenter, "createFromOptions")
+      .mockRejectedValueOnce(new Error("blocklisted"))
+      .mockResolvedValue({} as ImageSegmenter);
+
+    await new BackgroundEffectTransformer({}).init(
+      {} as VideoTransformerInitOptions,
+    );
+    expect(build.mock.calls.map(([, o]) => o.baseOptions?.delegate)).toEqual([
+      "GPU",
+      "CPU",
+    ]);
+  });
+
+  it("says whether a segmenter can be built at all", async () => {
+    vi.stubGlobal("OffscreenCanvas", class {});
+    const close = vi.fn();
+    const build = vi
+      .spyOn(ImageSegmenter, "createFromOptions")
+      .mockResolvedValue({ close } as unknown as ImageSegmenter);
+    expect(await canSegment()).toBe(true);
+    expect(close).toHaveBeenCalled();
+
+    build.mockRejectedValue(new Error("no model"));
+    expect(await canSegment()).toBe(false);
+  });
 });
@@ -11,11 +11,18 @@ import {
  VideoTransformer,
  type VideoTransformerInitOptions,
  type BackgroundOptions,
+  type SegmenterOptions,
 } from "@livekit/track-processors";
 import { logger } from "matrix-js-sdk/lib/logger";
 import { ImageSegmenter } from "@mediapipe/tasks-vision";

-import modelAssetPath from "../mediapipe/imageSegmenter/selfie_segmenter.tflite?url";
+// Only the SIMD build of MediaPipe's runtime: every browser Element Call
+// supports has WebAssembly SIMD. As files rather than data URLs, so that a
+// host's bundler serves them as assets of its own, and a browser fetches them
+// only when an effect is first switched on.
+import wasmLoaderPath from "@mediapipe/tasks-vision/vision_wasm_internal.js?url&no-inline";
+import wasmBinaryPath from "@mediapipe/tasks-vision/vision_wasm_internal.wasm?url&no-inline";
+import modelAssetPath from "../mediapipe/imageSegmenter/selfie_segmenter.tflite?url&no-inline";

 interface WasmFileset {
  /** The path to the Wasm loader script. */
@@ -24,26 +31,7 @@ interface WasmFileset {
  wasmBinaryPath: string;
 }

-// The MediaPipe package, by default, ships some alternative versions of the
-// WASM files which avoid SIMD for compatibility with older browsers. But SIMD
-// in WASM is actually fine by our support policy, so we include just the SIMD
-// versions.
-// It's really not ideal that we have to reference these internal files from
-// MediaPipe and depend on node_modules having this specific structure. It's
-// easy to see this breaking if our dependencies changed and MediaPipe were
-// no longer hoisted, or if we switched to another dependency loader such as
-// yarn PnP.
-// https://github.com/google-ai-edge/mediapipe/issues/5961
-const wasmFileset: WasmFileset = {
-  wasmLoaderPath: new URL(
-    "../../node_modules/@mediapipe/tasks-vision/wasm/vision_wasm_internal.js",
-    import.meta.url,
-  ).href,
-  wasmBinaryPath: new URL(
-    "../../node_modules/@mediapipe/tasks-vision/wasm/vision_wasm_internal.wasm",
-    import.meta.url,
-  ).href,
-};
+const wasmFileset: WasmFileset = { wasmLoaderPath, wasmBinaryPath };

 /**
 * Track processor that applies effects such as blurring to a user's background.
@@ -92,17 +80,10 @@ export class BackgroundEffectTransformer extends BackgroundTransformer {
      inputElement: inputVideo,
    });

-    this.imageSegmenter = await ImageSegmenter.createFromOptions(wasmFileset, {
-      baseOptions: {
-        modelAssetPath,
-        delegate: "GPU",
-        ...this.options.segmenterOptions,
-      },
-      canvas: this.canvas,
-      runningMode: "VIDEO",
-      outputCategoryMask: true,
-      outputConfidenceMasks: false,
-    });
+    this.imageSegmenter = await createSegmenter(
+      this.canvas,
+      this.options.segmenterOptions,
+    );

    // BackgroundTransformer's own init applies these, and this one replaces it.
    if (this.options.imagePath) {
@@ -116,3 +97,38 @@ export class BackgroundEffectTransformer extends BackgroundTransformer {
    this.gl?.setBackgroundDisabled(this.options.backgroundDisabled ?? false);
  }
 }
+
+/**
+ * Whether a segmenter can be built here, asked once before any camera depends
+ * on one: a pipeline that fails to build takes the camera down with it.
+ */
+export async function canSegment(): Promise<boolean> {
+  try {
+    (await createSegmenter(new OffscreenCanvas(1, 1))).close();
+    return true;
+  } catch (e) {
+    logger.warn("Background effects cannot run here", e);
+    return false;
+  }
+}
+
+/** On the GPU, or on the CPU where the GPU refuses, as a blocklisted one does. */
+async function createSegmenter(
+  canvas: HTMLCanvasElement | OffscreenCanvas | undefined,
+  segmenterOptions?: SegmenterOptions,
+): Promise<ImageSegmenter> {
+  const build = async (delegate: "GPU" | "CPU"): Promise<ImageSegmenter> =>
+    ImageSegmenter.createFromOptions(wasmFileset, {
+      baseOptions: { modelAssetPath, delegate, ...segmenterOptions },
+      canvas,
+      runningMode: "VIDEO",
+      outputCategoryMask: true,
+      outputConfidenceMasks: false,
+    });
+  try {
+    return await build("GPU");
+  } catch (e) {
+    logger.warn("Segmenting on the CPU, as the GPU refused", e);
+    return build("CPU");
+  }
+}
@@ -56,7 +56,8 @@ vi.mock("./backgroundImages", async () => {
  };
 });
 vi.mock("./BackgroundEffectTransformer", () => ({
-  BackgroundEffectTransformer: vi.fn(),
+  BackgroundEffectTransformer: vi.fn(function () {}),
+  canSegment: async (): Promise<boolean> => Promise.resolve(true),
 }));

 describe("BackgroundEffectsProvider", () => {
@@ -12,7 +12,10 @@ import {
 import { createContext, type FC, type JSX, use } from "react";

 import { backgroundEffect as backgroundEffectSetting } from "../settings/settings";
-import { BackgroundEffectTransformer } from "./BackgroundEffectTransformer";
+import {
+  BackgroundEffectTransformer,
+  canSegment,
+} from "./BackgroundEffectTransformer";
 import { addedBackgrounds } from "./backgroundImages";
 import { type Behavior } from "../state/Behavior";
 import { type ObservableScope } from "../state/ObservableScope";
@@ -56,6 +59,7 @@ export function createAppBackgroundEffects(
    effect$: backgroundEffectSetting.value$,
    setEffect: backgroundEffectSetting.setValue,
    added$: addedBackgrounds.added$,
+    canSegment,
    pipeline: new BackgroundProcessorWrapper(transformer, "background-effect"),
    transformer,
  });
@@ -5,8 +5,8 @@ SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-Element-Commercial
 Please see LICENSE in the repository root for full details.
 */

-import arc from "../graphics/background-arc.jpg?url";
-import glow from "../graphics/background-glow.jpg?url";
+import arc from "../graphics/background-arc.jpg?url&no-inline";
+import glow from "../graphics/background-glow.jpg?url&no-inline";

 export interface ShippedBackground {
  /** Stored in the user's settings when chosen, so it must not change. */
@@ -217,7 +217,7 @@ export function getBasicCallViewModelEnvironment(
    vm,
    muteStates,
    mediaDevices,
-    constant(false),
+    constant({ supported: false, processor: undefined }),
    "reactionId",
    { showControls: true, header: HeaderStyle.Standard },
  );
@@ -6,6 +6,7 @@ Please see LICENSE in the repository root for full details.
 */

 import { defineConfig } from "vite";
+import { posix } from "node:path";

 import { vitePluginsConfig } from "./vite.config.ts";
 import { scopeStylesToRoot } from "./component/build/scopeStylesToRoot.ts";
@@ -34,6 +35,25 @@ export default defineConfig(({ mode }) => {
    // whatever is in `public` — including the developer's own config.json, which
    // is not in the repository — into the output we would publish.
    publicDir: false,
+    // A library does not know where it will be served from, so every file it
+    // emits beside itself — the sounds, the pictures, MediaPipe's runtime — is
+    // referenced relative to the module's own URL. (A library build inlines
+    // each asset as a data URL unless its import says `no-inline`, so the
+    // imports do.)
+    base: "./",
+    experimental: {
+      // Relative to the chunk that references it, wherever that chunk lands,
+      // and spelled `./assets/…` rather than `assets/…`: a host's bundler reads
+      // a bare name as a package to look up.
+      renderBuiltUrl: (filename, { hostId, hostType }) => {
+        if (hostType !== "js") return { relative: true };
+        const relative = posix.relative(posix.dirname(hostId), filename);
+        const specifier = relative.startsWith(".") ? relative : `./${relative}`;
+        return {
+          runtime: `new URL(${JSON.stringify(specifier)}, import.meta.url).href`,
+        };
+      },
+    },
    // A host's document is not ours to style: everything in the stylesheet is
    // confined to the element Element Call is mounted in
    css: { postcss: { plugins: [scopeStylesToRoot()] } },
@@ -64,6 +84,15 @@ export default defineConfig(({ mode }) => {
        cssFileName: "element-call",
      },
      rollupOptions: {
+        output: {
+          // Hashed, as the app's are, so that a file a host serves as it is
+          // stays correct across versions of the package. Not the stylesheet,
+          // which the package exports by name (see `cssFileName`).
+          assetFileNames: ({ names }) =>
+            names.some((name) => name.endsWith(".css"))
+              ? "[name][extname]"
+              : "assets/[name]-[hash][extname]",
+        },
        // The host already has these, and a second copy of any of them does not
        // merely bloat the bundle: React would hold two sets of hooks, the
        // Matrix client would run two sync loops, and a second Compound would
Author	SHA1	Message	Date
Timo K.	704cbd65a6	Shorten component readme - At the end there were not that many unique facts in there. It now only lists the relevant bits without much text around it.	2026-11-10 14:31:12 +01:00
Timo	5f59283b4a	Merge pull request #4428 from element-hq/toger5/component-assets-as-files Ship the component's assets as files beside the module	2026-11-10 14:18:30 +01:00
Johannes Marbach	07dc08cd1c	Merge pull request #4440 from element-hq/johannes/embedded Make one site use the embedded build in playwright tests	2026-11-10 07:46:51 +01:00
Johannes Marbach	60d6029184	Delint	2026-11-10 07:38:23 +01:00
Johannes Marbach	f41bc2589a	Merge branch 'main' into johannes/embedded	2026-11-10 07:36:15 +01:00
Johannes Marbach	ceeb26b271	Remove test.slow()	2026-11-10 07:30:57 +01:00
Johannes Marbach	e822f967c1	Merge pull request #4442 from element-hq/johannes/shard Shard playwright tests	2026-11-10 07:22:10 +01:00
renovate[bot]	ec46b04b71	Update ghcr.io/element-hq/element-web:develop Docker digest to 37d88b4 (#4443 ) Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>	2026-11-10 03:24:30 +00:00
Johannes Marbach	4f3dc9af75	Shard playwright tests	2026-11-09 13:00:52 +01:00
renovate[bot]	38edc47f64	Update ghcr.io/element-hq/element-web:develop Docker digest to 0c39f7b (#4439 ) Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>	2026-11-09 09:29:45 +00:00
Johannes Marbach	ab4f0f0f77	Make one site use the embedded build in playwright tests	2026-11-09 10:05:04 +01:00
renovate[bot]	9411a56afc	Update ghcr.io/element-hq/element-web:develop Docker digest to f30be05 (#4438 ) Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>	2026-11-08 01:46:48 +00:00
Timo K.andClaude Fable 5.1	a5374791e4	Resolve asset references from the chunk that holds them The runtime path was `./` plus the asset's path from the package root, which is right for every chunk the build emits today, all of which sit at that root, and wrong for one emitted under a directory. Vite names the referencing chunk, so the path is now computed relative to it. The sound loader kept a branch for sounds inlined as data URLs, with a comment saying the component build produces them. No build does any more, and no host would inline a sound of this size, so the branch and its tests go. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>	2026-11-07 16:06:05 +01:00
renovate[bot]	a0601876d8	Update ghcr.io/element-hq/element-web:develop Docker digest to eada111 (#4435 ) Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>	2026-11-07 01:07:59 +00:00
fkwp	eef86fc06e	Merge pull request #4292 from element-hq/spec/background-effects/12-unbuildable-pipeline Offer no effects for the session where the pipeline cannot be built	2026-11-06 18:27:12 +01:00
Timo K.	cd5c367ce9	fix race in processor setup vs camera start	2026-11-06 16:54:42 +01:00
Timo K.andClaude Fable 5.1	190902eec0	Test the session where the pipeline cannot be built - The effects state's timeline through preparing, settling and attached, and that a failed trial build withholds the effects for the session while keeping the choice. - The footer offers nothing once the pipeline fails to build, and passes on the wait as the state's single wait field. - The call leaves the camera out of the join while the pipeline is prepared, publishes it once that is known, and turns on a camera turned off before the processor was chosen; the lobby opens no camera meanwhile. Both view models are driven by one state subject. - The transformer's trial build answers whether a segmenter can be made. - End to end, a browser whose pipeline cannot be built sees the effects withheld. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>	2026-11-06 16:54:42 +01:00
Timo K.andClaude Opus 5.5	88f748ec21	Offer no effects for the session where the pipeline cannot be built - A segmenter that can't be built, with the model unreachable or a GPU its delegate refuses, took the camera down with the pipeline: the SDK fails the track it was attaching to, and the lobby muted camera and microphone, on every load while the choice was kept. - One is now built on trial when an effect is first wanted, before the pipeline is attached to anything, on the CPU where the GPU refuses. Where it can't be built, the options are disabled with the existing notice for the rest of the session. The choice is kept: a browser update may fix what failed. - The view model takes that verdict from the pipeline's state rather than asking the SDK, so both say the same, and it no longer builds a WebGL context of its own to ask. - Until the trial is over, the lobby opens no camera and the call makes no camera track, so neither starts without the effect. Rebased onto main, where the Publisher is gone and the call view model publishes the camera through the SDK: - CallViewModel takes videoProcessorPreparing$. While it is true, the camera's mute handler waits before making a track, and the join leaves the camera out of its requests (initialPublishRequests); the view model publishes it once it is known whether the pipeline builds. The Publisher's test became CallViewModel tests. Cherry-picked from spec/background-effects/12-unbuildable-pipeline onto toger5/background-effect/camera-preview, where the camera menu's preview takes the camera from the footer's caller rather than from a track synced on the background-effects state. The footer factories take the pipeline's state again, as this commit needs its verdict and not only the settling flag; the synced camera and the preview's wrapper type are not brought along. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>	2026-11-06 16:54:42 +01:00
Timo	9639effd28	Merge pull request #4423 from element-hq/toger5/background-effect/camera-preview Camera Preview in video mute button context menu	2026-11-06 15:01:06 +00:00
Timo K.	f2f67019b4	Show the self preview in the video mute button context menu	2026-11-06 15:35:04 +01:00
Timo K.	51fac690f4	Hand the camera track to the footer view model	2026-11-06 15:35:04 +01:00
Timo K.andClaude Fable 5.1	797a86815b	Ship the component's assets as files beside the module The component module was 18 MB, of which 15 MB was MediaPipe's runtime as a base64 data URL, with the segmentation model, every sound and the background pictures inlined the same way: a library build inlines every asset unless its import says `no-inline`. Every host downloaded all of it before the first call, whether or not a background effect was ever switched on, and a host with a Content Security Policy without `script-src data:` could not run effects at all, since MediaPipe injects its loader as a script tag. The imports now say `no-inline`, and the build renders each reference as `new URL("./assets/…", import.meta.url)`, which webpack 5, Vite and Rollup follow as an asset of the host's own; the `./` is spelled out because a bare `assets/…` is read as a package name. The module is 2.3 MB. The runtime is fetched when an effect is first switched on and cached on its own. MediaPipe's files come through the package's `exports` rather than a path into node_modules. MediaPipe's JavaScript (160 kB) stays in the module: LiveKit's track processors import it statically, so a dynamic import moves nothing. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>	2026-11-06 14:39:25 +01:00
Johannes Marbach	1ae57e9647	Merge pull request #4420 from element-hq/johannes/mark-release-announcements-viewed Mark known release announcements as viewed	2026-11-06 13:50:54 +01:00
Timo	4b05ab503e	Merge pull request #4416 from element-hq/toger5/fix-flickering-device-id-mismatch fix camera restart due to device id mismatch	2026-11-06 13:45:04 +01:00
Johannes Marbach	ea977cecd9	Mark known release announcements as viewed	2026-11-06 11:31:01 +01:00
Johannes Marbach	aadbda39b1	Merge pull request #4405 from element-hq/renovate/npm-minimatch-vulnerability Update dependency minimatch to v11 [SECURITY]	2026-11-06 11:12:09 +01:00
renovate[bot]	6b56c7e683	Update dependency minimatch to v11 [SECURITY]	2026-11-06 09:35:24 +00:00
Timo K.	11e143c8ab	fix camera restart due to device id mismatch	2026-11-05 13:21:42 +01:00