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

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: [],
-    unassignedFaces: [],
+    people: faceData?.people ?? [],
+    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/mock-network/ocr-network.ts

@@ -0,0 +1,55 @@
+import { faker } from '@faker-js/faker';
+import type { AssetOcrResponseDto } from '@immich/sdk';
+import { BrowserContext } from '@playwright/test';
+
+export type MockOcrBox = {
+  text: string;
+  x1: number;
+  y1: number;
+  x2: number;
+  y2: number;
+  x3: number;
+  y3: number;
+  x4: number;
+  y4: number;
+};
+
+export const createMockOcrData = (assetId: string, boxes: MockOcrBox[]): AssetOcrResponseDto[] => {
+  return boxes.map((box) => ({
+    id: faker.string.uuid(),
+    assetId,
+    x1: box.x1,
+    y1: box.y1,
+    x2: box.x2,
+    y2: box.y2,
+    x3: box.x3,
+    y3: box.y3,
+    x4: box.x4,
+    y4: box.y4,
+    boxScore: 0.95,
+    textScore: 0.9,
+    text: box.text,
+  }));
+};
+
+export const setupOcrMockApiRoutes = async (
+  context: BrowserContext,
+  ocrDataByAssetId: Map<string, AssetOcrResponseDto[]>,
+) => {
+  await context.route('**/assets/*/ocr', async (route, request) => {
+    if (request.method() !== 'GET') {
+      return route.fallback();
+    }
+    const url = new URL(request.url());
+    const segments = url.pathname.split('/');
+    const assetIdIndex = segments.indexOf('assets') + 1;
+    const assetId = segments[assetIdIndex];
+
+    const ocrData = ocrDataByAssetId.get(assetId) ?? [];
+    return route.fulfill({
+      status: 200,
+      contentType: 'application/json',
+      json: ocrData,
+    });
+  });
+};

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

@@ -149,7 +149,7 @@ 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 +163,15 @@ test.describe('face-editor', () => {
     await expect(page.locator('#face-editor')).toBeHidden();
 
     expect(faceCreateCapture.requests).toHaveLength(1);
-    expect(faceCreateCapture.requests[0].assetId).toBe(asset.id);
-    expect(faceCreateCapture.requests[0].personId).toBe(personToTag.id);
+    const request = faceCreateCapture.requests[0];
+    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 +289,39 @@ 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 page.waitForTimeout(300);
+
+    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 +1,264 @@
+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[draggable="false"]');
+    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.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.locator('section div[role="button"]').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);
+  });
+});

e2e/src/ui/specs/asset-viewer/ocr.e2e-spec.ts

@@ -0,0 +1,300 @@
+import type { AssetOcrResponseDto, AssetResponseDto } from '@immich/sdk';
+import { expect, test } from '@playwright/test';
+import { toAssetResponseDto } from 'src/ui/generators/timeline';
+import {
+  createMockStack,
+  createMockStackAsset,
+  MockStack,
+  setupBrokenAssetMockApiRoutes,
+} from 'src/ui/mock-network/broken-asset-network';
+import { createMockOcrData, setupOcrMockApiRoutes } from 'src/ui/mock-network/ocr-network';
+import { assetViewerUtils } from '../timeline/utils';
+import { setupAssetViewerFixture } from './utils';
+
+test.describe.configure({ mode: 'parallel' });
+
+const PRIMARY_OCR_BOXES = [
+  { text: 'Hello World', x1: 0.1, y1: 0.1, x2: 0.4, y2: 0.1, x3: 0.4, y3: 0.15, x4: 0.1, y4: 0.15 },
+  { text: 'Immich Photo', x1: 0.2, y1: 0.3, x2: 0.6, y2: 0.3, x3: 0.6, y3: 0.36, x4: 0.2, y4: 0.36 },
+];
+
+const SECONDARY_OCR_BOXES = [
+  { text: 'Second Asset Text', x1: 0.15, y1: 0.2, x2: 0.55, y2: 0.2, x3: 0.55, y3: 0.26, x4: 0.15, y4: 0.26 },
+];
+
+test.describe('OCR bounding boxes', () => {
+  const fixture = setupAssetViewerFixture(920);
+
+  test.beforeEach(async ({ context }) => {
+    const primaryAssetDto = toAssetResponseDto(fixture.primaryAsset);
+    const ocrDataByAssetId = new Map<string, AssetOcrResponseDto[]>([
+      [primaryAssetDto.id, createMockOcrData(primaryAssetDto.id, PRIMARY_OCR_BOXES)],
+    ]);
+
+    await setupOcrMockApiRoutes(context, ocrDataByAssetId);
+  });
+
+  test('OCR bounding boxes appear when clicking OCR button', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    const ocrButton = page.getByLabel('Text recognition');
+    await expect(ocrButton).toBeVisible();
+    await ocrButton.click();
+
+    const ocrBoxes = page.locator('[data-viewer-content] [data-testid="ocr-box"]');
+    await expect(ocrBoxes).toHaveCount(2);
+
+    await expect(ocrBoxes.nth(0)).toContainText('Hello World');
+    await expect(ocrBoxes.nth(1)).toContainText('Immich Photo');
+  });
+
+  test('OCR bounding boxes toggle off on second click', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    const ocrButton = page.getByLabel('Text recognition');
+    await ocrButton.click();
+    await expect(page.locator('[data-viewer-content] [data-testid="ocr-box"]').first()).toBeVisible();
+
+    await ocrButton.click();
+    await expect(page.locator('[data-viewer-content] [data-testid="ocr-box"]')).toHaveCount(0);
+  });
+});
+
+test.describe('OCR with stacked assets', () => {
+  const fixture = setupAssetViewerFixture(921);
+  let mockStack: MockStack;
+  let primaryAssetDto: AssetResponseDto;
+  let secondAssetDto: AssetResponseDto;
+
+  test.beforeAll(async () => {
+    primaryAssetDto = toAssetResponseDto(fixture.primaryAsset);
+    secondAssetDto = createMockStackAsset(fixture.adminUserId);
+    secondAssetDto.originalFileName = 'second-ocr-asset.jpg';
+    mockStack = createMockStack(primaryAssetDto, [secondAssetDto], new Set());
+  });
+
+  test.beforeEach(async ({ context }) => {
+    await setupBrokenAssetMockApiRoutes(context, mockStack);
+
+    const ocrDataByAssetId = new Map<string, AssetOcrResponseDto[]>([
+      [primaryAssetDto.id, createMockOcrData(primaryAssetDto.id, PRIMARY_OCR_BOXES)],
+      [secondAssetDto.id, createMockOcrData(secondAssetDto.id, SECONDARY_OCR_BOXES)],
+    ]);
+
+    await setupOcrMockApiRoutes(context, ocrDataByAssetId);
+  });
+
+  test('different OCR boxes shown for different stacked assets', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    const ocrButton = page.getByLabel('Text recognition');
+    await expect(ocrButton).toBeVisible();
+    await ocrButton.click();
+
+    const ocrBoxes = page.locator('[data-viewer-content] [data-testid="ocr-box"]');
+    await expect(ocrBoxes).toHaveCount(2);
+    await expect(ocrBoxes.nth(0)).toContainText('Hello World');
+
+    const stackThumbnails = page.locator('#stack-slideshow [data-asset]');
+    await expect(stackThumbnails).toHaveCount(2);
+    await stackThumbnails.nth(1).click();
+
+    // refreshOcr() clears showOverlay when switching assets, so re-enable it
+    await expect(ocrBoxes).toHaveCount(0);
+    await expect(ocrButton).toBeVisible();
+    await ocrButton.click();
+
+    await expect(ocrBoxes).toHaveCount(1);
+    await expect(ocrBoxes.first()).toContainText('Second Asset Text');
+  });
+});
+
+test.describe('OCR boxes and zoom', () => {
+  const fixture = setupAssetViewerFixture(922);
+
+  test.beforeEach(async ({ context }) => {
+    const primaryAssetDto = toAssetResponseDto(fixture.primaryAsset);
+    const ocrDataByAssetId = new Map<string, AssetOcrResponseDto[]>([
+      [primaryAssetDto.id, createMockOcrData(primaryAssetDto.id, PRIMARY_OCR_BOXES)],
+    ]);
+
+    await setupOcrMockApiRoutes(context, ocrDataByAssetId);
+  });
+
+  test('OCR boxes scale with zoom', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    const ocrButton = page.getByLabel('Text recognition');
+    await expect(ocrButton).toBeVisible();
+    await ocrButton.click();
+
+    const ocrBox = page.locator('[data-viewer-content] [data-testid="ocr-box"]').first();
+    await expect(ocrBox).toBeVisible();
+
+    const initialBox = await ocrBox.boundingBox();
+    expect(initialBox).toBeTruthy();
+
+    const { width, height } = page.viewportSize()!;
+    await page.mouse.move(width / 2, height / 2);
+    await page.mouse.wheel(0, -3);
+
+    await expect(async () => {
+      const zoomedBox = await ocrBox.boundingBox();
+      expect(zoomedBox).toBeTruthy();
+      expect(zoomedBox!.width).toBeGreaterThan(initialBox!.width);
+      expect(zoomedBox!.height).toBeGreaterThan(initialBox!.height);
+    }).toPass({ timeout: 2000 });
+  });
+});
+
+test.describe('OCR text interaction', () => {
+  const fixture = setupAssetViewerFixture(923);
+
+  test.beforeEach(async ({ context }) => {
+    const primaryAssetDto = toAssetResponseDto(fixture.primaryAsset);
+    const ocrDataByAssetId = new Map<string, AssetOcrResponseDto[]>([
+      [primaryAssetDto.id, createMockOcrData(primaryAssetDto.id, PRIMARY_OCR_BOXES)],
+    ]);
+
+    await setupOcrMockApiRoutes(context, ocrDataByAssetId);
+  });
+
+  test('OCR text box has data-overlay-interactive attribute', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    await page.getByLabel('Text recognition').click();
+
+    const ocrBox = page.locator('[data-viewer-content] [data-testid="ocr-box"]').first();
+    await expect(ocrBox).toBeVisible();
+    await expect(ocrBox).toHaveAttribute('data-overlay-interactive');
+  });
+
+  test('OCR text box receives focus on click', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    await page.getByLabel('Text recognition').click();
+
+    const ocrBox = page.locator('[data-viewer-content] [data-testid="ocr-box"]').first();
+    await expect(ocrBox).toBeVisible();
+
+    await ocrBox.click();
+    await expect(ocrBox).toBeFocused();
+  });
+
+  test('dragging on OCR text box does not trigger image pan', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    await page.getByLabel('Text recognition').click();
+
+    const ocrBox = page.locator('[data-viewer-content] [data-testid="ocr-box"]').first();
+    await expect(ocrBox).toBeVisible();
+
+    const imgLocator = page.locator('[data-viewer-content] img[draggable="false"]');
+    const initialTransform = await imgLocator.evaluate((element) => {
+      return getComputedStyle(element.closest('[style*="transform"]') ?? element).transform;
+    });
+
+    const box = await ocrBox.boundingBox();
+    expect(box).toBeTruthy();
+    const centerX = box!.x + box!.width / 2;
+    const centerY = box!.y + box!.height / 2;
+
+    await page.mouse.move(centerX, centerY);
+    await page.mouse.down();
+    await page.mouse.move(centerX + 50, centerY + 30, { steps: 5 });
+    await page.mouse.up();
+
+    const afterTransform = await imgLocator.evaluate((element) => {
+      return getComputedStyle(element.closest('[style*="transform"]') ?? element).transform;
+    });
+    expect(afterTransform).toBe(initialTransform);
+  });
+
+  test('split touch gesture across zoom container does not trigger zoom', async ({ page }) => {
+    await page.goto(`/photos/${fixture.primaryAsset.id}`);
+    await assetViewerUtils.waitForViewerLoad(page, fixture.primaryAsset);
+
+    await page.getByLabel('Text recognition').click();
+    const ocrBox = page.locator('[data-viewer-content] [data-testid="ocr-box"]').first();
+    await expect(ocrBox).toBeVisible();
+
+    const imgLocator = page.locator('[data-viewer-content] img[draggable="false"]');
+    const initialTransform = await imgLocator.evaluate((element) => {
+      return getComputedStyle(element.closest('[style*="transform"]') ?? element).transform;
+    });
+
+    const viewerContent = page.locator('[data-viewer-content]');
+    const viewerBox = await viewerContent.boundingBox();
+    expect(viewerBox).toBeTruthy();
+
+    // Dispatch a synthetic split gesture: one touch inside the viewer, one outside
+    await page.evaluate(
+      ({ viewerCenterX, viewerCenterY, outsideY }) => {
+        const viewer = document.querySelector('[data-viewer-content]');
+        if (!viewer) {
+          return;
+        }
+
+        const createTouch = (id: number, x: number, y: number) => {
+          return new Touch({
+            identifier: id,
+            target: viewer,
+            clientX: x,
+            clientY: y,
+          });
+        };
+
+        const insideTouch = createTouch(0, viewerCenterX, viewerCenterY);
+        const outsideTouch = createTouch(1, viewerCenterX, outsideY);
+
+        const touchStartEvent = new TouchEvent('touchstart', {
+          touches: [insideTouch, outsideTouch],
+          targetTouches: [insideTouch],
+          changedTouches: [insideTouch, outsideTouch],
+          bubbles: true,
+          cancelable: true,
+        });
+
+        const touchMoveEvent = new TouchEvent('touchmove', {
+          touches: [createTouch(0, viewerCenterX, viewerCenterY - 30), createTouch(1, viewerCenterX, outsideY + 30)],
+          targetTouches: [createTouch(0, viewerCenterX, viewerCenterY - 30)],
+          changedTouches: [
+            createTouch(0, viewerCenterX, viewerCenterY - 30),
+            createTouch(1, viewerCenterX, outsideY + 30),
+          ],
+          bubbles: true,
+          cancelable: true,
+        });
+
+        const touchEndEvent = new TouchEvent('touchend', {
+          touches: [],
+          targetTouches: [],
+          changedTouches: [insideTouch, outsideTouch],
+          bubbles: true,
+          cancelable: true,
+        });
+
+        viewer.dispatchEvent(touchStartEvent);
+        viewer.dispatchEvent(touchMoveEvent);
+        viewer.dispatchEvent(touchEndEvent);
+      },
+      {
+        viewerCenterX: viewerBox!.x + viewerBox!.width / 2,
+        viewerCenterY: viewerBox!.y + viewerBox!.height / 2,
+        outsideY: 10, // near the top of the page, outside the viewer
+      },
+    );
+
+    const afterTransform = await imgLocator.evaluate((element) => {
+      return getComputedStyle(element.closest('[style*="transform"]') ?? element).transform;
+    });
+    expect(afterTransform).toBe(initialTransform);
+  });
+});

web/src/lib/actions/zoom-image.ts

@@ -1,11 +1,17 @@
 import { assetViewerManager } from '$lib/managers/asset-viewer-manager.svelte';
 import { createZoomImageWheel } from '@zoom-image/core';
 
-export const zoomImageAction = (node: HTMLElement, options?: { disabled?: boolean }) => {
+type TouchEventLike = {
+  touches: Iterable<{ clientX: number; clientY: number }> & { length: number };
+  targetTouches: ArrayLike<unknown>;
+};
+const asTouchEvent = (event: Event) => event as unknown as TouchEventLike;
+
+export const zoomImageAction = (node: HTMLElement, options?: { zoomTarget?: HTMLElement }) => {
   const zoomInstance = createZoomImageWheel(node, {
     maxZoom: 10,
     initialState: assetViewerManager.zoomState,
-    zoomTarget: null,
+    zoomTarget: options?.zoomTarget,
   });
 
   const unsubscribes = [
@@ -13,47 +19,124 @@ export const zoomImageAction = (node: HTMLElement, options?: { disabled?: boolea
     zoomInstance.subscribe(({ state }) => assetViewerManager.onZoomChange(state)),
   ];
 
-  const onInteractionStart = (event: Event) => {
-    if (options?.disabled) {
-      event.stopImmediatePropagation();
+  const controller = new AbortController();
+  const { signal } = controller;
+
+  node.addEventListener('pointerdown', () => assetViewerManager.cancelZoomAnimation(), { capture: true, signal });
+
+  // Intercept events in capture phase to prevent zoom-image from seeing interactions on
+  // overlay elements (e.g. OCR text boxes), preserving browser defaults like text selection.
+  const isOverlayEvent = (event: Event) => !!(event.target as HTMLElement).closest('[data-overlay-interactive]');
+  const isOverlayAtPoint = (x: number, y: number) =>
+    !!document.elementFromPoint(x, y)?.closest('[data-overlay-interactive]');
+
+  // Pointer event interception: track pointers that start on overlays and intercept the entire gesture.
+  const overlayPointers = new Set<number>();
+  const interceptedPointers = new Set<number>();
+  const interceptOverlayPointerDown = (event: PointerEvent) => {
+    if (isOverlayEvent(event) || isOverlayAtPoint(event.clientX, event.clientY)) {
+      overlayPointers.add(event.pointerId);
+      interceptedPointers.add(event.pointerId);
+      event.stopPropagation();
+    } else if (overlayPointers.size > 0) {
+      // Split gesture (e.g. pinch with one finger on overlay) — intercept entirely.
+      interceptedPointers.add(event.pointerId);
+      event.stopPropagation();
     }
-    assetViewerManager.cancelZoomAnimation();
   };
+  const interceptOverlayPointerEvent = (event: PointerEvent) => {
+    if (interceptedPointers.has(event.pointerId)) {
+      event.stopPropagation();
+    }
+  };
+  const interceptOverlayPointerEnd = (event: PointerEvent) => {
+    overlayPointers.delete(event.pointerId);
+    if (interceptedPointers.delete(event.pointerId)) {
+      event.stopPropagation();
+    }
+  };
+  node.addEventListener('pointerdown', interceptOverlayPointerDown, { capture: true, signal });
+  node.addEventListener('pointermove', interceptOverlayPointerEvent, { capture: true, signal });
+  node.addEventListener('pointerup', interceptOverlayPointerEnd, { capture: true, signal });
+  node.addEventListener('pointerleave', interceptOverlayPointerEnd, { capture: true, signal });
 
-  node.addEventListener('wheel', onInteractionStart, { capture: true });
-  node.addEventListener('pointerdown', onInteractionStart, { capture: true });
+  // Touch event interception for overlay touches or split gestures (pinch across container boundary).
+  // Once intercepted, stays intercepted until all fingers are lifted.
+  let touchGestureIntercepted = false;
+  const interceptOverlayTouchEvent = (event: Event) => {
+    if (touchGestureIntercepted) {
+      event.stopPropagation();
+      return;
+    }
+    const { touches, targetTouches } = asTouchEvent(event);
+    if (touches && targetTouches) {
+      if (touches.length > targetTouches.length) {
+        touchGestureIntercepted = true;
+        event.stopPropagation();
+        return;
+      }
+      for (const touch of touches) {
+        if (isOverlayAtPoint(touch.clientX, touch.clientY)) {
+          touchGestureIntercepted = true;
+          event.stopPropagation();
+          return;
+        }
+      }
+    } else if (isOverlayEvent(event)) {
+      event.stopPropagation();
+    }
+  };
+  const resetTouchGesture = (event: Event) => {
+    const { touches } = asTouchEvent(event);
+    if (touches.length === 0) {
+      touchGestureIntercepted = false;
+    }
+  };
+  node.addEventListener('touchstart', interceptOverlayTouchEvent, { capture: true, signal });
+  node.addEventListener('touchmove', interceptOverlayTouchEvent, { capture: true, signal });
+  node.addEventListener('touchend', resetTouchGesture, { capture: true, signal });
 
-  // Suppress Safari's synthetic dblclick on double-tap. Without this, zoom-image's touchstart
-  // handler zooms to maxZoom (10x), then Safari's synthetic dblclick triggers photo-viewer's
-  // handler which conflicts. Chrome does not fire synthetic dblclick on touch.
+  // Wheel and dblclick interception on overlay elements.
+  // Dblclick also intercepted for all touch double-taps (Safari fires synthetic dblclick
+  // on double-tap, which conflicts with zoom-image's touch zoom handler).
   let lastPointerWasTouch = false;
-  const trackPointerType = (event: PointerEvent) => {
-    lastPointerWasTouch = event.pointerType === 'touch';
-  };
-  const suppressTouchDblClick = (event: MouseEvent) => {
-    if (lastPointerWasTouch) {
-      event.stopImmediatePropagation();
-    }
-  };
-  node.addEventListener('pointerdown', trackPointerType, { capture: true });
-  node.addEventListener('dblclick', suppressTouchDblClick, { capture: true });
+  node.addEventListener('pointerdown', (event) => (lastPointerWasTouch = event.pointerType === 'touch'), {
+    capture: true,
+    signal,
+  });
+  node.addEventListener(
+    'wheel',
+    (event) => {
+      if (isOverlayEvent(event)) {
+        event.stopPropagation();
+      }
+    },
+    { capture: true, signal },
+  );
+  node.addEventListener(
+    'dblclick',
+    (event) => {
+      if (lastPointerWasTouch || isOverlayEvent(event)) {
+        event.stopImmediatePropagation();
+      }
+    },
+    { capture: true, signal },
+  );
 
-  // Allow zoomed content to render outside the container bounds
   node.style.overflow = 'visible';
-  // Prevent browser handling of touch gestures so zoom-image can manage them
   node.style.touchAction = 'none';
   return {
-    update(newOptions?: { disabled?: boolean }) {
+    update(newOptions?: { zoomTarget?: HTMLElement }) {
       options = newOptions;
+      if (newOptions?.zoomTarget !== undefined) {
+        zoomInstance.setState({ zoomTarget: newOptions.zoomTarget });
+      }
     },
     destroy() {
+      controller.abort();
       for (const unsubscribe of unsubscribes) {
         unsubscribe();
       }
-      node.removeEventListener('wheel', onInteractionStart, { capture: true });
-      node.removeEventListener('pointerdown', onInteractionStart, { capture: true });
-      node.removeEventListener('pointerdown', trackPointerType, { capture: true });
-      node.removeEventListener('dblclick', suppressTouchDblClick, { capture: true });
       zoomInstance.cleanup();
     },
   };

web/src/lib/components/AdaptiveImage.svelte

@@ -7,7 +7,7 @@
   import { assetViewerManager } from '$lib/managers/asset-viewer-manager.svelte';
   import { getAssetUrls } from '$lib/utils';
   import { AdaptiveImageLoader, type QualityList } from '$lib/utils/adaptive-image-loader.svelte';
-  import { scaleToCover, scaleToFit } from '$lib/utils/container-utils';
+  import { scaleToCover, scaleToFit, type Size } from '$lib/utils/container-utils';
   import { getAltText } from '$lib/utils/thumbnail-util';
   import { toTimelineAsset } from '$lib/utils/timeline-util';
   import type { AssetResponseDto, SharedLinkResponseDto } from '@immich/sdk';
@@ -17,10 +17,7 @@
     asset: AssetResponseDto;
     sharedLink?: SharedLinkResponseDto;
     objectFit?: 'contain' | 'cover';
-    container: {
-      width: number;
-      height: number;
-    };
+    container: Size;
     onUrlChange?: (url: string) => void;
     onImageReady?: () => void;
     onError?: () => void;
@@ -149,81 +146,66 @@
       (quality.preview === 'success' ? previewElement : undefined) ??
       (quality.thumbnail === 'success' ? thumbnailElement : undefined);
   });
-
-  const zoomTransform = $derived.by(() => {
-    const { currentZoom, currentPositionX, currentPositionY } = assetViewerManager.zoomState;
-    if (currentZoom === 1 && currentPositionX === 0 && currentPositionY === 0) {
-      return undefined;
-    }
-    return `translate(${currentPositionX}px, ${currentPositionY}px) scale(${currentZoom})`;
-  });
 </script>
 
 <div class="relative h-full w-full overflow-hidden will-change-transform" bind:this={ref}>
   {@render backdrop?.()}
 
-  <!-- pointer-events-none so events pass through to the container where zoom-image listens -->
-  <div
-    class="absolute inset-0 pointer-events-none"
-    style:transform={zoomTransform}
-    style:transform-origin={zoomTransform ? '0 0' : undefined}
-  >
-    <div class="absolute" style:left style:top style:width style:height>
-      {#if show.alphaBackground}
-        <AlphaBackground />
-      {/if}
+  <div class="absolute inset-0 pointer-events-none" style:left style:top style:width style:height>
+    {#if show.alphaBackground}
+      <AlphaBackground />
+    {/if}
 
-      {#if show.thumbhash}
-        {#if asset.thumbhash}
-          <!-- Thumbhash / spinner layer  -->
-          <canvas use:thumbhash={{ base64ThumbHash: asset.thumbhash }} class="h-full w-full absolute"></canvas>
-        {:else if show.spinner}
-          <DelayedLoadingSpinner />
-        {/if}
+    {#if show.thumbhash}
+      {#if asset.thumbhash}
+        <!-- Thumbhash / spinner layer  -->
+        <canvas use:thumbhash={{ base64ThumbHash: asset.thumbhash }} class="h-full w-full absolute"></canvas>
+      {:else if show.spinner}
+        <DelayedLoadingSpinner />
       {/if}
+    {/if}
 
-      {#if show.thumbnail}
-        <ImageLayer
-          {adaptiveImageLoader}
-          {width}
-          {height}
-          quality="thumbnail"
-          src={status.urls.thumbnail}
-          alt=""
-          role="presentation"
-          bind:ref={thumbnailElement}
-        />
-      {/if}
+    {#if show.thumbnail}
+      <ImageLayer
+        {adaptiveImageLoader}
+        {width}
+        {height}
+        quality="thumbnail"
+        src={status.urls.thumbnail}
+        alt=""
+        role="presentation"
+        bind:ref={thumbnailElement}
+      />
+    {/if}
 
-      {#if show.brokenAsset}
-        <BrokenAsset class="text-xl h-full w-full absolute" />
-      {/if}
+    {#if show.brokenAsset}
+      <BrokenAsset class="text-xl h-full w-full absolute" />
+    {/if}
 
-      {#if show.preview}
-        <ImageLayer
-          {adaptiveImageLoader}
-          {alt}
-          {width}
-          {height}
-          {overlays}
-          quality="preview"
-          src={status.urls.preview}
-          bind:ref={previewElement}
-        />
-      {/if}
+    {#if show.preview}
+      <ImageLayer
+        {adaptiveImageLoader}
+        {alt}
+        {width}
+        {height}
+        {overlays}
+        quality="preview"
+        src={status.urls.preview}
+        bind:ref={previewElement}
+      />
+    {/if}
 
-      {#if show.original}
-        <ImageLayer
-          {adaptiveImageLoader}
-          {alt}
-          {width}
-          {height}
-          {overlays}
-          quality="original"
-          src={status.urls.original}
-          bind:ref={originalElement}
-        />
-      {/if}
-    </div>
+    {#if show.original}
+      <ImageLayer
+        {adaptiveImageLoader}
+        {alt}
+        {width}
+        {height}
+        {overlays}
+        quality="original"
+        src={status.urls.original}
+        bind:ref={originalElement}
+      />
+    {/if}
   </div>
 </div>

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

@@ -176,6 +176,7 @@
   onDestroy(() => {
     activityManager.reset();
     assetViewerManager.closeEditor();
+    isFaceEditMode.value = false;
     syncAssetViewerOpenClass(false);
     preloadManager.destroy();
   });
@@ -358,15 +359,18 @@
     }
   };
 
+  const refreshOcr = async () => {
+    ocrManager.clear();
+    if (sharedLink) {
+      return;
+    }
+
+    await ocrManager.getAssetOcr(asset.id);
+  };
+
   const refresh = async () => {
     await refreshStack();
-    ocrManager.clear();
-    if (!sharedLink) {
-      if (previewStackedAsset) {
-        await ocrManager.getAssetOcr(previewStackedAsset.id);
-      }
-      await ocrManager.getAssetOcr(asset.id);
-    }
+    await refreshOcr();
   };
 
   $effect(() => {
@@ -375,6 +379,12 @@
     untrack(() => handlePromiseError(refresh()));
   });
 
+  $effect(() => {
+    // eslint-disable-next-line @typescript-eslint/no-unused-expressions
+    previewStackedAsset;
+    untrack(() => ocrManager.clear());
+  });
+
   let lastCursor = $state<AssetCursor>();
 
   $effect(() => {
@@ -460,7 +470,7 @@
 
 <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}
 >

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

@@ -7,6 +7,7 @@
   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';
@@ -122,6 +123,7 @@
 
   const handleRefreshPeople = async () => {
     asset = await getAssetInfo({ id: asset.id });
+    eventManager.emit('AssetUpdate', asset);
     showEditFaces = false;
   };
 
@@ -233,8 +235,8 @@
               href={Route.viewPerson(person, { previousRoute })}
               onfocus={() => ($boundingBoxesArray = people[index].faces)}
               onblur={() => ($boundingBoxesArray = [])}
-              onmouseover={() => ($boundingBoxesArray = people[index].faces)}
-              onmouseleave={() => ($boundingBoxesArray = [])}
+              onpointerover={() => ($boundingBoxesArray = people[index].faces)}
+              onpointerleave={() => ($boundingBoxesArray = [])}
             >
               <div class="relative">
                 <ImageThumbnail

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, getNaturalSize, mapContentRectToNatural } 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';
@@ -23,6 +25,7 @@
   let { htmlElement, 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 +35,8 @@
 
   let searchTerm = $state('');
   let faceBoxPosition = $state({ left: 0, top: 0, width: 0, height: 0 });
+  let userMovedRect = false;
+  let previousMetrics: ResizeContext | null = null;
 
   let filteredCandidates = $derived(
     searchTerm
@@ -57,7 +62,8 @@
       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 +81,100 @@
 
     canvas.add(faceRect);
     canvas.setActiveObject(faceRect);
-    setDefaultFaceRectanglePosition(faceRect);
   };
 
-  onMount(async () => {
-    setupCanvas();
-    await getPeople();
+  onMount(() => {
+    void 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({
-      width: containerWidth,
-      height: containerHeight,
-    });
+    const upperCanvas = canvas.upperCanvasEl;
+    const controller = new AbortController();
+    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(
+    computeContentMetrics(getNaturalSize(htmlElement), { 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) => {
-    const faceBox = faceRect.getBoundingRect();
-    return !(
-      0 > faceBox.left + faceBox.width ||
-      0 > faceBox.top + faceBox.height ||
-      canvas.width < faceBox.left ||
-      canvas.height < faceBox.top
-    );
-  };
+    faceRect.setCoords();
+    previousMetrics = { contentWidth, offsetX, offsetY };
+    canvas.renderAll();
+    positionFaceSelector();
+  });
 
   const cancel = () => {
     isFaceEditMode.value = false;
@@ -164,11 +204,12 @@
     const gap = 15;
     const padding = faceRect.padding ?? 0;
     const rawBox = faceRect.getBoundingRect();
+    const { currentZoom, currentPositionX, currentPositionY } = assetViewerManager.zoomState;
     const faceBox = {
-      left: rawBox.left - padding,
-      top: rawBox.top - padding,
-      width: rawBox.width + padding * 2,
-      height: rawBox.height + padding * 2,
+      left: (rawBox.left - padding) * currentZoom + currentPositionX,
+      top: (rawBox.top - padding) * currentZoom + currentPositionY,
+      width: (rawBox.width + padding * 2) * currentZoom,
+      height: (rawBox.height + padding * 2) * currentZoom,
     };
     const selectorWidth = faceSelectorEl.offsetWidth;
     const chromeHeight = faceSelectorEl.offsetHeight - scrollableListEl.offsetHeight;
@@ -178,20 +219,21 @@
     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 selectorRight = position.left + selectorWidth;
-      const selectorBottom = position.top + selectorHeight;
-      const faceRight = faceBox.left + faceBox.width;
-      const faceBottom = faceBox.top + faceBox.height;
+    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 overlapY = Math.max(0, Math.min(selectorBottom, faceBottom) - Math.max(position.top, faceBox.top));
+    const overlapArea = (position: { top: number; left: number }) => {
+      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 +255,89 @@
       }
     }
 
-    faceSelectorEl.style.top = `${bestPosition.top}px`;
-    faceSelectorEl.style.left = `${bestPosition.left}px`;
+    const containerRect = containerEl?.getBoundingClientRect();
+    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);
-      rect.on('scaling', positionFaceSelector);
+      const onUserMove = () => {
+        userMovedRect = true;
+        positionFaceSelector();
+      };
+      rect.on('moving', onUserMove);
+      rect.on('scaling', onUserMove);
       return () => {
-        rect.off('moving', positionFaceSelector);
-        rect.off('scaling', positionFaceSelector);
+        rect.off('moving', onUserMove);
+        rect.off('scaling', onUserMove);
       };
     }
   });
 
+  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) {
       return;
     }
 
-    const { left, top, width, height } = faceRect.getBoundingRect();
-    const { offsetX, offsetY, contentWidth, contentHeight } = imageContentMetrics;
+    const scaledWidth = faceRect.getScaledWidth();
+    const scaledHeight = faceRect.getScaledHeight();
     const natural = getNaturalSize(htmlElement);
 
-    const scaleX = natural.width / contentWidth;
-    const scaleY = natural.height / contentHeight;
-    const imageX = (left - offsetX) * scaleX;
-    const imageY = (top - offsetY) * scaleY;
+    const imageRect = mapContentRectToNatural(
+      {
+        left: faceRect.left - scaledWidth / 2,
+        top: faceRect.top - scaledHeight / 2,
+        width: scaledWidth,
+        height: scaledHeight,
+      },
+      imageContentMetrics,
+      natural,
+    );
 
     return {
       imageWidth: natural.width,
       imageHeight: natural.height,
-      x: Math.floor(imageX),
-      y: Math.floor(imageY),
-      width: Math.floor(width * scaleX),
-      height: Math.floor(height * scaleY),
+      x: Math.floor(imageRect.left),
+      y: Math.floor(imageRect.top),
+      width: Math.floor(imageRect.width),
+      height: Math.floor(imageRect.height),
     };
   };
 
@@ -282,10 +368,9 @@
       });
 
       await assetViewingStore.setAssetId(assetId);
+      isFaceEditMode.value = false;
     } catch (error) {
       handleError(error, 'Error tagging face');
-    } finally {
-      isFaceEditMode.value = false;
     }
   };
 </script>
@@ -294,6 +379,7 @@
 
 <div
   id="face-editor-data"
+  bind:this={containerEl}
   class="absolute start-0 top-0 z-5 h-full w-full overflow-hidden"
   data-face-left={faceBoxPosition.left}
   data-face-top={faceBoxPosition.top}
@@ -305,7 +391,9 @@
   <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 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>
 

web/src/lib/components/asset-viewer/ocr-bounding-box.svelte

@@ -1,4 +1,5 @@
 <script lang="ts">
+  import { mediaQueryManager } from '$lib/stores/media-query-manager.svelte';
   import type { OcrBox } from '$lib/utils/ocr-utils';
   import { calculateBoundingBoxMatrix, calculateFittedFontSize } from '$lib/utils/ocr-utils';
 
@@ -8,6 +9,7 @@
 
   let { ocrBox }: Props = $props();
 
+  const isTouch = $derived(mediaQueryManager.pointerCoarse);
   const dimensions = $derived(calculateBoundingBoxMatrix(ocrBox.points));
 
   const transform = $derived(`matrix3d(${dimensions.matrix.join(',')})`);
@@ -15,13 +17,23 @@
     calculateFittedFontSize(ocrBox.text, dimensions.width, dimensions.height, ocrBox.verticalMode) + 'px',
   );
 
+  const handleSelectStart = (event: Event) => {
+    const target = event.currentTarget as HTMLElement;
+    requestAnimationFrame(() => {
+      const selection = globalThis.getSelection();
+      if (selection) {
+        selection.selectAllChildren(target);
+      }
+    });
+  };
+
   const verticalStyle = $derived.by(() => {
     switch (ocrBox.verticalMode) {
       case 'cjk': {
-        return ' writing-mode: vertical-rl;';
+        return 'writing-mode: vertical-rl;';
       }
       case 'rotated': {
-        return ' writing-mode: vertical-rl; text-orientation: sideways;';
+        return 'writing-mode: vertical-rl; text-orientation: sideways;';
       }
       default: {
         return '';
@@ -30,17 +42,20 @@
   });
 </script>
 
-<div class="absolute left-0 top-0">
-  <div
-    class="absolute flex items-center justify-center text-transparent border-2 border-blue-500 bg-blue-500/10 pointer-events-auto cursor-text select-text transition-colors hover:z-1 hover:text-white hover:bg-black/60 hover:border-blue-600 hover:border-3 focus:z-1 focus:text-white focus:bg-black/60 focus:border-blue-600 focus:border-3 focus:outline-none {ocrBox.verticalMode ===
-    'none'
-      ? 'px-2 py-1 whitespace-nowrap'
-      : 'px-1 py-2'}"
-    style="font-size: {fontSize}; width: {dimensions.width}px; height: {dimensions.height}px; transform: {transform}; transform-origin: 0 0;{verticalStyle}"
-    tabindex="0"
-    role="button"
-    aria-label={ocrBox.text}
-  >
-    {ocrBox.text}
-  </div>
+<div
+  class="absolute left-0 top-0 flex items-center justify-center border-2 border-blue-500 pointer-events-auto cursor-text focus:z-1 focus:border-blue-600 focus:border-3 focus:outline-none {isTouch
+    ? 'text-white bg-black/60 select-all'
+    : 'select-text text-transparent bg-blue-500/10 transition-colors hover:z-1 hover:text-white hover:bg-black/60 hover:border-blue-600 hover:border-3'} {ocrBox.verticalMode ===
+  'none'
+    ? 'px-2 py-1 whitespace-nowrap'
+    : 'px-1 py-2'}"
+  style="font-size: {fontSize}; width: {dimensions.width}px; height: {dimensions.height}px; transform: {transform}; transform-origin: 0 0; touch-action: none; {verticalStyle}"
+  data-testid="ocr-box"
+  data-overlay-interactive
+  tabindex="0"
+  role="button"
+  aria-label={ocrBox.text}
+  onselectstart={isTouch ? handleSelectStart : undefined}
+>
+  {ocrBox.text}
 </div>

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

@@ -128,10 +128,8 @@
     }
 
     const boxes = getOcrBoundingBoxes(ocrData, {
-      contentWidth: viewer.state.textureData.panoData.croppedWidth,
-      contentHeight: viewer.state.textureData.panoData.croppedHeight,
-      offsetX: 0,
-      offsetY: 0,
+      width: viewer.state.textureData.panoData.croppedWidth,
+      height: viewer.state.textureData.panoData.croppedHeight,
     });
 
     for (const [index, box] of boxes.entries()) {

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

@@ -14,7 +14,7 @@
   import { SlideshowLook, SlideshowState, slideshowStore } from '$lib/stores/slideshow.store';
   import { handlePromiseError } from '$lib/utils';
   import { canCopyImageToClipboard, copyImageToClipboard } from '$lib/utils/asset-utils';
-  import { getNaturalSize, scaleToFit, type ContentMetrics } from '$lib/utils/container-utils';
+  import { getNaturalSize, scaleToFit, 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';
@@ -67,23 +67,15 @@
     height: containerHeight,
   });
 
-  const overlayMetrics = $derived.by((): ContentMetrics => {
+  const overlaySize = $derived.by((): Size => {
     if (!assetViewerManager.imgRef || !visibleImageReady) {
-      return { contentWidth: 0, contentHeight: 0, offsetX: 0, offsetY: 0 };
+      return { width: 0, height: 0 };
     }
 
-    const natural = getNaturalSize(assetViewerManager.imgRef);
-    const scaled = scaleToFit(natural, { width: containerWidth, height: containerHeight });
-
-    return {
-      contentWidth: scaled.width,
-      contentHeight: scaled.height,
-      offsetX: 0,
-      offsetY: 0,
-    };
+    return scaleToFit(getNaturalSize(assetViewerManager.imgRef), { width: containerWidth, height: containerHeight });
   });
 
-  const ocrBoxes = $derived(ocrManager.showOverlay ? getOcrBoundingBoxes(ocrManager.data, overlayMetrics) : []);
+  const ocrBoxes = $derived(ocrManager.showOverlay ? getOcrBoundingBoxes(ocrManager.data, overlaySize) : []);
 
   const onCopy = async () => {
     if (!canCopyImageToClipboard() || !assetViewerManager.imgRef) {
@@ -105,12 +97,6 @@
 
   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') {
@@ -151,6 +137,8 @@
     $slideshowState !== SlideshowState.None && $slideshowLook === SlideshowLook.BlurredBackground && !!asset.thumbhash,
   );
 
+  let adaptiveImage = $state<HTMLDivElement | undefined>();
+
   const faceToNameMap = $derived.by(() => {
     // eslint-disable-next-line svelte/prefer-svelte-reactivity
     const map = new Map<Faces, string>();
@@ -159,40 +147,18 @@
         map.set(face, person.name);
       }
     }
+    if (isFaceEditMode.value) {
+      for (const face of asset.unassignedFaces ?? []) {
+        map.set(face, '');
+      }
+    }
     return map;
   });
 
+  // Array needed for indexed access in the template (faces[index])
   const faces = $derived(Array.from(faceToNameMap.keys()));
-
-  const handleImageMouseMove = (event: MouseEvent) => {
-    $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, overlayMetrics);
-
-    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 = [];
-  };
+  const boundingBoxes = $derived(getBoundingBox(faces, overlaySize));
+  const activeBoundingBoxes = $derived(boundingBoxes.filter((box) => $boundingBoxesArray.some((f) => f.id === box.id)));
 </script>
 
 <AssetViewerEvents {onCopy} {onZoom} />
@@ -213,9 +179,7 @@
   bind:clientHeight={containerHeight}
   role="presentation"
   ondblclick={onZoom}
-  onmousemove={handleImageMouseMove}
-  onmouseleave={handleImageMouseLeave}
-  use:zoomImageAction={{ disabled: isFaceEditMode.value || ocrManager.showOverlay }}
+  use:zoomImageAction={{ zoomTarget: adaptiveImage }}
   {...useSwipe((event) => onSwipe?.(event))}
 >
   <AdaptiveImage
@@ -233,6 +197,7 @@
       onReady?.();
     }}
     bind:imgRef={assetViewerManager.imgRef}
+    bind:ref={adaptiveImage}
   >
     {#snippet backdrop()}
       {#if blurredSlideshow}
@@ -243,20 +208,38 @@
       {/if}
     {/snippet}
     {#snippet overlays()}
-      {#each getBoundingBox($boundingBoxesArray, overlayMetrics) as boundingbox, index (boundingbox.id)}
-        <div
-          class="absolute border-solid border-white border-3 rounded-lg"
-          style="top: {boundingbox.top}px; left: {boundingbox.left}px; height: {boundingbox.height}px; width: {boundingbox.width}px;"
-        ></div>
-        {#if faceToNameMap.get($boundingBoxesArray[index])}
+      {#if !isFaceEditMode.value}
+        {#each boundingBoxes as boundingbox, index (boundingbox.id)}
+          {@const face = faces[index]}
+          {@const name = faceToNameMap.get(face)}
+          <!-- svelte-ignore a11y_no_static_element_interactions -->
           <div
-            class="absolute bg-white/90 text-black px-2 py-1 rounded text-sm font-medium whitespace-nowrap pointer-events-none shadow-lg"
-            style="top: {boundingbox.top + boundingbox.height + 4}px; left: {boundingbox.left +
-              boundingbox.width}px; transform: translateX(-100%);"
-          >
-            {faceToNameMap.get($boundingBoxesArray[index])}
-          </div>
-        {/if}
+            class="absolute pointer-events-auto outline-none rounded-lg"
+            style="top: {boundingbox.top}px; left: {boundingbox.left}px; height: {boundingbox.height}px; width: {boundingbox.width}px;"
+            aria-label="{$t('person')}: {name || $t('unknown')}"
+            onpointerenter={() => ($boundingBoxesArray = [face])}
+            onpointerleave={() => ($boundingBoxesArray = [])}
+          ></div>
+        {/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 pointer-events-none"
+          style="top: {boundingbox.top}px; left: {boundingbox.left}px; height: {boundingbox.height}px; width: {boundingbox.width}px;"
+        >
+          {#if name}
+            <div
+              aria-hidden="true"
+              class="absolute bg-white/90 text-black px-2 py-1 rounded text-sm font-medium whitespace-nowrap shadow-lg"
+              style="top: {boundingbox.height + 4}px; right: 0;"
+            >
+              {name}
+            </div>
+          {/if}
+        </div>
       {/each}
 
       {#each ocrBoxes as ocrBox (ocrBox.id)}

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

@@ -232,8 +232,8 @@
               tabindex={index}
               class="absolute start-0 top-0 h-22.5 w-22.5 cursor-default"
               onfocus={() => ($boundingBoxesArray = [peopleWithFaces[index]])}
-              onmouseover={() => ($boundingBoxesArray = [peopleWithFaces[index]])}
-              onmouseleave={() => ($boundingBoxesArray = [])}
+              onpointerover={() => ($boundingBoxesArray = [peopleWithFaces[index]])}
+              onpointerleave={() => ($boundingBoxesArray = [])}
             >
               <div class="relative">
                 {#if selectedPersonToCreate[face.id]}

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

@@ -1,4 +1,14 @@
-import { getContentMetrics, getNaturalSize, scaleToFit } from '$lib/utils/container-utils';
+import {
+  computeContentMetrics,
+  getContentMetrics,
+  getNaturalSize,
+  mapContentRectToNatural,
+  mapContentToNatural,
+  mapNormalizedRectToContent,
+  mapNormalizedToContent,
+  scaleToCover,
+  scaleToFit,
+} from '$lib/utils/container-utils';
 
 const mockImage = (props: {
   naturalWidth: number;
@@ -92,3 +102,165 @@ describe('getNaturalSize', () => {
     expect(getNaturalSize(video)).toEqual({ width: 1920, height: 1080 });
   });
 });
+
+describe('scaleToCover', () => {
+  it('should scale up to cover container when image is smaller', () => {
+    expect(scaleToCover({ width: 400, height: 300 }, { width: 800, height: 600 })).toEqual({
+      width: 800,
+      height: 600,
+    });
+  });
+
+  it('should use height scale when image is wider than container', () => {
+    expect(scaleToCover({ width: 2000, height: 1000 }, { width: 800, height: 600 })).toEqual({
+      width: 1200,
+      height: 600,
+    });
+  });
+
+  it('should use width scale when image is taller than container', () => {
+    expect(scaleToCover({ width: 1000, height: 2000 }, { width: 800, height: 600 })).toEqual({
+      width: 800,
+      height: 1600,
+    });
+  });
+});
+
+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 };
+
+  it('should map top-left corner', () => {
+    expect(mapNormalizedToContent({ x: 0, y: 0 }, metrics)).toEqual({ x: 0, y: 100 });
+  });
+
+  it('should map bottom-right corner', () => {
+    expect(mapNormalizedToContent({ x: 1, y: 1 }, metrics)).toEqual({ x: 800, y: 500 });
+  });
+
+  it('should map center point', () => {
+    expect(mapNormalizedToContent({ x: 0.5, y: 0.5 }, metrics)).toEqual({ x: 400, y: 300 });
+  });
+
+  it('should apply offsets correctly for letterboxed content', () => {
+    const letterboxed = { contentWidth: 300, contentHeight: 600, offsetX: 250, offsetY: 0 };
+    expect(mapNormalizedToContent({ x: 0, y: 0 }, letterboxed)).toEqual({ x: 250, y: 0 });
+    expect(mapNormalizedToContent({ x: 1, y: 1 }, letterboxed)).toEqual({ x: 550, y: 600 });
+  });
+});
+
+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 };
+
+  it('should map a normalized rect to content pixel coordinates', () => {
+    const rect = mapNormalizedRectToContent({ x: 0.25, y: 0.25 }, { x: 0.75, y: 0.75 }, metrics);
+    expect(rect).toEqual({ left: 200, top: 200, width: 400, height: 200 });
+  });
+
+  it('should map full image rect', () => {
+    const rect = mapNormalizedRectToContent({ x: 0, y: 0 }, { x: 1, y: 1 }, metrics);
+    expect(rect).toEqual({ left: 0, top: 100, width: 800, height: 400 });
+  });
+
+  it('should handle letterboxed content with horizontal offsets', () => {
+    const letterboxed = { contentWidth: 300, contentHeight: 600, offsetX: 250, offsetY: 0 };
+    const rect = mapNormalizedRectToContent({ x: 0, y: 0 }, { x: 1, y: 1 }, letterboxed);
+    expect(rect).toEqual({ left: 250, top: 0, width: 300, height: 600 });
+  });
+});
+
+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

@@ -1,3 +1,27 @@
+// Coordinate spaces used throughout the viewer:
+//
+// "Normalized": 0–1 range, (0,0) = top-left, (1,1) = bottom-right. Resolution-independent.
+//   Example: OCR coordinates, or face coords after dividing by metadata dimensions.
+//
+// "Content": pixel position within the container after scaling (scaleToFit/scaleToCover)
+//   and centering. Used for DOM overlay positioning (face boxes, OCR text).
+//
+// "Natural": pixel position in the original full-resolution image file (e.g. 4000×3000).
+//   Used when cropping or drawing on the source image.
+//
+// "Metadata pixel space": coordinates from face detection / OCR models, in pixels relative
+//   to face.imageWidth/imageHeight. Divide by those dimensions to get normalized coords.
+
+export interface Point {
+  x: number;
+  y: number;
+}
+
+export interface Size {
+  width: number;
+  height: number;
+}
+
 export interface ContentMetrics {
   contentWidth: number;
   contentHeight: number;
@@ -5,10 +29,7 @@ export interface ContentMetrics {
   offsetY: number;
 }
 
-export const scaleToCover = (
-  dimensions: { width: number; height: number },
-  container: { width: number; height: number },
-): { width: number; height: number } => {
+export const scaleToCover = (dimensions: Size, container: Size): Size => {
   const scaleX = container.width / dimensions.width;
   const scaleY = container.height / dimensions.height;
   const scale = Math.max(scaleX, scaleY);
@@ -18,10 +39,7 @@ export const scaleToCover = (
   };
 };
 
-export const scaleToFit = (
-  dimensions: { width: number; height: number },
-  container: { width: number; height: number },
-): { width: number; height: number } => {
+export const scaleToFit = (dimensions: Size, container: Size): Size => {
   const scaleX = container.width / dimensions.width;
   const scaleY = container.height / dimensions.height;
   const scale = Math.min(scaleX, scaleY);
@@ -31,28 +49,83 @@ export const scaleToFit = (
   };
 };
 
-const getElementSize = (element: HTMLImageElement | HTMLVideoElement): { width: number; height: number } => {
+const getElementSize = (element: HTMLImageElement | HTMLVideoElement): Size => {
   if (element instanceof HTMLVideoElement) {
     return { width: element.clientWidth, height: element.clientHeight };
   }
   return { width: element.width, height: element.height };
 };
 
-export const getNaturalSize = (element: HTMLImageElement | HTMLVideoElement): { width: number; height: number } => {
+export const getNaturalSize = (element: HTMLImageElement | HTMLVideoElement): Size => {
   if (element instanceof HTMLVideoElement) {
     return { width: element.videoWidth, height: element.videoHeight };
   }
   return { width: element.naturalWidth, height: element.naturalHeight };
 };
 
-export const getContentMetrics = (element: HTMLImageElement | HTMLVideoElement): ContentMetrics => {
-  const natural = getNaturalSize(element);
-  const client = getElementSize(element);
-  const { width: contentWidth, height: contentHeight } = scaleToFit(natural, client);
+export function computeContentMetrics(
+  imageSize: Size,
+  containerSize: Size,
+  scaleFn: (dimensions: Size, container: Size) => Size = scaleToFit,
+) {
+  const { width: contentWidth, height: contentHeight } = scaleFn(imageSize, containerSize);
   return {
     contentWidth,
     contentHeight,
-    offsetX: (client.width - contentWidth) / 2,
-    offsetY: (client.height - contentHeight) / 2,
+    offsetX: (containerSize.width - contentWidth) / 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, metrics: ContentMetrics): Point {
+  return {
+    x: point.x * metrics.contentWidth + metrics.offsetX,
+    y: point.y * metrics.contentHeight + metrics.offsetY,
+  };
+}
+
+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 interface Rect {
+  top: number;
+  left: number;
+  width: number;
+  height: number;
+}
+
+export function mapNormalizedRectToContent(topLeft: Point, bottomRight: Point, metrics: ContentMetrics): Rect {
+  const tl = mapNormalizedToContent(topLeft, metrics);
+  const br = mapNormalizedToContent(bottomRight, metrics);
+  return {
+    top: tl.y,
+    left: tl.x,
+    width: br.x - tl.x,
+    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/ocr-utils.spec.ts

@@ -1,5 +1,5 @@
 import type { OcrBoundingBox } from '$lib/stores/ocr.svelte';
-import type { ContentMetrics } from '$lib/utils/container-utils';
+import type { Size } from '$lib/utils/container-utils';
 import { getOcrBoundingBoxes } from '$lib/utils/ocr-utils';
 
 describe('getOcrBoundingBoxes', () => {
@@ -21,9 +21,9 @@ describe('getOcrBoundingBoxes', () => {
         text: 'hello',
       },
     ];
-    const metrics: ContentMetrics = { contentWidth: 1000, contentHeight: 500, offsetX: 0, offsetY: 0 };
+    const imageSize: Size = { width: 1000, height: 500 };
 
-    const boxes = getOcrBoundingBoxes(ocrData, metrics);
+    const boxes = getOcrBoundingBoxes(ocrData, imageSize);
 
     expect(boxes).toHaveLength(1);
     expect(boxes[0].id).toBe('box1');
@@ -37,7 +37,7 @@ describe('getOcrBoundingBoxes', () => {
     ]);
   });
 
-  it('should apply offsets for letterboxed images', () => {
+  it('should map full-image box to full display area', () => {
     const ocrData: OcrBoundingBox[] = [
       {
         id: 'box1',
@@ -55,21 +55,20 @@ describe('getOcrBoundingBoxes', () => {
         text: 'test',
       },
     ];
-    const metrics: ContentMetrics = { contentWidth: 600, contentHeight: 400, offsetX: 100, offsetY: 50 };
+    const imageSize: Size = { width: 600, height: 400 };
 
-    const boxes = getOcrBoundingBoxes(ocrData, metrics);
+    const boxes = getOcrBoundingBoxes(ocrData, imageSize);
 
     expect(boxes[0].points).toEqual([
-      { x: 100, y: 50 },
-      { x: 700, y: 50 },
-      { x: 700, y: 450 },
-      { x: 100, y: 450 },
+      { x: 0, y: 0 },
+      { x: 600, y: 0 },
+      { x: 600, y: 400 },
+      { x: 0, y: 400 },
     ]);
   });
 
   it('should return empty array for empty input', () => {
-    const metrics: ContentMetrics = { contentWidth: 800, contentHeight: 600, offsetX: 0, offsetY: 0 };
-    expect(getOcrBoundingBoxes([], metrics)).toEqual([]);
+    expect(getOcrBoundingBoxes([], { width: 800, height: 600 })).toEqual([]);
   });
 
   it('should handle multiple boxes', () => {
@@ -105,9 +104,9 @@ describe('getOcrBoundingBoxes', () => {
         text: 'second',
       },
     ];
-    const metrics: ContentMetrics = { contentWidth: 200, contentHeight: 200, offsetX: 0, offsetY: 0 };
+    const imageSize: Size = { width: 200, height: 200 };
 
-    const boxes = getOcrBoundingBoxes(ocrData, metrics);
+    const boxes = getOcrBoundingBoxes(ocrData, imageSize);
 
     expect(boxes).toHaveLength(2);
     expect(boxes[0].text).toBe('first');

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

@@ -1,11 +1,7 @@
 import type { OcrBoundingBox } from '$lib/stores/ocr.svelte';
-import type { ContentMetrics } from '$lib/utils/container-utils';
+import { mapNormalizedToContent, type ContentMetrics, type Point, type Size } from '$lib/utils/container-utils';
 import { clamp } from 'lodash-es';
-
-export type Point = {
-  x: number;
-  y: number;
-};
+export type { Point } from '$lib/utils/container-utils';
 
 const distance = (p1: Point, p2: Point) => Math.hypot(p2.x - p1.x, p2.y - p1.y);
 
@@ -38,7 +34,7 @@ const getVerticalMode = (width: number, height: number, text: string): VerticalM
  * @param points - Array of 4 corner points of the bounding box
  * @returns 4x4 matrix to transform the div with text onto the polygon defined by the corner points, and size to set on the source div.
  */
-export const calculateBoundingBoxMatrix = (points: Point[]): { matrix: number[]; width: number; height: number } => {
+export const calculateBoundingBoxMatrix = (points: Point[]): Size & { matrix: number[] } => {
   const [topLeft, topRight, bottomRight, bottomLeft] = points;
 
   const width = Math.max(distance(topLeft, topRight), distance(bottomLeft, bottomRight));
@@ -163,7 +159,13 @@ export const calculateFittedFontSize = (
   return clamp(Math.min(scaleFromWidth, scaleFromHeight), MIN_FONT_SIZE, MAX_FONT_SIZE);
 };
 
-export const getOcrBoundingBoxes = (ocrData: OcrBoundingBox[], metrics: ContentMetrics): OcrBox[] => {
+export const getOcrBoundingBoxes = (ocrData: OcrBoundingBox[], imageSize: Size): OcrBox[] => {
+  const metrics: ContentMetrics = {
+    contentWidth: imageSize.width,
+    contentHeight: imageSize.height,
+    offsetX: 0,
+    offsetY: 0,
+  };
   const boxes: OcrBox[] = [];
   for (const ocr of ocrData) {
     const points = [
@@ -171,10 +173,7 @@ export const getOcrBoundingBoxes = (ocrData: OcrBoundingBox[], metrics: ContentM
       { x: ocr.x2, y: ocr.y2 },
       { x: ocr.x3, y: ocr.y3 },
       { x: ocr.x4, y: ocr.y4 },
-    ].map((point) => ({
-      x: point.x * metrics.contentWidth + metrics.offsetX,
-      y: point.y * metrics.contentHeight + metrics.offsetY,
-    }));
+    ].map((point) => mapNormalizedToContent(point, metrics));
 
     const boxWidth = Math.max(distance(points[0], points[1]), distance(points[3], points[2]));
     const boxHeight = Math.max(distance(points[0], points[3]), distance(points[1], points[2]));
@@ -188,7 +187,7 @@ export const getOcrBoundingBoxes = (ocrData: OcrBoundingBox[], metrics: ContentM
     });
   }
 
-  const rowThreshold = metrics.contentHeight * 0.02;
+  const rowThreshold = imageSize.height * 0.02;
   boxes.sort((a, b) => {
     const yDifference = a.points[0].y - b.points[0].y;
     if (Math.abs(yDifference) < rowThreshold) {

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

@@ -1,6 +1,6 @@
 import type { Faces } from '$lib/stores/people.store';
-import type { ContentMetrics } from '$lib/utils/container-utils';
-import { getBoundingBox } from '$lib/utils/people-utils';
+import type { Size } from '$lib/utils/container-utils';
+import { getBoundingBox, scaleFaceRectOnResize, type FaceRectState, type ResizeContext } from '$lib/utils/people-utils';
 
 const makeFace = (overrides: Partial<Faces> = {}): Faces => ({
   id: 'face-1',
@@ -16,21 +16,21 @@ const makeFace = (overrides: Partial<Faces> = {}): Faces => ({
 describe('getBoundingBox', () => {
   it('should scale face coordinates to display dimensions', () => {
     const face = makeFace();
-    const metrics: ContentMetrics = { contentWidth: 800, contentHeight: 600, offsetX: 0, offsetY: 0 };
+    const imageSize: Size = { width: 800, height: 600 };
 
-    const boxes = getBoundingBox([face], metrics);
+    const boxes = getBoundingBox([face], imageSize);
 
     expect(boxes).toHaveLength(1);
     expect(boxes[0]).toEqual({
       id: 'face-1',
-      top: Math.round(600 * (750 / 3000)),
-      left: Math.round(800 * (1000 / 4000)),
-      width: Math.round(800 * (2000 / 4000) - 800 * (1000 / 4000)),
-      height: Math.round(600 * (1500 / 3000) - 600 * (750 / 3000)),
+      top: 600 * (750 / 3000),
+      left: 800 * (1000 / 4000),
+      width: 800 * (2000 / 4000) - 800 * (1000 / 4000),
+      height: 600 * (1500 / 3000) - 600 * (750 / 3000),
     });
   });
 
-  it('should apply offsets for letterboxed display', () => {
+  it('should map full-image face to full display area', () => {
     const face = makeFace({
       imageWidth: 1000,
       imageHeight: 1000,
@@ -39,49 +39,21 @@ describe('getBoundingBox', () => {
       boundingBoxX2: 1000,
       boundingBoxY2: 1000,
     });
-    const metrics: ContentMetrics = { contentWidth: 600, contentHeight: 600, offsetX: 100, offsetY: 0 };
+    const imageSize: Size = { width: 600, height: 600 };
 
-    const boxes = getBoundingBox([face], metrics);
+    const boxes = getBoundingBox([face], imageSize);
 
     expect(boxes[0]).toEqual({
       id: 'face-1',
       top: 0,
-      left: 100,
+      left: 0,
       width: 600,
       height: 600,
     });
   });
 
-  it('should handle zoom by pre-scaled metrics', () => {
-    const face = makeFace({
-      imageWidth: 1000,
-      imageHeight: 1000,
-      boundingBoxX1: 0,
-      boundingBoxY1: 0,
-      boundingBoxX2: 500,
-      boundingBoxY2: 500,
-    });
-    const metrics: ContentMetrics = {
-      contentWidth: 1600,
-      contentHeight: 1200,
-      offsetX: -200,
-      offsetY: -100,
-    };
-
-    const boxes = getBoundingBox([face], metrics);
-
-    expect(boxes[0]).toEqual({
-      id: 'face-1',
-      top: -100,
-      left: -200,
-      width: 800,
-      height: 600,
-    });
-  });
-
   it('should return empty array for empty faces', () => {
-    const metrics: ContentMetrics = { contentWidth: 800, contentHeight: 600, offsetX: 0, offsetY: 0 };
-    expect(getBoundingBox([], metrics)).toEqual([]);
+    expect(getBoundingBox([], { width: 800, height: 600 })).toEqual([]);
   });
 
   it('should handle multiple faces', () => {
@@ -89,11 +61,103 @@ describe('getBoundingBox', () => {
       makeFace({ id: 'face-1', boundingBoxX1: 0, boundingBoxY1: 0, boundingBoxX2: 1000, boundingBoxY2: 1000 }),
       makeFace({ id: 'face-2', boundingBoxX1: 2000, boundingBoxY1: 1500, boundingBoxX2: 3000, boundingBoxY2: 2500 }),
     ];
-    const metrics: ContentMetrics = { contentWidth: 800, contentHeight: 600, offsetX: 0, offsetY: 0 };
 
-    const boxes = getBoundingBox(faces, metrics);
+    const boxes = getBoundingBox(faces, { width: 800, height: 600 });
 
     expect(boxes).toHaveLength(2);
     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

@@ -1,42 +1,58 @@
 import type { Faces } from '$lib/stores/people.store';
 import { getAssetMediaUrl } from '$lib/utils';
-import type { ContentMetrics } 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 interface BoundingBox {
-  id: string;
-  top: number;
-  left: number;
-  width: number;
-  height: number;
-}
+export type BoundingBox = Rect & { id: string };
 
-export const getBoundingBox = (faces: Faces[], metrics: ContentMetrics): BoundingBox[] => {
+export const getBoundingBox = (faces: Faces[], imageSize: Size): BoundingBox[] => {
+  const metrics: ContentMetrics = {
+    contentWidth: imageSize.width,
+    contentHeight: imageSize.height,
+    offsetX: 0,
+    offsetY: 0,
+  };
   const boxes: BoundingBox[] = [];
 
   for (const face of faces) {
-    const scaleX = metrics.contentWidth / face.imageWidth;
-    const scaleY = metrics.contentHeight / face.imageHeight;
+    const rect = mapNormalizedRectToContent(
+      { x: face.boundingBoxX1 / face.imageWidth, y: face.boundingBoxY1 / face.imageHeight },
+      { x: face.boundingBoxX2 / face.imageWidth, y: face.boundingBoxY2 / face.imageHeight },
+      metrics,
+    );
 
-    const coordinates = {
-      x1: scaleX * face.boundingBoxX1 + metrics.offsetX,
-      x2: scaleX * face.boundingBoxX2 + metrics.offsetX,
-      y1: scaleY * face.boundingBoxY1 + metrics.offsetY,
-      y2: scaleY * face.boundingBoxY2 + metrics.offsetY,
-    };
-
-    boxes.push({
-      id: face.id,
-      top: Math.round(coordinates.y1),
-      left: Math.round(coordinates.x1),
-      width: Math.round(coordinates.x2 - coordinates.x1),
-      height: Math.round(coordinates.y2 - coordinates.y1),
-    });
+    boxes.push({ id: face.id, ...rect });
   }
 
   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,