Release OpenScreen 1.4.4

docs/engineering/paddleocr-local-service.md

@@ -6,7 +6,7 @@ OpenScreen calls OCR through a local HTTP service. The default endpoint is:
 http://127.0.0.1:8866/ocr
 ```
 
-The app sends either `imageBase64` or `path` and expects OCR blocks:
+The app sends either `imageBase64` or `path`, plus optional `language` and `profile`, and expects OCR blocks:
 
 ```json
 {
@@ -38,7 +38,7 @@ If `paddle` is still missing after installing `paddleocr`, install the CPU Paddl
 ```powershell
 .\.venv-ocr\Scripts\Activate.ps1
 $env:PADDLEOCR_DEVICE="cpu"
-$env:PADDLEOCR_LANG="latin"
+$env:OPENSCREEN_OCR_PROFILE="vietnamese"
 npm run ocr:paddle
 ```
 
@@ -58,7 +58,8 @@ Expected healthy environment:
   "paddleocrInstalled": true,
   "paddleInstalled": true,
   "engineReady": false,
-  "defaultLanguage": "latin"
+  "defaultLanguage": "vi,en",
+  "defaultProfile": "vietnamese"
 }
 ```
 
@@ -67,7 +68,10 @@ Expected healthy environment:
 ## Configuration
 
 - `PADDLEOCR_DEVICE`: `cpu`, `gpu:0`, or another PaddleOCR device string.
-- `PADDLEOCR_LANG`: defaults to `latin`; this is preferred for Vietnamese UI text because it uses a Latin-script recognition model.
+- `OPENSCREEN_OCR_PROFILE`: `fast`, `vietnamese`, or `hybrid`. The default `vietnamese` profile upscales and sharpens focused UI screenshots before OCR.
+- `OPENSCREEN_GUIDE_OCR_LANGUAGE`: defaults to `vi,en`.
+- `PADDLEOCR_LANG`: optional hard override. Leave unset for the app profile/language settings to work.
 - `PADDLEOCR_VERSION`: defaults to `PP-OCRv5`.
 - `PADDLEOCR_USE_MOBILE`: defaults to `1`; set to `0` to use the default/server models.
+- `PADDLEOCR_REC_MODEL`: optional recognizer model override. The bundled profile uses `latin_PP-OCRv5_mobile_rec`, which supports Vietnamese Latin-script text.
 - `OPENSCREEN_GUIDE_OCR_URL`: OpenScreen OCR endpoint override; defaults to `http://127.0.0.1:8866`.

electron/guide/ai/deepseekSettingsStore.test.ts

@@ -0,0 +1,66 @@
+import fs from "node:fs/promises";
+import os from "node:os";
+import path from "node:path";
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
+import { DeepSeekSettingsStore } from "./deepseekSettingsStore";
+
+const tempDirs: string[] = [];
+const originalOcrProfile = process.env.OPENSCREEN_GUIDE_OCR_PROFILE;
+const originalOcrLanguage = process.env.OPENSCREEN_GUIDE_OCR_LANGUAGE;
+
+beforeEach(() => {
+	delete process.env.OPENSCREEN_GUIDE_OCR_PROFILE;
+	delete process.env.OPENSCREEN_GUIDE_OCR_LANGUAGE;
+});
+
+afterEach(async () => {
+	restoreEnv("OPENSCREEN_GUIDE_OCR_PROFILE", originalOcrProfile);
+	restoreEnv("OPENSCREEN_GUIDE_OCR_LANGUAGE", originalOcrLanguage);
+	await Promise.all(tempDirs.splice(0).map((dir) => fs.rm(dir, { recursive: true, force: true })));
+});
+
+function restoreEnv(name: string, value: string | undefined): void {
+	if (value === undefined) {
+		delete process.env[name];
+		return;
+	}
+	process.env[name] = value;
+}
+
+async function createStore(): Promise<DeepSeekSettingsStore> {
+	const dir = await fs.mkdtemp(path.join(os.tmpdir(), "openscreen-guide-settings-"));
+	tempDirs.push(dir);
+	return new DeepSeekSettingsStore(path.join(dir, "guide-ai-settings.json"));
+}
+
+describe("DeepSeekSettingsStore OCR settings", () => {
+	it("defaults to the Vietnamese enhanced OCR profile", async () => {
+		const store = await createStore();
+
+		await expect(store.getOcrConfig()).resolves.toEqual({
+			profile: "vietnamese",
+			language: "vi,en",
+		});
+	});
+
+	it("persists OCR profile changes alongside DeepSeek settings", async () => {
+		const store = await createStore();
+
+		const status = await store.save({
+			deepseekApiKeyEnvName: "DEEPSEEK_API_KEY",
+			baseUrl: "https://api.deepseek.com",
+			model: "deepseek-chat",
+			ocrProfile: "hybrid",
+			ocrLanguage: "vi,en",
+		});
+
+		expect(status.ocr).toMatchObject({
+			profile: "hybrid",
+			language: "vi,en",
+		});
+		await expect(store.getOcrConfig()).resolves.toEqual({
+			profile: "hybrid",
+			language: "vi,en",
+		});
+	});
+});

electron/guide/ai/deepseekSettingsStore.ts

@@ -1,6 +1,10 @@
 import fs from "node:fs/promises";
 import path from "node:path";
-import type { GuideAiSettings, SaveGuideAiSettingsInput } from "../../../src/guide/contracts";
+import type {
+	GuideAiSettings,
+	GuideOcrProfile,
+	SaveGuideAiSettingsInput,
+} from "../../../src/guide/contracts";
 
 export interface DeepSeekGuideConfig {
 	apiKey?: string;
@@ -12,8 +16,22 @@ export interface DeepSeekGuideConfigProvider {
 	getDeepSeekConfig(): Promise<DeepSeekGuideConfig>;
 }
 
+export interface GuideOcrConfig {
+	profile: GuideOcrProfile;
+	language: string;
+}
+
+export interface GuideOcrConfigProvider {
+	getOcrConfig(): Promise<GuideOcrConfig>;
+}
+
 interface PersistedGuideAiSettings {
 	schemaVersion: 1;
+	ocr?: {
+		profile?: GuideOcrProfile;
+		language?: string;
+		updatedAt?: string;
+	};
 	deepseek?: {
 		apiKeyEnvName?: string;
 		baseUrl?: string;
@@ -25,8 +43,10 @@ interface PersistedGuideAiSettings {
 const DEFAULT_DEEPSEEK_API_KEY_ENV_NAME = "DEEPSEEK_API_KEY";
 const DEFAULT_DEEPSEEK_BASE_URL = "https://api.deepseek.com";
 const DEFAULT_DEEPSEEK_MODEL = "deepseek-chat";
+const DEFAULT_OCR_PROFILE: GuideOcrProfile = "vietnamese";
+const DEFAULT_OCR_LANGUAGE = "vi,en";
 
-export class DeepSeekSettingsStore implements DeepSeekGuideConfigProvider {
+export class DeepSeekSettingsStore implements DeepSeekGuideConfigProvider, GuideOcrConfigProvider {
 	constructor(private readonly filePath: string) {}
 
 	async getStatus(): Promise<GuideAiSettings> {
@@ -35,6 +55,13 @@ export class DeepSeekSettingsStore implements DeepSeekGuideConfigProvider {
 		const activeApiKey = process.env[apiKeyEnvName];
 
 		return {
+			ocr: {
+				profile: normalizeOcrProfile(raw?.ocr?.profile ?? process.env.OPENSCREEN_GUIDE_OCR_PROFILE),
+				language: normalizeOcrLanguage(
+					raw?.ocr?.language ?? process.env.OPENSCREEN_GUIDE_OCR_LANGUAGE,
+				),
+				updatedAt: raw?.ocr?.updatedAt,
+			},
 			deepseek: {
 				hasApiKey: Boolean(activeApiKey),
 				apiKeyEnvName,
@@ -49,7 +76,14 @@ export class DeepSeekSettingsStore implements DeepSeekGuideConfigProvider {
 
 	async save(input: SaveGuideAiSettingsInput): Promise<GuideAiSettings> {
 		const current = (await this.readSettings()) ?? { schemaVersion: 1 };
+		const currentOcr = current.ocr ?? {};
 		const currentDeepSeek = current.deepseek ?? {};
+		const nextOcr = {
+			...currentOcr,
+			profile: normalizeOcrProfile(input.ocrProfile ?? currentOcr.profile),
+			language: normalizeOcrLanguage(input.ocrLanguage ?? currentOcr.language),
+			updatedAt: new Date().toISOString(),
+		};
 		const nextDeepSeek = {
 			...currentDeepSeek,
 			baseUrl: normalizeBaseUrl(input.baseUrl ?? currentDeepSeek.baseUrl),
@@ -65,6 +99,7 @@ export class DeepSeekSettingsStore implements DeepSeekGuideConfigProvider {
 
 		await this.writeSettings({
 			schemaVersion: 1,
+			ocr: nextOcr,
 			deepseek: nextDeepSeek,
 		});
 		return await this.getStatus();
@@ -80,6 +115,16 @@ export class DeepSeekSettingsStore implements DeepSeekGuideConfigProvider {
 		};
 	}
 
+	async getOcrConfig(): Promise<GuideOcrConfig> {
+		const raw = await this.readSettings();
+		return {
+			profile: normalizeOcrProfile(raw?.ocr?.profile ?? process.env.OPENSCREEN_GUIDE_OCR_PROFILE),
+			language: normalizeOcrLanguage(
+				raw?.ocr?.language ?? process.env.OPENSCREEN_GUIDE_OCR_LANGUAGE,
+			),
+		};
+	}
+
 	private async readSettings(): Promise<PersistedGuideAiSettings | null> {
 		try {
 			const content = await fs.readFile(this.filePath, "utf-8");
@@ -120,6 +165,11 @@ function normalizePersistedSettings(input: unknown): PersistedGuideAiSettings |
 	}
 	return {
 		schemaVersion: 1,
+		ocr: {
+			profile: normalizeOcrProfile(raw.ocr?.profile),
+			language: normalizeOcrLanguage(raw.ocr?.language),
+			updatedAt: raw.ocr?.updatedAt,
+		},
 		deepseek: {
 			apiKeyEnvName: normalizeEnvName(raw.deepseek?.apiKeyEnvName),
 			baseUrl: raw.deepseek?.baseUrl,
@@ -155,3 +205,19 @@ function normalizeBaseUrl(value: string | undefined): string {
 function normalizeModel(value: string | undefined): string {
 	return value?.trim() || DEFAULT_DEEPSEEK_MODEL;
 }
+
+function normalizeOcrProfile(value: string | undefined): GuideOcrProfile {
+	if (value === "fast" || value === "vietnamese" || value === "hybrid") {
+		return value;
+	}
+	return DEFAULT_OCR_PROFILE;
+}
+
+function normalizeOcrLanguage(value: string | undefined): string {
+	const normalized = value
+		?.split(",")
+		.map((part) => part.trim().toLowerCase())
+		.filter(Boolean)
+		.join(",");
+	return normalized || DEFAULT_OCR_LANGUAGE;
+}

electron/guide/guideStore.ts

@@ -34,7 +34,10 @@ import {
 	DeepSeekGuideClientError,
 	type GuideDraftClient,
 } from "./ai/deepseekGuideClient";
-import type { DeepSeekGuideConfigProvider } from "./ai/deepseekSettingsStore";
+import type {
+	DeepSeekGuideConfigProvider,
+	GuideOcrConfigProvider,
+} from "./ai/deepseekSettingsStore";
 import { type GuidePaths, normalizeGuideRecordingId, resolveGuidePaths } from "./guidePaths";
 import { createFocusedOcrSnapshot, remapFocusedOcrBlocks } from "./ocr/focusedOcrSnapshot";
 import { DefaultGuideOcrClient, type GuideOcrClient } from "./ocr/paddleOcrClient";
@@ -70,6 +73,7 @@ export interface GuideStoreDependencies {
 	ocrClient?: GuideOcrClient;
 	draftClient?: GuideDraftClient;
 	deepSeekConfigProvider?: DeepSeekGuideConfigProvider;
+	ocrConfigProvider?: GuideOcrConfigProvider;
 	focusOcrSnapshots?: boolean;
 }
 
@@ -255,7 +259,9 @@ export class GuideStore {
 			throw new GuideStoreError("guide-invalid-input", "No guide snapshots are available for OCR.");
 		}
 
-		const ocrClient = this.dependencies.ocrClient ?? new DefaultGuideOcrClient();
+		const ocrClient =
+			this.dependencies.ocrClient ??
+			DefaultGuideOcrClient.fromConfig(await this.dependencies.ocrConfigProvider?.getOcrConfig());
 		const shouldFocusOcrSnapshots =
 			this.dependencies.focusOcrSnapshots ?? this.dependencies.ocrClient === undefined;
 		const eventsById = new Map(session.events.map((event) => [event.id, event]));

electron/guide/ocr/bundledOcrService.ts

@@ -156,8 +156,10 @@ function startOcrServiceProcess(
 			OPENSCREEN_OCR_PORT: DEFAULT_OCR_PORT,
 			PADDLEOCR_DEVICE: process.env.PADDLEOCR_DEVICE ?? "cpu",
 			PADDLEOCR_ENABLE_MKLDNN: process.env.PADDLEOCR_ENABLE_MKLDNN ?? "0",
-			PADDLEOCR_LANG: process.env.PADDLEOCR_LANG ?? "latin",
+			PADDLEOCR_LANG: process.env.PADDLEOCR_LANG ?? "",
 			PADDLEOCR_USE_MOBILE: process.env.PADDLEOCR_USE_MOBILE ?? "1",
+			OPENSCREEN_OCR_PROFILE:
+				process.env.OPENSCREEN_OCR_PROFILE ?? process.env.OPENSCREEN_GUIDE_OCR_PROFILE ?? "",
 			PADDLE_PDX_ENABLE_MKLDNN_BYDEFAULT: process.env.PADDLE_PDX_ENABLE_MKLDNN_BYDEFAULT ?? "False",
 			PADDLE_PDX_CACHE_HOME: process.env.PADDLE_PDX_CACHE_HOME ?? runtimePaths.paddlexCachePath,
 			PADDLE_PDX_DISABLE_MODEL_SOURCE_CHECK:

electron/guide/ocr/paddleOcrClient.test.ts

@@ -1,8 +1,12 @@
-import { describe, expect, it } from "vitest";
+import fs from "node:fs/promises";
+import os from "node:os";
+import path from "node:path";
+import { afterEach, describe, expect, it, vi } from "vitest";
 import type { GuideSnapshot, OcrBlock } from "../../../src/guide/contracts";
 import {
 	DefaultGuideOcrClient,
 	normalizeOcrResponse,
+	PaddleOcrHttpClient,
 	parseWindowsOcrPayload,
 } from "./paddleOcrClient";
 
@@ -16,6 +20,10 @@ const snapshot: GuideSnapshot = {
 	height: 800,
 };
 
+afterEach(() => {
+	vi.unstubAllGlobals();
+});
+
 describe("normalizeOcrResponse", () => {
 	it("normalizes pixel boxes into guide OCR blocks", () => {
 		const blocks = normalizeOcrResponse(
@@ -67,6 +75,35 @@ describe("normalizeOcrResponse", () => {
 	});
 });
 
+describe("PaddleOcrHttpClient", () => {
+	it("sends the selected OCR profile to the local service", async () => {
+		const tempDir = await fs.mkdtemp(path.join(os.tmpdir(), "openscreen-ocr-client-"));
+		const imagePath = path.join(tempDir, "step.png");
+		await fs.writeFile(imagePath, Buffer.from([137, 80, 78, 71]));
+		const requests: unknown[] = [];
+		vi.stubGlobal(
+			"fetch",
+			vi.fn(async (_url: string, init?: RequestInit) => {
+				requests.push(JSON.parse(String(init?.body ?? "{}")));
+				return new Response(JSON.stringify({ blocks: [] }), {
+					status: 200,
+					headers: { "content-type": "application/json" },
+				});
+			}),
+		);
+
+		const client = new PaddleOcrHttpClient("https://ocr.example.test", "vi,en", "hybrid");
+		await client.recognize({ ...snapshot, path: imagePath });
+
+		expect(requests[0]).toMatchObject({
+			language: "vi,en",
+			profile: "hybrid",
+			path: imagePath,
+		});
+		await fs.rm(tempDir, { recursive: true, force: true });
+	});
+});
+
 describe("DefaultGuideOcrClient", () => {
 	it("falls back when the HTTP OCR service is unavailable", async () => {
 		const fallbackBlock: OcrBlock = {

electron/guide/ocr/paddleOcrClient.ts

@@ -1,7 +1,7 @@
 import { execFile } from "node:child_process";
 import fs from "node:fs/promises";
 import { promisify } from "node:util";
-import type { GuideSnapshot, OcrBlock } from "../../../src/guide/contracts";
+import type { GuideOcrProfile, GuideSnapshot, OcrBlock } from "../../../src/guide/contracts";
 import { ensureBundledOcrServiceRunning } from "./bundledOcrService";
 
 const execFileAsync = promisify(execFile);
@@ -10,6 +10,11 @@ export interface GuideOcrClient {
 	recognize(snapshot: GuideSnapshot): Promise<OcrBlock[]>;
 }
 
+export interface GuideOcrClientConfig {
+	profile: GuideOcrProfile;
+	language: string;
+}
+
 interface PaddleOcrResponseBlock {
 	text?: unknown;
 	confidence?: unknown;
@@ -21,7 +26,8 @@ interface PaddleOcrResponseBlock {
 export class PaddleOcrHttpClient implements GuideOcrClient {
 	constructor(
 		private readonly baseUrl = process.env.OPENSCREEN_GUIDE_OCR_URL ?? "http://127.0.0.1:8866",
-		private readonly language = process.env.OPENSCREEN_GUIDE_OCR_LANGUAGE ?? "vi,en",
+		private readonly language = normalizeOcrLanguage(process.env.OPENSCREEN_GUIDE_OCR_LANGUAGE),
+		private readonly profile = normalizeOcrProfile(process.env.OPENSCREEN_GUIDE_OCR_PROFILE),
 	) {}
 
 	async recognize(snapshot: GuideSnapshot): Promise<OcrBlock[]> {
@@ -36,6 +42,7 @@ export class PaddleOcrHttpClient implements GuideOcrClient {
 					imageBase64,
 					path: snapshot.path,
 					language: this.language,
+					profile: this.profile,
 				}),
 			});
 		} catch (error) {
@@ -54,7 +61,9 @@ export class PaddleOcrHttpClient implements GuideOcrClient {
 }
 
 export class WindowsOcrClient implements GuideOcrClient {
-	constructor(private readonly language = process.env.OPENSCREEN_GUIDE_OCR_LANGUAGE ?? "vi,en") {}
+	constructor(
+		private readonly language = normalizeOcrLanguage(process.env.OPENSCREEN_GUIDE_OCR_LANGUAGE),
+	) {}
 
 	async recognize(snapshot: GuideSnapshot): Promise<OcrBlock[]> {
 		if (process.platform !== "win32") {
@@ -96,6 +105,14 @@ export class WindowsOcrClient implements GuideOcrClient {
 }
 
 export class DefaultGuideOcrClient implements GuideOcrClient {
+	static fromConfig(config?: Partial<GuideOcrClientConfig>): DefaultGuideOcrClient {
+		const normalizedConfig = normalizeOcrClientConfig(config);
+		return new DefaultGuideOcrClient(
+			new PaddleOcrHttpClient(undefined, normalizedConfig.language, normalizedConfig.profile),
+			new WindowsOcrClient(normalizedConfig.language),
+		);
+	}
+
 	constructor(
 		private readonly httpClient = new PaddleOcrHttpClient(),
 		private readonly windowsClient = new WindowsOcrClient(),
@@ -119,6 +136,31 @@ export class DefaultGuideOcrClient implements GuideOcrClient {
 	}
 }
 
+function normalizeOcrClientConfig(
+	config: Partial<GuideOcrClientConfig> | undefined,
+): GuideOcrClientConfig {
+	return {
+		profile: normalizeOcrProfile(config?.profile ?? process.env.OPENSCREEN_GUIDE_OCR_PROFILE),
+		language: normalizeOcrLanguage(config?.language ?? process.env.OPENSCREEN_GUIDE_OCR_LANGUAGE),
+	};
+}
+
+function normalizeOcrProfile(value: string | undefined): GuideOcrProfile {
+	if (value === "fast" || value === "vietnamese" || value === "hybrid") {
+		return value;
+	}
+	return "vietnamese";
+}
+
+function normalizeOcrLanguage(value: string | undefined): string {
+	const normalized = value
+		?.split(",")
+		.map((part) => part.trim().toLowerCase())
+		.filter(Boolean)
+		.join(",");
+	return normalized || "vi,en";
+}
+
 export function parseWindowsOcrPayload(stdout: string): unknown {
 	const normalized = stdout.replace(/^\uFEFF/, "").trim();
 	try {

electron/ipc/handlers.ts

@@ -1732,7 +1732,7 @@ export function registerIpcHandlers(
 		const sources = await desktopCapturer.getSources(opts);
 		lastEnumeratedSources = new Map(sources.map((source) => [source.id, source]));
 		let screenSourceIndex = 0;
-		return sources.map((source) => {
+		const processedSources = sources.map((source) => {
 			const isScreenSource = source.id.startsWith("screen:");
 			const sourceIndex = isScreenSource
 				? (parseDesktopCapturerScreenIndex(source.id) ?? screenSourceIndex)
@@ -1760,6 +1760,43 @@ export function registerIpcHandlers(
 				bounds,
 			};
 		});
+		const screenDisplays = screen.getAllDisplays();
+		const mappedDisplayIds = new Set(
+			processedSources
+				.filter((source) => source.id.startsWith("screen:") && typeof source.displayId === "number")
+				.map((source) => source.displayId),
+		);
+		const fallbackScreenSources = screenDisplays
+			.map((display, displayIndex) => ({ display, displayIndex }))
+			.filter(({ display }) => !mappedDisplayIds.has(display.id))
+			.map(({ display, displayIndex }) => {
+				const bounds = toSourceBounds(display.bounds);
+				return {
+					id: `screen:${displayIndex}:fallback:${display.id}`,
+					name: `Screen ${displayIndex + 1}`,
+					display_id: String(display.id),
+					thumbnail: null,
+					appIcon: null,
+					displayId: display.id,
+					displayIndex,
+					screenIndex: displayIndex,
+					displayLabel: `Display ${displayIndex + 1} - ${bounds.width}x${bounds.height} @ ${bounds.x},${bounds.y}`,
+					bounds,
+				};
+			});
+		if (fallbackScreenSources.length > 0) {
+			console.warn("[desktop-capturer] added fallback display sources", {
+				capturerScreens: processedSources.filter((source) => source.id.startsWith("screen:"))
+					.length,
+				electronDisplays: screenDisplays.length,
+				fallbackScreens: fallbackScreenSources.map((source) => ({
+					id: source.id,
+					displayId: source.displayId,
+					bounds: source.bounds,
+				})),
+			});
+		}
+		return [...processedSources, ...fallbackScreenSources];
 	});
 
 	ipcMain.handle("select-source", async (_, source: SelectedSource) => {
@@ -2637,6 +2674,7 @@ export function registerIpcHandlers(
 	);
 	const guideStore = new GuideStore(RECORDINGS_DIR, {
 		deepSeekConfigProvider: guideAiSettingsStore,
+		ocrConfigProvider: guideAiSettingsStore,
 	});
 	registerGuideMarkerHotkey(guideStore);
 	registerGuideIpcHandlers(ipcMain, guideStore, guideAiSettingsStore, {

package-lock.json

@@ -1,12 +1,12 @@
 {
 	"name": "openscreen",
-	"version": "1.4.2",
+	"version": "1.4.4",
 	"lockfileVersion": 3,
 	"requires": true,
 	"packages": {
 		"": {
 			"name": "openscreen",
-			"version": "1.4.2",
+			"version": "1.4.4",
 			"dependencies": {
 				"@fix-webm-duration/fix": "^1.0.1",
 				"@pixi/filter-drop-shadow": "^5.2.0",

package.json

@@ -1,7 +1,7 @@
 {
 	"name": "openscreen",
 	"private": true,
-	"version": "1.4.2",
+	"version": "1.4.4",
 	"type": "module",
 	"packageManager": "npm@10.9.4",
 	"engines": {

src/components/launch/SourceSelector.tsx

@@ -65,7 +65,13 @@ export function SourceSelector() {
 		fetchSources();
 	}, []);
 
-	const screenSources = sources.filter((s) => s.id.startsWith("screen:"));
+	const screenSources = sources
+		.filter((s) => s.id.startsWith("screen:"))
+		.sort(
+			(left, right) =>
+				(left.displayIndex ?? left.screenIndex ?? Number.MAX_SAFE_INTEGER) -
+				(right.displayIndex ?? right.screenIndex ?? Number.MAX_SAFE_INTEGER),
+		);
 	const windowSources = sources.filter((s) => s.id.startsWith("window:"));
 
 	const handleSourceSelect = (source: DesktopSource) => setSelectedSource(source);
@@ -96,11 +102,17 @@ export function SourceSelector() {
 				onClick={() => handleSourceSelect(source)}
 			>
 				<div className="relative mb-1.5 overflow-hidden rounded-lg border border-white/[0.06] bg-black/30">
+					{source.thumbnail ? (
 						<img
-						src={source.thumbnail || ""}
+							src={source.thumbnail}
 							alt={source.name}
 							className="w-full aspect-video object-cover"
 						/>
+					) : (
+						<div className="flex aspect-video w-full items-center justify-center bg-zinc-950 text-center text-[11px] font-medium text-zinc-400">
+							{source.displayLabel ?? source.name}
+						</div>
+					)}
 					{isSelected && (
 						<div className="absolute right-1.5 top-1.5">
 							<div className={styles.checkBadge}>

src/components/video-editor/guide/GuidePanel.tsx

@@ -7,6 +7,7 @@ import type {
 	GuideAiProvider,
 	GuideAiSettings,
 	GuideLanguage,
+	GuideOcrProfile,
 	GuideSession,
 } from "@/guide/contracts";
 import { captureGuideSnapshots } from "@/guide/snapshot/extractGuideSnapshots";
@@ -42,13 +43,19 @@ const COPY = {
 		captureStep: "Capture step",
 		captureLabel: "Manual capture",
 		settings: "Settings",
+		guideSettings: "Guide settings",
 		apiKey: "API key env",
 		apiKeyPlaceholder: "DEEPSEEK_API_KEY",
 		baseUrl: "Base URL",
 		model: "Model",
+		ocrProfile: "OCR profile",
+		ocrLanguage: "OCR languages",
+		ocrFast: "Fast Latin",
+		ocrVietnamese: "Vietnamese Enhanced",
+		ocrHybrid: "Hybrid Vi + Latin",
 		saveSettings: "Save",
 		clearKey: "Reset env",
-		keySaved: "DeepSeek settings saved.",
+		settingsSaved: "Guide settings saved.",
 		keyMissing: "Set a DeepSeek API key environment variable before generating with DeepSeek.",
 		keyConfigured: "Env ready",
 		keyNotConfigured: "Env value missing",
@@ -78,13 +85,19 @@ const COPY = {
 		captureStep: "Chụp bước",
 		captureLabel: "Chụp thủ công",
 		settings: "Cài đặt",
+		guideSettings: "Guide settings",
 		apiKey: "API key env",
 		apiKeyPlaceholder: "DEEPSEEK_API_KEY",
 		baseUrl: "Base URL",
 		model: "Model",
+		ocrProfile: "OCR profile",
+		ocrLanguage: "OCR languages",
+		ocrFast: "Fast Latin",
+		ocrVietnamese: "Vietnamese Enhanced",
+		ocrHybrid: "Hybrid Vi + Latin",
 		saveSettings: "Lưu",
 		clearKey: "Reset env",
-		keySaved: "Đã lưu cài đặt DeepSeek.",
+		settingsSaved: "Da luu cai dat guide.",
 		keyMissing: "Hãy set biến môi trường DeepSeek API key trước khi tạo draft bằng DeepSeek.",
 		keyConfigured: "Env ready",
 		keyNotConfigured: "Chưa thấy giá trị env",
@@ -108,6 +121,8 @@ export function GuidePanel({ recordingId, videoPath, videoSourcePath }: GuidePan
 	const [deepSeekApiKeyEnvName, setDeepSeekApiKeyEnvName] = useState("DEEPSEEK_API_KEY");
 	const [deepSeekBaseUrl, setDeepSeekBaseUrl] = useState("https://api.deepseek.com");
 	const [deepSeekModel, setDeepSeekModel] = useState("deepseek-chat");
+	const [ocrProfile, setOcrProfile] = useState<GuideOcrProfile>("vietnamese");
+	const [ocrLanguage, setOcrLanguage] = useState("vi,en");
 	const [message, setMessage] = useState<string | null>(null);
 
 	const isBusy = busyAction !== null;
@@ -138,6 +153,8 @@ export function GuidePanel({ recordingId, videoPath, videoSourcePath }: GuidePan
 		setDeepSeekBaseUrl(result.data.deepseek.baseUrl);
 		setDeepSeekModel(result.data.deepseek.model);
 		setDeepSeekApiKeyEnvName(result.data.deepseek.apiKeyEnvName);
+		setOcrProfile(result.data.ocr.profile);
+		setOcrLanguage(result.data.ocr.language);
 	}, []);
 
 	useEffect(() => {
@@ -269,6 +286,8 @@ export function GuidePanel({ recordingId, videoPath, videoSourcePath }: GuidePan
 				deepseekApiKeyEnvName: deepSeekApiKeyEnvName,
 				baseUrl: deepSeekBaseUrl,
 				model: deepSeekModel,
+				ocrProfile,
+				ocrLanguage,
 			});
 			if (!result.success) {
 				throw new Error(result.error);
@@ -277,7 +296,9 @@ export function GuidePanel({ recordingId, videoPath, videoSourcePath }: GuidePan
 			setDeepSeekApiKeyEnvName(result.data.deepseek.apiKeyEnvName);
 			setDeepSeekBaseUrl(result.data.deepseek.baseUrl);
 			setDeepSeekModel(result.data.deepseek.model);
-			toast.success(copy.keySaved);
+			setOcrProfile(result.data.ocr.profile);
+			setOcrLanguage(result.data.ocr.language);
+			toast.success(copy.settingsSaved);
 		} catch (error) {
 			const text = error instanceof Error ? error.message : String(error);
 			setMessage(text);
@@ -285,7 +306,14 @@ export function GuidePanel({ recordingId, videoPath, videoSourcePath }: GuidePan
 		} finally {
 			setSettingsBusy(false);
 		}
-	}, [copy.keySaved, deepSeekApiKeyEnvName, deepSeekBaseUrl, deepSeekModel]);
+	}, [
+		copy.settingsSaved,
+		deepSeekApiKeyEnvName,
+		deepSeekBaseUrl,
+		deepSeekModel,
+		ocrLanguage,
+		ocrProfile,
+	]);
 
 	const handleClearDeepSeekKey = useCallback(async () => {
 		if (!window.electronAPI?.guide?.saveAiSettings) {
@@ -298,13 +326,17 @@ export function GuidePanel({ recordingId, videoPath, videoSourcePath }: GuidePan
 				clearDeepseekApiKeyEnvName: true,
 				baseUrl: deepSeekBaseUrl,
 				model: deepSeekModel,
+				ocrProfile,
+				ocrLanguage,
 			});
 			if (!result.success) {
 				throw new Error(result.error);
 			}
 			setAiSettings(result.data);
 			setDeepSeekApiKeyEnvName(result.data.deepseek.apiKeyEnvName);
-			toast.success(copy.keySaved);
+			setOcrProfile(result.data.ocr.profile);
+			setOcrLanguage(result.data.ocr.language);
+			toast.success(copy.settingsSaved);
 		} catch (error) {
 			const text = error instanceof Error ? error.message : String(error);
 			setMessage(text);
@@ -312,7 +344,7 @@ export function GuidePanel({ recordingId, videoPath, videoSourcePath }: GuidePan
 		} finally {
 			setSettingsBusy(false);
 		}
-	}, [copy.keySaved, deepSeekBaseUrl, deepSeekModel]);
+	}, [copy.settingsSaved, deepSeekBaseUrl, deepSeekModel, ocrLanguage, ocrProfile]);
 
 	const handleGenerateGuide = useCallback(() => {
 		void runAction("generate", async () => {
@@ -455,7 +487,7 @@ export function GuidePanel({ recordingId, videoPath, videoSourcePath }: GuidePan
 						<div className="flex items-center justify-between gap-2">
 							<div className="min-w-0">
 								<div className="truncate text-[11px] font-semibold text-slate-100">
-									{copy.deepseek} {copy.settings}
+									{copy.guideSettings}
 								</div>
 								<div className="truncate text-[10px] text-slate-500">
 									{aiSettings?.deepseek.hasApiKey
@@ -470,6 +502,33 @@ export function GuidePanel({ recordingId, videoPath, videoSourcePath }: GuidePan
 							</span>
 						</div>
 
+						<div className="grid grid-cols-2 gap-1.5">
+							<label className="block min-w-0 text-[10px] font-medium text-slate-400">
+								{copy.ocrProfile}
+								<select
+									value={ocrProfile}
+									onChange={(event) => setOcrProfile(event.target.value as GuideOcrProfile)}
+									disabled={settingsBusy}
+									className="mt-1 h-8 w-full rounded-md border border-white/[0.08] bg-black/20 px-2 text-[11px] text-slate-100 outline-none"
+								>
+									<option value="vietnamese">{copy.ocrVietnamese}</option>
+									<option value="hybrid">{copy.ocrHybrid}</option>
+									<option value="fast">{copy.ocrFast}</option>
+								</select>
+							</label>
+							<label className="block min-w-0 text-[10px] font-medium text-slate-400">
+								{copy.ocrLanguage}
+								<input
+									type="text"
+									value={ocrLanguage}
+									onChange={(event) => setOcrLanguage(event.target.value)}
+									placeholder="vi,en"
+									disabled={settingsBusy}
+									className="mt-1 h-8 w-full rounded-md border border-white/[0.08] bg-black/20 px-2 text-[11px] text-slate-100 outline-none placeholder:text-slate-600"
+								/>
+							</label>
+						</div>
+
 						<label className="block text-[10px] font-medium text-slate-400">
 							{copy.apiKey}
 							<input

src/guide/contracts.ts

@@ -9,6 +9,7 @@ export type GuideTargetRole = "button" | "menu" | "tab" | "field" | "link" | "un
 export type GuideLanguage = "vi" | "en";
 export type GuideAiProvider = "deepseek" | "local";
 export type GuideSecretStorage = "environment" | "none";
+export type GuideOcrProfile = "fast" | "vietnamese" | "hybrid";
 
 export type GuideSessionStatus =
 	| "recording"
@@ -178,6 +179,11 @@ export interface GenerateGuideDraftInput {
 }
 
 export interface GuideAiSettings {
+	ocr: {
+		profile: GuideOcrProfile;
+		language: string;
+		updatedAt?: string;
+	};
 	deepseek: {
 		hasApiKey: boolean;
 		apiKeyEnvName: string;
@@ -194,6 +200,8 @@ export interface SaveGuideAiSettingsInput {
 	clearDeepseekApiKeyEnvName?: boolean;
 	baseUrl?: string;
 	model?: string;
+	ocrProfile?: GuideOcrProfile;
+	ocrLanguage?: string;
 }
 
 export interface SaveGuideInput {

tools/ocr/paddle_ocr_service.py

@@ -5,6 +5,7 @@ import importlib.util
 import os
 import sys
 import tempfile
+from dataclasses import dataclass
 from pathlib import Path
 from threading import Lock
 from typing import Any
@@ -17,6 +18,65 @@ app = FastAPI(title="OpenScreen PaddleOCR service")
 
 _engines: dict[str, Any] = {}
 _engine_lock = Lock()
+_LATIN_RECOGNITION_LANGS = {
+    "af",
+    "az",
+    "bs",
+    "ca",
+    "cs",
+    "cy",
+    "da",
+    "de",
+    "en",
+    "es",
+    "et",
+    "eu",
+    "fi",
+    "fr",
+    "ga",
+    "gl",
+    "hr",
+    "hu",
+    "id",
+    "is",
+    "it",
+    "ku",
+    "la",
+    "latin",
+    "lb",
+    "lt",
+    "lv",
+    "mi",
+    "ms",
+    "mt",
+    "nl",
+    "no",
+    "oc",
+    "pi",
+    "pl",
+    "pt",
+    "qu",
+    "rm",
+    "ro",
+    "rs_latin",
+    "rslatin",
+    "sk",
+    "sl",
+    "sq",
+    "sv",
+    "sw",
+    "tl",
+    "tr",
+    "uz",
+    "vi",
+}
+
+
+@dataclass(frozen=True)
+class PreparedImage:
+    path: str
+    scale: float = 1.0
+    should_delete: bool = False
 
 
 class OcrRequest(BaseModel):
@@ -24,6 +84,7 @@ class OcrRequest(BaseModel):
     path: str | None = None
     imagePath: str | None = None
     language: str | None = None
+    profile: str | None = None
 
 
 @app.get("/health")
@@ -33,7 +94,9 @@ def health() -> dict[str, Any]:
         "paddleocrInstalled": importlib.util.find_spec("paddleocr") is not None,
         "paddleInstalled": importlib.util.find_spec("paddle") is not None,
         "engineReady": bool(_engines),
-        "defaultLanguage": os.getenv("PADDLEOCR_LANG", "latin"),
+        "defaultLanguage": os.getenv("PADDLEOCR_LANG") or "vi,en",
+        "defaultProfile": os.getenv("OPENSCREEN_OCR_PROFILE") or "vietnamese",
+        "loadedEngines": sorted(_engines.keys()),
     }
 
 
@@ -41,8 +104,12 @@ def health() -> dict[str, Any]:
 async def ocr(request: OcrRequest) -> dict[str, Any]:
     image_path, should_delete = _resolve_image_path(request)
     try:
-        engine = _get_engine(request.language)
-        blocks = await run_in_threadpool(_recognize_blocks, engine, image_path)
+        blocks = await run_in_threadpool(
+            _recognize_profile_blocks,
+            image_path,
+            request.language,
+            request.profile,
+        )
         return {"blocks": blocks}
     finally:
         if should_delete:
@@ -73,8 +140,7 @@ def _resolve_image_path(request: OcrRequest) -> tuple[str, bool]:
     return handle.name, True
 
 
-def _get_engine(language: str | None) -> Any:
-    paddle_lang = _resolve_paddle_language(language)
+def _get_engine(paddle_lang: str) -> Any:
     cache_key = f"{paddle_lang}|{os.getenv('PADDLEOCR_DEVICE', 'cpu')}"
     with _engine_lock:
         if cache_key not in _engines:
@@ -105,13 +171,17 @@ def _create_engine(paddle_lang: str) -> Any:
         "enable_mkldnn": os.getenv("PADDLEOCR_ENABLE_MKLDNN", "0") == "1",
         "use_doc_orientation_classify": False,
         "use_doc_unwarping": False,
-        "use_textline_orientation": False,
+        "use_textline_orientation": os.getenv("PADDLEOCR_USE_TEXTLINE_ORIENTATION", "0") == "1",
     }
     if os.getenv("PADDLEOCR_USE_MOBILE", "1") != "0":
         modern_kwargs.update(
             {
-                "text_detection_model_name": "PP-OCRv5_mobile_det",
-                "text_recognition_model_name": _mobile_recognition_model(paddle_lang),
+                "text_detection_model_name": os.getenv(
+                    "PADDLEOCR_DET_MODEL",
+                    "PP-OCRv5_mobile_det",
+                ),
+                "text_recognition_model_name": os.getenv("PADDLEOCR_REC_MODEL")
+                or _mobile_recognition_model(paddle_lang),
             }
         )
 
@@ -150,23 +220,236 @@ def _patch_paddlex_frozen_ocr_extra_gate() -> None:
     deps._openscreen_ocr_extra_patch = True
 
 
-def _resolve_paddle_language(language: str | None) -> str:
-    explicit = os.getenv("PADDLEOCR_LANG")
+def _recognize_profile_blocks(
+    image_path: str,
+    language: str | None,
+    profile: str | None,
+) -> list[dict[str, Any]]:
+    ocr_profile = _resolve_ocr_profile(profile)
+    languages = _resolve_paddle_languages(language, ocr_profile)
+    prepared = _prepare_image_for_profile(image_path, ocr_profile)
+    try:
+        blocks: list[dict[str, Any]] = []
+        for paddle_lang in languages:
+            engine = _get_engine(paddle_lang)
+            recognized = _recognize_blocks(engine, prepared.path)
+            blocks.extend(_scale_blocks(recognized, prepared.scale))
+        return _merge_blocks(blocks)
+    finally:
+        if prepared.should_delete:
+            Path(prepared.path).unlink(missing_ok=True)
+
+
+def _resolve_ocr_profile(profile: str | None) -> str:
+    explicit = (os.getenv("OPENSCREEN_OCR_PROFILE") or "").strip().lower()
+    value = explicit or (profile or "").strip().lower()
+    if value in {"fast", "vietnamese", "hybrid"}:
+        return value
+    return "vietnamese"
+
+
+def _resolve_paddle_languages(language: str | None, profile: str) -> list[str]:
+    explicit = (os.getenv("PADDLEOCR_LANG") or "").strip().lower()
     if explicit:
-        return explicit
+        return [explicit]
 
     language_value = (language or "vi,en").lower()
-    if "vi" in language_value or "latin" in language_value:
+    has_vietnamese = "vi" in _split_language_tags(language_value)
+    if profile == "fast":
+        return [_resolve_primary_paddle_language(language_value, prefer_vietnamese=False)]
+    if profile == "hybrid":
+        languages = ["vi"] if has_vietnamese else []
+        languages.append("latin")
+        return _dedupe_languages(languages)
+    return [_resolve_primary_paddle_language(language_value, prefer_vietnamese=True)]
+
+
+def _split_language_tags(language: str) -> set[str]:
+    return {part.strip().lower() for part in language.split(",") if part.strip()}
+
+
+def _dedupe_languages(languages: list[str]) -> list[str]:
+    seen: set[str] = set()
+    result: list[str] = []
+    for language in languages:
+        if language not in seen:
+            seen.add(language)
+            result.append(language)
+    return result
+
+
+def _resolve_primary_paddle_language(language_value: str, *, prefer_vietnamese: bool) -> str:
+    tags = _split_language_tags(language_value)
+    if prefer_vietnamese and "vi" in tags:
+        return "vi"
+    if "latin" in tags or "vi" in tags or "en" in tags:
         return "latin"
-    if "en" in language_value:
-        return "en"
-    return language_value.split(",")[0].strip() or "latin"
+    for tag in tags:
+        return tag
+    return "latin"
+
+
+def _prepare_image_for_profile(image_path: str, profile: str) -> PreparedImage:
+    if profile == "fast":
+        return PreparedImage(image_path)
+
+    try:
+        from PIL import Image, ImageEnhance, ImageOps
+    except Exception:
+        return PreparedImage(image_path)
+
+    try:
+        with Image.open(image_path) as source:
+            image = source.convert("RGB")
+    except Exception:
+        return PreparedImage(image_path)
+
+    scale = _resolve_enhancement_scale(image.width, image.height)
+    if scale <= 1:
+        return PreparedImage(image_path)
+
+    resampling = getattr(getattr(Image, "Resampling", Image), "LANCZOS")
+    enhanced = image.resize((round(image.width * scale), round(image.height * scale)), resampling)
+    enhanced = ImageOps.autocontrast(enhanced)
+    enhanced = ImageEnhance.Contrast(enhanced).enhance(1.25)
+    enhanced = ImageEnhance.Sharpness(enhanced).enhance(1.35)
+
+    handle = tempfile.NamedTemporaryFile(prefix="openscreen-ocr-enhanced-", suffix=".png", delete=False)
+    try:
+        handle.close()
+        enhanced.save(handle.name, format="PNG")
+        return PreparedImage(handle.name, scale=scale, should_delete=True)
+    except Exception:
+        Path(handle.name).unlink(missing_ok=True)
+        return PreparedImage(image_path)
+
+
+def _resolve_enhancement_scale(width: int, height: int) -> float:
+    try:
+        requested_scale = float(os.getenv("OPENSCREEN_OCR_ENHANCE_SCALE", "2"))
+    except ValueError:
+        requested_scale = 2.0
+    scale = max(1.0, min(3.0, requested_scale))
+    try:
+        max_side = int(os.getenv("OPENSCREEN_OCR_ENHANCE_MAX_SIDE", "2400"))
+    except ValueError:
+        max_side = 2400
+    largest_side = max(width, height)
+    if largest_side <= 0:
+        return 1.0
+    return max(1.0, min(scale, max_side / largest_side))
+
+
+def _scale_blocks(blocks: list[dict[str, Any]], scale: float) -> list[dict[str, Any]]:
+    if scale <= 1:
+        return blocks
+
+    scaled_blocks: list[dict[str, Any]] = []
+    for block in blocks:
+        box = block.get("box")
+        if not isinstance(box, dict) or not _box_uses_pixels(box):
+            scaled_blocks.append(block)
+            continue
+        scaled_box = {
+            "x": float(box["x"]) / scale,
+            "y": float(box["y"]) / scale,
+            "width": float(box["width"]) / scale,
+            "height": float(box["height"]) / scale,
+        }
+        scaled_blocks.append({**block, "box": scaled_box})
+    return scaled_blocks
+
+
+def _box_uses_pixels(box: dict[str, Any]) -> bool:
+    try:
+        x = float(box["x"])
+        y = float(box["y"])
+        width = float(box["width"])
+        height = float(box["height"])
+    except (KeyError, TypeError, ValueError):
+        return False
+    return x > 1 or y > 1 or width > 1 or height > 1 or x + width > 1 or y + height > 1
+
+
+def _merge_blocks(blocks: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    merged: list[dict[str, Any]] = []
+    for block in sorted(blocks, key=_block_quality, reverse=True):
+        box = block.get("box")
+        if not isinstance(box, dict):
+            continue
+        overlapping_index = next(
+            (
+                index
+                for index, existing in enumerate(merged)
+                if _box_iou(box, existing.get("box")) >= 0.62
+            ),
+            None,
+        )
+        if overlapping_index is None:
+            merged.append(block)
+            continue
+        if _block_quality(block) > _block_quality(merged[overlapping_index]):
+            merged[overlapping_index] = block
+    return sorted(merged, key=lambda block: _box_sort_key(block.get("box")))
+
+
+def _block_quality(block: dict[str, Any]) -> float:
+    text = str(block.get("text") or "")
+    score = _score_to_float(block.get("confidence"))
+    if _has_vietnamese_diacritics(text):
+        score += 0.08
+    if len(text) >= 2:
+        score += min(0.04, len(text) * 0.002)
+    return score
+
+
+def _has_vietnamese_diacritics(text: str) -> bool:
+    return any(
+        character
+        in "ăâđêôơưĂÂĐÊÔƠƯáàảãạắằẳẵặấầẩẫậéèẻẽẹếềểễệíìỉĩịóòỏõọốồổỗộớờởỡợúùủũụứừửữựýỳỷỹỵ"
+        for character in text
+    )
+
+
+def _box_iou(left: Any, right: Any) -> float:
+    if not isinstance(left, dict) or not isinstance(right, dict):
+        return 0.0
+    try:
+        left_x = float(left["x"])
+        left_y = float(left["y"])
+        left_width = float(left["width"])
+        left_height = float(left["height"])
+        right_x = float(right["x"])
+        right_y = float(right["y"])
+        right_width = float(right["width"])
+        right_height = float(right["height"])
+    except (KeyError, TypeError, ValueError):
+        return 0.0
+
+    intersection_left = max(left_x, right_x)
+    intersection_top = max(left_y, right_y)
+    intersection_right = min(left_x + left_width, right_x + right_width)
+    intersection_bottom = min(left_y + left_height, right_y + right_height)
+    intersection_width = max(0.0, intersection_right - intersection_left)
+    intersection_height = max(0.0, intersection_bottom - intersection_top)
+    intersection_area = intersection_width * intersection_height
+    if intersection_area <= 0:
+        return 0.0
+    union_area = left_width * left_height + right_width * right_height - intersection_area
+    return intersection_area / union_area if union_area > 0 else 0.0
+
+
+def _box_sort_key(box: Any) -> tuple[float, float]:
+    if not isinstance(box, dict):
+        return (0.0, 0.0)
+    try:
+        return (float(box["y"]), float(box["x"]))
+    except (KeyError, TypeError, ValueError):
+        return (0.0, 0.0)
 
 
 def _mobile_recognition_model(paddle_lang: str) -> str:
-    if paddle_lang == "en":
-        return "en_PP-OCRv5_mobile_rec"
-    if paddle_lang == "latin":
+    if paddle_lang in _LATIN_RECOGNITION_LANGS:
         return "latin_PP-OCRv5_mobile_rec"
     return "PP-OCRv5_mobile_rec"