@pixeldojo/mcp 0.30.0 → 0.31.0

package/dist/index.js

@@ -549,7 +549,7 @@ var init_generate = __esm({
     InputSchema = z.object({
       prompt: z.string().min(1).max(8e3).describe('Prompt describing what to generate. Write @Name to use a character or element the user saved in Character Studio ("@Mara on a beach"): its photos are attached automatically on models that take reference images. pixeldojo_library source=characters lists them.'),
       model: z.string().optional().describe(
-        `Model apiId. Recommended (the models the PixelDojo studios use): "seedream-5" for images (the default), "seedance-2-5" for video (prompt only or a starting image), "wan-3-0-video" for video the others refuse. Use another model only when the user names one. Full catalog: https://pixeldojo.ai/api/v1/models; one model's fields: https://pixeldojo.ai/llm/{apiId}.txt. Omit for seedream-5, which makes still images only; name a video model to get a video.`
+        `Model apiId. Recommended (the models the PixelDojo studios use): "seedream-5" for images (the default), "seedance-2-5" for video (prompt only or a starting image), "kling-4-0" for a whole scene up to 30 s written as a shot list (dialogue in quotes, exact on-screen text), "minimax-h3-turbo" for the cheapest video from a photo or up to 9 reference images (it needs at least one picture), "wan-3-0-video" for video the others refuse. Use another model only when the user names one. Full catalog: https://pixeldojo.ai/api/v1/models; one model's fields: https://pixeldojo.ai/llm/{apiId}.txt. Omit for seedream-5, which makes still images only; name a video model to get a video.`
       ),
       preset: z.string().optional().describe(
         "Preset slug that seeds the running model's settings; explicit fields and params override it. Each preset belongs to one model (browse them by model at https://pixeldojo.ai/presets), so set `model` to that preset's model: a slug the running model has no preset for is rejected with a 400 before any charge."
@@ -588,7 +588,7 @@ var init_generate = __esm({
         prompt: { type: "string", description: 'Prompt describing what to generate. Write @Name to use a character or element the user saved in Character Studio ("@Mara on a beach"): its photos are attached automatically on models that take reference images. pixeldojo_library source=characters lists them.' },
         model: {
           type: "string",
-          description: `Model apiId. Recommended (the models the PixelDojo studios use): "seedream-5" for images (the default), "seedance-2-5" for video (prompt only or a starting image), "wan-3-0-video" for video the others refuse. Use another model only when the user names one. Full catalog: https://pixeldojo.ai/api/v1/models; one model's fields: https://pixeldojo.ai/llm/{apiId}.txt. Omit for seedream-5, which makes still images only; name a video model to get a video.`
+          description: `Model apiId. Recommended (the models the PixelDojo studios use): "seedream-5" for images (the default), "seedance-2-5" for video (prompt only or a starting image), "kling-4-0" for a whole scene up to 30 s written as a shot list (dialogue in quotes, exact on-screen text), "minimax-h3-turbo" for the cheapest video from a photo or up to 9 reference images (it needs at least one picture), "wan-3-0-video" for video the others refuse. Use another model only when the user names one. Full catalog: https://pixeldojo.ai/api/v1/models; one model's fields: https://pixeldojo.ai/llm/{apiId}.txt. Omit for seedream-5, which makes still images only; name a video model to get a video.`
         },
         preset: {
           type: "string",
@@ -636,8 +636,8 @@ var init_generate = __esm({
     };
     generateSkill = {
       name: "pixeldojo_generate",
-      description: "Generate an image or video from a prompt with any model in the PixelDojo catalog. With no `model`, a prompt runs on seedream-5, which makes still images only. For a video, name seedance-2-5, which works from a prompt alone or a starting image. If a provider refuses the content, retry once on wan-3-0-video (video) or seedream-5 (images), the least-restricted models. For a multi-shot film or a music video synced to a song use pixeldojo_film; to change, move or extend an existing clip use pixeldojo_video. These are the models the PixelDojo studios use; name another only when the user asks for it. References: `image_url`/`image_urls` for stills, `video_url` for a source clip, `image_path`/`image_paths` for absolute local files (uploaded to 24h temp storage first). A clip or an audio track URL with no `model` runs on seedance-2-reference; a named model that cannot read a clip is swapped for it and the result says so, while a named model with no audio input drops the track. Model-specific settings go in `params` alongside an explicit `model`. Returns asset URLs, or a jobId for pixeldojo_status when the job outlasts the 30s wait.",
-      hostedDescription: "Generate an image or video from a prompt with any model in the PixelDojo catalog. With no `model`, a prompt runs on seedream-5, which makes still images only. For a video, name seedance-2-5, which works from a prompt alone or a starting image. If a provider refuses the content, retry once on wan-3-0-video (video) or seedream-5 (images), the least-restricted models. For a multi-shot film or a music video synced to a song use pixeldojo_film; to change, move or extend an existing clip use pixeldojo_video. These are the models the PixelDojo studios use; name another only when the user asks for it. References: `image_url`/`image_urls` for stills and `video_url` for a source clip, all public http(s) URLs. A clip or an audio track URL with no `model` runs on seedance-2-reference; a named model that cannot read a clip is swapped for it and the result says so, while a named model with no audio input drops the track. Model-specific settings go in `params` alongside an explicit `model`. Returns asset URLs, or a jobId for pixeldojo_status when the job outlasts the 30s wait.",
+      description: "Generate an image or video from a prompt with any model in the PixelDojo catalog. With no `model`, a prompt runs on seedream-5, which makes still images only. For a video, name seedance-2-5, which works from a prompt alone or a starting image; kling-4-0 for a multi-shot scene up to 30 s with dialogue; minimax-h3-turbo for the lowest price when you have a photo or reference images (it refuses a prompt alone). If a provider refuses the content, retry once on wan-3-0-video (video) or seedream-5 (images), the least-restricted models. For a multi-shot film or a music video synced to a song use pixeldojo_film; to change, move or extend an existing clip use pixeldojo_video. These are the models the PixelDojo studios use; name another only when the user asks for it. References: `image_url`/`image_urls` for stills, `video_url` for a source clip, `image_path`/`image_paths` for absolute local files (uploaded to 24h temp storage first). A clip or an audio track URL with no `model` runs on seedance-2-reference; a named model that cannot read a clip is swapped for it and the result says so, while a named model with no audio input drops the track. Model-specific settings go in `params` alongside an explicit `model`. Returns asset URLs, or a jobId for pixeldojo_status when the job outlasts the 30s wait.",
+      hostedDescription: "Generate an image or video from a prompt with any model in the PixelDojo catalog. With no `model`, a prompt runs on seedream-5, which makes still images only. For a video, name seedance-2-5, which works from a prompt alone or a starting image; kling-4-0 for a multi-shot scene up to 30 s with dialogue; minimax-h3-turbo for the lowest price when you have a photo or reference images (it refuses a prompt alone). If a provider refuses the content, retry once on wan-3-0-video (video) or seedream-5 (images), the least-restricted models. For a multi-shot film or a music video synced to a song use pixeldojo_film; to change, move or extend an existing clip use pixeldojo_video. These are the models the PixelDojo studios use; name another only when the user asks for it. References: `image_url`/`image_urls` for stills and `video_url` for a source clip, all public http(s) URLs. A clip or an audio track URL with no `model` runs on seedance-2-reference; a named model that cannot read a clip is swapped for it and the result says so, while a named model with no audio input drops the track. Model-specific settings go in `params` alongside an explicit `model`. Returns asset URLs, or a jobId for pixeldojo_status when the job outlasts the 30s wait.",
       inputZod: InputSchema,
       inputJsonSchema: InputJsonSchema,
       allowedModels: null,
@@ -1051,6 +1051,8 @@ var init_storyboard = __esm({
     ALLOWED2 = [
       "veo-3.1-fast",
       "veo-3.1-standard",
+      // Kling 4.0 (2026-10-24): a brief written as a shot list renders as one clip, up to 30 s.
+      "kling-4-0",
       "kling-video-v3-standard-text",
       "seedance-2-high",
       VIDEO_REFERENCE_MODEL
@@ -2442,6 +2444,27 @@ function buildVideoPayload(verb, apiId, i) {
         return { body: { prompt, video_url: i.videoUrl, video_duration: vd }, note: droppedImages };
       case "happyhorse-1.0-video-edit":
         return { body: { prompt, video_url: i.videoUrl, video_duration: vd, reference_urls: i.images } };
+      case "kling-4-0":
+        return {
+          body: {
+            prompt: /@video\s*1/i.test(prompt) ? prompt : `Edit @Video1: ${prompt}`,
+            video: i.videoUrl,
+            reference_images: i.images,
+            duration: clamp(Math.ceil(vd), 3, 30),
+            resolution: "1080p"
+          }
+        };
+      case "minimax-h3-turbo":
+        return {
+          body: {
+            prompt: `Recreate <Video 1> exactly: the same people, the same motion and timing, the same camera move and framing, with one change: ${prompt}`,
+            reference_videos: [i.videoUrl],
+            reference_video_audio: true,
+            resolution: "720p",
+            duration: clamp(Math.ceil(vd ?? 5), 5, 15)
+          },
+          note: droppedImages
+        };
       default:
         return { body: { prompt, video_url: i.videoUrl, video_duration: vd, image_urls: i.images } };
     }
@@ -2570,6 +2593,8 @@ var init_video = __esm({
         "seedance-2-5-video-edit",
         "flux-video-edit",
         "kling-video-edit",
+        "kling-4-0",
+        "minimax-h3-turbo",
         "wan-video-edit",
         "seedance-video-edit",
         "happyhorse-1.0-video-edit",
@@ -2589,9 +2614,10 @@ var init_video = __esm({
       "happyhorse-1.0-video-edit",
       "xai-video-edit",
       "kling-motion-control",
-      "wan-video-character-swap"
+      "wan-video-character-swap",
+      "kling-4-0"
     ]);
-    NO_IMAGE_ENGINES = /* @__PURE__ */ new Set(["flux-video-edit", "xai-video-edit"]);
+    NO_IMAGE_ENGINES = /* @__PURE__ */ new Set(["flux-video-edit", "xai-video-edit", "minimax-h3-turbo"]);
     InputSchema11 = z.object({
       verb: z.enum(VERBS),
       prompt: z.string().max(8e3).optional(),
@@ -2946,6 +2972,14 @@ function describeDrama(v, o = DRAMA_OPTS) {
       text: [`Planned (${o.idField}: ${id}), ${v.format === "9:16" ? "vertical 9:16" : "widescreen 16:9"} (to switch, create it again in the other format before producing):`, v.plan ?? "", "", `Nothing has been charged yet. ${cost} Show this to the user and, once they agree, call ${next2("produce")}. (In Sensei: put it in pixeldojo_plan as ONE step with skill "${music ? "music_video" : "drama"}" and params {"${o.idField}": "${id}"}, which prices it from this quote.)`].join("\n")
     };
   }
+  const gist = (note) => {
+    const first = note.split(/(?<=\.) /)[0];
+    return first.length > 160 ? `${first.slice(0, 157).trimEnd()}...` : first;
+  };
+  const rewrites = (b) => {
+    const shots2 = Array.from(new Set(Array.from((b.redo_films ?? "").matchAll(/Shot (\d+) \(/g), (m) => m[1])));
+    return shots2.length ? ` (the redo rewrites shot${shots2.length > 1 ? "s" : ""} ${shots2.join(", ")})` : "";
+  };
   if (v.status === "notes") {
     const flagged = v.beats.filter((b) => b.note && (b.needs_redo ?? true));
     const other = v.beats.filter((b) => b.note && b.needs_redo === false);
@@ -2953,17 +2987,17 @@ function describeDrama(v, o = DRAMA_OPTS) {
       kind: "text",
       text: [
         `Every ${o.unit} is filmed. Before the cut, the review flagged ${o.unit}s ${flagged.map((b) => b.beat).join(", ")}:`,
-        ...flagged.map((b) => `- ${o.unit} ${b.beat}: ${b.note}${b.redo_films ? ` The redo films: ${b.redo_films}` : ""}`),
-        other.length ? `Also noted, no redo needed: ${other.map((b) => `${o.unit} ${b.beat}: ${b.note}`).join(" ")}` : "",
+        ...flagged.map((b) => `- ${o.unit} ${b.beat}: ${gist(b.note)}${rewrites(b)}`),
+        other.length ? `Also noted, no redo needed: ${other.map((b) => `${o.unit} ${b.beat}: ${gist(b.note)}`).join(" ")}` : "",
         simplify,
-        `Show these to the user and ask which they want: redo those ${o.unit}s with these notes (${o.tool} with {"${o.idField}": "${id}", "action": "redo", "beats": [${flagged.map((b) => b.beat).join(", ")}], "notes": [the notes, same order]}, each at its own price), or cut it as it is (action "cut", free). Nothing more is filmed until they choose.`
+        `Show these to the user and ask which they want: redo those ${o.unit}s with the review's fixes (${o.tool} with {"${o.idField}": "${id}", "action": "redo", "beats": [${flagged.map((b) => b.beat).join(", ")}]}, no notes needed, each at its own price; add notes only for the user's own changes), or cut it as it is (action "cut", free). Nothing more is filmed until they choose.`
       ].filter(Boolean).join("\n")
     };
   }
   if (v.status === "done" && v.final_url) {
     const redo = v.beats.filter((b) => (b.note || b.missing_lines.length) && (b.needs_redo ?? true));
-    const notes = redo.map((b) => `${o.unit} ${b.beat}: ${b.note ?? `did not say ${b.missing_lines.map((l) => `"${l}"`).join(", ")}`}${b.redo_films ? ` The redo films: ${b.redo_films}` : ""}`);
-    const other = v.beats.filter((b) => b.note && b.needs_redo === false).map((b) => `${o.unit} ${b.beat}: ${b.note}`);
+    const notes = redo.map((b) => `${o.unit} ${b.beat}: ${b.note ? gist(b.note) : `did not say ${b.missing_lines.map((l) => `"${l}"`).join(", ")}`}`);
+    const other = v.beats.filter((b) => b.note && b.needs_redo === false).map((b) => `${o.unit} ${b.beat}: ${gist(b.note)}`);
     return {
       kind: "assets",
       text: [
@@ -2979,7 +3013,7 @@ function describeDrama(v, o = DRAMA_OPTS) {
   const doing = music ? { scripting: "listening to the song and writing the plan", shots: "filming the sections", cutting: "cutting it to the song" }[v.status] ?? STEP[v.status] ?? v.status : STEP[v.status] ?? v.status;
   const ready = v.beats.filter((b) => b.url).length;
   const total = v.beats.length || (music ? v.sections?.length ?? 0 : 5);
-  const again = v.rendering.map((n) => v.beats.find((b) => b.beat === n)).filter((b) => Boolean(b?.url)).map((b) => `${o.unit} ${b.beat} is being filmed again${b.missing_lines.length ? ` (it dropped the line "${b.missing_lines[0]}")` : b.note ? ` (the review asked: ${b.note.split(/(?<=\.) /)[0]})` : ""}`);
+  const again = v.rendering.map((n) => v.beats.find((b) => b.beat === n)).filter((b) => Boolean(b?.url)).map((b) => `${o.unit} ${b.beat} is being filmed again${b.filming_note ? ` (your note: ${gist(b.filming_note)})` : b.missing_lines.length ? ` (it dropped the line "${b.missing_lines[0]}")` : b.note ? ` (the review asked: ${gist(b.note)})` : ""}`);
   return {
     kind: "text",
     text: `Still ${doing}${v.status === "shots" ? ` (${ready} of ${total} ${o.unit}s ready${v.rendering.length ? `, filming ${o.unit} ${v.rendering.join(", ")}` : ""})` : ""}.${again.length ? ` ${again.join("; ")}.` : ""} It takes about 10 to 15 minutes in all. Call ${next2("status")} in about a minute.`,
@@ -3331,7 +3365,7 @@ var init_kata = __esm({
         },
         photos: {
           type: "array",
-          items: { type: "object", properties: { url: { type: "string" }, note: { type: "string", description: 'What it is (e.g. "the product").' } }, required: ["url"] },
+          items: { type: "object", properties: { url: { type: "string", description: "Public URL of the photo." }, note: { type: "string", description: 'What it is (e.g. "the product").' } }, required: ["url"] },
           description: "Create: up to 4 products or photos to put in the edit, on the PixelDojo CDN."
         },
         instructions: { type: "string", description: "Create (optional): what to change, e.g. the setting, the words on screen, where the product goes." },
@@ -4422,11 +4456,11 @@ var init_marketing = __esm({
           action: { type: "string", enum: ["get", "import", "update"], description: "Default get." },
           url: { type: "string", description: "import: the brand website, e.g. yourbrand.com." },
           save: { type: "boolean", description: "import: false returns the kit without saving it. Default true." },
-          name: { type: "string" },
-          tagline: { type: "string" },
+          name: { type: "string", description: "Brand name." },
+          tagline: { type: "string", description: "Short brand tagline." },
           voice: { type: "string", description: "How the brand talks, one or two sentences." },
           colors: { type: "array", items: { type: "string" }, maxItems: 5, description: "#RRGGBB, lead color first." },
-          fonts: { type: "array", items: { type: "string" }, maxItems: 2 },
+          fonts: { type: "array", items: { type: "string" }, maxItems: 2, description: "Up to two brand font names." },
           selling_points: { type: "array", items: { type: "string" }, maxItems: 5, description: "Short product claims, 2-7 words each." },
           logo_url: { type: "string", format: "uri", description: "A logo on PixelDojo (upload it first); PNG with transparency is best." }
         }
@@ -4479,8 +4513,8 @@ var init_marketing = __esm({
         description: `image: an image template id from templates (e.g. stacked-shout, pure-white, ice-float). motion: ${MOTION.join(", ")}.`
       },
       product_url: { type: "string", format: "uri", description: "image: product page; name, description and photos are read from it." },
-      product_name: { type: "string" },
-      product_description: { type: "string" },
+      product_name: { type: "string", description: "Product name when there is no product_url (with product_image_urls)." },
+      product_description: { type: "string", description: "Short product description when there is no product_url." },
       product_image_urls: { type: "array", items: { type: "string", format: "uri" }, maxItems: 5, description: "Product photos. motion uses the first, cut out automatically." },
       product_image_paths: { type: "array", items: { type: "string" }, maxItems: 5, description: "Local product photos instead of URLs; uploaded first." },
       character_image_urls: { type: "array", items: { type: "string", format: "uri" }, maxItems: 2, description: "image: your own character (e.g. from pixeldojo_library source characters) for templates with a person." },
@@ -5117,7 +5151,7 @@ function makeRedactedLogger(target = console) {
 }
 
 // src/server.ts
-var PACKAGE_VERSION = "0.30.0" ;
+var PACKAGE_VERSION = "0.31.0" ;
 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) {