@vibeframe/cli 0.104.2 → 0.105.1

package/dist/index.js

@@ -17106,6 +17106,53 @@ function pickReferenceImageUrl(input) {
     return input;
   return void 0;
 }
+function isFalFileInput(value) {
+  return value.startsWith("http://") || value.startsWith("https://") || value.startsWith("data:");
+}
+async function normaliseReferences(input, client) {
+  if (!input)
+    return [];
+  const references = [];
+  for (const ref of input) {
+    if (!isFalFileInput(ref.url))
+      continue;
+    references.push({
+      ...ref,
+      url: ref.url.startsWith("data:") ? await uploadDataUri(client, ref.url) : ref.url
+    });
+  }
+  return references;
+}
+async function uploadDataUri(client, dataUri) {
+  const blob = dataUriToBlob(dataUri);
+  return client.storage.upload(blob, {
+    lifecycle: { expiresIn: "1h" }
+  });
+}
+function dataUriToBlob(dataUri) {
+  const match = /^data:([^;,]+)?(;base64)?,(.*)$/s.exec(dataUri);
+  if (!match)
+    throw new Error("Invalid data URI reference.");
+  const mimeType = match[1] || "application/octet-stream";
+  const isBase64 = Boolean(match[2]);
+  const payload = match[3] ?? "";
+  const buffer = isBase64 ? Buffer.from(payload, "base64") : Buffer.from(decodeURIComponent(payload), "utf-8");
+  return new Blob([new Uint8Array(buffer)], { type: mimeType });
+}
+function groupReferences(references) {
+  const image_urls = [];
+  const video_urls = [];
+  const audio_urls = [];
+  for (const ref of references) {
+    if (ref.kind === "image")
+      image_urls.push(ref.url);
+    if (ref.kind === "video")
+      video_urls.push(ref.url);
+    if (ref.kind === "audio")
+      audio_urls.push(ref.url);
+  }
+  return { image_urls, video_urls, audio_urls };
+}
 function normaliseAspect(value) {
   if (!value)
     return "auto";
@@ -17128,7 +17175,7 @@ function normaliseDuration(value) {
     return "auto";
   return Math.max(4, Math.min(15, Math.round(value)));
 }
-var import_client, ENDPOINT_TEXT_TO_VIDEO, ENDPOINT_IMAGE_TO_VIDEO, DEFAULT_VARIANT, VALID_RESOLUTIONS, VALID_ASPECTS, FalProvider, falProvider;
+var import_client, ENDPOINT_TEXT_TO_VIDEO, ENDPOINT_IMAGE_TO_VIDEO, ENDPOINT_REFERENCE_TO_VIDEO, DEFAULT_VARIANT, VALID_RESOLUTIONS, VALID_ASPECTS, FalProvider, falProvider;
 var init_FalProvider = __esm({
   "../ai-providers/dist/fal/FalProvider.js"() {
     "use strict";
@@ -17141,6 +17188,10 @@ var init_FalProvider = __esm({
       "seedance-2.0": "bytedance/seedance-2.0/image-to-video",
       "seedance-2.0-fast": "bytedance/seedance-2.0/fast/image-to-video"
     };
+    ENDPOINT_REFERENCE_TO_VIDEO = {
+      "seedance-2.0": "bytedance/seedance-2.0/reference-to-video",
+      "seedance-2.0-fast": "bytedance/seedance-2.0/fast/reference-to-video"
+    };
     DEFAULT_VARIANT = "seedance-2.0";
     VALID_RESOLUTIONS = ["480p", "720p", "1080p"];
     VALID_ASPECTS = ["21:9", "16:9", "4:3", "1:1", "3:4", "9:16", "auto"];
@@ -17149,7 +17200,7 @@ var init_FalProvider = __esm({
         this.id = "seedance";
         this.name = "fal.ai (Seedance 2.0)";
         this.description = "fal.ai hosting ByteDance Seedance 2.0 \u2014 Artificial Analysis #2 on both text-to-video and image-to-video leaderboards";
-        this.capabilities = ["text-to-video", "image-to-video"];
+        this.capabilities = ["text-to-video", "image-to-video", "reference-to-video"];
         this.iconUrl = "/icons/fal.svg";
         this.isAvailable = true;
       }
@@ -17188,9 +17239,11 @@ var init_FalProvider = __esm({
             error: `Unknown Seedance variant: ${variant}. Valid: ${Object.keys(ENDPOINT_TEXT_TO_VIDEO).join(", ")}.`
           };
         }
-        const referenceImage = pickReferenceImageUrl(options?.referenceImage);
+        const references = await normaliseReferences(options?.references, this.client);
+        const hasReferences = references.length > 0;
+        const referenceImage = hasReferences ? void 0 : pickReferenceImageUrl(options?.referenceImage);
         const isImageToVideo = !!referenceImage;
-        const endpointId = isImageToVideo ? ENDPOINT_IMAGE_TO_VIDEO[variant] : ENDPOINT_TEXT_TO_VIDEO[variant];
+        const endpointId = hasReferences ? ENDPOINT_REFERENCE_TO_VIDEO[variant] : isImageToVideo ? ENDPOINT_IMAGE_TO_VIDEO[variant] : ENDPOINT_TEXT_TO_VIDEO[variant];
         const aspect = normaliseAspect(options?.aspectRatio);
         const resolution = normaliseResolution(options?.resolution);
         const duration = normaliseDuration(options?.duration);
@@ -17200,13 +17253,26 @@ var init_FalProvider = __esm({
           resolution,
           duration
         };
-        if (referenceImage)
+        if (hasReferences) {
+          const grouped = groupReferences(references);
+          if (grouped.image_urls.length > 0)
+            input.image_urls = grouped.image_urls;
+          if (grouped.video_urls.length > 0)
+            input.video_urls = grouped.video_urls;
+          if (grouped.audio_urls.length > 0)
+            input.audio_urls = grouped.audio_urls;
+        } else if (referenceImage) {
           input.image_url = referenceImage;
+        }
         if (options?.negativePrompt)
           input.negative_prompt = options.negativePrompt;
         if (typeof options?.seed === "number")
           input.seed = options.seed;
-        if (options?.lastFrame)
+        if (typeof options?.generateAudio === "boolean")
+          input.generate_audio = options.generateAudio;
+        if (options?.endUserId)
+          input.end_user_id = options.endUserId;
+        if (!hasReferences && options?.lastFrame)
           input.end_image_url = options.lastFrame;
         try {
           const out = await this.client.subscribe(endpointId, { input, logs: false });
@@ -17253,14 +17319,15 @@ var init_fal = __esm({
       // Review this alias at the 1.0 cut.
       aliases: ["fal"],
       models: ["seedance-2.0", "seedance-2.0-fast"],
-      capabilities: ["text-to-video", "image-to-video", "native-audio"],
+      capabilities: ["text-to-video", "image-to-video", "reference-to-video", "native-audio"],
       apiKey: "fal",
       kinds: ["video"],
       resolverPriority: { video: 1 },
       commandsUnlocked: [
         "generate video -p seedance (Seedance 2.0 via fal.ai \u2014 default since v0.57)",
         "generate video -p seedance --seedance-model fast (lower-latency variant)",
-        "generate video -p seedance -i <image> (image-to-video)"
+        "generate video -p seedance -i <image> (image-to-video)",
+        "generate video -p seedance --ref-images <images...> (reference-to-video)"
       ]
     });
   }
@@ -26478,7 +26545,7 @@ YAML cues that drive narration, backdrop generation, and timing.
 
 \`\`\`yaml
 narration: "Introduce the promise in one crisp sentence."
-backdrop: "Cinematic abstract technology backdrop, precise light, premium editorial feel"
+backdrop: "Topic-aligned editorial background plate, abstract visual system, no readable text, no logos, no consumer products, clean negative space for HTML overlays"
 duration: 4
 \`\`\`
 
@@ -26489,7 +26556,7 @@ screen and one spoken breath.
 
 \`\`\`yaml
 narration: "Show the mechanism or proof point that makes the promise believable."
-backdrop: "Layered interface details, subtle motion trails, high-contrast product storytelling"
+backdrop: "Topic-aligned analytical background plate, abstract dashboard structure, no readable text, no product photos, no shoes, no unrelated objects"
 duration: 4
 \`\`\`
 
@@ -26500,7 +26567,7 @@ before/after.
 
 \`\`\`yaml
 narration: "Close with the action the viewer should remember."
-backdrop: "Resolved hero frame, confident final composition, clean negative space"
+backdrop: "Resolved editorial background plate, confident final composition, clean negative space, no readable text, no logos, no unrelated products"
 duration: 4
 \`\`\`
 
@@ -26553,6 +26620,33 @@ consult this file \u2014 run the generate command directly.
 Browse named styles: \`vibe scene list-styles\`. Re-seed from one with
 \`vibe scene init . --visual-style "Swiss Pulse"\` (idempotent).
 
+## Brief and local media
+
+\`brief.md\` is raw intent, not a strict schema. It may contain messy notes,
+pasted research, links, product requirements, or a one-line idea. \`vibe init
+--from brief.md\` uses it only to seed \`STORYBOARD.md\` and \`DESIGN.md\`;
+after init, those two files are the working source of truth.
+
+Use \`media/\` for user-provided source files: product photos, screenshots,
+logos, B-roll, recorded narration, or reference clips. Keep those inputs
+inside this project so build references stay project-relative. Do not put user
+media in \`references/\`; that directory is reserved for local composition
+rules installed by VibeFrame.
+
+When a beat should reuse a local file, reference it from \`STORYBOARD.md\`
+with a project-relative path:
+
+\`\`\`yaml
+backdrop: "media/product-shot.png" # existing still image
+video: "media/broll.mp4"           # existing video/B-roll
+narration: "media/voice.wav"       # existing recorded narration
+asset: "media/logo.png"            # generic local asset reference
+\`\`\`
+
+Use text cues when you want VibeFrame to generate an asset. Use path cues
+when you want VibeFrame to reuse a local file. Avoid absolute paths or parent
+directory references; copy files into \`media/\` first.
+
 ## Provider keys and project scope
 
 Use VibeFrame CLI generation for project assets:
@@ -26600,9 +26694,11 @@ the framework-level minimum, not the cinematic craft layer.
 
 - \`DESIGN.md\` \u2014 visual identity contract (palette, type, motion, transitions)
 - \`STORYBOARD.md\` \u2014 per-beat narration/backdrop/duration cues for \`vibe build\`
+- \`media/\` \u2014 user-provided source files (photos, logos, clips, voice recordings)
 - \`index.html\` \u2014 root composition (timeline)
 - \`compositions/scene-*.html\` \u2014 per-scene HTML authored by you or the agent
-- \`assets/\` \u2014 shared media (narration audio, images, video)
+- \`assets/\` \u2014 generated/canonical build media (narration audio, images, video)
+- \`references/\` \u2014 composition rule docs installed by VibeFrame, not user media
 - \`transcript.json\` \u2014 Whisper word-level transcript (if narration exists)
 - \`hyperframes.json\` \u2014 HF registry config (speak to both toolchains)
 - \`vibe.config.json\` \u2014 canonical VibeFrame config (providers, budget)
@@ -26633,7 +26729,11 @@ npx hyperframes render
    \`\`\`
 4. Videos use \`muted\` with a separate \`<audio>\` element for the audio track.
 5. Sub-compositions use \`data-composition-src="compositions/file.html"\`.
-6. Only deterministic logic \u2014 no \`Date.now()\`, \`Math.random()\`, or network fetches.
+6. For render-stable text, do not apply continuous \`scale\`, \`x\`, \`y\`, or
+   \`filter\` tweens to \`.scene-content\` or any ancestor containing live text.
+   Animate background/media layers instead; text/cards should enter briefly and
+   then hold still at their final CSS positions.
+7. Only deterministic logic \u2014 no \`Date.now()\`, \`Math.random()\`, or network fetches.
 
 ## Linting \u2014 run after changes
 
@@ -26746,7 +26846,11 @@ async function scaffoldSceneProject(opts) {
     if (await pathExists(metaPath)) {
       skipped2.push(metaPath);
     } else {
-      await writeFile4(metaPath, JSON.stringify(buildHyperframesMeta(name, now), null, 2) + "\n", "utf-8");
+      await writeFile4(
+        metaPath,
+        JSON.stringify(buildHyperframesMeta(name, now), null, 2) + "\n",
+        "utf-8"
+      );
       created.push(metaPath);
     }
     const rootPath = resolve7(dir, "index.html");
@@ -26761,11 +26865,7 @@ async function scaffoldSceneProject(opts) {
   if (await pathExists(vibeConfigJsonPath)) {
     skipped2.push(vibeConfigJsonPath);
   } else {
-    await writeFile4(
-      vibeConfigJsonPath,
-      projectConfigJson({ name, aspect }),
-      "utf-8"
-    );
+    await writeFile4(vibeConfigJsonPath, projectConfigJson({ name, aspect }), "utf-8");
     created.push(vibeConfigJsonPath);
   }
   const vibePath = resolve7(dir, "vibe.project.yaml");
@@ -26804,11 +26904,7 @@ ${existing}`, "utf-8");
   if (await pathExists(designPath)) {
     skipped2.push(designPath);
   } else {
-    await writeFile4(
-      designPath,
-      buildDesignMd({ name, style: opts.visualStyle }),
-      "utf-8"
-    );
+    await writeFile4(designPath, buildDesignMd({ name, style: opts.visualStyle }), "utf-8");
     created.push(designPath);
   }
   const storyboardPath2 = resolve7(dir, "STORYBOARD.md");
@@ -29061,13 +29157,18 @@ Requirements (non-negotiable):
   producer's seek lands past the timeline's natural end and visibility state
   goes stale \u2014 the hold phase renders BLACK. Anchor the timeline to the full
   beat duration via either:
-    1. A subtle idle motion spanning 0\u2192duration on a parent element, e.g.
-       \`tl.fromTo(".scene-content", { scale: 1.0 }, { scale: 1.015, duration: <beat>, ease: "none" }, 0);\`
+    1. A subtle idle motion spanning 0\u2192duration on a background/media layer,
+       e.g. \`tl.fromTo(".backdrop", { scale: 1.0 }, { scale: 1.015, duration: <beat>, ease: "none" }, 0);\`
        (Ken-Burns, breathing opacity, gradient drift \u2014 should be barely
        perceptible so it doesn't compete with entry/exit beats).
     2. OR an explicit \`tl.set(target, { ...natural state... }, <beat - 0.001>)\`
        anchor at the end.
   This is the #2 source of "text disappears mid-beat" bugs after \`.clip\` sizing.
+- Do not apply continuous \`scale\`, \`x\`, \`y\`, \`filter\`, or other transform
+  tweens to \`.scene-content\` or any ancestor that contains live text/cards.
+  Animate the backdrop/media plane instead; let text enter briefly, then hold
+  still at its final CSS position. Continuous transforms on text ancestors can
+  create subpixel shimmer in screenshot-captured renders.
 - Timed children inside the composition have \`class="clip"\` plus
   \`data-start\`, \`data-duration\`, \`data-track-index\`.
 - If \`assets/backdrop-${ctx.beat.id}.png\` exists, use that local file as the
@@ -29137,7 +29238,9 @@ Reference shape (verbatim \u2014 match this skeleton exactly, no DOCTYPE / html
       const tl = gsap.timeline({ paused: true });
       // Idle motion spanning full beat duration \u2014 required to keep timeline
       // length aligned with data-duration (otherwise hold phase goes black).
-      tl.fromTo(".scene-content", { scale: 1.0 }, { scale: 1.015, duration: <sec>, ease: "none" }, 0);
+      // Keep continuous motion on the background/media layer so live text does
+      // not shimmer from subpixel resampling.
+      tl.fromTo(".backdrop", { scale: 1.0 }, { scale: 1.015, duration: <sec>, ease: "none" }, 0);
       // entry tweens
       window.__timelines["${compositionId}"] = tl;
     </script>
@@ -33809,6 +33912,9 @@ async function executeVideoGenerate(options) {
     prompt: prompt3,
     provider = "kling",
     image,
+    refImages,
+    refVideos,
+    refAudio,
     duration = 5,
     ratio = "16:9",
     seed,
@@ -33817,6 +33923,7 @@ async function executeVideoGenerate(options) {
     resolution,
     veoModel = "3.1-fast",
     seedanceModel = "quality",
+    generateAudio,
     output,
     wait = true,
     apiKey
@@ -33855,8 +33962,13 @@ async function executeVideoGenerate(options) {
     if (provider === "seedance" || provider === "fal") {
       const fal = new FalProvider();
       await fal.initialize({ apiKey: key });
+      const references = await prepareSeedanceReferences({
+        refImages,
+        refVideos,
+        refAudio
+      });
       let falImage = referenceImage;
-      if (falImage && falImage.startsWith("data:")) {
+      if (falImage && falImage.startsWith("data:") && references.length === 0) {
         const uploadHost = await resolveUploadHost();
         const upload = await uploadHost.uploadImage(referenceImageBuffer, {
           filename: image,
@@ -33867,11 +33979,14 @@ async function executeVideoGenerate(options) {
       const model = seedanceModel === "fast" || seedanceModel === "seedance-2.0-fast" ? "seedance-2.0-fast" : "seedance-2.0";
       const result = await fal.generateVideo(prompt3, {
         prompt: prompt3,
-        referenceImage: falImage,
+        referenceImage: references.length > 0 ? void 0 : falImage,
+        references: references.length > 0 ? references : void 0,
         duration,
         aspectRatio: ratio,
         negativePrompt: negative,
-        model
+        model,
+        resolution,
+        generateAudio
       });
       if (result.status === "failed")
         return { success: false, error: result.error || "Seedance generation failed" };
@@ -34048,6 +34163,54 @@ async function executeVideoGenerate(options) {
     };
   }
 }
+async function prepareSeedanceReferences(opts) {
+  const references = [];
+  for (const sourcePath of opts.refImages ?? []) {
+    references.push({
+      kind: "image",
+      url: await fileInputToUrlOrDataUri(sourcePath, "image/png"),
+      sourcePath
+    });
+  }
+  for (const sourcePath of opts.refVideos ?? []) {
+    references.push({
+      kind: "video",
+      url: await fileInputToUrlOrDataUri(sourcePath, "video/mp4"),
+      sourcePath
+    });
+  }
+  for (const sourcePath of opts.refAudio ?? []) {
+    references.push({
+      kind: "audio",
+      url: await fileInputToUrlOrDataUri(sourcePath, "audio/mpeg"),
+      sourcePath
+    });
+  }
+  return references;
+}
+async function fileInputToUrlOrDataUri(input, fallbackMimeType) {
+  if (input.startsWith("http://") || input.startsWith("https://") || input.startsWith("data:")) {
+    return input;
+  }
+  const absPath = resolve23(process.cwd(), input);
+  const buffer = await readFile15(absPath);
+  return `data:${mimeTypeForPath(input, fallbackMimeType)};base64,${buffer.toString("base64")}`;
+}
+function mimeTypeForPath(path2, fallback2) {
+  const ext = path2.toLowerCase().split(".").pop();
+  const mimeTypes2 = {
+    jpg: "image/jpeg",
+    jpeg: "image/jpeg",
+    png: "image/png",
+    gif: "image/gif",
+    webp: "image/webp",
+    mp4: "video/mp4",
+    mov: "video/quicktime",
+    mp3: "audio/mpeg",
+    wav: "audio/wav"
+  };
+  return mimeTypes2[ext || ""] || fallback2;
+}
 async function executeVideoStatus(options) {
   const {
     taskId,
@@ -36301,10 +36464,10 @@ var init_provider_resolver = __esm({
 
 // src/commands/_shared/openai-image.ts
 function resolveOpenAIImageModel(modelAlias) {
-  const isGptImage2 = modelAlias === "2" || modelAlias === "gpt-image-2";
+  const isGptImage15 = modelAlias === "1.5" || modelAlias === "gpt-image-1.5";
   return {
-    openaiModel: isGptImage2 ? "gpt-image-2" : void 0,
-    modelLabel: isGptImage2 ? "GPT Image 2" : "GPT Image 1.5"
+    openaiModel: isGptImage15 ? "gpt-image-1.5" : "gpt-image-2",
+    modelLabel: isGptImage15 ? "GPT Image 1.5" : "GPT Image 2"
   };
 }
 async function executeOpenAIImageGenerate(prompt3, options, ctx) {
@@ -36341,7 +36504,7 @@ function registerImageCommand(parent) {
     "1:1"
   ).option("--quality <quality>", "Quality: standard, hd (openai only)", "standard").option("--style <style>", "Style: vivid, natural (openai only)", "vivid").option("--count <n>", "Number of images to generate", "1").option(
     "-m, --model <model>",
-    "Model. Gemini: flash, 3.1-flash, latest, pro. OpenAI: 1.5 (default), 2 (gpt-image-2)"
+    "Model. Gemini: flash, 3.1-flash, latest, pro. OpenAI: 2 (default), 1.5"
   ).option("--dry-run", "Preview parameters without executing").addHelpText(
     "after",
     `
@@ -37775,8 +37938,8 @@ function registerVideoCommand(parent) {
     "quality"
   ).option("--negative <prompt>", "Negative prompt - what to avoid (Kling/Veo)").option("--resolution <res>", "Video resolution: 720p, 1080p, 4k (Veo only)").option("--last-frame <path>", "Last frame image for frame interpolation (Veo only)").option(
     "--ref-images <paths...>",
-    "Reference images for character consistency (Veo 3.1 only, max 3)"
-  ).option("--person <mode>", "Person generation: allow_all, allow_adult (Veo only)").option("--veo-model <model>", "Veo model: 3.0, 3.1, 3.1-fast (default: 3.1-fast)", "3.1-fast").option(
+    "Reference images for Seedance reference-to-video or Veo character consistency"
+  ).option("--ref-videos <paths...>", "Reference videos for Seedance reference-to-video").option("--ref-audio <paths...>", "Reference audio for Seedance reference-to-video").option("--no-generate-audio", "Disable native audio when the provider supports it").option("--person <mode>", "Person generation: allow_all, allow_adult (Veo only)").option("--veo-model <model>", "Veo model: 3.0, 3.1, 3.1-fast (default: 3.1-fast)", "3.1-fast").option(
     "--runway-model <model>",
     "Runway model: gen4.5 (default, text+image-to-video), gen4_turbo (image-to-video only)",
     "gen4.5"
@@ -37919,7 +38082,11 @@ Examples:
               negative: options.negative,
               resolution: options.resolution,
               veoModel: options.veoModel,
-              seedanceModel: options.seedanceModel
+              seedanceModel: options.seedanceModel,
+              refImages: options.refImages,
+              refVideos: options.refVideos,
+              refAudio: options.refAudio,
+              generateAudio: options.generateAudio
             }
           }
         });
@@ -38206,8 +38373,13 @@ Examples:
       } else if (provider === "seedance") {
         const fal = new FalProvider();
         await fal.initialize({ apiKey });
+        const seedanceReferences = await prepareSeedanceReferences2({
+          refImages: options.refImages,
+          refVideos: options.refVideos,
+          refAudio: options.refAudio
+        });
         let falImage = referenceImage;
-        if (falImage && falImage.startsWith("data:")) {
+        if (falImage && falImage.startsWith("data:") && seedanceReferences.length === 0) {
           try {
             const uploadHost = await resolveUploadHost();
             spinner2.text = `Uploading image via ${uploadHost.provider} for Seedance...`;
@@ -38230,11 +38402,14 @@ Examples:
         const falModel = seedanceModel === "fast" || seedanceModel === "seedance-2.0-fast" ? "seedance-2.0-fast" : "seedance-2.0";
         result = await fal.generateVideo(prompt3, {
           prompt: prompt3,
-          referenceImage: falImage,
+          referenceImage: seedanceReferences.length > 0 ? void 0 : falImage,
+          references: seedanceReferences.length > 0 ? seedanceReferences : void 0,
           duration: options.duration ? parseInt(options.duration) : void 0,
           aspectRatio: options.ratio,
           negativePrompt: options.negative,
-          model: falModel
+          model: falModel,
+          resolution: options.resolution,
+          generateAudio: options.generateAudio
         });
         finalResult = result;
       }
@@ -38289,6 +38464,54 @@ Examples:
     }
   });
 }
+async function prepareSeedanceReferences2(opts) {
+  const references = [];
+  for (const sourcePath of opts.refImages ?? []) {
+    references.push({
+      kind: "image",
+      url: await fileInputToUrlOrDataUri2(sourcePath, "image/png"),
+      sourcePath
+    });
+  }
+  for (const sourcePath of opts.refVideos ?? []) {
+    references.push({
+      kind: "video",
+      url: await fileInputToUrlOrDataUri2(sourcePath, "video/mp4"),
+      sourcePath
+    });
+  }
+  for (const sourcePath of opts.refAudio ?? []) {
+    references.push({
+      kind: "audio",
+      url: await fileInputToUrlOrDataUri2(sourcePath, "audio/mpeg"),
+      sourcePath
+    });
+  }
+  return references;
+}
+async function fileInputToUrlOrDataUri2(input, fallbackMimeType) {
+  if (input.startsWith("http://") || input.startsWith("https://") || input.startsWith("data:")) {
+    return input;
+  }
+  const absPath = resolve31(process.cwd(), input);
+  const buffer = await readFile18(absPath);
+  return `data:${mimeTypeForPath2(input, fallbackMimeType)};base64,${buffer.toString("base64")}`;
+}
+function mimeTypeForPath2(path2, fallback2) {
+  const ext = path2.toLowerCase().split(".").pop();
+  const mimeTypes2 = {
+    jpg: "image/jpeg",
+    jpeg: "image/jpeg",
+    png: "image/png",
+    gif: "image/gif",
+    webp: "image/webp",
+    mp4: "video/mp4",
+    mov: "video/quicktime",
+    mp3: "audio/mpeg",
+    wav: "audio/wav"
+  };
+  return mimeTypes2[ext || ""] || fallback2;
+}
 async function recordVideoNoWaitJob(opts) {
   return createAndWriteJobRecord({
     jobType: "generate-video",
@@ -58370,6 +58593,32 @@ var init_build_cache = __esm({
   }
 });
 
+// src/commands/_shared/build-backdrop-prompt.ts
+function augmentBackdropPrompt(cue) {
+  const trimmed = cue.trim();
+  const lower = trimmed.toLowerCase();
+  const requestsTextOrMarks = /\b(text|typography|title|headline|label|caption|logo|logos|wordmark|brand mark|brand marks)\b/.test(lower);
+  const forbidsTextOrMarks = /\b(no|without|avoid)\s+(readable\s+)?(text|typography|titles?|headlines?|labels?|captions?|brand\s+logos?|logos?|wordmarks?|brand\s+marks?)\b/.test(
+    lower
+  );
+  const allowsTextOrMarks = requestsTextOrMarks && !forbidsTextOrMarks;
+  const overlayContract = allowsTextOrMarks ? "The image is a video background or end-card plate; do not add any text, logos, charts, or UI beyond what the scene cue explicitly requests." : "The image is a background only; HTML overlays will provide all final text, charts, logos, and UI labels.";
+  const textRule = allowsTextOrMarks ? "If text, logos, or brand marks are explicitly requested, keep them minimal, legible, and do not invent extras." : "No readable text, labels, UI copy, logos, brand marks, watermarks, or invented typography.";
+  return [
+    "Create a 16:9 video background plate for a HyperFrames scene.",
+    overlayContract,
+    `Scene cue: ${trimmed}`,
+    textRule,
+    "Avoid unrelated consumer product photography, shoes, packaging, food, people, celebrity faces, advertisements, and random objects unless explicitly requested by the scene cue.",
+    "Leave generous negative space for overlay text and cards. Keep the result topic-aligned, editorial, cinematic, and non-distracting."
+  ].join(" ");
+}
+var init_build_backdrop_prompt = __esm({
+  "src/commands/_shared/build-backdrop-prompt.ts"() {
+    "use strict";
+  }
+});
+
 // src/commands/_shared/build-plan.ts
 import { existsSync as existsSync44 } from "node:fs";
 import { readFile as readFile40 } from "node:fs/promises";
@@ -58463,6 +58712,7 @@ async function createBuildPlan(opts) {
     const voice = stringOrUndefined3(cue.voice) ?? resolved.voice;
     const narrationText = stringOrUndefined3(cue.narration);
     const backdropPrompt = stringOrUndefined3(cue.backdrop);
+    const augmentedBackdropPrompt = backdropPrompt ? augmentBackdropPrompt(backdropPrompt) : null;
     const videoPrompt = stringOrUndefined3(cue.video);
     const musicPrompt = stringOrUndefined3(cue.music);
     const genericReference = resolveGenericAssetReference(projectDir, cue.asset);
@@ -58482,9 +58732,9 @@ async function createBuildPlan(opts) {
       voice,
       ext: resolved.narration.resolved === "elevenlabs" ? "mp3" : "wav"
     }) : null;
-    const backdropCache = backdropPrompt && !backdropReference ? backdropCacheDescriptor({
+    const backdropCache = augmentedBackdropPrompt && !backdropReference ? backdropCacheDescriptor({
       beatId: beat.id,
-      cue: backdropPrompt,
+      cue: augmentedBackdropPrompt,
       provider: resolved.image.resolved,
       quality: imageQuality,
       size: imageSize2,
@@ -59014,6 +59264,7 @@ var init_build_plan = __esm({
     init_build_asset_reference();
     init_build_cache();
     init_build_asset_metadata();
+    init_build_backdrop_prompt();
     init_composer_resolve();
     init_storyboard_parse();
     init_project_config();
@@ -60428,8 +60679,9 @@ async function dispatchNarration(beat, ctx) {
 async function dispatchBackdrop(beat, ctx) {
   const reference = assetReferenceForBeat(ctx.projectDir, "backdrop", beat);
   if (reference) return referencePrimitiveOutcome("backdrop", beat, ctx, reference);
-  const prompt3 = stringOrUndefined4(beat.cues?.backdrop);
-  if (!prompt3) return { status: "no-cue" };
+  const cue = stringOrUndefined4(beat.cues?.backdrop);
+  if (!cue) return { status: "no-cue" };
+  const prompt3 = augmentBackdropPrompt(cue);
   const rel = `assets/backdrop-${beat.id}.png`;
   const abs = join22(ctx.projectDir, rel);
   const size = ctx.imageSize ?? "1536x1024";
@@ -61407,6 +61659,7 @@ var init_scene_build = __esm({
     init_root_sync();
     init_build_cache();
     init_build_asset_metadata();
+    init_build_backdrop_prompt();
     init_ai_video();
     init_music();
     init_status_jobs();
@@ -68156,6 +68409,9 @@ var init_generate2 = __esm({
           "Video provider (default: seedance when FAL_API_KEY is configured, otherwise first configured provider)"
         ),
         image: external_exports.string().optional().describe("Reference image path for image-to-video"),
+        refImages: external_exports.array(external_exports.string()).optional().describe("Reference images for Seedance reference-to-video"),
+        refVideos: external_exports.array(external_exports.string()).optional().describe("Reference videos for Seedance reference-to-video"),
+        refAudio: external_exports.array(external_exports.string()).optional().describe("Reference audio files for Seedance reference-to-video"),
         duration: external_exports.number().optional().describe("Duration in seconds (default: 5; Seedance accepts 4-15)"),
         ratio: external_exports.string().optional().describe("Aspect ratio: 16:9, 9:16, 1:1 (default: 16:9)"),
         mode: external_exports.string().optional().describe("Kling mode: std or pro"),
@@ -68164,6 +68420,7 @@ var init_generate2 = __esm({
         veoModel: external_exports.string().optional().describe("Veo model: 3.0, 3.1, 3.1-fast"),
         runwayModel: external_exports.string().optional().describe("Runway model: gen4.5, gen4_turbo"),
         seedanceModel: external_exports.string().optional().describe("Seedance variant: quality or fast (fal.ai only)"),
+        generateAudio: external_exports.boolean().optional().describe("Generate native synchronized audio when supported"),
         output: external_exports.string().optional().describe("Output file path (downloads video)"),
         wait: external_exports.boolean().optional().describe("Wait for completion (default: true)")
       }),
@@ -71397,7 +71654,7 @@ init_source();
 // package.json
 var package_default = {
   name: "@vibeframe/cli",
-  version: "0.104.2",
+  version: "0.105.1",
   description: "VibeFrame CLI - natural language video editing from the terminal",
   private: false,
   license: "MIT",
@@ -79966,7 +80223,7 @@ function fourBeatDraft(opts) {
       title: scenic ? "First light" : "Hook",
       duration: opts.durations[0],
       narration: scenic ? `The first light reaches ${opts.product} before the day begins.` : `${opts.product} starts with one clear promise.`,
-      backdrop: scenic ? `misty mountain ridge at sunrise, golden light, fog layers, cinematic ${opts.theme}` : `polished opening frame for ${opts.product}, ${opts.theme}, clear hero subject`,
+      backdrop: scenic ? `misty mountain ridge at sunrise, golden light, fog layers, cinematic ${opts.theme}` : `topic-aligned editorial background plate for ${opts.product}, ${opts.theme}, no readable text, no logos, no unrelated consumer products, clean negative space for HTML overlays`,
       motion: scenic ? "quiet title reveal, soft parallax, slow camera drift" : "large readable headline, restrained camera push, immediate value reveal",
       body: scenic ? "Show the mountain identity immediately. Keep text minimal." : "Make the value obvious in the first beat. Avoid setup that delays the payoff."
     },
@@ -79975,7 +80232,7 @@ function fourBeatDraft(opts) {
       title: scenic ? "The path appears" : "Proof",
       duration: opts.durations[1],
       narration: scenic ? "A clear path appears when every step is visible." : "The proof appears when each step is visible.",
-      backdrop: scenic ? "ridge trail emerging through fog, warm sunrise edge light, high contrast" : `concrete proof frame for ${opts.product}, visible workflow, before and after contrast`,
+      backdrop: scenic ? "ridge trail emerging through fog, warm sunrise edge light, high contrast" : `topic-aligned proof background plate for ${opts.product}, abstract workflow structure, no readable text, no product photos, no shoes, no unrelated objects`,
       motion: scenic ? "thin line traces the path, small labels fade in" : "interface panels or proof points slide into place with precise transitions",
       body: "Connect the visual proof to the promise in one focused moment."
     },
@@ -79993,7 +80250,7 @@ function fourBeatDraft(opts) {
       title: scenic ? "Day one" : "Close",
       duration: opts.durations[3],
       narration: scenic ? `${opts.product}. Day one begins here.` : `${opts.product} is ready to show.`,
-      backdrop: scenic ? "wide mountain ridge at peak golden hour, valley filled with warm light" : `resolved final frame for ${opts.product}, clean negative space, confident lockup`,
+      backdrop: scenic ? "wide mountain ridge at peak golden hour, valley filled with warm light" : `resolved editorial background plate for ${opts.product}, clean negative space, confident lockup, no readable text, no logos, no unrelated products`,
       motion: "clean final title, slow settling motion, gentle fade to the final frame",
       body: "End with a quiet branded title and no new idea."
     }
@@ -80006,7 +80263,7 @@ function threeBeatDraft(opts) {
       title: "Hook",
       duration: opts.durations[0],
       narration: `${opts.product} starts with one clear promise.`,
-      backdrop: `polished opening frame for ${opts.product}, ${opts.theme}, clear hero subject`,
+      backdrop: `topic-aligned editorial background plate for ${opts.product}, ${opts.theme}, no readable text, no logos, no unrelated consumer products, clean negative space for HTML overlays`,
       motion: "large readable headline, restrained camera push, immediate value reveal",
       body: "Make the value obvious in one beat."
     },
@@ -80015,7 +80272,7 @@ function threeBeatDraft(opts) {
       title: "Proof",
       duration: opts.durations[1],
       narration: "The proof appears when each step is visible.",
-      backdrop: `concrete proof frame for ${opts.product}, visible workflow, before and after contrast`,
+      backdrop: `topic-aligned proof background plate for ${opts.product}, abstract workflow structure, no readable text, no product photos, no shoes, no unrelated objects`,
       motion: "layered panels, highlighted cause and effect, precise transitions",
       body: "Show the mechanism, metric, or before/after that makes the promise believable."
     },
@@ -80024,7 +80281,7 @@ function threeBeatDraft(opts) {
       title: "Close",
       duration: opts.durations[2],
       narration: `${opts.product} is ready to show.`,
-      backdrop: `resolved final frame for ${opts.product}, clean negative space, confident lockup`,
+      backdrop: `resolved editorial background plate for ${opts.product}, clean negative space, confident lockup, no readable text, no logos, no unrelated products`,
       motion: "clean final title, subtle parallax, gentle fade to the final frame",
       body: "End on the product name, offer, command, or memorable final line."
     }
@@ -80257,6 +80514,12 @@ principles, type system, and visual-identity hard-gate. The same skill
 governs \`vibe scene lint\` so your authored HTML and the linter stay in
 agreement.
 
+For render-stable text, do not apply continuous \`scale\`, \`x\`, \`y\`,
+\`filter\`, or other transform tweens to \`.scene-content\` or any ancestor
+containing live text/cards. Put ambient zoom/parallax on background or media
+layers only; text should enter briefly and then hold still at its final CSS
+position.
+
 To retro-install on a project scaffolded before this command existed, run
 \`vibe scene install-skill [--host all]\`.
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vibeframe/cli",
-  "version": "0.104.2",
+  "version": "0.105.1",
   "description": "VibeFrame CLI - natural language video editing from the terminal",
   "private": false,
   "license": "MIT",
@@ -183,8 +183,8 @@
     "tsx": "^4.21.0",
     "typescript": "^5.3.3",
     "vitest": "^1.2.2",
-    "@vibeframe/ai-providers": "0.104.2",
-    "@vibeframe/core": "0.104.2"
+    "@vibeframe/ai-providers": "0.105.1",
+    "@vibeframe/core": "0.105.1"
   },
   "engines": {
     "node": ">=20"