@pixeldojo/mcp 0.20.0 → 0.21.0

package/dist/cli.js

@@ -1248,6 +1248,67 @@ var init_upscale = __esm({
   }
 });
 async function run6(input, ctx) {
+  const submitted = await ctx.client.run(MODEL, {
+    prompt: "",
+    video: input.video_url,
+    ...input.style ? { caption_style: input.style } : {},
+    ...input.highlight_color ? { highlight_color: input.highlight_color } : {}
+  });
+  const result = await pollJobUntilDoneOrHandoff(
+    submitted.jobId,
+    (id) => ctx.client.getJob(id),
+    ctx.buildStatusUrl,
+    { now: ctx.pollOverrides?.now, sleep: ctx.pollOverrides?.sleep }
+  );
+  return pollResultToSkillResult(
+    result,
+    (urls) => urls.length === 0 ? `Job ${submitted.jobId} completed but returned no assets.` : `Captioned. Asset: ${urls[0]}`
+  );
+}
+var MODEL, STYLES, InputSchema6, InputJsonSchema6, captionSkill;
+var init_caption = __esm({
+  "src/skills/caption.ts"() {
+    init_poll();
+    init_types();
+    MODEL = "video-autocaption";
+    STYLES = ["pop", "highlighter", "clean"];
+    InputSchema6 = z.object({
+      video_url: z.string().url().describe("URL of the video to caption. It must have speech."),
+      style: z.enum(STYLES).optional(),
+      highlight_color: z.string().optional()
+    });
+    InputJsonSchema6 = {
+      type: "object",
+      required: ["video_url"],
+      properties: {
+        video_url: {
+          type: "string",
+          format: "uri",
+          description: "The video to caption (MP4 or MOV). It must have speech; a silent clip is refused uncharged."
+        },
+        style: {
+          type: "string",
+          enum: STYLES,
+          description: "pop (default): bold caps, the spoken word lit and growing. highlighter: the spoken word on a block of the color. clean: sentence case on a dark box."
+        },
+        highlight_color: {
+          type: "string",
+          description: "The highlight as #RRGGBB (a brand color) or a name like yellow, green, pink. Defaults to yellow."
+        }
+      }
+    };
+    captionSkill = {
+      name: "pixeldojo_caption",
+      description: "Burn animated word-by-word captions onto a video with speech, the short-form look for TikTok, Reels and Shorts: the spoken word lights up, placed above the platform buttons. Pass video_url; optional style (pop, highlighter, clean) and highlight_color (#RRGGBB for a brand color). 5 credits.",
+      inputZod: InputSchema6,
+      inputJsonSchema: InputJsonSchema6,
+      allowedModels: [MODEL],
+      defaultModel: MODEL,
+      run: run6
+    };
+  }
+});
+async function run7(input, ctx) {
   const job = await ctx.client.getJob(input.jobId);
   if (job.status === "completed") {
     const urls = extractAssetUrls2(job);
@@ -1311,13 +1372,13 @@ function extractAssetUrls2(job) {
   if (output?.video) return [output.video];
   return [];
 }
-var InputSchema6, InputJsonSchema6, STUCK_HINT_MS, ABANDON_HINT_MS, statusSkill;
+var InputSchema7, InputJsonSchema7, STUCK_HINT_MS, ABANDON_HINT_MS, statusSkill;
 var init_status = __esm({
   "src/skills/status.ts"() {
-    InputSchema6 = z.object({
+    InputSchema7 = z.object({
       jobId: z.string().min(1).max(200).describe("Job id returned from a prior pixeldojo_* skill handoff.")
     });
-    InputJsonSchema6 = {
+    InputJsonSchema7 = {
       type: "object",
       properties: {
         jobId: {
@@ -1332,16 +1393,16 @@ var init_status = __esm({
     statusSkill = {
       name: "pixeldojo_status",
       description: 'Check once, without waiting, on a jobId that a generation tool returned instead of an asset after its 30s wait. Free. Returns asset URLs when complete, the failure reason if it failed, or the current state and job age if still running. Campaign ids go to pixeldojo_campaign_status; films report progress through pixeldojo_film with action "status".',
-      inputZod: InputSchema6,
-      inputJsonSchema: InputJsonSchema6,
+      inputZod: InputSchema7,
+      inputJsonSchema: InputJsonSchema7,
       // status is special — it doesn't dispatch to a model
       allowedModels: [],
       defaultModel: "",
-      run: run6
+      run: run7
     };
   }
 });
-async function run7(input, ctx) {
+async function run8(input, ctx) {
   const product = await ctx.client.extractProduct(input.url);
   const summary = [
     `Product: ${product.name}`,
@@ -1358,13 +1419,13 @@ function truncate(input, max) {
   if (input.length <= max) return input;
   return input.slice(0, max - 3) + "...";
 }
-var InputSchema7, InputJsonSchema7, fromUrlSkill;
+var InputSchema8, InputJsonSchema8, fromUrlSkill;
 var init_from_url = __esm({
   "src/skills/from-url.ts"() {
-    InputSchema7 = z.object({
+    InputSchema8 = z.object({
       url: z.string().min(1).max(2048).describe("Product page URL. Shopify, BigCommerce, WooCommerce, or any page with JSON-LD / OpenGraph product metadata.")
     });
-    InputJsonSchema7 = {
+    InputJsonSchema8 = {
       type: "object",
       properties: {
         url: {
@@ -1377,12 +1438,12 @@ var init_from_url = __esm({
     fromUrlSkill = {
       name: "pixeldojo_from_url",
       description: "Extract a product profile from a product page URL: name, description, image URLs (the first three are listed), and the extraction method used. Works on Shopify, BigCommerce, WooCommerce, or any page with JSON-LD / OpenGraph product metadata. Free. pixeldojo_campaign and pixeldojo_ad accept a product URL and run this extraction themselves; call this directly to check the product data first, or to reuse its images in other tools.",
-      inputZod: InputSchema7,
-      inputJsonSchema: InputJsonSchema7,
+      inputZod: InputSchema8,
+      inputJsonSchema: InputJsonSchema8,
       // No model dispatch — this skill calls a server-side scrape, not a generation.
       allowedModels: [],
       defaultModel: "",
-      run: run7
+      run: run8
     };
   }
 });
@@ -1945,7 +2006,7 @@ async function redoShots(input, ctx) {
     followUp: keepGoing(filmId, 30, { soundtrack: input.soundtrack })
   };
 }
-async function run8(input, ctx) {
+async function run9(input, ctx) {
   const action = input.action === "plan" && input.film_id ? input.cast?.length ? "cast" : "status" : input.action;
   if (action === "plan") return plan(input, ctx);
   if (!input.film_id) return { kind: "text", text: "produce/status/cast/redo need the film_id returned by the plan step." };
@@ -1965,7 +2026,7 @@ async function run8(input, ctx) {
   if (action === "redo") return redoShots(input, ctx);
   return advance(input.film_id, input.soundtrack, input.music_job_id, ctx);
 }
-var STORY_LOOKS, MUSIC_VIDEO_LOOKS, LOOKS, MUSIC_CLIP_ENGINE_ID, InputSchema8, InputJsonSchema8, shots, benignDispatchRefusal, count, frameDone, RETRY_CLEARS, REFUSED, isRefused, PLANNER_POLL_MS, PLANNER_MAX_POLLS, isMusicVideo, needsSlice, stillPlanningText, LOOK_MUSIC, ARC, filmSkill;
+var STORY_LOOKS, MUSIC_VIDEO_LOOKS, LOOKS, MUSIC_CLIP_ENGINE_ID, InputSchema9, InputJsonSchema9, shots, benignDispatchRefusal, count, frameDone, RETRY_CLEARS, REFUSED, isRefused, PLANNER_POLL_MS, PLANNER_MAX_POLLS, isMusicVideo, needsSlice, stillPlanningText, LOOK_MUSIC, ARC, filmSkill;
 var init_film = __esm({
   "src/skills/film.ts"() {
     init_upload_helpers();
@@ -1973,7 +2034,7 @@ var init_film = __esm({
     MUSIC_VIDEO_LOOKS = ["rap-video", "neon-club", "stage-performance", "retro-camcorder", "y2k-pop", "moody-rnb", "rock-grit", "dream-pop", "k-pop"];
     LOOKS = [...STORY_LOOKS, ...MUSIC_VIDEO_LOOKS];
     MUSIC_CLIP_ENGINE_ID = "minimax-h3-reference";
-    InputSchema8 = z.object({
+    InputSchema9 = z.object({
       idea: z.string().max(2e3).optional().describe("One-to-three sentence film idea. The studio writes the screenplay (free). Provide idea OR script for planning."),
       script: z.string().max(2e4).optional().describe("A full screenplay to plan instead of an idea (sluglines, action lines, dialogue)."),
       look: z.enum(LOOKS).optional().describe("Visual style carried through frames, clips, and soundtrack. Default live-action, or rap-video for a music video."),
@@ -2001,7 +2062,7 @@ var init_film = __esm({
       soundtrack: z.enum(["scene-synced", "flat", "none"]).default("scene-synced").describe("scene-synced composes one musical section per scene, pinned to the scene lengths (default). flat is a single cohesive bed. none skips music."),
       music_job_id: z.string().max(80).optional().describe("The music_job_id from the previous status reply while the score is composing, so that score is attached. A status call without it, once all clips are ready and no score is attached, starts and charges for a new score.")
     });
-    InputJsonSchema8 = {
+    InputJsonSchema9 = {
       type: "object",
       properties: {
         idea: { type: "string", description: `Plan step: one-to-three sentence film idea. The studio writes the screenplay (free). Provide idea or script. Keep the user's own words for who each character is, including their gender and pronouns ("she", "he") and species; the writer has nothing else to go on.` },
@@ -2115,11 +2176,11 @@ var init_film = __esm({
       name: "pixeldojo_film",
       description: "Make a finished short film from one idea: screenplay, storyboard frames (AI-reviewed), video clips with spoken dialogue, scene-synced soundtrack, stitched export. Or a music video: pass song_path (a local file) or song_url (your own PixelDojo upload from pixeldojo_upload, never a third-party link) with optional lyrics; the film is cut to that song, each clip lip-synced to its slice on the studio music video engine, and no score is composed: the export plays the song. Every named character must be cast (cast: your own images, a portrait prompt, or a Character Studio character) before frames render; cast entries can also replace or remove someone. Phased: plan is free; cast; produce starts charged generation; status advances production (poll every ~30s); redo renders a new take of chosen shots and status re-exports.",
       hostedDescription: "Make a finished short film from one idea: screenplay, storyboard frames (AI-reviewed), video clips with spoken dialogue, scene-synced soundtrack, stitched export. Or a music video: pass song_url (your own PixelDojo upload made with POST /api/v1/upload, never a third-party link) with optional lyrics; the film is cut to that song, each clip lip-synced to its slice on the studio music video engine, and no score is composed: the export plays the song. Every named character must be cast (cast: your own images, a portrait prompt, or a Character Studio character) before frames render; cast entries can also replace or remove someone. Phased: plan is free; cast; produce starts charged generation; status advances production (poll every ~30s); redo renders a new take of chosen shots and status re-exports.",
-      inputZod: InputSchema8,
-      inputJsonSchema: InputJsonSchema8,
+      inputZod: InputSchema9,
+      inputJsonSchema: InputJsonSchema9,
       allowedModels: [],
       defaultModel: "",
-      run: run8
+      run: run9
     };
   }
 });
@@ -2206,7 +2267,7 @@ function buildVideoPayload(verb, apiId, i) {
       };
   }
 }
-async function run9(input, ctx) {
+async function run10(input, ctx) {
   const paramsGate = paramsWithoutModel(input.params, input.model);
   if (paramsGate) return { kind: "text", text: paramsGate };
   const verb = input.verb;
@@ -2269,7 +2330,7 @@ async function run9(input, ctx) {
     notes.join(" ")
   );
 }
-var VIDEO_STUDIO_ENGINES, VERBS, BACKWARD_ENGINES, clamp, NEEDS_VIDEO_DURATION, NO_IMAGE_ENGINES, InputSchema9, engineList, InputJsonSchema9, BASE_DESCRIPTION, videoSkill;
+var VIDEO_STUDIO_ENGINES, VERBS, BACKWARD_ENGINES, clamp, NEEDS_VIDEO_DURATION, NO_IMAGE_ENGINES, InputSchema10, engineList, InputJsonSchema10, BASE_DESCRIPTION, videoSkill;
 var init_video = __esm({
   "src/skills/video.ts"() {
     init_poll();
@@ -2303,7 +2364,7 @@ var init_video = __esm({
       "wan-video-character-swap"
     ]);
     NO_IMAGE_ENGINES = /* @__PURE__ */ new Set(["flux-video-edit", "xai-video-edit"]);
-    InputSchema9 = z.object({
+    InputSchema10 = z.object({
       verb: z.enum(VERBS),
       prompt: z.string().max(8e3).optional(),
       video_url: z.string().url().optional(),
@@ -2318,7 +2379,7 @@ var init_video = __esm({
       params: ParamsInput
     });
     engineList = (verb) => `${verb}: ${VIDEO_STUDIO_ENGINES[verb].join(", ")} (default ${VIDEO_STUDIO_ENGINES[verb][0]})`;
-    InputJsonSchema9 = {
+    InputJsonSchema10 = {
       type: "object",
       properties: {
         verb: {
@@ -2379,11 +2440,11 @@ var init_video = __esm({
       name: "pixeldojo_video",
       description: `${BASE_DESCRIPTION} Local files: video_path / image_path (uploaded to 24h temp storage first).`,
       hostedDescription: `${BASE_DESCRIPTION} Media must be public http(s) URLs.`,
-      inputZod: InputSchema9,
-      inputJsonSchema: InputJsonSchema9,
+      inputZod: InputSchema10,
+      inputJsonSchema: InputJsonSchema10,
       allowedModels: Array.from(new Set(VERBS.flatMap((v) => VIDEO_STUDIO_ENGINES[v]))),
       defaultModel: VIDEO_STUDIO_ENGINES.change[0],
-      run: run9
+      run: run10
     };
   }
 });
@@ -2558,28 +2619,28 @@ async function produce2(id, ctx) {
   const musicNote = ep.musicEnabled && ep.music?.status === "failed" ? " The music bed failed, so this cut has no music." : "";
   return { kind: "text", text: `Export started, free${prior}.${musicNote} ${next(id, "produce", "~60s")}` };
 }
-async function run10(input, ctx) {
+async function run11(input, ctx) {
   const action = input.action === "create" && input.episode_id ? "status" : input.action;
   if (action === "create") return create(input, ctx);
   if (!input.episode_id) return { kind: "text", text: "produce and status need the episode_id returned by the create step." };
   if (action === "produce") return produce2(input.episode_id, ctx);
   return status(input.episode_id, ctx);
 }
-var STYLES, CATEGORIES, ENGINES, MIN_SECONDS, MAX_SECONDS, MAX_IN_FLIGHT, InputSchema10, InputJsonSchema10, call, next, needsRender, PLAN_POLL_MS, PLAN_MAX_POLLS, stillPlanning, episodeSkill;
+var STYLES2, CATEGORIES, ENGINES, MIN_SECONDS, MAX_SECONDS, MAX_IN_FLIGHT, InputSchema11, InputJsonSchema11, call, next, needsRender, PLAN_POLL_MS, PLAN_MAX_POLLS, stillPlanning, episodeSkill;
 var init_episode = __esm({
   "src/skills/episode.ts"() {
-    STYLES = ["editorial-motion", "paper-diorama", "hand-drawn", "storybook", "poster-vector", "soft-3d"];
+    STYLES2 = ["editorial-motion", "paper-diorama", "hand-drawn", "storybook", "poster-vector", "soft-3d"];
     CATEGORIES = ["education", "history", "kids", "storytelling", "custom"];
     ENGINES = ["omni-draft", "omni-final", "still-motion"];
     MIN_SECONDS = 60;
     MAX_SECONDS = 300;
     MAX_IN_FLIGHT = 4;
-    InputSchema10 = z.object({
+    InputSchema11 = z.object({
       action: z.enum(["create", "produce", "status"]).default("create"),
       topic: z.string().min(8).max(400).optional(),
       episode_id: z.string().max(40).optional(),
       category: z.enum(CATEGORIES).default("education"),
-      style: z.enum(STYLES).default("editorial-motion"),
+      style: z.enum(STYLES2).default("editorial-motion"),
       aspect: z.enum(["16:9", "9:16"]).default("16:9"),
       length_seconds: z.number().int().min(MIN_SECONDS).max(MAX_SECONDS).default(120),
       voice: z.string().min(1).max(80).default("Wise_Woman"),
@@ -2587,7 +2648,7 @@ var init_episode = __esm({
       music: z.boolean().default(true),
       title: z.string().max(120).optional()
     });
-    InputJsonSchema10 = {
+    InputJsonSchema11 = {
       type: "object",
       properties: {
         action: {
@@ -2598,7 +2659,7 @@ var init_episode = __esm({
         topic: { type: "string", description: "Create step: what the episode is about, 8 to 400 characters." },
         episode_id: { type: "string", description: "The episode_id from the create step. Required for produce and status." },
         category: { type: "string", enum: [...CATEGORIES], description: "Create step: tone of the script. Default education." },
-        style: { type: "string", enum: [...STYLES], description: "Create step: visual style of every scene. Default editorial-motion." },
+        style: { type: "string", enum: [...STYLES2], description: "Create step: visual style of every scene. Default editorial-motion." },
         aspect: { type: "string", enum: ["16:9", "9:16"], description: "Create step: 16:9 landscape (default) or 9:16 vertical." },
         length_seconds: { type: "integer", minimum: MIN_SECONDS, maximum: MAX_SECONDS, description: "Create step: target length in seconds. Default 120." },
         voice: { type: "string", description: "Create step: narrator voice id from the text-to-speech tool (e.g. Wise_Woman, Deep_Voice_Man, Friendly_Person). Default Wise_Woman." },
@@ -2620,15 +2681,15 @@ var init_episode = __esm({
     episodeSkill = {
       name: "pixeldojo_episode",
       description: "Make a finished narrated faceless video (explainer, history, kids story) from one topic: script, scene visuals, AI narration, music bed, stitched export saved to the Library. Phased: create is free and returns a credit quote; produce is charged and advances one step per call (poll every ~30s until it returns the video URL); status is free.",
-      inputZod: InputSchema10,
-      inputJsonSchema: InputJsonSchema10,
+      inputZod: InputSchema11,
+      inputJsonSchema: InputJsonSchema11,
       allowedModels: [],
       defaultModel: "",
-      run: run10
+      run: run11
     };
   }
 });
-async function run11(input, ctx) {
+async function run12(input, ctx) {
   const productImages = [
     ...input.product_image_urls ?? [],
     ...await Promise.all((input.product_image_paths ?? []).map((p) => uploadLocalFile(p, ctx)))
@@ -2671,13 +2732,13 @@ async function run11(input, ctx) {
     (urls) => urls.length === 0 ? `Ad job ${submitted.jobId} completed but returned no assets.` : `Ad rendered (style: ${styleLabel}${submitted.ad?.product?.name ? `, product: ${submitted.ad.product.name}` : ""}). Video: ${urls[0]}`
   );
 }
-var InputSchema11, InputJsonSchema11, adSkill;
+var InputSchema12, InputJsonSchema12, adSkill;
 var init_ad = __esm({
   "src/skills/ad.ts"() {
     init_types();
     init_poll();
     init_upload_helpers();
-    InputSchema11 = z.object({
+    InputSchema12 = z.object({
       style: z.string().max(80).optional().describe(
         "Ad style slug (default ugc-testimonial). Catalog: GET /api/v1/marketing/presets \u2014 e.g. ugc-testimonial, unboxing-reveal, unboxing-asmr, virtual-try-on, street-style, product-showcase, macro-details, crush-test, hyper-motion, before-after, day-in-the-life, the-gift, tv-spot."
       ),
@@ -2703,7 +2764,7 @@ var init_ad = __esm({
         path: ["product_url"]
       }
     );
-    InputJsonSchema11 = {
+    InputJsonSchema12 = {
       type: "object",
       properties: {
         style: {
@@ -2790,16 +2851,16 @@ var init_ad = __esm({
       name: "pixeldojo_ad",
       description: "Marketing Studio: turn a product into a rendered video ad with voiceover. Requires `product_url` (name, description and images are extracted), or `product_name` plus product_image_urls / product_image_paths. Pick a `style` slug (default ugc-testimonial; full catalog at GET https://pixeldojo.ai/api/v1/marketing/presets), and optionally add presenter reference images, a location, a hook, and the exact spoken script. Returns the video URL, or a jobId to check with pixeldojo_status.",
       hostedDescription: "Marketing Studio: turn a product into a rendered video ad with voiceover. Requires `product_url` (name, description and images are extracted), or `product_name` plus product_image_urls (public http(s) URLs). Pick a `style` slug (default ugc-testimonial; full catalog at GET https://pixeldojo.ai/api/v1/marketing/presets), and optionally add presenter reference images, a location, a hook, and the exact spoken script. Returns the video URL, or a jobId to check with pixeldojo_status.",
-      inputZod: InputSchema11,
-      inputJsonSchema: InputJsonSchema11,
+      inputZod: InputSchema12,
+      inputJsonSchema: InputJsonSchema12,
       // Composition skill — the server routes to the ad engine internally.
       allowedModels: [],
       defaultModel: "",
-      run: run11
+      run: run12
     };
   }
 });
-async function run12(input, ctx) {
+async function run13(input, ctx) {
   const client = ctx.client;
   const result = await client.runCampaign(input);
   if (result.kind === "campaign") {
@@ -2826,10 +2887,10 @@ async function run12(input, ctx) {
     statusUrl: result.statusUrl ?? ctx.buildStatusUrl(result.campaignId)
   };
 }
-var InputSchema12, InputJsonSchema12, campaignSkill;
+var InputSchema13, InputJsonSchema13, campaignSkill;
 var init_campaign = __esm({
   "src/skills/campaign.ts"() {
-    InputSchema12 = z.object({
+    InputSchema13 = z.object({
       productUrl: z.string().url().optional().describe("Product page URL. We extract name + description + images automatically."),
       product: z.object({
         name: z.string().min(1).max(200),
@@ -2845,7 +2906,7 @@ var init_campaign = __esm({
       message: "productUrl or product is required",
       path: ["productUrl"]
     });
-    InputJsonSchema12 = {
+    InputJsonSchema13 = {
       type: "object",
       properties: {
         productUrl: {
@@ -2896,17 +2957,17 @@ var init_campaign = __esm({
     campaignSkill = {
       name: "pixeldojo_campaign",
       description: "Generate a coherent product launch package (hero image + N lifestyle variants + optional vertical video) from one URL or product profile. Composes hero / lifestyle / video models server-side; returns assets in one call or a handoff with campaignId to poll via pixeldojo_campaign_status.",
-      inputZod: InputSchema12,
-      inputJsonSchema: InputJsonSchema12,
+      inputZod: InputSchema13,
+      inputJsonSchema: InputJsonSchema13,
       // Composition skill — internal sub-jobs route to per-tool models, no
       // single allowed-models list applies here.
       allowedModels: [],
       defaultModel: "",
-      run: run12
+      run: run13
     };
   }
 });
-async function run13(input, ctx) {
+async function run14(input, ctx) {
   const client = ctx.client;
   const result = await client.getCampaign(input.campaignId);
   if (result.kind === "campaign") {
@@ -2936,13 +2997,13 @@ async function run13(input, ctx) {
     statusUrl: result.statusUrl ?? ctx.buildStatusUrl(result.campaignId)
   };
 }
-var InputSchema13, InputJsonSchema13, campaignStatusSkill;
+var InputSchema14, InputJsonSchema14, campaignStatusSkill;
 var init_campaign_status = __esm({
   "src/skills/campaign-status.ts"() {
-    InputSchema13 = z.object({
+    InputSchema14 = z.object({
       campaignId: z.string().min(1).max(200).describe("Campaign id returned from a prior pixeldojo_campaign handoff.")
     });
-    InputJsonSchema13 = {
+    InputJsonSchema14 = {
       type: "object",
       properties: {
         campaignId: {
@@ -2955,15 +3016,15 @@ var init_campaign_status = __esm({
     campaignStatusSkill = {
       name: "pixeldojo_campaign_status",
       description: "Poll a campaign by campaignId returned from pixeldojo_campaign. Returns assets when every sub-job is done, or a handoff describing what is still in flight.",
-      inputZod: InputSchema13,
-      inputJsonSchema: InputJsonSchema13,
+      inputZod: InputSchema14,
+      inputJsonSchema: InputJsonSchema14,
       allowedModels: [],
       defaultModel: "",
-      run: run13
+      run: run14
     };
   }
 });
-async function run14(input, ctx) {
+async function run15(input, ctx) {
   let url;
   try {
     url = await uploadLocalFile(input.path, ctx);
@@ -2979,16 +3040,16 @@ async function run14(input, ctx) {
 (temp storage; auto-deletes in ~24h. Pass this URL as image_url, video_url for a clip, or song_url on pixeldojo_film for a song, to any pixeldojo tool that takes one.)`
   };
 }
-var InputSchema14, InputJsonSchema14, uploadSkill;
+var InputSchema15, InputJsonSchema15, uploadSkill;
 var init_upload = __esm({
   "src/skills/upload.ts"() {
     init_upload_helpers();
-    InputSchema14 = z.object({
+    InputSchema15 = z.object({
       path: z.string().min(1).max(4096).describe(
         "Absolute local file path to upload. JPEG, PNG, WebP, GIF (\u226450MB), MP4, WebM, MOV (\u2264200MB) or MP3, WAV, M4A, AAC, OGG, FLAC audio (\u226450MB). Returns a public URL on the temp CDN; storage auto-expires after 24 hours."
       )
     });
-    InputJsonSchema14 = {
+    InputJsonSchema15 = {
       type: "object",
       properties: {
         path: {
@@ -3001,15 +3062,15 @@ var init_upload = __esm({
     uploadSkill = {
       name: "pixeldojo_upload",
       description: "Upload a local image, video or audio file (JPEG, PNG, WebP, GIF up to 50MB; MP4, WebM, MOV up to 200MB; MP3, WAV, M4A, AAC, OGG, FLAC up to 50MB) and get back a public URL that expires after 24 hours. Free. Pass an audio URL as song_url to pixeldojo_film for a music video. Pass the URL as image_url (stills) or video_url (clips) to tools that take hosted URLs only, such as pixeldojo_character, pixeldojo_storyboard and pixeldojo_upscale. pixeldojo_generate, pixeldojo_edit and pixeldojo_ad take local paths directly and need no separate upload.",
-      inputZod: InputSchema14,
-      inputJsonSchema: InputJsonSchema14,
+      inputZod: InputSchema15,
+      inputJsonSchema: InputJsonSchema15,
       allowedModels: [],
       defaultModel: "",
-      run: run14
+      run: run15
     };
   }
 });
-async function run15(input, ctx) {
+async function run16(input, ctx) {
   const requested = input.model ?? DEFAULT_MODEL6;
   if (!ALLOWED4.includes(requested)) {
     return {
@@ -3047,14 +3108,14 @@ async function run15(input, ctx) {
     (urls) => urls.length === 0 ? `Job ${submitted.jobId} completed but returned no assets.` : `Audio ready. Asset: ${urls[0]}`
   );
 }
-var ALLOWED4, DEFAULT_MODEL6, InputSchema15, InputJsonSchema15, audioSkill;
+var ALLOWED4, DEFAULT_MODEL6, InputSchema16, InputJsonSchema16, audioSkill;
 var init_audio = __esm({
   "src/skills/audio.ts"() {
     init_poll();
     init_types();
     ALLOWED4 = ["seed-audio", "text-to-speech"];
     DEFAULT_MODEL6 = "seed-audio";
-    InputSchema15 = z.object({
+    InputSchema16 = z.object({
       prompt: z.string().min(1).describe(
         "seed-audio: a script to speak, or a scene description (dialogue, sound effects, music), up to 2048 characters; refer to audio_urls clips in order as @Audio1..@Audio3. text-to-speech: the exact text to speak, up to 10,000 characters."
       ),
@@ -3064,7 +3125,7 @@ var init_audio = __esm({
       output_format: z.string().optional().describe("Output format. seed-audio: mp3 (default), wav, pcm, ogg_opus. text-to-speech: mp3 (default), wav, flac, pcm."),
       speed: z.number().min(0.5).max(2).optional().describe("Speech speed multiplier, 0.5-2 (default 1). Both models.")
     });
-    InputJsonSchema15 = {
+    InputJsonSchema16 = {
       type: "object",
       properties: {
         prompt: {
@@ -3102,15 +3163,15 @@ var init_audio = __esm({
     audioSkill = {
       name: "pixeldojo_audio",
       description: "Generate audio: speech, multi-character dialogue, sound effects, and background music from one prompt (Seed Audio 1.0 by default; plain text-to-speech available). Voice cloning takes hosted reference clip URLs in audio_urls; local audio files are not accepted. Returns an audio URL, or a jobId for pixeldojo_status.",
-      inputZod: InputSchema15,
-      inputJsonSchema: InputJsonSchema15,
+      inputZod: InputSchema16,
+      inputJsonSchema: InputJsonSchema16,
       allowedModels: [...ALLOWED4],
       defaultModel: DEFAULT_MODEL6,
-      run: run15
+      run: run16
     };
   }
 });
-async function run16(input, ctx) {
+async function run17(input, ctx) {
   if (input.source === "characters") {
     const { characters } = await ctx.client.listCharacters({ q: input.q, limit: input.limit ?? 25 });
     if (characters.length === 0) {
@@ -3161,17 +3222,17 @@ ${lines.join("\n")}
 Pass any URL back into pixeldojo_generate, pixeldojo_edit, pixeldojo_character, or pixeldojo_upscale as a reference input.`
   };
 }
-var InputSchema16, InputJsonSchema16, librarySkill;
+var InputSchema17, InputJsonSchema17, librarySkill;
 var init_library = __esm({
   "src/skills/library.ts"() {
-    InputSchema16 = z.object({
+    InputSchema17 = z.object({
       source: z.enum(["media", "characters"]).optional().describe("media (default): saved and recent results. characters: saved Character Studio characters and elements, usable as @Name in prompts."),
       q: z.string().optional().describe("Substring match against the prompt that produced the asset."),
       model: z.string().optional().describe("Restrict to a single model (apiId)."),
       modality: z.enum(["image", "video", "audio"]).optional().describe("Restrict to one output type."),
       limit: z.number().int().min(1).max(50).optional().describe("Max results (default 10).")
     });
-    InputJsonSchema16 = {
+    InputJsonSchema17 = {
       type: "object",
       properties: {
         source: {
@@ -3196,15 +3257,15 @@ var init_library = __esm({
     librarySkill = {
       name: "pixeldojo_library",
       description: "Search your Library, newest first. source=media (default): saved media (permanent) and recent API/MCP generations (deleted 1 hour after they finish), as URLs to reuse as references. source=characters: saved Character Studio characters and elements, each usable as @Name in any prompt.",
-      inputZod: InputSchema16,
-      inputJsonSchema: InputJsonSchema16,
+      inputZod: InputSchema17,
+      inputJsonSchema: InputJsonSchema17,
       allowedModels: [],
       defaultModel: "",
-      run: run16
+      run: run17
     };
   }
 });
-async function run17(input, ctx) {
+async function run18(input, ctx) {
   const res = await ctx.client.saveMedia({
     url: input.url,
     prompt: input.prompt,
@@ -3218,17 +3279,17 @@ async function run17(input, ctx) {
 It will appear at https://pixeldojo.ai/my-images and in pixeldojo_library searches.`
   };
 }
-var InputSchema17, InputJsonSchema17, saveSkill;
+var InputSchema18, InputJsonSchema18, saveSkill;
 var init_save = __esm({
   "src/skills/save.ts"() {
-    InputSchema17 = z.object({
+    InputSchema18 = z.object({
       url: z.string().url().describe("The asset URL to save (image or video)."),
       prompt: z.string().optional().describe("The prompt that produced it \u2014 makes it findable in the library later."),
       tool_name: z.string().optional().describe("Tool/model slug that produced it."),
       modality: z.enum(["image", "video", "audio"]).optional(),
       job_id: z.string().optional().describe("The jobId that produced this asset, when known.")
     });
-    InputJsonSchema17 = {
+    InputJsonSchema18 = {
       type: "object",
       properties: {
         url: { type: "string", format: "uri", description: "The result URL to keep (image or video from a PixelDojo job)." },
@@ -3242,15 +3303,15 @@ var init_save = __esm({
     saveSkill = {
       name: "pixeldojo_save",
       description: `Save a generated image or video into the user's My Media library permanently ("save this", "keep it"). Results otherwise expire 1 hour after they finish. Pass the result URL; it is safety-checked before it is kept.`,
-      inputZod: InputSchema17,
-      inputJsonSchema: InputJsonSchema17,
+      inputZod: InputSchema18,
+      inputJsonSchema: InputJsonSchema18,
       allowedModels: [],
       defaultModel: "",
-      run: run17
+      run: run18
     };
   }
 });
-async function run18(input, ctx) {
+async function run19(input, ctx) {
   const res = await ctx.client.saveCharacter({
     name: input.name,
     kind: input.kind ?? "character",
@@ -3262,17 +3323,17 @@ async function run18(input, ctx) {
     text: `Saved ${res.kind} "${res.name}" to Character Studio (${res.image_urls.length} image${res.image_urls.length === 1 ? "" : "s"}). Use it in any prompt as ${res.mention}, for example "${res.mention} walking on a beach at sunset". It also appears at https://pixeldojo.ai/characters.`
   };
 }
-var KINDS, InputSchema18, InputJsonSchema18, saveCharacterSkill;
+var KINDS, InputSchema19, InputJsonSchema19, saveCharacterSkill;
 var init_save_character = __esm({
   "src/skills/save-character.ts"() {
     KINDS = ["character", "product", "outfit", "location", "object", "style"];
-    InputSchema18 = z.object({
+    InputSchema19 = z.object({
       name: z.string().min(1).max(60).describe("What to call it. Used as @Name in prompts."),
       kind: z.enum(KINDS).optional().describe("Default character."),
       image_urls: z.array(z.string().url()).max(8).optional().describe("3 to 5 photos from different angles works best for a character."),
       description: z.string().max(1500).optional().describe("Appearance notes, or the scene text for an element.")
     });
-    InputJsonSchema18 = {
+    InputJsonSchema19 = {
       type: "object",
       properties: {
         name: { type: "string", description: "What to call it. Used as @Name in prompts afterwards." },
@@ -3290,11 +3351,11 @@ var init_save_character = __esm({
     saveCharacterSkill = {
       name: "pixeldojo_save_character",
       description: "Save a character (a person, from 3-5 photos) or an element (product, outfit, location, prop, style) to the user's Character Studio library. Afterwards write @Name in any prompt to use it. List saved ones with pixeldojo_library source=characters.",
-      inputZod: InputSchema18,
-      inputJsonSchema: InputJsonSchema18,
+      inputZod: InputSchema19,
+      inputJsonSchema: InputJsonSchema19,
       allowedModels: [],
       defaultModel: "",
-      run: run18
+      run: run19
     };
   }
 });
@@ -3543,6 +3604,7 @@ __export(skills_exports, {
   audioSkill: () => audioSkill,
   campaignSkill: () => campaignSkill,
   campaignStatusSkill: () => campaignStatusSkill,
+  captionSkill: () => captionSkill,
   characterSkill: () => characterSkill,
   editSkill: () => editSkill,
   episodeSkill: () => episodeSkill,
@@ -3590,6 +3652,7 @@ var init_skills = __esm({
     init_character();
     init_storyboard();
     init_upscale();
+    init_caption();
     init_status();
     init_from_url();
     init_film();
@@ -3610,6 +3673,7 @@ var init_skills = __esm({
       characterSkill,
       storyboardSkill,
       upscaleSkill,
+      captionSkill,
       audioSkill,
       librarySkill,
       saveSkill,
@@ -3649,6 +3713,7 @@ var init_skills = __esm({
       pixeldojo_character: render("Consistent character image"),
       pixeldojo_storyboard: render("Storyboard preview video"),
       pixeldojo_upscale: render("Upscale image or video"),
+      pixeldojo_caption: render("Caption video"),
       pixeldojo_audio: render("Generate music, speech or sound"),
       pixeldojo_video: render("Change, animate or extend video"),
       pixeldojo_ad: render("Make video ad"),
@@ -3719,7 +3784,7 @@ function makeRedactedLogger(target = console) {
 }
 
 // src/server.ts
-var PACKAGE_VERSION = "0.20.0" ;
+var PACKAGE_VERSION = "0.21.0" ;
 var SERVER_INSTRUCTIONS = "Pixel Dojo generates images, video, audio and short films from one catalog of AI models, billed in credits from the connected Pixel Dojo account. A job that runs past the 30-second wait returns a jobId to check later; check it rather than submitting again, since every new submission is charged. Generated asset URLs are deleted one hour after they are created: pixeldojo_save keeps one in the user's My Media permanently (after a safety check), and finished films stay in the account library. In any prompt, @Name uses a character, product, outfit, location, prop or style the user saved in Character Studio: its photos are attached automatically. pixeldojo_library source=characters lists them, and pixeldojo_save_character saves a new one from 3 to 5 photos.";
 function buildServer(options) {
   if (!options.apiKey) {
@@ -3891,6 +3956,7 @@ function printInit() {
       `#    pixeldojo_character      Consistent characters across shots.`,
       `#    pixeldojo_storyboard     Multi-shot scenes from one brief.`,
       `#    pixeldojo_upscale        Enhance any image or video.`,
+      `#    pixeldojo_caption        Animated word-by-word captions on a video.`,
       `#    pixeldojo_audio          Voices + effects + music in one clip.`,
       `#    pixeldojo_library        Search your saved media + generations.`,
       `#    pixeldojo_*_workflow     Save / run / list multi-step chains.`,