@pixeldojo/mcp 0.34.1 → 0.36.0

package/README.md

@@ -46,7 +46,7 @@ Restart your agent. You'll see 32 new tools.
 
 | Skill | What it does |
 |---|---|
-| `pixeldojo_generate` | Any prompt → image or video. Defaults to seedream-5 (stills); pass any apiId, e.g. seedance-2-5 for video. The recommended models are the ones PixelDojo's studios use. |
+| `pixeldojo_generate` | Any prompt → image or video. Defaults to seedream-5-5 (stills); pass any apiId, e.g. seedance-2-5-flash for video (seedance-2-5 for 1080p). The recommended models are the ones PixelDojo's studios use. |
 | `pixeldojo_edit` | Change an existing image with a text instruction (multi-image capable). |
 | `pixeldojo_character` | Consistent characters across shots. |
 | `pixeldojo_storyboard` | One video shot from a scene brief (use `pixeldojo_film` for multi-shot). |

package/dist/cli.cjs

@@ -563,7 +563,7 @@ var init_generate = __esm({
     InputSchema = zod.z.object({
       prompt: zod.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: zod.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), "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 prompt alone, a photo or up to 9 reference images, "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-5-5" for images (the default), "seedance-2-5-flash" for video (prompt only or a starting image, 480p or 720p), "seedance-2-5" for 1080p video, "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 prompt alone, a photo or up to 9 reference images, "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-5-5, which makes still images only; name a video model to get a video.`
       ),
       preset: zod.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({
         "Multiple absolute local file paths. Each is uploaded to PixelDojo (24h temp storage) and used as a reference, same rules as image_urls."
       )
     });
-    DEFAULT_MODEL = "seedream-5";
+    DEFAULT_MODEL = "seedream-5-5";
     VIDEO_REFERENCE_MODEL = "seedance-2-reference";
     VIDEO_REFERENCE_SLOTS = {
       image: { name: "reference_images", isArray: true },
@@ -602,7 +602,7 @@ 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), "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 prompt alone, a photo or up to 9 reference images, "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-5-5" for images (the default), "seedance-2-5-flash" for video (prompt only or a starting image, 480p or 720p), "seedance-2-5" for 1080p video, "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 prompt alone, a photo or up to 9 reference images, "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-5-5, which makes still images only; name a video model to get a video.`
         },
         preset: {
           type: "string",
@@ -650,8 +650,8 @@ 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; kling-4-0 for a multi-shot scene up to 30 s with dialogue; minimax-h3-turbo for the lowest price, from a prompt alone, a photo or reference images. 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, from a prompt alone, a photo or reference images. 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-5-5, which makes still images only. For a video, name seedance-2-5-flash, which works from a prompt alone or a starting image at 480p or 720p; seedance-2-5 for 1080p; kling-4-0 for a multi-shot scene up to 30 s with dialogue; minimax-h3-turbo for the lowest price, from a prompt alone, a photo or reference images. 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-5-5, which makes still images only. For a video, name seedance-2-5-flash, which works from a prompt alone or a starting image at 480p or 720p; seedance-2-5 for 1080p; kling-4-0 for a multi-shot scene up to 30 s with dialogue; minimax-h3-turbo for the lowest price, from a prompt alone, a photo or reference images. 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,
@@ -931,6 +931,7 @@ var init_character = __esm({
     init_schema_snap();
     init_params();
     ALLOWED = [
+      "seedream-5-5",
       "seedream-5",
       "seedream-4",
       "seedream-5-flash",
@@ -938,20 +939,23 @@ var init_character = __esm({
       "wan-2.7-image-edit",
       "qwen-3-pro",
       "qwen-image-2.0-pro",
-      "p-h3-image"
+      "p-h3-image",
+      "minimax-h3-image"
     ];
-    DEFAULT_MODEL3 = "seedream-5";
+    DEFAULT_MODEL3 = "seedream-5-5";
     MAX_REFERENCES = 9;
     REQUIRES_REFERENCE = /* @__PURE__ */ new Set(["wan-2.7-image-edit"]);
     FALLBACK_REF_KEY = {
       "seedream-5": "image_urls",
+      "seedream-5-5": "image_urls",
       "seedream-4": "image_urls",
       "seedream-5-flash": "image_urls",
       "google-nano-banana": "image_input",
       "wan-2.7-image-edit": "image_urls",
       "qwen-3-pro": "image",
       "qwen-image-2.0-pro": "reference_images",
-      "p-h3-image": "reference_images"
+      "p-h3-image": "reference_images",
+      "minimax-h3-image": "reference_images"
     };
     InputSchema3 = zod.z.object({
       prompt: zod.z.string().min(1).max(8e3),
@@ -988,7 +992,7 @@ var init_character = __esm({
         },
         aspect_ratio: {
           type: "string",
-          description: 'e.g. "1:1", "16:9", "9:16". Snapped to the nearest ratio the model supports, and dropped for models with no aspect_ratio field (including the default seedream-5, which sizes by image_size: set that in params with an explicit model).'
+          description: 'e.g. "1:1", "16:9", "9:16". Snapped to the nearest ratio the model supports, and dropped for models with no aspect_ratio field (including the default seedream-5-5, which sizes by image_size: set that in params with an explicit model).'
         },
         params: PARAMS_JSON_SCHEMA
       },
@@ -996,7 +1000,7 @@ var init_character = __esm({
     };
     characterSkill = {
       name: "pixeldojo_character",
-      description: "Generate or edit consistent characters across shots. Pass reference_image_urls (up to 9) to lock identity \u2014 send every photo of the person you have, since more angles hold the face far better than one. Defaults to seedream-5.",
+      description: "Generate or edit consistent characters across shots. Pass reference_image_urls (up to 9) to lock identity \u2014 send every photo of the person you have, since more angles hold the face far better than one. Defaults to seedream-5-5.",
       inputZod: InputSchema3,
       inputJsonSchema: InputJsonSchema3,
       allowedModels: [...ALLOWED],
@@ -2990,7 +2994,7 @@ function describeDrama(v, o = DRAMA_OPTS) {
   if (v.status === "failed" && v.error?.startsWith("Prohibited content")) return { kind: "text", text: `${v.error} Tell the user this idea is prohibited content; do not rewrite it to get around the refusal.` };
   if (v.status === "failed") return { kind: "text", text: `The ${what} stopped: ${v.error ?? "unknown error"}. Each render that failed was refunded.${v.beats.some((b) => b.url) ? ` Re-film ${o.unit}s with action "redo".` : ""}` };
   if (v.status === "quote") {
-    const cost = music ? `Filming costs about ${v.estimate_credits} credits (cast portraits and ${v.sections?.length ?? "several"} sections lip-synced to the song).` : `Filming costs about ${v.estimate_credits} credits (cast portraits, scenes of up to 30 seconds with their dialogue and sound${v.score === false ? ", no music" : ", and one score over the finished edit"}); every scene is checked, and a scene that drops a line or shows a clear fault waits for the user's OK before it is filmed again, at the scene's own price (a render that fails is tried again free).`;
+    const cost = music ? `Filming costs about ${v.estimate_credits} credits (cast portraits and ${v.sections?.length ?? "several"} sections lip-synced to the song).` : `Filming costs about ${v.estimate_credits} credits (cast portraits, scenes of up to 30 seconds with ${v.dialogue === false ? "their own sound and no dialogue" : "their dialogue and sound"}${v.score === false ? ", no music" : ", and one score over the finished edit"}); every scene is checked, and a scene that drops a line or shows a clear fault waits for the user's OK before it is filmed again, at the scene's own price (a render that fails is tried again free).`;
     return {
       kind: "text",
       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")
@@ -3077,7 +3081,8 @@ var init_drama = __esm({
       length_sec: zod.z.number().int().min(24).max(180).multipleOf(6).optional(),
       scenes: zod.z.number().int().min(1).max(15).optional(),
       score: zod.z.boolean().optional(),
-      engine: zod.z.enum(["seedance-2-5", "wan-3-0-video", "kling-4-0"]).optional(),
+      dialogue: zod.z.boolean().optional(),
+      engine: zod.z.enum(["seedance-2-5", "seedance-2-5-flash", "wan-3-0-video", "kling-4-0"]).optional(),
       drama_id: zod.z.string().max(60).optional(),
       beats: zod.z.array(zod.z.number().int().min(1).max(15)).min(1).max(15).optional(),
       notes: zod.z.array(zod.z.string().max(300)).max(15).optional()
@@ -3125,8 +3130,9 @@ var init_drama = __esm({
         },
         length_sec: { type: "integer", minimum: 24, maximum: 180, multipleOf: 6, description: "Create (optional): the film's length in seconds, 24 to 180 in steps of 6. Default 30. If the user has not said how long, ask (30 seconds, 1, 2 or 3 minutes; the price grows with the length) or leave it out for 30 seconds. The writer splits the length into scenes of up to 30 s, one per place and cast." },
         scenes: { type: "integer", minimum: 1, maximum: 15, description: "Create (optional, older): the length in 12-second units (5 = 60 s). Prefer length_sec." },
-        engine: { type: "string", enum: ["seedance-2-5", "wan-3-0-video", "kling-4-0"], description: "Create (optional): the video model every scene films on, from the cast pictures. seedance-2-5 (default), wan-3-0-video (about a third of the price), kling-4-0. Pass it only when the user names one. Adult work always runs on wan-3-0-video." },
+        engine: { type: "string", enum: ["seedance-2-5", "seedance-2-5-flash", "wan-3-0-video", "kling-4-0"], description: "Create (optional): the video model every scene films on, from the cast pictures. seedance-2-5 (default), seedance-2-5-flash (the same model family, faster and about a third cheaper, 720p), wan-3-0-video (about a third of the price), kling-4-0. Pass it only when the user names one. Adult work always runs on wan-3-0-video." },
         score: { type: "boolean", description: "Create (optional): one score laid over the finished edit, on by default and included in the price. Pass false only when the user wants no music (dialogue and ambience only)." },
+        dialogue: { type: "boolean", description: "Create (optional): pass false for a film with no dialogue, where nobody speaks and the story is told in pictures, sound and the score. Use it when the user asks for no dialogue, a silent or wordless film, or visuals with music only; otherwise leave it out (characters speak)." },
         drama_id: { type: "string", description: "From create. Required for produce, status and redo." },
         beats: { type: "array", items: { type: "integer", minimum: 1, maximum: 15 }, description: "Redo: the beat numbers to film again." },
         notes: { type: "array", items: { type: "string" }, description: "Redo (optional): what to fix in each of those beats, in the same order." }
@@ -3142,10 +3148,10 @@ var init_drama = __esm({
     DRAMA_OPTS = { tool: "pixeldojo_drama", idField: "drama_id", unit: "beat" };
     dramaSkill = {
       name: "pixeldojo_drama",
-      description: "Make a micro-drama episode (30 seconds by default, up to 3 minutes with length_sec; hook, twist, cliffhanger; optionally a classic genre such as billionaire, werewolf or mafia), short film (a complete story) or movie trailer from an idea, vertical 9:16 or 16:9, in any of Film Studio's looks (live-action, anime, 3D animation, noir, horror, sci-fi, western, romance, comedy and more). Scenes of up to 30 seconds (one per place and cast), each filmed in one pass with several cuts, on Seedance 2.5, WAN 3 or Kling 4, the cast's faces kept from portraits and every line spoken and lip-synced, cut together with their own sound and one score over the finished edit, with free motion graphics (title card, name tags, end screen; each can be turned off), and saved to the Library. Not for music videos (pixeldojo_music_video) create is free and returns the plan and the exact price (about 190 credits for a regular 30 second film, about 380 for a minute; less on the wan-3-0-video engine and for mature work); produce films it; then call status until it is finished (10-15 minutes).",
+      description: "Make a micro-drama episode (30 seconds by default, up to 3 minutes with length_sec; hook, twist, cliffhanger; optionally a classic genre such as billionaire, werewolf or mafia), short film (a complete story) or movie trailer from an idea, vertical 9:16 or 16:9, in any of Film Studio's looks (live-action, anime, 3D animation, noir, horror, sci-fi, western, romance, comedy and more). Scenes of up to 30 seconds (one per place and cast), each filmed in one pass with several cuts, on Seedance 2.5, WAN 3 or Kling 4, the cast's faces kept from portraits and every line spoken and lip-synced (or no dialogue at all with dialogue: false), cut together with their own sound and one score over the finished edit, with free motion graphics (title card, name tags, end screen; each can be turned off), and saved to the Library. Not for music videos (pixeldojo_music_video) create is free and returns the plan and the exact price (about 190 credits for a regular 30 second film, about 380 for a minute; less on the wan-3-0-video engine and for mature work); produce films it; then call status until it is finished (10-15 minutes).",
       inputZod: InputSchema13,
       inputJsonSchema: InputJsonSchema13,
-      allowedModels: ["seedance-2-5", "seedream-4"],
+      allowedModels: ["seedance-2-5", "seedance-2-5-flash", "seedream-4"],
       defaultModel: "seedance-2-5",
       async run(input, ctx) {
         if (input.action === "create") {
@@ -3161,6 +3167,7 @@ var init_drama = __esm({
             ...input.scenes ? { scenes: input.scenes } : {},
             ...input.engine ? { engine: input.engine } : {},
             ...input.score === false ? { score: false } : {},
+            ...input.dialogue === false ? { dialogue: false } : {},
             ...Object.fromEntries(
               ["title_card", "name_tags", "end_screen", "end_text", "brand_color", "logo_url"].filter((k) => input[k] !== void 0).map((k) => [k, input[k]])
             ),
@@ -3369,6 +3376,7 @@ var init_kata = __esm({
       photos: zod.z.array(zod.z.object({ url: zod.z.string().url(), note: zod.z.string().max(80).optional() })).max(4).optional(),
       instructions: zod.z.string().max(1500).optional(),
       look: zod.z.string().max(80).optional(),
+      engine: zod.z.enum(["seedance-2-5", "seedance-2-5-flash"]).optional(),
       kata_id: zod.z.string().max(60).optional()
     });
     InputJsonSchema15 = {
@@ -3401,6 +3409,7 @@ var init_kata = __esm({
         },
         instructions: { type: "string", description: "Create (optional): what to change, e.g. the setting, the words on screen, where the product goes." },
         look: { type: "string", description: 'Create (optional): style transfer, one look for every new shot, e.g. "hand-drawn anime".' },
+        engine: { type: "string", enum: ["seedance-2-5", "seedance-2-5-flash"], description: "Create (optional): the video model the new clips film on. seedance-2-5 (default) or seedance-2-5-flash (faster, about a third cheaper). Pass it only when the user asks." },
         kata_id: { type: "string", description: "From create. Required for produce and status." }
       }
     };
@@ -3422,7 +3431,7 @@ var init_kata = __esm({
       description: `Kata: recreate a reference edit (a viral short, an ad, a montage, a trailer, 4 to 60 s) with the user's own characters, products and photos. It keeps the same pacing and cut points to the frame, the transitions, the camera moves, the caption style and the music feel (an original score on the same tempo, never the reference's audio); dance moves come across by motion transfer, and a look can restyle every shot. Start from a reference_url or a Kata library preset (action "presets"). create is free and returns the plan and the price; produce makes it; then call status until it is finished. Saved to the Library.`,
       inputZod: InputSchema15,
       inputJsonSchema: InputJsonSchema15,
-      allowedModels: ["seedance-2-5", "google-nano-banana", "kling-motion-control"],
+      allowedModels: ["seedance-2-5", "seedance-2-5-flash", "google-nano-banana", "kling-motion-control"],
       defaultModel: "seedance-2-5",
       async run(input, ctx) {
         if (input.action === "presets") {
@@ -3441,7 +3450,8 @@ var init_kata = __esm({
             ...input.cast?.length ? { cast: input.cast } : {},
             ...input.photos?.length ? { photos: input.photos } : {},
             ...input.instructions ? { instructions: input.instructions } : {},
-            ...input.look ? { look: input.look } : {}
+            ...input.look ? { look: input.look } : {},
+            ...input.engine ? { engine: input.engine } : {}
           });
           if (!created.data.id) return { kind: "text", text: `Could not start the Kata: ${errorText3(created.data, created.status)}` };
           const planned = await waitForPlan3(ctx, created.data.id);
@@ -5199,7 +5209,7 @@ function makeRedactedLogger(target = console) {
 }
 
 // src/server.ts
-var PACKAGE_VERSION = "0.34.1" ;
+var PACKAGE_VERSION = "0.36.0" ;
 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) {