@effectnode/media 0.10.0 → 0.11.0

package/dist/backend/movie-backend/agent/prompt/ltx.txt

@@ -0,0 +1,20 @@
+This LTX Prompting Guide outlines the exact structural brief required to control LTX video generation models (including LTX-2.3 and LTX-2.5). LTX is highly prompt-sensitive and requires literal, chronological, and cinematic instructions formatted as a single continuous paragraph under 200 words. 
+
+
+Core Prompt Structure
+To generate professional-grade clips, construct your prompt like a cinematographer's shot list using these six core elements in sequential order: 
+Establish the Shot: Open with the camera framing, lens type, and cinematography terms matching the genre (e.g., Close-up shot, Wide anamorphic lens, Overhead shot). 
+Define the Characters: Detail age, hairstyle, clothing, and posture. Express emotions through concrete physical cues (e.g., trembling hands) rather than abstract emotional labels. 
+Describe the Action: Write the main action as a natural sequence of present-tense verbs flowing chronologically from beginning to end. 
+Set the Scene: Detail the environmental background, surface textures, lighting conditions, and color palette to establish mood. 
+Identify Camera Movements: Specify how the camera moves (e.g., dolly in, jib up, pan left). Describing how the subject looks after the movement helps the model complete the motion accurately. 
+Describe the Audio: LTX models generate synchronized native audio. Describe ambient sounds, music, or speech. Put spoken dialogue inside "quotation marks" and specify languages or accents if needed. 
+
+
+
+Golden Rules for LTX Prompting
+Write a Single Paragraph: Do not use bullet points or line breaks inside your final prompt; merge all elements into one fluid paragraph.
+Keep it Under 200 Words: Excessively long descriptions dilute the model's attention span and compromise prompt adherence.
+Be Literal and Precise: Avoid poetic writing or abstract metaphors. State exactly what should be rendered on screen.
+Use Hierarchical Structuring: If using an LLM to expand your prompt, structure the data strictly by Shot Type ➔ Subject ➔ Action ➔ Camera Movement ➔ Lighting ➔ Mood.
+Leverage Negative Space: Explicitly describe what is absent if you need a clean shot (e.g., "clean composition with minimal background elements, no distracting motion"). 

package/dist/backend/movie-backend/core.js

@@ -575,7 +575,15 @@ async function installPythonDependencies() {
         }
         const dotsTtsFolder = join(pythonAppSrcDir, "dots-tts-mlx");
         if (!existsSync(dotsTtsFolder)) {
-            const cloneCMD = await runCommand("git", ["clone", "https://github.com/sb1992/dots-tts-mlx.git", "dots-tts-mlx"], { cwd: pythonAppSrcDir });
+            const cloneCMD = await runCommand("git", [
+                "clone",
+                "--depth",
+                "1",
+                "--branch",
+                "v0.2.0",
+                "https://github.com/sb1992/dots-tts-mlx",
+                "dots-tts-mlx",
+            ], { cwd: pythonAppSrcDir });
             if (!cloneCMD.success) {
                 console.error("Failed to clone dots-tts-mlx:", cloneCMD.error);
                 return false;

package/dist/backend/movie-backend/render-media.js

@@ -224,40 +224,27 @@ async function getDotsTtsBin() {
     }
     return "dots-tts";
 }
-/** Resolve the `mlx_whisper` executable installed via `uv tool install mlx-whisper`. */
-async function getMlxWhisperBin() {
-    const candidates = [
-        join(homedir(), ".local", "bin", "mlx_whisper"),
-        "/opt/homebrew/bin/mlx_whisper",
-        "/usr/local/bin/mlx_whisper",
-    ];
-    for (const p of candidates) {
-        if (existsSync(p))
-            return p;
-    }
-    return "mlx_whisper";
-}
 /** The cloned dots-tts-mlx project directory. */
 const DOTS_TTS_FOLDER = join(APP_DATA_DIR, "python-src", "dots-tts-mlx");
 /** Base directory where dots-tts MLX weights live (inside the project folder). */
 const DOTS_TTS_WEIGHTS_DIR = join(DOTS_TTS_FOLDER, "dots-tts-mlx-weights");
 /**
- * Resolve a dots-tts `--model` value. Accepts `./dots-tts-mlx-weights/mf-int4`
- * (relative to the dots-tts-mlx folder), a bare variant (`mf-int4`), or an
+ * Resolve a dots-tts `--model` value. Accepts `./dots-tts-mlx-weights/int4`
+ * (relative to the dots-tts-mlx folder), a bare variant (`int4`), or an
  * explicit path.
  */
 function resolveDotsTtsModel(model) {
     if (!model)
-        return join(DOTS_TTS_WEIGHTS_DIR, "mf-int4");
+        return join(DOTS_TTS_WEIGHTS_DIR, "int4");
     if (model.startsWith("./"))
         return join(DOTS_TTS_FOLDER, model.slice(2));
     if (model.includes("/"))
         return model;
     return join(DOTS_TTS_WEIGHTS_DIR, model);
 }
-/** True when the dots-tts `mf-int4` weights have been downloaded. */
+/** True when the dots-tts `int4` weights have been downloaded. */
 function isDotsTtsModelDownloaded() {
-    return existsSync(join(DOTS_TTS_WEIGHTS_DIR, "mf-int4"));
+    return existsSync(join(DOTS_TTS_WEIGHTS_DIR, "int4"));
 }
 /** Resolve the ffmpeg binary installed via Homebrew (Apple Silicon then Intel). */
 async function getFfmpegBin() {
@@ -979,8 +966,7 @@ export async function generateAdvancedVoiceClone(projectId, params, onLog) {
         .replace(/[^a-zA-Z0-9_-]/g, "_")
         .slice(0, 64) || "voice";
     const model = resolveDotsTtsModel(String(params.model || ""));
-    // dots-tts and mlx_whisper both expect a WAV reference — convert mp3/other
-    // formats to 16-bit PCM WAV first.
+    // dots-tts expects a WAV reference — convert mp3/other formats to 16-bit PCM WAV.
     let refAudio = resolvedRef;
     if (!resolvedRef.toLowerCase().endsWith(".wav")) {
         const tempDir = join(TEMP_DIR, String(projectId));
@@ -1005,19 +991,6 @@ export async function generateAdvancedVoiceClone(projectId, params, onLog) {
         }
         refAudio = convertedRef;
     }
-    // dots-tts requires `--ref-text` (the reference audio's transcript). Transcribe
-    // the WAV reference with mlx_whisper to supply it.
-    const whisperBin = await getMlxWhisperBin();
-    const transcribeDir = join(TEMP_DIR, String(projectId));
-    ensureDir(transcribeDir);
-    const whisperResult = await runCommand([whisperBin, refAudio, "--output-dir", transcribeDir], { onLog });
-    const refTxtPath = join(transcribeDir, (refAudio.split(sep).pop() || "ref").replace(/\.[^.]+$/, "") + ".txt");
-    const refText = whisperResult.success && existsSync(refTxtPath)
-        ? readFileSync(refTxtPath, "utf-8").replace(/\s+/g, " ").trim()
-        : "";
-    if (!refText) {
-        return { error: "Failed to transcribe reference audio (mlx_whisper)" };
-    }
     const dotsTtsBin = await getDotsTtsBin();
     // Mirror the voice-clone tab's timestamped-folder layout so outputs never
     // overwrite each other.
@@ -1032,14 +1005,14 @@ export async function generateAdvancedVoiceClone(projectId, params, onLog) {
         cleanText || "please provide text",
         "--ref-audio",
         refAudio,
-        "--ref-text",
-        refText,
         "--language",
         language,
         "--out-path",
         outDir,
         "--out-prefix",
         prefix,
+        "--max-generate-length",
+        "3000",
     ]);
     const result = await runCommand([
         dotsTtsBin,
@@ -1049,14 +1022,14 @@ export async function generateAdvancedVoiceClone(projectId, params, onLog) {
         cleanText || "please provide text",
         "--ref-audio",
         refAudio,
-        "--ref-text",
-        refText,
         "--language",
         language,
         "--out-path",
         outDir,
         "--out-prefix",
         prefix,
+        "--max-generate-length",
+        "3000",
     ], { cwd: dirname(dotsTtsBin), onLog });
     if (!result.success) {
         return { error: result.output || "Advanced voice clone failed" };
@@ -1072,7 +1045,6 @@ export async function generateAdvancedVoiceClone(projectId, params, onLog) {
     const meta = {
         id: voiceId,
         transcript: cleanText,
-        refText,
         language,
         model,
         refAudioFilename: params.refAudioPath,
@@ -2612,14 +2584,14 @@ export async function renderMediaRoutes({ app, getUvPath, }) {
             ensureDir(DOTS_TTS_FOLDER);
             send("progress", {
                 status: "starting",
-                label: "Downloading dots-tts model (mf-int4)...",
+                label: "Downloading dots-tts model (int4)...",
             });
             const proc = spawn([
                 "hf",
                 "download",
                 "shraey/dots-tts-mlx",
                 "--include",
-                "mf-int4/*",
+                "int4/*",
                 "--local-dir",
                 "./dots-tts-mlx-weights",
             ], { cwd: DOTS_TTS_FOLDER, stdout: "pipe", stderr: "pipe" });

package/frontend/src/movie-app/components/EditorTabs/SetupAiModelTab.tsx

@@ -217,7 +217,7 @@ const DOWNLOAD_MODELS: {
   },
   {
     id: "dots-tts",
-    name: "shraey/dots-tts-mlx (mf-int4)",
+    name: "shraey/dots-tts-mlx (int4)",
     desc: "Advanced voice clone (dots-tts)",
   },
 ];

package/frontend/src/movie-app/components/EditorTabs/UpscaleTab.tsx

@@ -264,6 +264,7 @@ export default function UpscaleTab({ projectId }: Props) {
         <div className="flex flex-wrap gap-2">
           {[
             { value: "1x" as UpscaleMode, label: "Refine (1x)" },
+            { value: "2x" as UpscaleMode, label: "Refine (2x)" },
             { value: "1000" as UpscaleMode, label: "Upscale to 1000px" },
             { value: "1500" as UpscaleMode, label: "Upscale to 1500px" },
             { value: "2000" as UpscaleMode, label: "Upscale to 2000px" },

package/frontend/src/movie-app/stores/advancedVoiceCloneStore.ts

@@ -79,7 +79,7 @@ export const useAdvancedVoiceCloneStore = create<AdvancedVoiceCloneStore>(
     text: "",
     language: "YUE",
     outPrefix: "voice",
-    model: "./dots-tts-mlx-weights/mf-int4",
+    model: "./dots-tts-mlx-weights/int4",
     refAudio: null,
     audios: [],
     audiosLoading: false,
@@ -127,9 +127,7 @@ export const useAdvancedVoiceCloneStore = create<AdvancedVoiceCloneStore>(
           uploading: false,
           refAudio: {
             filename: data.filename,
-            url: resolveUrl(
-              `/api/files?path=${encodeURIComponent(data.path)}`,
-            ),
+            url: resolveUrl(`/api/files?path=${encodeURIComponent(data.path)}`),
           },
         });
         void get().fetchAudios(projectId);
@@ -193,7 +191,7 @@ export const useAdvancedVoiceCloneStore = create<AdvancedVoiceCloneStore>(
         text: "",
         language: "YUE",
         outPrefix: "voice",
-        model: "./dots-tts-mlx-weights/mf-int4",
+        model: "./dots-tts-mlx-weights/int4",
         refAudio: null,
         audios: [],
         audiosLoading: false,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@effectnode/media",
-  "version": "0.10.0",
+  "version": "0.11.0",
   "description": "Start a full-stack media app: Vite + React + TypeScript frontend, Express backend with REST + WebSocket API",
   "license": "MIT",
   "author": "",