@koda-sl/baker-cli 0.185.2 → 0.186.0

package/dist/cli.js

@@ -29,6 +29,7 @@ import {
   isPersistedAssetRef,
   looksLikeHttpUrl,
   maxInputReferences,
+  maxInputSlot,
   nearestClipDuration,
   nearestSupportedAspectRatio,
   nearestSupportedImageSize,
@@ -46,7 +47,7 @@ import {
   toModelSafeImage,
   ulid,
   validateCanvasDeep
-} from "./chunk-3TXZKSAR.js";
+} from "./chunk-OWDTSPY4.js";
 import {
   csvOrJson,
   daysAgoIso,
@@ -3487,9 +3488,10 @@ var imagesIngestResponseSchema = z11.object({
 import { z as z12 } from "zod";
 var studioGenerationStatusSchema = z12.enum(["running", "completed", "partial", "failed"]);
 var studioVariantStatusSchema = z12.enum(["pending", "ready", "failed"]);
-var studioClipModeSchema = z12.enum(["first_frame", "references"]);
+var studioClipModeSchema = z12.enum(["first_frame", "references", "text"]);
 var videoGenerateModelSchema = z12.enum([
   "bytedance/seedance-2.0",
+  "bytedance/seedance-2.5",
   "google/gemini-omni-flash",
   "google/veo-3.1",
   "google/veo-3.1-fast",
@@ -3504,7 +3506,9 @@ var MAX_STUDIO_IMAGES = 8;
 var MAX_STUDIO_CLIPS = 3;
 var MAX_STUDIO_PROMPT_CHARS = 32e3;
 var MAX_STUDIO_IMAGE_REFERENCES = 16;
-var MAX_STUDIO_CLIP_IMAGES = 9;
+var MAX_STUDIO_CLIP_IMAGES = 30;
+var MAX_STUDIO_CLIP_REFS = 10;
+var MAX_STUDIO_CLIP_SECONDS = 30;
 var aspectRatioSchema = z12.string().regex(/^[1-9]\d*:[1-9]\d*$/, "Write the shape as width:height, like 1:1, 4:5 or 16:9");
 var seedSchema = z12.coerce.number().int().min(0).max(4294967295);
 var studioGenerateRequestSchema = z12.object({
@@ -3537,15 +3541,39 @@ var studioGenerateRequestSchema = z12.object({
   backgroundRgbColor: rgbTriple.optional()
 });
 var studioAnimateRequestSchema = z12.object({
-  /** What MOVES. The look is already fixed by the still this starts from. */
+  /**
+   * What MOVES, when a still fixes the look. With no still it is the whole shot,
+   * because the clip is then generated from the prompt alone.
+   */
   prompt: z12.string().min(1).max(MAX_STUDIO_PROMPT_CHARS),
-  imageIds: z12.array(z12.string()).min(1).max(MAX_STUDIO_CLIP_IMAGES),
+  /**
+   * The still(s) the clip is built from. MAY BE EMPTY — every curated video model
+   * also does text-to-video, so a prompt on its own is a complete request. The
+   * CLI still requires one unless the caller asks for text-to-video explicitly,
+   * because "I forgot the image" and "I meant no image" look identical here.
+   */
+  imageIds: z12.array(z12.string()).max(MAX_STUDIO_CLIP_IMAGES),
+  /**
+   * Clips this render builds on — extend, edit, or carry the motion of. Addressed
+   * by the batch and slot they were rendered in, because a studio take is a plain
+   * public file the provider can fetch while a library video is a Mux asset that
+   * is not. Seedance 2.5 only; the backend refuses them on any other model.
+   */
+  clipRefs: z12.array(z12.object({ generationId: z12.string(), slot: z12.coerce.number().int().min(0) })).max(MAX_STUDIO_CLIP_REFS).optional(),
+  /**
+   * The same thing from the VIDEO LIBRARY — footage the company uploaded or
+   * synced rather than rendered here. Ids, not URLs: a library video is a Mux
+   * asset with no downloadable file until one is encoded for it, which the
+   * backend does on first use.
+   */
+  videoIds: z12.array(z12.string()).max(MAX_STUDIO_CLIP_REFS).optional(),
   clipMode: studioClipModeSchema.optional(),
   model: videoGenerateModelSchema.optional(),
   aspectRatio: aspectRatioSchema.optional(),
-  // 15s is the longest clip any curated model renders; which lengths the chosen
-  // one actually offers is a per-model enum settled server-side.
-  duration: z12.coerce.number().int().min(1).max(15).optional(),
+  // 30s is the longest clip any curated model renders (Seedance 2.5); which
+  // lengths the chosen one actually offers is a per-model enum settled
+  // server-side, and most of the roster stops well short of this.
+  duration: z12.coerce.number().int().min(1).max(MAX_STUDIO_CLIP_SECONDS).optional(),
   resolution: z12.string().optional(),
   count: z12.coerce.number().int().min(1).max(MAX_STUDIO_CLIPS).optional(),
   generateAudio: z12.boolean().optional(),
@@ -3554,7 +3582,9 @@ var studioAnimateRequestSchema = z12.object({
   personGeneration: z12.enum(["allow_all", "allow_adult"]).optional(),
   enhancePrompt: z12.boolean().optional(),
   conditioningScale: z12.coerce.number().optional(),
-  cfgScale: z12.coerce.number().min(0).max(1).optional()
+  cfgScale: z12.coerce.number().min(0).max(1).optional(),
+  /** Seedance 2.5 only. Left unset the lane sends `false` — see `resolveClipRequest`. */
+  watermark: z12.boolean().optional()
 });
 var studioVariantSchema = z12.object({
   slot: z12.number(),
@@ -3592,7 +3622,8 @@ var studioClipSettingsSchema = z12.object({
   personGeneration: z12.string().optional(),
   enhancePrompt: z12.boolean().optional(),
   conditioningScale: z12.number().optional(),
-  cfgScale: z12.number().optional()
+  cfgScale: z12.number().optional(),
+  watermark: z12.boolean().optional()
 });
 var studioGenerationSchema = z12.object({
   generationId: z12.string(),
@@ -34862,23 +34893,33 @@ var MODEL_LIST = VIDEO_MODEL_IDS;
 var DEFAULT_MAX_WAIT_MS = 6e5;
 registerSchema({
   command: "studio.animate",
-  description: "Turn a still into a clip. The look is already fixed by the image, so the prompt describes MOVEMENT \u2014 camera, subject, timing. --image takes library image ids, local file paths and/or image URLs (paths and URLs are added to the media library first). A rendered clip is NOT in the video library until `baker studio keep` approves it.",
+  description: "Render a clip. With --image the look is already fixed, so the prompt describes MOVEMENT \u2014 camera, subject, timing; with --from text there is no image and the prompt is the whole shot. --image takes library image ids, local file paths and/or image URLs (paths and URLs are added to the media library first). A rendered clip is NOT in the video library until `baker studio keep` approves it.",
   args: {
     prompt: {
       type: "string",
-      description: "What MOVES \u2014 camera, subject, timing. Kling reads 2,500 characters and Seedance 4,000; over the cap the request is refused rather than trimmed.",
+      description: "What MOVES \u2014 camera, subject, timing. Kling reads 2,500 characters and Seedance 2.0 4,000; over the cap the request is refused rather than trimmed. Seedance 2.5 publishes no cap, which is why it is the one to reach for when the brief is a whole multi-shot sequence.",
       required: true
     },
     image: {
       type: "string",
-      description: "Comma-separated source images: library image ids, local file paths, and/or image URLs. With the default --from frame this must be exactly ONE image (the opening frame) \u2014 several are refused rather than silently reduced to the first. With --from references, up to 9 on Seedance.",
-      required: true
+      description: "Comma-separated source images: library image ids, local file paths, and/or image URLs. Required unless --from text. With the default --from frame this must be exactly ONE image (the opening frame) \u2014 several are refused rather than silently reduced to the first. With --from references, up to 9 on Seedance 2.0 and 30 on Seedance 2.5.",
+      required: false
     },
     from: {
       type: "string",
-      description: "frame (default) animates a single image as the opening frame; references uses several images as visual guidance instead. Mutually exclusive at every provider, and reference mode is Seedance/Gemini Omni only \u2014 Veo and Kling are frame-only.",
+      description: "What the clip is built FROM. frame (default) animates a single image as the opening frame; references uses several assets as visual guidance; text renders the prompt with no image at all. frame and references are mutually exclusive at every provider, and reference mode is Seedance/Gemini Omni only \u2014 Veo and Kling are frame-only. Every model does text.",
       required: false,
-      enum: ["frame", "references"]
+      enum: ["frame", "references", "text"]
+    },
+    "from-clip": {
+      type: "string",
+      description: "Comma-separated Studio takes to BUILD ON, each written batchId:slot (e.g. j57abc\u2026:0) \u2014 this is how you extend a clip, edit one, or carry its motion into a new shot. Seedance 2.5 only, up to 10 source clips in total, and requires --from references.",
+      required: false
+    },
+    "from-video": {
+      type: "string",
+      description: "The same, but from the VIDEO LIBRARY: comma-separated video ids (see `baker videos library`) to build on. Use this for footage the client uploaded or synced rather than something the Studio rendered. Counts against the same 10 source clips as --from-clip. The first use of a given video spends a minute encoding a reusable copy and is refused with a 'try again shortly' message; every use after that is instant.",
+      required: false
     },
     model: {
       type: "string",
@@ -34896,7 +34937,7 @@ registerSchema({
       // Left unset this SNAPS onto the model's set rather than failing: the lane
       // default is 5s and Veo renders only 4/6/8, so an unstated length becomes 4
       // there. An explicit length the model cannot render is refused by name.
-      description: `Clip length in seconds \u2014 per-model. Unset it and the lane default (5s) snaps onto what the model renders. Across the roster: ${CLIP_DURATIONS.join(", ")}`,
+      description: `Clip length in seconds \u2014 per-model. Unset it and the lane default (5s) snaps onto what the model renders. Past 15s only bytedance/seedance-2.5 will do it, out to 30. Across the roster: ${CLIP_DURATIONS.join(", ")}`,
       required: false
     },
     quality: {
@@ -34949,6 +34990,11 @@ registerSchema({
       description: "How closely the clip follows the prompt, 0-1 (Kling only; higher = closer to the prompt)",
       required: false
     },
+    watermark: {
+      type: "boolean",
+      description: "Burn the provider's watermark into the picture. Seedance 2.5 only, and OFF by default because it cannot be removed after the render \u2014 pass it only if a client actually wants the mark.",
+      required: false
+    },
     wait: {
       type: "string",
       description: "Pass false to get the batch id back immediately instead of waiting for the takes",
@@ -34975,8 +35021,31 @@ function clipKnobs(args) {
   if (typeof args["enhance-prompt"] === "boolean") knobs.enhancePrompt = args["enhance-prompt"];
   knobs.conditioningScale = num2(args["conditioning-scale"]);
   knobs.cfgScale = num2(args["cfg-scale"]);
+  if (args.watermark === true) knobs.watermark = true;
   return knobs;
 }
+function parseVideoIds(value) {
+  if (typeof value !== "string") return void 0;
+  const ids = value.split(",").map((part) => part.trim()).filter(Boolean);
+  return ids.length > 0 ? ids : void 0;
+}
+function parseClipRefs(value) {
+  if (typeof value !== "string" || value.trim() === "") return void 0;
+  const refs = [];
+  for (const entry of value.split(",").map((part) => part.trim())) {
+    if (!entry) continue;
+    const separator = entry.lastIndexOf(":");
+    const generationId = separator === -1 ? "" : entry.slice(0, separator);
+    const slot = Number(entry.slice(separator + 1));
+    if (!generationId || !Number.isInteger(slot) || slot < 0) {
+      throw new Error(
+        `--from-clip takes batchId:slot pairs \u2014 "${entry}" is not one. Read a batch with \`baker studio get <id>\` to see its slots.`
+      );
+    }
+    refs.push({ generationId, slot });
+  }
+  return refs.length > 0 ? refs : void 0;
+}
 function clipImageLimitFor(args) {
   const model = args.model ?? DEFAULT_VIDEO_GENERATE_MODEL;
   const label = MODEL_REGISTRY.video_generate[model]?.label ?? model;
@@ -34999,9 +35068,18 @@ function clipImageLimitFor(args) {
     refuse: (count) => `${label} takes up to ${max} reference images \u2014 you passed ${count}. Drop the weakest ones.`
   };
 }
+function clipModeFor(from) {
+  if (from === "references") return "references";
+  if (from === "text") return "text";
+  return "first_frame";
+}
 function buildAnimateBody(args, prompt, imageIds) {
   const body = { prompt, imageIds, ...clipKnobs(args) };
-  if (args.from) body.clipMode = args.from === "references" ? "references" : "first_frame";
+  const clipRefs = parseClipRefs(args["from-clip"]);
+  if (clipRefs) body.clipRefs = clipRefs;
+  const videoIds = parseVideoIds(args["from-video"]);
+  if (videoIds) body.videoIds = videoIds;
+  if (args.from) body.clipMode = clipModeFor(args.from);
   if (args.model) body.model = args.model;
   if (args["aspect-ratio"]) body.aspectRatio = args["aspect-ratio"];
   body.duration = num2(args.duration);
@@ -35021,7 +35099,12 @@ function costHintsFor(body) {
     // ceiling against an actual 5s — exactly double.
     duration: body.duration ?? nearestClipDuration(model, DEFAULT_CLIP_DURATION_S),
     resolution: body.resolution,
-    generateAudio: body.generateAudio
+    generateAudio: body.generateAudio,
+    buildsOnSourceClip: (body.clipRefs?.length ?? 0) + (body.videoIds?.length ?? 0) > 0,
+    // Reference images might carry a face, and a face sends the render to the
+    // dearer provider. Scoped to reference mode because that is the only shape
+    // that gets re-routed — a frame-started clip never does.
+    mayRouteToFal: body.clipMode === "references" && body.imageIds.length > 0
   });
   const total = perTake * count;
   const hints = [
@@ -35037,7 +35120,7 @@ function costHintsFor(body) {
 var animateCommand = defineCommand163({
   meta: {
     name: "animate",
-    description: "Turn a still into a clip. The image fixes the look, so the prompt describes MOVEMENT \u2014 what the camera does, what the subject does, in what order.\n\nA rendered clip is NOT usable anywhere until you keep it: `baker studio keep <id> --slot N` is what puts it in the video library. Takes nobody keeps are never ingested, which is what makes a rejected batch cheap.\n\nExamples:\n  baker studio animate 'slow push in, model turns to camera and smiles' --image j57abc123def456ghi789\n  baker studio animate 'handheld drift right, steam rising from the cup' --image './out/hero.png' --duration 6 --quality 1080p\n  baker studio animate 'product rotates once on a turntable' --image j57abc\u2026,j57def\u2026 --from references"
+    description: "Render a clip. With an image the look is already fixed, so the prompt describes MOVEMENT \u2014 what the camera does, what the subject does, in what order. With --from text there is no image and the prompt is the whole shot.\n\nA rendered clip is NOT usable anywhere until you keep it: `baker studio keep <id> --slot N` is what puts it in the video library. Takes nobody keeps are never ingested, which is what makes a rejected batch cheap.\n\nFor anything longer than 15 seconds, or to build on footage that already exists, use --model bytedance/seedance-2.5: it renders 4-30s and is the only model that reads an existing clip or an existing soundtrack.\n\nExamples:\n  baker studio animate 'slow push in, model turns to camera and smiles' --image j57abc123def456ghi789\n  baker studio animate 'handheld drift right, steam rising from the cup' --image './out/hero.png' --duration 6 --quality 1080p\n  baker studio animate 'product rotates once on a turntable' --image j57abc\u2026,j57def\u2026 --from references\n  baker studio animate 'she keeps walking, camera stays with her, then she stops and looks up' --image j57abc\u2026 --from references --from-clip j57batch\u2026:0 --model bytedance/seedance-2.5 --duration 20\n  baker studio animate 'hold on the product, then a slow push-in' --from references --from-video j57vid\u2026 --model bytedance/seedance-2.5\n  baker studio animate 'slow drone pull-back over a solar farm at golden hour, no people' --from text --model bytedance/seedance-2.5 --duration 12"
   },
   args: {
     prompt: { type: "positional", description: "What MOVES", required: false },
@@ -35046,7 +35129,17 @@ var animateCommand = defineCommand163({
       description: "Comma-separated library image ids, local file paths, and/or image URLs",
       required: false
     },
-    from: { type: "string", description: "frame (default) | references", required: false },
+    from: { type: "string", description: "frame (default) | references | text", required: false },
+    "from-clip": {
+      type: "string",
+      description: "Studio takes to build on, batchId:slot (Seedance 2.5, needs --from references)",
+      required: false
+    },
+    "from-video": {
+      type: "string",
+      description: "Library video ids to build on (Seedance 2.5, needs --from references)",
+      required: false
+    },
     model: {
       type: "string",
       description: `Model id (default ${DEFAULT_VIDEO_GENERATE_MODEL})`,
@@ -35079,6 +35172,11 @@ var animateCommand = defineCommand163({
     },
     "conditioning-scale": { type: "string", description: "Frame adherence", required: false },
     "cfg-scale": { type: "string", description: "Prompt adherence 0-1", required: false },
+    watermark: {
+      type: "boolean",
+      description: "Burn in the provider watermark (Seedance 2.5, off by default)",
+      required: false
+    },
     wait: { type: "string", description: "false to return the batch id immediately", required: false },
     "max-wait": { type: "string", description: "Max ms to wait (default 600000, 0 = no wait)", required: false },
     full: { type: "boolean", description: "Include settings, references and attribution", required: false }
@@ -35090,17 +35188,18 @@ var animateCommand = defineCommand163({
         writeJson({ ok: false, error: { code: "VALIDATION_ERROR", message: "A prompt is required" } });
         process.exit(1);
       }
-      if (!args.image) {
+      if (!args.image && args.from !== "text") {
         writeJson({
           ok: false,
           error: {
             code: "VALIDATION_ERROR",
-            message: "--image is required: a clip animates a still, so it needs the image to start from"
+            message: "--image is required unless you pass --from text: a clip normally animates a still, so it needs the image to start from",
+            fix: "Pass --image <id|path|url>, or --from text to render from the prompt alone."
           }
         });
         process.exit(1);
       }
-      const { imageIds } = await resolveLibraryImageIds(args.image, clipImageLimitFor(args));
+      const { imageIds } = args.image ? await resolveLibraryImageIds(args.image, clipImageLimitFor(args)) : { imageIds: [] };
       const body = buildAnimateBody(args, prompt, imageIds);
       const costHints = costHintsFor(body);
       const start = await apiPost("/api/studio/animate", body);
@@ -35619,6 +35718,7 @@ function imageCard(model) {
   };
 }
 function videoCard(model) {
+  const sourceClips = maxInputSlot("video_generate", model, "video");
   return {
     id: model,
     label: MODEL_REGISTRY.video_generate[model]?.label,
@@ -35630,6 +35730,11 @@ function videoCard(model) {
     // Only meaningful in reference mode; a first-frame clip always takes exactly
     // one image, whatever the model's reference slot allows.
     referenceMode: supportsReferenceToVideo(model) ? `up to ${maxInputReferences("video_generate", model) ?? "several"} reference images` : "first frame only \u2014 this model cannot take reference images",
+    // Stated on EVERY card, not just the one model that has it: "can this build
+    // on a clip I already made" is the question that decides whether extend and
+    // edit are possible at all, and a field that appears only when the answer is
+    // yes reads as "unknown" everywhere else.
+    buildsOnClips: sourceClips ? `up to ${sourceClips} source clips \u2014 extend, edit, or carry their motion (pass with --from-clip)` : "no \u2014 this model cannot build on an existing clip",
     knobs: knobsFor("video_generate", model)
   };
 }
@@ -35679,7 +35784,7 @@ Examples:
       data: cards,
       meta: { count: cards.length },
       hints: [
-        kind === "image" ? "Shapes snap to the nearest the model renders; every other setting is refused if the model does not have it." : "A clip starting from one exact frame uses ONE image. Only Seedance and Gemini Omni take several as references."
+        kind === "image" ? "Shapes snap to the nearest the model renders; every other setting is refused if the model does not have it." : "A clip starting from one exact frame uses ONE image. Only Seedance and Gemini Omni take several as references, and only Seedance 2.5 goes past 15s or builds on an existing clip."
       ]
     });
   }