feat(web): face overlay hover UX and face editor zoom preservation

Change-Id: I7164305d7764bec54fa06b8738cd97fd6a6a6964

refactor(web): use asset metadata for face editor image dimensions instead of DOM

The face editor previously read naturalWidth/naturalHeight from the DOM
element via a $effect + load event listener. This was fragile on slow
hardware (ARM CI) because imgRef changes as AdaptiveImage progresses
through quality levels, and the DOM element's natural dimensions could
be 0 during transitions.

Now the face editor receives imageSize as a prop from the parent, derived
from the asset's metadata dimensions which are always available immediately.

Change-Id: Id4c3a59110feff4c50f429bbd744eac46a6a6964
Change-Id: I7164305d7764bec54fa06b8738cd97fd6a6a6964

e2e/src/ui/generators/timeline.ts

@@ -20,7 +20,7 @@ export {
   toColumnarFormat,
 } from './timeline/rest-response';
 
-export type { Changes } from './timeline/rest-response';
+export type { Changes, FaceData } from './timeline/rest-response';
 
 export { randomImage, randomImageFromString, randomPreview, randomThumbnail } from './timeline/images';
 

e2e/src/ui/generators/timeline/rest-response.ts

@@ -7,8 +7,10 @@ import {
   AssetVisibility,
   UserAvatarColor,
   type AlbumResponseDto,
+  type AssetFaceWithoutPersonResponseDto,
   type AssetResponseDto,
   type ExifResponseDto,
+  type PersonWithFacesResponseDto,
   type TimeBucketAssetResponseDto,
   type TimeBucketsResponseDto,
   type UserResponseDto,
@@ -284,7 +286,16 @@ const createDefaultOwner = (ownerId: string) => {
  * Convert a TimelineAssetConfig to a full AssetResponseDto
  * This matches the response from GET /api/assets/:id
  */
-export function toAssetResponseDto(asset: MockTimelineAsset, owner?: UserResponseDto): AssetResponseDto {
+export type FaceData = {
+  people: PersonWithFacesResponseDto[];
+  unassignedFaces: AssetFaceWithoutPersonResponseDto[];
+};
+
+export function toAssetResponseDto(
+  asset: MockTimelineAsset,
+  owner?: UserResponseDto,
+  faceData?: FaceData,
+): AssetResponseDto {
   const now = new Date().toISOString();
 
   // Default owner if not provided
@@ -338,8 +349,8 @@ export function toAssetResponseDto(asset: MockTimelineAsset, owner?: UserRespons
     exifInfo,
     livePhotoVideoId: asset.livePhotoVideoId,
     tags: [],
-    people: [],
+    people: faceData?.people ?? [],
-    unassignedFaces: [],
+    unassignedFaces: faceData?.unassignedFaces ?? [],
     stack: asset.stack,
     isOffline: false,
     hasMetadata: true,

e2e/src/ui/mock-network/face-editor-network.ts

@@ -1,5 +1,6 @@
+import type { AssetFaceResponseDto, AssetResponseDto, PersonWithFacesResponseDto, SourceType } from '@immich/sdk';
 import { BrowserContext } from '@playwright/test';
-import { randomThumbnail } from 'src/ui/generators/timeline';
+import { type FaceData, randomThumbnail } from 'src/ui/generators/timeline';
 
 // Minimal valid H.264 MP4 (8x8px, 1 frame) that browsers can decode to get videoWidth/videoHeight
 const MINIMAL_MP4_BASE64 =
@@ -125,3 +126,84 @@ export const setupFaceEditorMockApiRoutes = async (
     });
   });
 };
+
+export type MockFaceSpec = {
+  personId: string;
+  personName: string;
+  faceId: string;
+  boundingBoxX1: number;
+  boundingBoxY1: number;
+  boundingBoxX2: number;
+  boundingBoxY2: number;
+};
+
+const toPersonResponseDto = (spec: MockFaceSpec) => ({
+  id: spec.personId,
+  name: spec.personName,
+  birthDate: null,
+  isHidden: false,
+  thumbnailPath: `/upload/thumbs/${spec.personId}.jpeg`,
+  updatedAt: '2025-01-01T00:00:00.000Z',
+});
+
+const toBoundingBox = (spec: MockFaceSpec, imageWidth: number, imageHeight: number) => ({
+  id: spec.faceId,
+  imageWidth,
+  imageHeight,
+  boundingBoxX1: spec.boundingBoxX1,
+  boundingBoxY1: spec.boundingBoxY1,
+  boundingBoxX2: spec.boundingBoxX2,
+  boundingBoxY2: spec.boundingBoxY2,
+});
+
+export const createMockFaceData = (specs: MockFaceSpec[], imageWidth: number, imageHeight: number): FaceData => {
+  const people: PersonWithFacesResponseDto[] = specs.map((spec) => ({
+    ...toPersonResponseDto(spec),
+    faces: [toBoundingBox(spec, imageWidth, imageHeight)],
+  }));
+
+  return { people, unassignedFaces: [] };
+};
+
+export const createMockAssetFaces = (
+  specs: MockFaceSpec[],
+  imageWidth: number,
+  imageHeight: number,
+): AssetFaceResponseDto[] => {
+  return specs.map((spec) => ({
+    ...toBoundingBox(spec, imageWidth, imageHeight),
+    person: toPersonResponseDto(spec),
+    sourceType: 'machine-learning' as SourceType,
+  }));
+};
+
+export const setupGetFacesMockApiRoute = async (context: BrowserContext, faces: AssetFaceResponseDto[]) => {
+  await context.route('**/api/faces?*', async (route, request) => {
+    if (request.method() !== 'GET') {
+      return route.fallback();
+    }
+    return route.fulfill({
+      status: 200,
+      contentType: 'application/json',
+      json: faces,
+    });
+  });
+};
+
+export const setupFaceOverlayMockApiRoutes = async (context: BrowserContext, assetDto: AssetResponseDto) => {
+  await context.route('**/api/assets/*', async (route, request) => {
+    if (request.method() !== 'GET') {
+      return route.fallback();
+    }
+    const url = new URL(request.url());
+    const assetId = url.pathname.split('/').at(-1);
+    if (assetId !== assetDto.id) {
+      return route.fallback();
+    }
+    return route.fulfill({
+      status: 200,
+      contentType: 'application/json',
+      json: assetDto,
+    });
+  });
+};

e2e/src/ui/specs/asset-viewer/face-editor.e2e-spec.ts

@@ -10,16 +10,21 @@ import { assetViewerUtils } from '../timeline/utils';
 import { setupAssetViewerFixture } from './utils';
 
 const waitForSelectorTransition = async (page: Page) => {
-  await page.waitForFunction(
+  await expect(page.locator('#face-editor-data')).toHaveAttribute('data-face-width', /^[1-9]/, { timeout: 10_000 });
-    () => {
+  await page.locator('#face-selector').evaluate(
-      const selector = document.querySelector('#face-selector') as HTMLElement | null;
+    (el) =>
-      if (!selector) {
+      new Promise<void>((resolve) => {
-        return false;
+        requestAnimationFrame(() =>
-      }
+          requestAnimationFrame(() => {
-      return selector.getAnimations({ subtree: false }).every((animation) => animation.playState === 'finished');
+            const animations = el.getAnimations();
-    },
+            if (animations.length === 0) {
-    undefined,
+              resolve();
-    { timeout: 1000, polling: 50 },
+              return;
+            }
+            void Promise.all(animations.map((a) => a.finished)).then(() => resolve());
+          }),
+        );
+      }),
   );
 };
 
@@ -95,7 +100,7 @@ test.describe('face-editor', () => {
     await page.mouse.down();
     await page.mouse.move(centerX + deltaX, centerY + deltaY, { steps: 5 });
     await page.mouse.up();
-    await page.waitForTimeout(300);
+    await waitForSelectorTransition(page);
   };
 
   test('Face editor opens with person list', async ({ page }) => {
@@ -149,7 +154,7 @@ test.describe('face-editor', () => {
     await expect(page.getByRole('dialog')).toBeVisible();
   });
 
-  test('Confirming tag calls createFace API and closes editor', async ({ page }) => {
+  test('Confirming tag calls createFace API with valid coordinates and closes editor', async ({ page }) => {
     const asset = selectRandom(fixture.assets, rng);
     await openFaceEditor(page, asset);
 
@@ -163,8 +168,15 @@ test.describe('face-editor', () => {
     await expect(page.locator('#face-editor')).toBeHidden();
 
     expect(faceCreateCapture.requests).toHaveLength(1);
-    expect(faceCreateCapture.requests[0].assetId).toBe(asset.id);
+    const request = faceCreateCapture.requests[0];
-    expect(faceCreateCapture.requests[0].personId).toBe(personToTag.id);
+    expect(request.assetId).toBe(asset.id);
+    expect(request.personId).toBe(personToTag.id);
+    expect(request.x).toBeGreaterThanOrEqual(0);
+    expect(request.y).toBeGreaterThanOrEqual(0);
+    expect(request.width).toBeGreaterThan(0);
+    expect(request.height).toBeGreaterThan(0);
+    expect(request.x + request.width).toBeLessThanOrEqual(request.imageWidth);
+    expect(request.y + request.height).toBeLessThanOrEqual(request.imageHeight);
   });
 
   test('Cancel button closes face editor', async ({ page }) => {
@@ -282,4 +294,39 @@ test.describe('face-editor', () => {
     expect(afterDrag.left).toBeGreaterThan(beforeDrag.left + 50);
     expect(afterDrag.top).toBeGreaterThan(beforeDrag.top + 20);
   });
+
+  test('Cancel on confirmation dialog keeps face editor open', async ({ page }) => {
+    const asset = selectRandom(fixture.assets, rng);
+    await openFaceEditor(page, asset);
+
+    const personToTag = mockPeople[0];
+    await page.locator('#face-selector').getByText(personToTag.name).click();
+
+    await expect(page.getByRole('dialog')).toBeVisible();
+    await page
+      .getByRole('dialog')
+      .getByRole('button', { name: /cancel/i })
+      .click();
+
+    await expect(page.getByRole('dialog')).toBeHidden();
+    await expect(page.locator('#face-selector')).toBeVisible();
+    await expect(page.locator('#face-editor')).toBeVisible();
+    expect(faceCreateCapture.requests).toHaveLength(0);
+  });
+
+  test('Clicking on face rect center does not reposition it', async ({ page }) => {
+    const asset = selectRandom(fixture.assets, rng);
+    await openFaceEditor(page, asset);
+
+    const beforeClick = await getFaceBoxRect(page);
+    const centerX = beforeClick.left + beforeClick.width / 2;
+    const centerY = beforeClick.top + beforeClick.height / 2;
+
+    await page.mouse.click(centerX, centerY);
+    await waitForSelectorTransition(page);
+
+    const afterClick = await getFaceBoxRect(page);
+    expect(Math.abs(afterClick.left - beforeClick.left)).toBeLessThan(3);
+    expect(Math.abs(afterClick.top - beforeClick.top)).toBeLessThan(3);
+  });
 });

e2e/src/ui/specs/asset-viewer/face-overlay.e2e-spec.ts

@@ -0,0 +1,340 @@
+import { expect, test } from '@playwright/test';
+import { toAssetResponseDto } from 'src/ui/generators/timeline';
+import {
+  createMockAssetFaces,
+  createMockFaceData,
+  createMockPeople,
+  type MockFaceSpec,
+  setupFaceEditorMockApiRoutes,
+  setupFaceOverlayMockApiRoutes,
+  setupGetFacesMockApiRoute,
+} from 'src/ui/mock-network/face-editor-network';
+import { assetViewerUtils } from '../timeline/utils';
+import { ensureDetailPanelVisible, setupAssetViewerFixture } from './utils';
+
+test.describe.configure({ mode: 'parallel' });
+
+const FACE_SPECS: MockFaceSpec[] = [
+  {
+    personId: 'person-alice',
+    personName: 'Alice Johnson',
+    faceId: 'face-alice',
+    boundingBoxX1: 1000,
+    boundingBoxY1: 500,
+    boundingBoxX2: 1500,
+    boundingBoxY2: 1200,
+  },
+  {
+    personId: 'person-bob',
+    personName: 'Bob Smith',
+    faceId: 'face-bob',
+    boundingBoxX1: 2000,
+    boundingBoxY1: 800,
+    boundingBoxX2: 2400,
+    boundingBoxY2: 1600,
+  },
+];
+
+const setupFaceMocks = async (
+  context: import('@playwright/test').BrowserContext,
+  fixture: ReturnType<typeof setupAssetViewerFixture>,
+) => {
+  const mockPeople = createMockPeople(4);
+  const faceData = createMockFaceData(
+    FACE_SPECS,
+    fixture.primaryAssetDto.width ?? 3000,
+    fixture.primaryAssetDto.height ?? 4000,
+  );
+  const assetDtoWithFaces = toAssetResponseDto(fixture.primaryAsset, undefined, faceData);
+  await setupFaceOverlayMockApiRoutes(context, assetDtoWithFaces);
+  await setupFaceEditorMockApiRoutes(context, mockPeople, { requests: [] });
+};
+
+test.describe('face overlay bounding boxes', () => {
+  const fixture = setupAssetViewerFixture(901);
+
+  test.beforeEach(async ({ context }) => {
+    await setupFaceMocks(context, fixture);
+  });
+
+  test('face overlay divs render with correct aria labels', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    const aliceOverlay = page.getByLabel('Person: Alice Johnson');
+    const bobOverlay = page.getByLabel('Person: Bob Smith');
+
+    await expect(aliceOverlay).toBeVisible();
+    await expect(bobOverlay).toBeVisible();
+  });
+
+  test('face overlay shows border on hover', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    const aliceOverlay = page.getByLabel('Person: Alice Johnson');
+    await expect(aliceOverlay).toBeVisible();
+
+    const activeBorder = page.locator('[data-viewer-content] .border-solid.border-white.border-3');
+    await expect(activeBorder).toHaveCount(0);
+
+    await aliceOverlay.hover();
+    await expect(activeBorder).toHaveCount(1);
+  });
+
+  test('face name tooltip appears on hover', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    const aliceOverlay = page.getByLabel('Person: Alice Johnson');
+    await expect(aliceOverlay).toBeVisible();
+
+    await aliceOverlay.hover();
+
+    const nameTooltip = page.locator('[data-viewer-content]').getByText('Alice Johnson');
+    await expect(nameTooltip).toBeVisible();
+  });
+
+  test('face overlays hidden in face edit mode', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    const aliceOverlay = page.getByLabel('Person: Alice Johnson');
+    await expect(aliceOverlay).toBeVisible();
+
+    await ensureDetailPanelVisible(page);
+    await page.getByLabel('Tag people').click();
+    await page.locator('#face-selector').waitFor({ state: 'visible' });
+
+    await expect(aliceOverlay).toBeHidden();
+  });
+
+  test('face overlay hover works after exiting face edit mode', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    const aliceOverlay = page.getByLabel('Person: Alice Johnson');
+    await expect(aliceOverlay).toBeVisible();
+
+    await ensureDetailPanelVisible(page);
+    await page.getByLabel('Tag people').click();
+    await page.locator('#face-selector').waitFor({ state: 'visible' });
+    await expect(aliceOverlay).toBeHidden();
+
+    await page.getByRole('button', { name: /cancel/i }).click();
+    await expect(page.locator('#face-selector')).toBeHidden();
+
+    await expect(aliceOverlay).toBeVisible();
+
+    const activeBorder = page.locator('[data-viewer-content] .border-solid.border-white.border-3');
+    await expect(activeBorder).toHaveCount(0);
+    await aliceOverlay.hover();
+    await expect(activeBorder).toHaveCount(1);
+  });
+});
+
+test.describe('zoom and face editor interaction', () => {
+  const fixture = setupAssetViewerFixture(902);
+
+  test.beforeEach(async ({ context }) => {
+    await setupFaceMocks(context, fixture);
+  });
+
+  test('zoom is preserved when entering face edit mode', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    const { width, height } = page.viewportSize()!;
+    await page.mouse.move(width / 2, height / 2);
+    await page.mouse.wheel(0, -1);
+
+    const imgLocator = page.locator('[data-viewer-content] img[data-testid="preview"]');
+    await expect(async () => {
+      const transform = await imgLocator.evaluate((element) => {
+        return getComputedStyle(element.closest('[style*="transform"]') ?? element).transform;
+      });
+      expect(transform).not.toBe('none');
+      expect(transform).not.toBe('');
+    }).toPass({ timeout: 2000 });
+
+    await ensureDetailPanelVisible(page);
+    await page.getByLabel('Tag people').click();
+    await page.locator('#face-selector').waitFor({ state: 'visible' });
+
+    await expect(page.locator('#face-editor')).toBeVisible();
+
+    const afterTransform = await imgLocator.evaluate((element) => {
+      return getComputedStyle(element.closest('[style*="transform"]') ?? element).transform;
+    });
+    expect(afterTransform).not.toBe('none');
+  });
+
+  test('modifier+drag pans zoomed image without repositioning face rect', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    const { width, height } = page.viewportSize()!;
+    await page.mouse.move(width / 2, height / 2);
+    for (let i = 0; i < 10; i++) {
+      await page.mouse.wheel(0, -3);
+    }
+
+    const imgLocator = page.locator('[data-viewer-content] img[data-testid="preview"]');
+    await expect(async () => {
+      const transform = await imgLocator.evaluate((element) => {
+        return getComputedStyle(element.closest('[style*="transform"]') ?? element).transform;
+      });
+      expect(transform).not.toBe('none');
+    }).toPass({ timeout: 2000 });
+
+    await ensureDetailPanelVisible(page);
+    await page.getByLabel('Tag people').click();
+    await page.locator('#face-selector').waitFor({ state: 'visible' });
+
+    const dataEl = page.locator('#face-editor-data');
+    await expect(dataEl).toHaveAttribute('data-face-width', /^[1-9]/);
+    const beforeLeft = Number(await dataEl.getAttribute('data-face-left'));
+    const beforeTop = Number(await dataEl.getAttribute('data-face-top'));
+    const transformBefore = await imgLocator.evaluate((element) => {
+      return getComputedStyle(element.closest('[style*="transform"]') ?? element).transform;
+    });
+
+    const panModifier = await page.evaluate(() =>
+      /Mac|iPhone|iPad|iPod/.test(navigator.userAgent) ? 'Meta' : 'Control',
+    );
+    await page.keyboard.down(panModifier);
+
+    // Verify face editor becomes transparent to pointer events
+    await expect(async () => {
+      const pe = await dataEl.evaluate((el) => getComputedStyle(el).pointerEvents);
+      expect(pe).toBe('none');
+    }).toPass({ timeout: 2000 });
+
+    await page.mouse.move(width / 2, height / 2);
+    await page.mouse.down();
+    await page.mouse.move(width / 2 + 100, height / 2 + 50, { steps: 5 });
+    await page.mouse.up();
+    await page.keyboard.up(panModifier);
+
+    const transformAfter = await imgLocator.evaluate((element) => {
+      return getComputedStyle(element.closest('[style*="transform"]') ?? element).transform;
+    });
+    expect(transformAfter).not.toBe(transformBefore);
+
+    // Extract translate values from matrix(a, b, c, d, tx, ty)
+    const parseTranslate = (matrix: string) => {
+      const values =
+        matrix
+          .match(/matrix\((.+)\)/)?.[1]
+          .split(',')
+          .map(Number) ?? [];
+      return { tx: values[4], ty: values[5] };
+    };
+    const panBefore = parseTranslate(transformBefore);
+    const panAfter = parseTranslate(transformAfter);
+    const panDeltaX = panAfter.tx - panBefore.tx;
+    const panDeltaY = panAfter.ty - panBefore.ty;
+
+    // Face rect screen position should have moved by the same amount as the pan
+    // (it follows the image), NOT been repositioned by a click
+    const afterLeft = Number(await dataEl.getAttribute('data-face-left'));
+    const afterTop = Number(await dataEl.getAttribute('data-face-top'));
+    const faceDeltaX = afterLeft - beforeLeft;
+    const faceDeltaY = afterTop - beforeTop;
+    expect(Math.abs(faceDeltaX - panDeltaX)).toBeLessThan(3);
+    expect(Math.abs(faceDeltaY - panDeltaY)).toBeLessThan(3);
+  });
+});
+
+test.describe('face overlay via detail panel interaction', () => {
+  const fixture = setupAssetViewerFixture(903);
+
+  test.beforeEach(async ({ context }) => {
+    await setupFaceMocks(context, fixture);
+  });
+
+  test('hovering person in detail panel shows face overlay border', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    await ensureDetailPanelVisible(page);
+
+    const personLink = page.locator('#detail-panel a').filter({ hasText: 'Alice Johnson' });
+    await expect(personLink).toBeVisible();
+
+    const activeBorder = page.locator('[data-viewer-content] .border-solid.border-white.border-3');
+    await expect(activeBorder).toHaveCount(0);
+
+    await personLink.hover();
+    await expect(activeBorder).toHaveCount(1);
+  });
+
+  test('touch pointer on person in detail panel shows face overlay border', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    await ensureDetailPanelVisible(page);
+
+    const personLink = page.locator('#detail-panel a').filter({ hasText: 'Alice Johnson' });
+    await expect(personLink).toBeVisible();
+
+    const activeBorder = page.locator('[data-viewer-content] .border-solid.border-white.border-3');
+    await expect(activeBorder).toHaveCount(0);
+
+    // Simulate a touch-type pointerover (the fix changed from onmouseover to onpointerover,
+    // which fires for touch pointers unlike mouseover)
+    await personLink.dispatchEvent('pointerover', { pointerType: 'touch' });
+    await expect(activeBorder).toHaveCount(1);
+  });
+
+  test('hovering person in detail panel works after exiting face edit mode', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    await ensureDetailPanelVisible(page);
+    await page.getByLabel('Tag people').click();
+    await page.locator('#face-selector').waitFor({ state: 'visible' });
+
+    await page.getByRole('button', { name: /cancel/i }).click();
+    await expect(page.locator('#face-selector')).toBeHidden();
+
+    const personLink = page.locator('#detail-panel a').filter({ hasText: 'Alice Johnson' });
+    await expect(personLink).toBeVisible();
+
+    const activeBorder = page.locator('[data-viewer-content] .border-solid.border-white.border-3');
+    await personLink.hover();
+    await expect(activeBorder).toHaveCount(1);
+  });
+});
+
+test.describe('face overlay via edit faces side panel', () => {
+  const fixture = setupAssetViewerFixture(904);
+
+  test.beforeEach(async ({ context }) => {
+    await setupFaceMocks(context, fixture);
+
+    const assetFaces = createMockAssetFaces(
+      FACE_SPECS,
+      fixture.primaryAssetDto.width ?? 3000,
+      fixture.primaryAssetDto.height ?? 4000,
+    );
+    await setupGetFacesMockApiRoute(context, assetFaces);
+  });
+
+  test('hovering person in edit faces panel shows face overlay border', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    await ensureDetailPanelVisible(page);
+    await page.getByLabel('Edit people').click();
+
+    const faceThumbnail = page.getByTestId('face-thumbnail').first();
+    await expect(faceThumbnail).toBeVisible();
+
+    const activeBorder = page.locator('[data-viewer-content] .border-solid.border-white.border-3');
+    await expect(activeBorder).toHaveCount(0);
+
+    await faceThumbnail.hover();
+    await expect(activeBorder).toHaveCount(1);
+  });
+});

i18n/en.json

@@ -1275,6 +1275,7 @@
   "hide_schema": "Hide schema",
   "hide_text_recognition": "Hide text recognition",
   "hide_unnamed_people": "Hide unnamed people",
+  "hold_key_to_pan": "Hold {key} to pan",
   "home_page_add_to_album_conflicts": "Added {added} assets to album {album}. {failed} assets are already in the album.",
   "home_page_add_to_album_err_local": "Can not add local assets to albums yet, skipping",
   "home_page_add_to_album_success": "Added {added} assets to album {album}.",

test-results/.last-run.json

@@ -0,0 +1,4 @@
+{
+  "status": "failed",
+  "failedTests": []
+}

web/src/lib/components/AdaptiveImage.svelte

@@ -23,6 +23,8 @@
     onError?: () => void;
     ref?: HTMLDivElement;
     imgRef?: HTMLImageElement;
+    imgNaturalSize?: Size;
+    imgScaledSize?: Size;
     backdrop?: Snippet;
     overlays?: Snippet;
   };
@@ -31,6 +33,10 @@
     ref = $bindable(),
     // eslint-disable-next-line no-useless-assignment
     imgRef = $bindable(),
+    // eslint-disable-next-line no-useless-assignment
+    imgNaturalSize = $bindable(),
+    // eslint-disable-next-line no-useless-assignment
+    imgScaledSize = $bindable(),
     asset,
     sharedLink,
     objectFit = 'contain',
@@ -98,9 +104,21 @@
     return { width: 1, height: 1 };
   });
 
-  const { width, height, left, top } = $derived.by(() => {
+  $effect(() => {
+    imgNaturalSize = imageDimensions;
+  });
+
+  const scaledDimensions = $derived.by(() => {
     const scaleFn = objectFit === 'cover' ? scaleToCover : scaleToFit;
-    const { width, height } = scaleFn(imageDimensions, container);
+    return scaleFn(imageDimensions, container);
+  });
+
+  $effect(() => {
+    imgScaledSize = scaledDimensions;
+  });
+
+  const { width, height, left, top } = $derived.by(() => {
+    const { width, height } = scaledDimensions;
     return {
       width: width + 'px',
       height: height + 'px',

web/src/lib/components/asset-viewer/asset-viewer.svelte

@@ -59,7 +59,7 @@
     previousAsset?: AssetResponseDto;
   };
 
-  interface Props {
+  type Props = {
     cursor: AssetCursor;
     showNavigation?: boolean;
     withStacked?: boolean;
@@ -72,7 +72,7 @@
     onUndoDelete?: OnUndoDelete;
     onClose?: (asset: AssetResponseDto) => void;
     onRandom?: () => Promise<{ id: string } | undefined>;
-  }
+  };
 
   let {
     cursor,
@@ -176,6 +176,7 @@
   onDestroy(() => {
     activityManager.reset();
     assetViewerManager.closeEditor();
+    isFaceEditMode.value = false;
     syncAssetViewerOpenClass(false);
     preloadManager.destroy();
   });
@@ -290,6 +291,9 @@
 
   const handleStackedAssetMouseEvent = (isMouseOver: boolean, stackedAsset: AssetResponseDto) => {
     previewStackedAsset = isMouseOver ? stackedAsset : undefined;
+    if (isMouseOver) {
+      isFaceEditMode.value = false;
+    }
   };
 
   const handlePreAction = (action: Action) => {
@@ -358,15 +362,18 @@
     }
   };
 
+  const refreshOcr = async () => {
+    ocrManager.clear();
+    if (sharedLink) {
+      return;
+    }
+
+    await ocrManager.getAssetOcr(asset.id);
+  };
+
   const refresh = async () => {
     await refreshStack();
-    ocrManager.clear();
+    await refreshOcr();
-    if (!sharedLink) {
-      if (previewStackedAsset) {
-        await ocrManager.getAssetOcr(previewStackedAsset.id);
-      }
-      await ocrManager.getAssetOcr(asset.id);
-    }
   };
 
   $effect(() => {
@@ -375,6 +382,12 @@
     untrack(() => handlePromiseError(refresh()));
   });
 
+  $effect(() => {
+    // eslint-disable-next-line @typescript-eslint/no-unused-expressions
+    previewStackedAsset;
+    untrack(() => ocrManager.clear());
+  });
+
   let lastCursor = $state<AssetCursor>();
 
   $effect(() => {
@@ -460,7 +473,7 @@
 
 <section
   id="immich-asset-viewer"
-  class="fixed start-0 top-0 grid size-full grid-cols-4 grid-rows-[64px_1fr] overflow-hidden bg-black"
+  class="fixed start-0 top-0 grid size-full grid-cols-4 grid-rows-[64px_1fr] overflow-hidden bg-black touch-none"
   use:focusTrap
   bind:this={assetViewerHtmlElement}
 >
@@ -612,6 +625,7 @@
               onClick={() => {
                 cursor.current = stackedAsset;
                 previewStackedAsset = undefined;
+                isFaceEditMode.value = false;
               }}
               onMouseEvent={({ isMouseOver }) => handleStackedAssetMouseEvent(isMouseOver, stackedAsset)}
               readonly

web/src/lib/components/asset-viewer/detail-panel.svelte

@@ -7,10 +7,11 @@
   import { timeToLoadTheMap } from '$lib/constants';
   import { assetViewerManager } from '$lib/managers/asset-viewer-manager.svelte';
   import { authManager } from '$lib/managers/auth-manager.svelte';
+  import { eventManager } from '$lib/managers/event-manager.svelte';
   import { featureFlagsManager } from '$lib/managers/feature-flags-manager.svelte';
   import AssetChangeDateModal from '$lib/modals/AssetChangeDateModal.svelte';
   import { Route } from '$lib/route';
-  import { isFaceEditMode } from '$lib/stores/face-edit.svelte';
+  import { isEditFacesPanelOpen, isFaceEditMode } from '$lib/stores/face-edit.svelte';
   import { boundingBoxesArray } from '$lib/stores/people.store';
   import { locale } from '$lib/stores/preferences.store';
   import { preferences, user } from '$lib/stores/user.store';
@@ -49,15 +50,15 @@
   import UserAvatar from '../shared-components/user-avatar.svelte';
   import AlbumListItemDetails from './album-list-item-details.svelte';
 
-  interface Props {
+  type Props = {
     asset: AssetResponseDto;
     currentAlbum?: AlbumResponseDto | null;
-  }
+  };
 
   let { asset, currentAlbum = null }: Props = $props();
 
   let showAssetPath = $state(false);
-  let showEditFaces = $state(false);
+  let showEditFaces = $derived(isEditFacesPanelOpen.value);
   let isOwner = $derived($user?.id === asset.ownerId);
   let people = $derived(asset.people || []);
   let unassignedFaces = $derived(asset.unassignedFaces || []);
@@ -106,7 +107,7 @@
       return;
     }
 
-    showEditFaces = false;
+    isEditFacesPanelOpen.value = false;
     previousId = asset.id;
   });
 
@@ -122,7 +123,8 @@
 
   const handleRefreshPeople = async () => {
     asset = await getAssetInfo({ id: asset.id });
-    showEditFaces = false;
+    eventManager.emit('AssetUpdate', asset);
+    isEditFacesPanelOpen.value = false;
   };
 
   const getAssetFolderHref = (asset: AssetResponseDto) => {
@@ -219,7 +221,7 @@
               shape="round"
               color="secondary"
               variant="ghost"
-              onclick={() => (showEditFaces = true)}
+              onclick={() => (isEditFacesPanelOpen.value = true)}
             />
           {/if}
         </div>
@@ -228,13 +230,14 @@
       <div class="mt-2 flex flex-wrap gap-2">
         {#each people as person, index (person.id)}
           {#if showingHiddenPeople || !person.isHidden}
+            {@const isHighlighted = people[index].faces.some((f) => $boundingBoxesArray.some((b) => b.id === f.id))}
             <a
-              class="w-22"
+              class="group w-22 outline-none"
               href={Route.viewPerson(person, { previousRoute })}
               onfocus={() => ($boundingBoxesArray = people[index].faces)}
               onblur={() => ($boundingBoxesArray = [])}
-              onmouseover={() => ($boundingBoxesArray = people[index].faces)}
+              onpointerover={() => ($boundingBoxesArray = people[index].faces)}
-              onmouseleave={() => ($boundingBoxesArray = [])}
+              onpointerleave={() => ($boundingBoxesArray = [])}
             >
               <div class="relative">
                 <ImageThumbnail
@@ -246,6 +249,8 @@
                   widthStyle="90px"
                   heightStyle="90px"
                   hidden={person.isHidden}
+                  highlighted={isHighlighted}
+                  class="group-focus-visible:outline-2 group-focus-visible:outline-offset-2 group-focus-visible:outline-immich-primary dark:group-focus-visible:outline-immich-dark-primary"
                 />
               </div>
               <p class="mt-1 truncate font-medium" title={person.name}>{person.name}</p>
@@ -574,7 +579,7 @@
   <PersonSidePanel
     assetId={asset.id}
     assetType={asset.type}
-    onClose={() => (showEditFaces = false)}
+    onClose={() => (isEditFacesPanelOpen.value = false)}
     onRefresh={handleRefreshPeople}
   />
 {/if}

web/src/lib/components/asset-viewer/face-editor/face-editor.svelte

@@ -1,10 +1,12 @@
 <script lang="ts">
   import ImageThumbnail from '$lib/components/assets/thumbnail/image-thumbnail.svelte';
+  import { assetViewerManager } from '$lib/managers/asset-viewer-manager.svelte';
   import { assetViewingStore } from '$lib/stores/asset-viewing.store';
   import { isFaceEditMode } from '$lib/stores/face-edit.svelte';
   import { getPeopleThumbnailUrl } from '$lib/utils';
-  import { getNaturalSize, scaleToFit } from '$lib/utils/container-utils';
+  import { computeContentMetrics, mapContentRectToNatural, type Size } from '$lib/utils/container-utils';
   import { handleError } from '$lib/utils/handle-error';
+  import { scaleFaceRectOnResize, type ResizeContext } from '$lib/utils/people-utils';
   import { createFace, getAllPeople, type PersonResponseDto } from '@immich/sdk';
   import { shortcut } from '$lib/actions/shortcut';
   import { Button, Input, modalManager, toastManager } from '@immich/ui';
@@ -12,17 +14,19 @@
   import { clamp } from 'lodash-es';
   import { onMount } from 'svelte';
   import { t } from 'svelte-i18n';
+  import { fade } from 'svelte/transition';
 
   type Props = {
-    htmlElement: HTMLImageElement | HTMLVideoElement;
+    imageSize: Size;
     containerWidth: number;
     containerHeight: number;
     assetId: string;
   };
 
-  let { htmlElement, containerWidth, containerHeight, assetId }: Props = $props();
+  let { imageSize, containerWidth, containerHeight, assetId }: Props = $props();
 
   let canvasEl: HTMLCanvasElement | undefined = $state();
+  let containerEl: HTMLDivElement | undefined = $state();
   let canvas: Canvas | undefined = $state();
   let faceRect: Rect | undefined = $state();
   let faceSelectorEl: HTMLDivElement | undefined = $state();
@@ -32,6 +36,9 @@
 
   let searchTerm = $state('');
   let faceBoxPosition = $state({ left: 0, top: 0, width: 0, height: 0 });
+  let userMovedRect = false;
+  let previousMetrics: ResizeContext | null = null;
+  let panModifierHeld = $state(false);
 
   let filteredCandidates = $derived(
     searchTerm
@@ -53,11 +60,12 @@
   };
 
   const setupCanvas = () => {
-    if (!canvasEl || !htmlElement) {
+    if (!canvasEl) {
       return;
     }
 
-    canvas = new Canvas(canvasEl);
+    canvas = new Canvas(canvasEl, { width: containerWidth, height: containerHeight });
+    canvas.selection = false;
     configureControlStyle();
 
     // eslint-disable-next-line tscompat/tscompat
@@ -75,66 +83,103 @@
 
     canvas.add(faceRect);
     canvas.setActiveObject(faceRect);
-    setDefaultFaceRectanglePosition(faceRect);
   };
 
-  onMount(async () => {
+  onMount(() => {
-    setupCanvas();
+    void getPeople();
-    await getPeople();
   });
 
-  const imageContentMetrics = $derived.by(() => {
-    const natural = getNaturalSize(htmlElement);
-    const container = { width: containerWidth, height: containerHeight };
-    const { width: contentWidth, height: contentHeight } = scaleToFit(natural, container);
-    return {
-      contentWidth,
-      contentHeight,
-      offsetX: (containerWidth - contentWidth) / 2,
-      offsetY: (containerHeight - contentHeight) / 2,
-    };
-  });
-
-  const setDefaultFaceRectanglePosition = (faceRect: Rect) => {
-    const { offsetX, offsetY } = imageContentMetrics;
-
-    faceRect.set({
-      top: offsetY + 200,
-      left: offsetX + 200,
-    });
-
-    faceRect.setCoords();
-    positionFaceSelector();
-  };
-
   $effect(() => {
     if (!canvas) {
       return;
     }
 
-    canvas.setDimensions({
+    const upperCanvas = canvas.upperCanvasEl;
-      width: containerWidth,
+    const controller = new AbortController();
-      height: containerHeight,
+    const { signal } = controller;
-    });
 
-    if (!faceRect) {
+    const stopIfOnTarget = (event: PointerEvent) => {
+      if (canvas?.findTarget(event).target) {
+        event.stopPropagation();
+      }
+    };
+
+    const handlePointerDown = (event: PointerEvent) => {
+      if (!canvas) {
+        return;
+      }
+      if (canvas.findTarget(event).target) {
+        event.stopPropagation();
+        return;
+      }
+      if (faceRect) {
+        event.stopPropagation();
+        const pointer = canvas.getScenePoint(event);
+        faceRect.set({ left: pointer.x, top: pointer.y });
+        faceRect.setCoords();
+        userMovedRect = true;
+        canvas.renderAll();
+        positionFaceSelector();
+      }
+    };
+
+    upperCanvas.addEventListener('pointerdown', handlePointerDown, { signal });
+    upperCanvas.addEventListener('pointermove', stopIfOnTarget, { signal });
+    upperCanvas.addEventListener('pointerup', stopIfOnTarget, { signal });
+
+    return () => {
+      controller.abort();
+    };
+  });
+
+  const imageContentMetrics = $derived.by(() => {
+    if (imageSize.width === 0 || imageSize.height === 0) {
+      return { contentWidth: 0, contentHeight: 0, offsetX: 0, offsetY: 0 };
+    }
+    return computeContentMetrics(imageSize, { width: containerWidth, height: containerHeight });
+  });
+
+  const setDefaultFaceRectanglePosition = (faceRect: Rect) => {
+    const { offsetX, offsetY, contentWidth, contentHeight } = imageContentMetrics;
+
+    faceRect.set({
+      top: offsetY + contentHeight / 2 - 56,
+      left: offsetX + contentWidth / 2 - 56,
+    });
+  };
+
+  $effect(() => {
+    const { offsetX, offsetY, contentWidth } = imageContentMetrics;
+
+    if (contentWidth === 0) {
       return;
     }
 
-    if (!isFaceRectIntersectingCanvas(faceRect, canvas)) {
+    const isFirstRun = previousMetrics === null;
+
+    if (isFirstRun && !canvas) {
+      setupCanvas();
+    }
+
+    if (!canvas || !faceRect) {
+      return;
+    }
+
+    if (!isFirstRun) {
+      canvas.setDimensions({ width: containerWidth, height: containerHeight });
+    }
+
+    if (!isFirstRun && userMovedRect && previousMetrics) {
+      faceRect.set(scaleFaceRectOnResize(faceRect, previousMetrics, { contentWidth, offsetX, offsetY }));
+    } else {
       setDefaultFaceRectanglePosition(faceRect);
     }
-  });
 
-  const isFaceRectIntersectingCanvas = (faceRect: Rect, canvas: Canvas) => {
+    faceRect.setCoords();
-    const faceBox = faceRect.getBoundingRect();
+    previousMetrics = { contentWidth, offsetX, offsetY };
-    return !(
+    canvas.renderAll();
-      0 > faceBox.left + faceBox.width ||
+    positionFaceSelector();
-      0 > faceBox.top + faceBox.height ||
+  });
-      canvas.width < faceBox.left ||
-      canvas.height < faceBox.top
-    );
-  };
 
   const cancel = () => {
     isFaceEditMode.value = false;
@@ -164,11 +209,15 @@
     const gap = 15;
     const padding = faceRect.padding ?? 0;
     const rawBox = faceRect.getBoundingRect();
+    if (Number.isNaN(rawBox.left) || Number.isNaN(rawBox.width)) {
+      return;
+    }
+    const { currentZoom, currentPositionX, currentPositionY } = assetViewerManager.zoomState;
     const faceBox = {
-      left: rawBox.left - padding,
+      left: (rawBox.left - padding) * currentZoom + currentPositionX,
-      top: rawBox.top - padding,
+      top: (rawBox.top - padding) * currentZoom + currentPositionY,
-      width: rawBox.width + padding * 2,
+      width: (rawBox.width + padding * 2) * currentZoom,
-      height: rawBox.height + padding * 2,
+      height: (rawBox.height + padding * 2) * currentZoom,
     };
     const selectorWidth = faceSelectorEl.offsetWidth;
     const chromeHeight = faceSelectorEl.offsetHeight - scrollableListEl.offsetHeight;
@@ -178,20 +227,21 @@
     const clampTop = (top: number) => clamp(top, gap, containerHeight - selectorHeight - gap);
     const clampLeft = (left: number) => clamp(left, gap, containerWidth - selectorWidth - gap);
 
-    const overlapArea = (position: { top: number; left: number }) => {
+    const faceRight = faceBox.left + faceBox.width;
-      const selectorRight = position.left + selectorWidth;
+    const faceBottom = faceBox.top + faceBox.height;
-      const selectorBottom = position.top + selectorHeight;
-      const faceRight = faceBox.left + faceBox.width;
-      const faceBottom = faceBox.top + faceBox.height;
 
-      const overlapX = Math.max(0, Math.min(selectorRight, faceRight) - Math.max(position.left, faceBox.left));
+    const overlapArea = (position: { top: number; left: number }) => {
-      const overlapY = Math.max(0, Math.min(selectorBottom, faceBottom) - Math.max(position.top, faceBox.top));
+      const overlapX = Math.max(
+        0,
+        Math.min(position.left + selectorWidth, faceRight) - Math.max(position.left, faceBox.left),
+      );
+      const overlapY = Math.max(
+        0,
+        Math.min(position.top + selectorHeight, faceBottom) - Math.max(position.top, faceBox.top),
+      );
       return overlapX * overlapY;
     };
 
-    const faceBottom = faceBox.top + faceBox.height;
-    const faceRight = faceBox.left + faceBox.width;
-
     const positions = [
       { top: clampTop(faceBottom + gap), left: clampLeft(faceBox.left) },
       { top: clampTop(faceBox.top - selectorHeight - gap), left: clampLeft(faceBox.left) },
@@ -213,45 +263,139 @@
       }
     }
 
-    faceSelectorEl.style.top = `${bestPosition.top}px`;
+    const containerRect = containerEl?.getBoundingClientRect();
-    faceSelectorEl.style.left = `${bestPosition.left}px`;
+    const offsetTop = containerRect?.top ?? 0;
+    const offsetLeft = containerRect?.left ?? 0;
+    faceSelectorEl.style.top = `${bestPosition.top + offsetTop}px`;
+    faceSelectorEl.style.left = `${bestPosition.left + offsetLeft}px`;
     scrollableListEl.style.height = `${listHeight}px`;
-    faceBoxPosition = { left: faceBox.left, top: faceBox.top, width: faceBox.width, height: faceBox.height };
+    faceBoxPosition = faceBox;
   };
 
+  $effect(() => {
+    if (!canvas) {
+      return;
+    }
+
+    const { currentZoom, currentPositionX, currentPositionY } = assetViewerManager.zoomState;
+    canvas.setViewportTransform([currentZoom, 0, 0, currentZoom, currentPositionX, currentPositionY]);
+    canvas.renderAll();
+    positionFaceSelector();
+  });
+
   $effect(() => {
     const rect = faceRect;
     if (rect) {
-      rect.on('moving', positionFaceSelector);
+      const onUserMove = () => {
-      rect.on('scaling', positionFaceSelector);
+        userMovedRect = true;
+        positionFaceSelector();
+      };
+      rect.on('moving', onUserMove);
+      rect.on('scaling', onUserMove);
       return () => {
-        rect.off('moving', positionFaceSelector);
+        rect.off('moving', onUserMove);
-        rect.off('scaling', positionFaceSelector);
+        rect.off('scaling', onUserMove);
       };
     }
   });
 
+  const isMac = typeof navigator !== 'undefined' && /Mac|iPhone|iPad|iPod/.test(navigator.userAgent);
+  const panModifierKey = isMac ? 'Meta' : 'Control';
+  const panModifierLabel = isMac ? '⌘' : 'Ctrl';
+  const isZoomed = $derived(assetViewerManager.zoom > 1);
+
+  $effect(() => {
+    if (!containerEl) {
+      return;
+    }
+    const element = containerEl;
+    const parent = element.parentElement;
+
+    const activate = () => {
+      panModifierHeld = true;
+      element.style.pointerEvents = 'none';
+      if (parent) {
+        parent.style.cursor = 'move';
+      }
+    };
+
+    const deactivate = () => {
+      panModifierHeld = false;
+      element.style.pointerEvents = '';
+      if (parent) {
+        parent.style.cursor = '';
+      }
+    };
+
+    const onKeyDown = (event: KeyboardEvent) => {
+      if (event.key === panModifierKey) {
+        activate();
+      }
+    };
+    const onKeyUp = (event: KeyboardEvent) => {
+      if (event.key === panModifierKey) {
+        deactivate();
+      }
+    };
+
+    document.addEventListener('keydown', onKeyDown);
+    document.addEventListener('keyup', onKeyUp);
+    window.addEventListener('blur', deactivate);
+
+    return () => {
+      document.removeEventListener('keydown', onKeyDown);
+      document.removeEventListener('keyup', onKeyUp);
+      window.removeEventListener('blur', deactivate);
+      deactivate();
+    };
+  });
+
+  const trapEvents = (node: HTMLElement) => {
+    const stop = (e: Event) => e.stopPropagation();
+    const eventTypes = ['keydown', 'pointerdown', 'pointermove', 'pointerup'] as const;
+    for (const type of eventTypes) {
+      node.addEventListener(type, stop);
+    }
+
+    // Move to body so the selector isn't affected by the zoom transform on the container
+    document.body.append(node);
+
+    return {
+      destroy() {
+        for (const type of eventTypes) {
+          node.removeEventListener(type, stop);
+        }
+        node.remove();
+      },
+    };
+  };
+
   const getFaceCroppedCoordinates = () => {
-    if (!faceRect || !htmlElement) {
+    if (!faceRect || imageSize.width === 0 || imageSize.height === 0) {
       return;
     }
 
-    const { left, top, width, height } = faceRect.getBoundingRect();
+    const scaledWidth = faceRect.getScaledWidth();
-    const { offsetX, offsetY, contentWidth, contentHeight } = imageContentMetrics;
+    const scaledHeight = faceRect.getScaledHeight();
-    const natural = getNaturalSize(htmlElement);
 
-    const scaleX = natural.width / contentWidth;
+    const imageRect = mapContentRectToNatural(
-    const scaleY = natural.height / contentHeight;
+      {
-    const imageX = (left - offsetX) * scaleX;
+        left: faceRect.left - scaledWidth / 2,
-    const imageY = (top - offsetY) * scaleY;
+        top: faceRect.top - scaledHeight / 2,
+        width: scaledWidth,
+        height: scaledHeight,
+      },
+      imageContentMetrics,
+      imageSize,
+    );
 
     return {
-      imageWidth: natural.width,
+      imageWidth: imageSize.width,
-      imageHeight: natural.height,
+      imageHeight: imageSize.height,
-      x: Math.floor(imageX),
+      x: Math.floor(imageRect.left),
-      y: Math.floor(imageY),
+      y: Math.floor(imageRect.top),
-      width: Math.floor(width * scaleX),
+      width: Math.floor(imageRect.width),
-      height: Math.floor(height * scaleY),
+      height: Math.floor(imageRect.height),
     };
   };
 
@@ -282,10 +426,9 @@
       });
 
       await assetViewingStore.setAssetId(assetId);
+      isFaceEditMode.value = false;
     } catch (error) {
       handleError(error, 'Error tagging face');
-    } finally {
-      isFaceEditMode.value = false;
     }
   };
 </script>
@@ -294,8 +437,8 @@
 
 <div
   id="face-editor-data"
+  bind:this={containerEl}
   class="absolute start-0 top-0 z-5 h-full w-full overflow-hidden"
-  data-overlay-interactive
   data-face-left={faceBoxPosition.left}
   data-face-top={faceBoxPosition.top}
   data-face-width={faceBoxPosition.width}
@@ -306,7 +449,9 @@
   <div
     id="face-selector"
     bind:this={faceSelectorEl}
-    class="absolute top-[calc(50%-250px)] start-[calc(50%-125px)] max-w-[250px] w-[250px] bg-white dark:bg-immich-dark-gray dark:text-immich-dark-fg backdrop-blur-sm px-2 py-4 rounded-xl border border-gray-200 dark:border-gray-800 transition-[top,left] duration-200 ease-out"
+    class="fixed z-20 w-[min(200px,45vw)] min-w-48 bg-white dark:bg-immich-dark-gray dark:text-immich-dark-fg backdrop-blur-sm px-2 py-4 rounded-xl border border-gray-200 dark:border-gray-800 transition-[top,left] duration-200 ease-out"
+    use:trapEvents
+    onwheel={(e) => e.stopPropagation()}
   >
     <p class="text-center text-sm">{$t('select_person_to_tag')}</p>
 
@@ -347,4 +492,15 @@
 
     <Button size="small" fullWidth onclick={cancel} color="danger" class="mt-2">{$t('cancel')}</Button>
   </div>
+
+  {#if isZoomed && !panModifierHeld}
+    <div
+      transition:fade={{ duration: 200 }}
+      class="absolute bottom-4 inset-s-1/2 -translate-x-1/2 pointer-events-none z-10"
+    >
+      <p class="bg-black/60 text-white text-xs px-3 py-1.5 rounded-full whitespace-nowrap">
+        {$t('hold_key_to_pan', { values: { key: panModifierLabel } })}
+      </p>
+    </div>
+  {/if}
 </div>

web/src/lib/components/asset-viewer/photo-viewer.svelte

@@ -8,13 +8,13 @@
   import AssetViewerEvents from '$lib/components/AssetViewerEvents.svelte';
   import { assetViewerManager } from '$lib/managers/asset-viewer-manager.svelte';
   import { castManager } from '$lib/managers/cast-manager.svelte';
-  import { isFaceEditMode } from '$lib/stores/face-edit.svelte';
+  import { isEditFacesPanelOpen, isFaceEditMode } from '$lib/stores/face-edit.svelte';
   import { ocrManager } from '$lib/stores/ocr.svelte';
   import { boundingBoxesArray, type Faces } from '$lib/stores/people.store';
   import { SlideshowLook, SlideshowState, slideshowStore } from '$lib/stores/slideshow.store';
   import { handlePromiseError } from '$lib/utils';
   import { canCopyImageToClipboard, copyImageToClipboard } from '$lib/utils/asset-utils';
-  import { getNaturalSize, scaleToFit, type Size } from '$lib/utils/container-utils';
+  import type { Size } from '$lib/utils/container-utils';
   import { handleError } from '$lib/utils/handle-error';
   import { getOcrBoundingBoxes } from '$lib/utils/ocr-utils';
   import { getBoundingBox } from '$lib/utils/people-utils';
@@ -67,13 +67,10 @@
     height: containerHeight,
   });
 
-  const overlaySize = $derived.by((): Size => {
+  let imageDimensions = $state<Size>({ width: 0, height: 0 });
-    if (!assetViewerManager.imgRef || !visibleImageReady) {
+  let scaledDimensions = $state<Size>({ width: 0, height: 0 });
-      return { width: 0, height: 0 };
-    }
 
-    return scaleToFit(getNaturalSize(assetViewerManager.imgRef), { width: containerWidth, height: containerHeight });
+  const overlaySize = $derived(visibleImageReady ? scaledDimensions : { width: 0, height: 0 });
-  });
 
   const ocrBoxes = $derived(ocrManager.showOverlay ? getOcrBoundingBoxes(ocrManager.data, overlaySize) : []);
 
@@ -97,12 +94,6 @@
 
   const onPlaySlideshow = () => ($slideshowState = SlideshowState.PlaySlideshow);
 
-  $effect(() => {
-    if (isFaceEditMode.value && assetViewerManager.zoom > 1) {
-      onZoom();
-    }
-  });
-
   // TODO move to action + command palette
   const onCopyShortcut = (event: KeyboardEvent) => {
     if (globalThis.getSelection()?.type === 'Range') {
@@ -147,46 +138,22 @@
 
   const faceToNameMap = $derived.by(() => {
     // eslint-disable-next-line svelte/prefer-svelte-reactivity
-    const map = new Map<Faces, string>();
+    const map = new Map<Faces, string | undefined>();
     for (const person of asset.people ?? []) {
       for (const face of person.faces ?? []) {
         map.set(face, person.name);
       }
     }
+    for (const face of asset.unassignedFaces ?? []) {
+      map.set(face, undefined);
+    }
     return map;
   });
 
+  // Array needed for indexed access in the template (faces[index])
   const faces = $derived(Array.from(faceToNameMap.keys()));
-
+  const boundingBoxes = $derived(getBoundingBox(faces, overlaySize));
-  const handleImageMouseMove = (event: MouseEvent) => {
+  const activeBoundingBoxes = $derived(boundingBoxes.filter((box) => $boundingBoxesArray.some((f) => f.id === box.id)));
-    $boundingBoxesArray = [];
-    if (!assetViewerManager.imgRef || !element || isFaceEditMode.value || ocrManager.showOverlay) {
-      return;
-    }
-
-    const natural = getNaturalSize(assetViewerManager.imgRef);
-    const scaled = scaleToFit(natural, container);
-    const { currentZoom, currentPositionX, currentPositionY } = assetViewerManager.zoomState;
-
-    const contentOffsetX = (container.width - scaled.width) / 2;
-    const contentOffsetY = (container.height - scaled.height) / 2;
-
-    const containerRect = element.getBoundingClientRect();
-    const mouseX = (event.clientX - containerRect.left - contentOffsetX * currentZoom - currentPositionX) / currentZoom;
-    const mouseY = (event.clientY - containerRect.top - contentOffsetY * currentZoom - currentPositionY) / currentZoom;
-
-    const faceBoxes = getBoundingBox(faces, overlaySize);
-
-    for (const [index, box] of faceBoxes.entries()) {
-      if (mouseX >= box.left && mouseX <= box.left + box.width && mouseY >= box.top && mouseY <= box.top + box.height) {
-        $boundingBoxesArray.push(faces[index]);
-      }
-    }
-  };
-
-  const handleImageMouseLeave = () => {
-    $boundingBoxesArray = [];
-  };
 </script>
 
 <AssetViewerEvents {onCopy} {onZoom} />
@@ -207,8 +174,6 @@
   bind:clientHeight={containerHeight}
   role="presentation"
   ondblclick={onZoom}
-  onmousemove={handleImageMouseMove}
-  onmouseleave={handleImageMouseLeave}
   use:zoomImageAction={{ zoomTarget: adaptiveImage }}
   {...useSwipe((event) => onSwipe?.(event))}
 >
@@ -227,6 +192,8 @@
       onReady?.();
     }}
     bind:imgRef={assetViewerManager.imgRef}
+    bind:imgNaturalSize={imageDimensions}
+    bind:imgScaledSize={scaledDimensions}
     bind:ref={adaptiveImage}
   >
     {#snippet backdrop()}
@@ -238,20 +205,40 @@
       {/if}
     {/snippet}
     {#snippet overlays()}
-      {#each getBoundingBox($boundingBoxesArray, overlaySize) as boundingbox, index (boundingbox.id)}
+      {#if !isFaceEditMode.value}
+        {#each boundingBoxes as boundingbox, index (boundingbox.id)}
+          {@const face = faces[index]}
+          {@const name = faceToNameMap.get(face)}
+          {#if name !== undefined || isEditFacesPanelOpen.value}
+            <!-- svelte-ignore a11y_no_static_element_interactions -->
+            <div
+              class="absolute pointer-events-auto outline-none rounded-lg"
+              style="top: {boundingbox.top}px; left: {boundingbox.left}px; height: {boundingbox.height}px; width: {boundingbox.width}px;"
+              aria-label="{$t('person')}: {name ?? $t('unknown')}"
+              onpointerenter={() => ($boundingBoxesArray = [face])}
+              onpointerleave={() => ($boundingBoxesArray = [])}
+            ></div>
+          {/if}
+        {/each}
+      {/if}
+
+      {#each activeBoundingBoxes as boundingbox (boundingbox.id)}
+        {@const face = faces.find((f) => f.id === boundingbox.id)}
+        {@const name = face ? faceToNameMap.get(face) : undefined}
         <div
-          class="absolute border-solid border-white border-3 rounded-lg"
+          class="absolute border-solid border-white border-3 rounded-lg pointer-events-none"
           style="top: {boundingbox.top}px; left: {boundingbox.left}px; height: {boundingbox.height}px; width: {boundingbox.width}px;"
-        ></div>
+        >
-        {#if faceToNameMap.get($boundingBoxesArray[index])}
+          {#if name}
-          <div
+            <div
-            class="absolute bg-white/90 text-black px-2 py-1 rounded text-sm font-medium whitespace-nowrap pointer-events-none shadow-lg"
+              aria-hidden="true"
-            style="top: {boundingbox.top + boundingbox.height + 4}px; left: {boundingbox.left +
+              class="absolute bg-white/90 text-black px-2 py-1 rounded text-sm font-medium whitespace-nowrap shadow-lg"
-              boundingbox.width}px; transform: translateX(-100%);"
+              style="top: {boundingbox.height + 4}px; right: 0;"
-          >
+            >
-            {faceToNameMap.get($boundingBoxesArray[index])}
+              {name}
-          </div>
+            </div>
-        {/if}
+          {/if}
+        </div>
       {/each}
 
       {#each ocrBoxes as ocrBox (ocrBox.id)}
@@ -261,6 +248,6 @@
   </AdaptiveImage>
 
   {#if isFaceEditMode.value && assetViewerManager.imgRef}
-    <FaceEditor htmlElement={assetViewerManager.imgRef} {containerWidth} {containerHeight} assetId={asset.id} />
+    <FaceEditor imageSize={imageDimensions} {containerWidth} {containerHeight} assetId={asset.id} />
   {/if}
 </div>

web/src/lib/components/asset-viewer/video-native-viewer.svelte

@@ -11,14 +11,16 @@
     videoViewerVolume,
   } from '$lib/stores/preferences.store';
   import { getAssetMediaUrl, getAssetPlaybackUrl } from '$lib/utils';
+  import type { Size } from '$lib/utils/container-utils';
   import { AssetMediaSize } from '@immich/sdk';
   import { LoadingSpinner } from '@immich/ui';
   import { onDestroy, onMount } from 'svelte';
   import { useSwipe, type SwipeCustomEvent } from 'svelte-gestures';
   import { fade } from 'svelte/transition';
 
-  interface Props {
+  type Props = {
     assetId: string;
+    imageSize: Size;
     loopVideo: boolean;
     cacheKey: string | null;
     playOriginalVideo: boolean;
@@ -27,10 +29,11 @@
     onVideoEnded?: () => void;
     onVideoStarted?: () => void;
     onClose?: () => void;
-  }
+  };
 
   let {
     assetId,
+    imageSize,
     loopVideo,
     cacheKey,
     playOriginalVideo,
@@ -173,7 +176,7 @@
       {/if}
 
       {#if isFaceEditMode.value}
-        <FaceEditor htmlElement={videoPlayer} {containerWidth} {containerHeight} {assetId} />
+        <FaceEditor {imageSize} {containerWidth} {containerHeight} {assetId} />
       {/if}
     {/if}
   </div>

web/src/lib/components/asset-viewer/video-wrapper-viewer.svelte

@@ -4,7 +4,7 @@
   import { ProjectionType } from '$lib/constants';
   import type { AssetResponseDto } from '@immich/sdk';
 
-  interface Props {
+  type Props = {
     asset: AssetResponseDto;
     assetId?: string;
     projectionType: string | null | undefined;
@@ -16,7 +16,7 @@
     onNextAsset?: () => void;
     onVideoEnded?: () => void;
     onVideoStarted?: () => void;
-  }
+  };
 
   let {
     asset,
@@ -42,6 +42,7 @@
     {loopVideo}
     {cacheKey}
     assetId={effectiveAssetId}
+    imageSize={{ width: asset.width ?? 1, height: asset.height ?? 1 }}
     {playOriginalVideo}
     {onPreviousAsset}
     {onNextAsset}

web/src/lib/components/assets/thumbnail/image-thumbnail.svelte

@@ -16,6 +16,7 @@
     circle?: boolean;
     hidden?: boolean;
     border?: boolean;
+    highlighted?: boolean;
     hiddenIconClass?: string;
     class?: ClassValue;
     brokenAssetClass?: ClassValue;
@@ -34,6 +35,7 @@
     circle = false,
     hidden = false,
     border = false,
+    highlighted = false,
     hiddenIconClass = 'text-white',
     onComplete = undefined,
     class: imageClass = '',
@@ -83,6 +85,10 @@
   />
 {/if}
 
+{#if highlighted}
+  <span class={['absolute inset-0 pointer-events-none border-2 border-white', sharedClasses]} {style}></span>
+{/if}
+
 {#if hidden}
   <div class="absolute start-1/2 top-1/2 translate-x-[-50%] translate-y-[-50%] transform">
     <!-- TODO fix `title` type -->

web/src/lib/components/faces-page/person-side-panel.svelte

@@ -27,12 +27,12 @@
   import ImageThumbnail from '../assets/thumbnail/image-thumbnail.svelte';
   import AssignFaceSidePanel from './assign-face-side-panel.svelte';
 
-  interface Props {
+  type Props = {
     assetId: string;
     assetType: AssetTypeEnum;
     onClose: () => void;
     onRefresh: () => void;
-  }
+  };
 
   let { assetId, assetType, onClose, onRefresh }: Props = $props();
 
@@ -58,6 +58,8 @@
   let automaticRefreshTimeout: ReturnType<typeof setTimeout>;
 
   const thumbnailWidth = '90px';
+  const focusHighlightClass =
+    'group-focus-visible:outline-2 group-focus-visible:outline-offset-2 group-focus-visible:outline-immich-primary dark:group-focus-visible:outline-immich-dark-primary';
 
   async function loadPeople() {
     const timeout = setTimeout(() => (isShowLoadingPeople = true), timeBeforeShowLoadingSpinner);
@@ -226,14 +228,16 @@
       {:else}
         {#each peopleWithFaces as face, index (face.id)}
           {@const personName = face.person ? face.person?.name : $t('face_unassigned')}
+          {@const isHighlighted = $boundingBoxesArray.some((f) => f.id === face.id)}
           <div class="relative h-29 w-24">
             <div
               role="button"
               tabindex={index}
-              class="absolute start-0 top-0 h-22.5 w-22.5 cursor-default"
+              data-testid="face-thumbnail"
+              class="group absolute inset-s-0 top-0 h-22.5 w-22.5 cursor-default outline-none"
               onfocus={() => ($boundingBoxesArray = [peopleWithFaces[index]])}
-              onmouseover={() => ($boundingBoxesArray = [peopleWithFaces[index]])}
+              onpointerover={() => ($boundingBoxesArray = [peopleWithFaces[index]])}
-              onmouseleave={() => ($boundingBoxesArray = [])}
+              onpointerleave={() => ($boundingBoxesArray = [])}
             >
               <div class="relative">
                 {#if selectedPersonToCreate[face.id]}
@@ -245,6 +249,8 @@
                     title={$t('new_person')}
                     widthStyle={thumbnailWidth}
                     heightStyle={thumbnailWidth}
+                    highlighted={isHighlighted}
+                    class={focusHighlightClass}
                   />
                 {:else if selectedPersonToReassign[face.id]}
                   <ImageThumbnail
@@ -259,6 +265,8 @@
                     widthStyle={thumbnailWidth}
                     heightStyle={thumbnailWidth}
                     hidden={selectedPersonToReassign[face.id].isHidden}
+                    highlighted={isHighlighted}
+                    class={focusHighlightClass}
                   />
                 {:else if face.person}
                   <ImageThumbnail
@@ -270,6 +278,8 @@
                     widthStyle={thumbnailWidth}
                     heightStyle={thumbnailWidth}
                     hidden={face.person.isHidden}
+                    highlighted={isHighlighted}
+                    class={focusHighlightClass}
                   />
                 {:else}
                   {#await zoomImageToBase64(face, assetId, assetType, assetViewerManager.imgRef)}
@@ -281,6 +291,8 @@
                       title={$t('face_unassigned')}
                       widthStyle="90px"
                       heightStyle="90px"
+                      highlighted={isHighlighted}
+                      class={focusHighlightClass}
                     />
                   {:then data}
                     <ImageThumbnail
@@ -291,6 +303,8 @@
                       title={$t('face_unassigned')}
                       widthStyle="90px"
                       heightStyle="90px"
+                      highlighted={isHighlighted}
+                      class={focusHighlightClass}
                     />
                   {/await}
                 {/if}

web/src/lib/stores/face-edit.svelte.ts

@@ -1 +1,2 @@
 export const isFaceEditMode = $state({ value: false });
+export const isEditFacesPanelOpen = $state({ value: false });

web/src/lib/utils/container-utils.spec.ts

@@ -1,6 +1,9 @@
 import {
+  computeContentMetrics,
   getContentMetrics,
   getNaturalSize,
+  mapContentRectToNatural,
+  mapContentToNatural,
   mapNormalizedRectToContent,
   mapNormalizedToContent,
   scaleToCover,
@@ -123,6 +126,53 @@ describe('scaleToCover', () => {
   });
 });
 
+describe('computeContentMetrics', () => {
+  it('should compute metrics with scaleToFit by default', () => {
+    expect(computeContentMetrics({ width: 2000, height: 1000 }, { width: 800, height: 600 })).toEqual({
+      contentWidth: 800,
+      contentHeight: 400,
+      offsetX: 0,
+      offsetY: 100,
+    });
+  });
+
+  it('should accept scaleToCover as scale function', () => {
+    expect(computeContentMetrics({ width: 2000, height: 1000 }, { width: 800, height: 600 }, scaleToCover)).toEqual({
+      contentWidth: 1200,
+      contentHeight: 600,
+      offsetX: -200,
+      offsetY: 0,
+    });
+  });
+
+  it('should compute zero offsets when aspect ratios match', () => {
+    expect(computeContentMetrics({ width: 1600, height: 900 }, { width: 800, height: 450 })).toEqual({
+      contentWidth: 800,
+      contentHeight: 450,
+      offsetX: 0,
+      offsetY: 0,
+    });
+  });
+});
+
+// Coordinate space glossary:
+//
+// "Normalized" coordinates: values in the 0–1 range, where (0,0) is the top-left
+// of the image and (1,1) is the bottom-right. Resolution-independent.
+//
+// "Content" coordinates: pixel positions within the container, after the image
+// has been scaled (scaleToFit/scaleToCover) and offset (centered). This is what
+// CSS and DOM layout use for positioning overlays like face boxes and OCR text.
+//
+// "Natural" coordinates: pixel positions in the original image file at its full
+// resolution (e.g. 4000×3000). Used when cropping or drawing on the source image.
+//
+// "Metadata pixel space": the coordinate system used by face detection / OCR
+// models, where positions are in pixels relative to the image dimensions stored
+// in metadata (face.imageWidth/imageHeight). These may differ from the natural
+// dimensions if the image was resized. To convert to normalized, divide by
+// the metadata dimensions (e.g. face.boundingBoxX1 / face.imageWidth).
+
 describe('mapNormalizedToContent', () => {
   const metrics = { contentWidth: 800, contentHeight: 400, offsetX: 0, offsetY: 100 };
 
@@ -152,6 +202,31 @@ describe('mapNormalizedToContent', () => {
   });
 });
 
+describe('mapContentToNatural', () => {
+  const metrics = { contentWidth: 800, contentHeight: 400, offsetX: 0, offsetY: 100 };
+  const natural = { width: 4000, height: 2000 };
+
+  it('should map content origin to natural origin', () => {
+    expect(mapContentToNatural({ x: 0, y: 100 }, metrics, natural)).toEqual({ x: 0, y: 0 });
+  });
+
+  it('should map content bottom-right to natural bottom-right', () => {
+    expect(mapContentToNatural({ x: 800, y: 500 }, metrics, natural)).toEqual({ x: 4000, y: 2000 });
+  });
+
+  it('should map content center to natural center', () => {
+    expect(mapContentToNatural({ x: 400, y: 300 }, metrics, natural)).toEqual({ x: 2000, y: 1000 });
+  });
+
+  it('should be the inverse of mapNormalizedToContent', () => {
+    const normalized = { x: 0.3, y: 0.7 };
+    const contentPoint = mapNormalizedToContent(normalized, metrics);
+    const naturalPoint = mapContentToNatural(contentPoint, metrics, natural);
+    expect(naturalPoint.x).toBeCloseTo(normalized.x * natural.width);
+    expect(naturalPoint.y).toBeCloseTo(normalized.y * natural.height);
+  });
+});
+
 describe('mapNormalizedRectToContent', () => {
   const metrics = { contentWidth: 800, contentHeight: 400, offsetX: 0, offsetY: 100 };
 
@@ -177,3 +252,28 @@ describe('mapNormalizedRectToContent', () => {
     expect(rect).toEqual({ left: 200, top: 100, width: 400, height: 200 });
   });
 });
+
+describe('mapContentRectToNatural', () => {
+  const metrics = { contentWidth: 800, contentHeight: 400, offsetX: 0, offsetY: 100 };
+  const natural = { width: 4000, height: 2000 };
+
+  it('should map a content rect to natural image coordinates', () => {
+    const rect = mapContentRectToNatural({ left: 200, top: 200, width: 400, height: 200 }, metrics, natural);
+    expect(rect).toEqual({ left: 1000, top: 500, width: 2000, height: 1000 });
+  });
+
+  it('should map full content rect to full natural dimensions', () => {
+    const rect = mapContentRectToNatural({ left: 0, top: 100, width: 800, height: 400 }, metrics, natural);
+    expect(rect).toEqual({ left: 0, top: 0, width: 4000, height: 2000 });
+  });
+
+  it('should be the inverse of mapNormalizedRectToContent', () => {
+    const normalized = { topLeft: { x: 0.2, y: 0.3 }, bottomRight: { x: 0.8, y: 0.9 } };
+    const contentRect = mapNormalizedRectToContent(normalized.topLeft, normalized.bottomRight, metrics);
+    const naturalRect = mapContentRectToNatural(contentRect, metrics, natural);
+    expect(naturalRect.left).toBeCloseTo(normalized.topLeft.x * natural.width);
+    expect(naturalRect.top).toBeCloseTo(normalized.topLeft.y * natural.height);
+    expect(naturalRect.width).toBeCloseTo((normalized.bottomRight.x - normalized.topLeft.x) * natural.width);
+    expect(naturalRect.height).toBeCloseTo((normalized.bottomRight.y - normalized.topLeft.y) * natural.height);
+  });
+});

web/src/lib/utils/container-utils.ts

@@ -63,16 +63,24 @@ export const getNaturalSize = (element: HTMLImageElement | HTMLVideoElement): Si
   return { width: element.naturalWidth, height: element.naturalHeight };
 };
 
-export const getContentMetrics = (element: HTMLImageElement | HTMLVideoElement): ContentMetrics => {
+export function computeContentMetrics(
-  const natural = getNaturalSize(element);
+  imageSize: Size,
-  const client = getElementSize(element);
+  containerSize: Size,
-  const { width: contentWidth, height: contentHeight } = scaleToFit(natural, client);
+  scaleFn: (dimensions: Size, container: Size) => Size = scaleToFit,
+) {
+  const { width: contentWidth, height: contentHeight } = scaleFn(imageSize, containerSize);
   return {
     contentWidth,
     contentHeight,
-    offsetX: (client.width - contentWidth) / 2,
+    offsetX: (containerSize.width - contentWidth) / 2,
-    offsetY: (client.height - contentHeight) / 2,
+    offsetY: (containerSize.height - contentHeight) / 2,
   };
+}
+
+export const getContentMetrics = (element: HTMLImageElement | HTMLVideoElement): ContentMetrics => {
+  const natural = getNaturalSize(element);
+  const client = getElementSize(element);
+  return computeContentMetrics(natural, client);
 };
 
 export function mapNormalizedToContent(point: Point, sizeOrMetrics: Size | ContentMetrics): Point {
@@ -88,6 +96,13 @@ export function mapNormalizedToContent(point: Point, sizeOrMetrics: Size | Conte
   };
 }
 
+export function mapContentToNatural(point: Point, metrics: ContentMetrics, naturalSize: Size): Point {
+  return {
+    x: ((point.x - metrics.offsetX) / metrics.contentWidth) * naturalSize.width,
+    y: ((point.y - metrics.offsetY) / metrics.contentHeight) * naturalSize.height,
+  };
+}
+
 export type Rect = {
   top: number;
   left: number;
@@ -109,3 +124,18 @@ export function mapNormalizedRectToContent(
     height: br.y - tl.y,
   };
 }
+
+export function mapContentRectToNatural(rect: Rect, metrics: ContentMetrics, naturalSize: Size): Rect {
+  const topLeft = mapContentToNatural({ x: rect.left, y: rect.top }, metrics, naturalSize);
+  const bottomRight = mapContentToNatural(
+    { x: rect.left + rect.width, y: rect.top + rect.height },
+    metrics,
+    naturalSize,
+  );
+  return {
+    top: topLeft.y,
+    left: topLeft.x,
+    width: bottomRight.x - topLeft.x,
+    height: bottomRight.y - topLeft.y,
+  };
+}

web/src/lib/utils/people-utils.spec.ts

@@ -1,6 +1,6 @@
 import type { Faces } from '$lib/stores/people.store';
 import type { Size } from '$lib/utils/container-utils';
-import { getBoundingBox } from '$lib/utils/people-utils';
+import { getBoundingBox, scaleFaceRectOnResize, type FaceRectState, type ResizeContext } from '$lib/utils/people-utils';
 
 const makeFace = (overrides: Partial<Faces> = {}): Faces => ({
   id: 'face-1',
@@ -68,3 +68,96 @@ describe('getBoundingBox', () => {
     expect(boxes[0].left).toBeLessThan(boxes[1].left);
   });
 });
+
+describe('scaleFaceRectOnResize', () => {
+  const makeRect = (overrides: Partial<FaceRectState> = {}): FaceRectState => ({
+    left: 300,
+    top: 400,
+    scaleX: 1,
+    scaleY: 1,
+    ...overrides,
+  });
+
+  const makePrevious = (overrides: Partial<ResizeContext> = {}): ResizeContext => ({
+    offsetX: 100,
+    offsetY: 50,
+    contentWidth: 800,
+    ...overrides,
+  });
+
+  it('should preserve relative position when container doubles in size', () => {
+    const rect = makeRect({ left: 300, top: 250 });
+    const previous = makePrevious({ offsetX: 100, offsetY: 50, contentWidth: 800 });
+
+    const result = scaleFaceRectOnResize(rect, previous, { offsetX: 200, offsetY: 100, contentWidth: 1600 });
+
+    // imageRelLeft = (300 - 100) * 2 = 400, new left = 200 + 400 = 600
+    // imageRelTop = (250 - 50) * 2 = 400, new top = 100 + 400 = 500
+    expect(result.left).toBe(600);
+    expect(result.top).toBe(500);
+    expect(result.scaleX).toBe(2);
+    expect(result.scaleY).toBe(2);
+  });
+
+  it('should preserve relative position when container halves in size', () => {
+    const rect = makeRect({ left: 300, top: 250 });
+    const previous = makePrevious({ offsetX: 100, offsetY: 50, contentWidth: 800 });
+
+    const result = scaleFaceRectOnResize(rect, previous, { offsetX: 50, offsetY: 25, contentWidth: 400 });
+
+    // imageRelLeft = (300 - 100) * 0.5 = 100, new left = 50 + 100 = 150
+    // imageRelTop = (250 - 50) * 0.5 = 100, new top = 25 + 100 = 125
+    expect(result.left).toBe(150);
+    expect(result.top).toBe(125);
+    expect(result.scaleX).toBe(0.5);
+    expect(result.scaleY).toBe(0.5);
+  });
+
+  it('should handle no change in dimensions', () => {
+    const rect = makeRect({ left: 300, top: 250, scaleX: 1.5, scaleY: 1.5 });
+    const previous = makePrevious({ offsetX: 100, offsetY: 50, contentWidth: 800 });
+
+    const result = scaleFaceRectOnResize(rect, previous, { offsetX: 100, offsetY: 50, contentWidth: 800 });
+
+    expect(result.left).toBe(300);
+    expect(result.top).toBe(250);
+    expect(result.scaleX).toBe(1.5);
+    expect(result.scaleY).toBe(1.5);
+  });
+
+  it('should handle offset changes without content width change', () => {
+    const rect = makeRect({ left: 300, top: 250 });
+    const previous = makePrevious({ offsetX: 100, offsetY: 50, contentWidth: 800 });
+
+    const result = scaleFaceRectOnResize(rect, previous, { offsetX: 150, offsetY: 75, contentWidth: 800 });
+
+    // scale = 1, imageRelLeft = 200, imageRelTop = 200
+    // new left = 150 + 200 = 350, new top = 75 + 200 = 275
+    expect(result.left).toBe(350);
+    expect(result.top).toBe(275);
+    expect(result.scaleX).toBe(1);
+    expect(result.scaleY).toBe(1);
+  });
+
+  it('should compound existing scale factors', () => {
+    const rect = makeRect({ left: 300, top: 250, scaleX: 2, scaleY: 3 });
+    const previous = makePrevious({ contentWidth: 800 });
+
+    const result = scaleFaceRectOnResize(rect, previous, { ...previous, contentWidth: 1600 });
+
+    expect(result.scaleX).toBe(4);
+    expect(result.scaleY).toBe(6);
+  });
+
+  it('should handle rect at image origin (top-left of content area)', () => {
+    const rect = makeRect({ left: 100, top: 50 });
+    const previous = makePrevious({ offsetX: 100, offsetY: 50, contentWidth: 800 });
+
+    const result = scaleFaceRectOnResize(rect, previous, { offsetX: 200, offsetY: 100, contentWidth: 1600 });
+
+    // imageRelLeft = (100 - 100) * 2 = 0, new left = 200
+    // imageRelTop = (50 - 50) * 2 = 0, new top = 100
+    expect(result.left).toBe(200);
+    expect(result.top).toBe(100);
+  });
+});

web/src/lib/utils/people-utils.ts

@@ -1,6 +1,6 @@
 import type { Faces } from '$lib/stores/people.store';
 import { getAssetMediaUrl } from '$lib/utils';
-import { mapNormalizedRectToContent, type Rect, type Size } from '$lib/utils/container-utils';
+import { mapNormalizedRectToContent, type ContentMetrics, type Rect, type Size } from '$lib/utils/container-utils';
 import { AssetTypeEnum, type AssetFaceResponseDto } from '@immich/sdk';
 
 export type BoundingBox = Rect & { id: string };
@@ -21,6 +21,32 @@ export const getBoundingBox = (faces: Faces[], imageSize: Size): BoundingBox[] =
   return boxes;
 };
 
+export type FaceRectState = {
+  left: number;
+  top: number;
+  scaleX: number;
+  scaleY: number;
+};
+
+export type ResizeContext = Pick<ContentMetrics, 'contentWidth' | 'offsetX' | 'offsetY'>;
+
+export const scaleFaceRectOnResize = (
+  faceRect: FaceRectState,
+  previous: ResizeContext,
+  current: ResizeContext,
+): FaceRectState => {
+  const scale = current.contentWidth / previous.contentWidth;
+  const imageRelativeLeft = (faceRect.left - previous.offsetX) * scale;
+  const imageRelativeTop = (faceRect.top - previous.offsetY) * scale;
+
+  return {
+    left: current.offsetX + imageRelativeLeft,
+    top: current.offsetY + imageRelativeTop,
+    scaleX: faceRect.scaleX * scale,
+    scaleY: faceRect.scaleY * scale,
+  };
+};
+
 export const zoomImageToBase64 = async (
   face: AssetFaceResponseDto,
   assetId: string,