@pixeldojo/mcp 0.16.2 → 0.17.0

package/dist/index.js

@@ -1453,6 +1453,39 @@ async function filmApi(ctx, method, path, body) {
   const data = await res.json().catch(() => ({}));
   return { status: res.status, data };
 }
+async function applyCast(ctx, filmId, cast) {
+  let charged = false;
+  const lines = await Promise.all(cast.map(async (c) => {
+    const r = await filmApi(
+      ctx,
+      "POST",
+      `/api/films/${filmId}/cast`,
+      {
+        role: c.name,
+        ...c.character_id ? { characterId: c.character_id } : {},
+        ...c.image_urls ? { imageUrls: c.image_urls } : {},
+        ...c.prompt ? { prompt: c.prompt } : {}
+      }
+    );
+    if (r.status === 200 && r.data.creditCost) charged = true;
+    if (r.status !== 200) return `Could not cast ${c.name}: ${r.data.error ?? r.status}`;
+    if (r.data.alreadyCast) return `${c.name} was already cast; nothing changed.`;
+    return `Cast ${c.name}${r.data.creditCost ? ` (portrait, ${r.data.creditCost} credits)` : ""}${r.data.boundShots ? `, in ${r.data.boundShots} shot(s)` : ""}.`;
+  }));
+  return { lines, charged };
+}
+async function uncastRolesOf(ctx, filmId) {
+  const r = await filmApi(ctx, "GET", `/api/films/${filmId}/cast`);
+  return r.data.uncastRoles ?? [];
+}
+function castNeededText(filmId, roles) {
+  const example = roles.map((r) => ({ name: r.name, prompt: r.suggestedPrompt || `describe ${r.name}` }));
+  return [
+    `Cast these characters before production (every named character needs a cast member so faces, ages and species hold in every shot):`,
+    ...roles.map((r) => `  ${r.name}${r.suggestedPrompt ? `: ${r.suggestedPrompt}` : ""}`),
+    `Call pixeldojo_film with {"film_id": "${filmId}", "action": "cast", "cast": ${JSON.stringify(example)}}, giving each one a prompt (a portrait is generated, charged), image_urls of your own photos, or a character_id from Character Studio.`
+  ].join("\n");
+}
 function progressLine(f) {
   const all = shots(f);
   const frames = all.filter((sh) => ["frame-ready", "approved", "clip-generating", "clip-ready"].includes(sh.status) || sh.frameUrl).length;
@@ -1526,32 +1559,44 @@ async function plan(input, ctx) {
   });
   if (!created.data.film) return { kind: "text", text: `Could not create the film: ${created.data.error ?? created.status}` };
   const id = created.data.film.id;
+  const castStarted = Date.now();
+  const cast = input.cast?.length ? await applyCast(ctx, id, input.cast) : { lines: [], charged: false };
+  const pollBudget = Math.max(0, PLANNER_MAX_POLLS - Math.ceil((Date.now() - castStarted) / PLANNER_POLL_MS));
+  const reply = (text) => ({
+    kind: "text",
+    text: [
+      ...cast.lines,
+      cast.charged ? text.replace(/Nothing has been charged( yet)?\./g, "Only the cast portraits above have been charged.") : text
+    ].join("\n")
+  });
   let planned;
   if (music) {
     const brief = [input.idea?.trim() || input.script.trim(), input.lyrics?.trim() ? `Lyrics:
 ${input.lyrics.trim()}` : ""].filter(Boolean).join("\n\n");
-    planned = await runPlanner(ctx, id, `/api/films/${id}/plan`, { script: brief });
+    planned = await runPlanner(ctx, id, `/api/films/${id}/plan`, { script: brief }, pollBudget);
   } else {
     let script = input.script?.trim() ?? "";
     if (!script) {
-      const wrote = await runPlanner(ctx, id, `/api/films/${id}/script`, { idea: input.idea.trim() });
-      if ("pending" in wrote) return { kind: "text", text: stillPlanningText(id, "script") };
-      if ("error" in wrote) return { kind: "text", text: `Script writing failed: ${wrote.error}` };
+      const wrote = await runPlanner(ctx, id, `/api/films/${id}/script`, { idea: input.idea.trim() }, pollBudget);
+      if ("pending" in wrote) return reply(stillPlanningText(id, "script"));
+      if ("error" in wrote) return reply(`Script writing failed: ${wrote.error}`);
       script = wrote.film.script ?? "";
-      if (!script) return { kind: "text", text: "Script writing failed: the studio returned an empty script." };
+      if (!script) return reply("Script writing failed: the studio returned an empty script.");
     }
     planned = await runPlanner(ctx, id, `/api/films/${id}/plan`, { script });
   }
-  if ("pending" in planned) return { kind: "text", text: stillPlanningText(id) };
-  if ("error" in planned) return { kind: "text", text: `Shot planning failed: ${planned.error}` };
+  if ("pending" in planned) return reply(stillPlanningText(id));
+  if ("error" in planned) return reply(`Shot planning failed: ${planned.error}`);
   const film = planned.film;
   if (needsSlice(film)) {
     const err = await sliceSong(ctx, film);
-    if (err) return { kind: "text", text: `${plannedText(film, ctx)}
+    if (err) return reply(`${plannedText(film, ctx)}
 
-${err} Call action "status" to retry the cut.` };
+${err} Call action "status" to retry the cut.`);
   }
-  return { kind: "text", text: plannedText(film, ctx) };
+  const uncast = await uncastRolesOf(ctx, id);
+  return reply([plannedText(film, ctx), uncast.length ? `
+${castNeededText(id, uncast)}` : ""].filter(Boolean).join("\n"));
 }
 async function resumePlanning(film, ctx) {
   if (film.scenes.length > 0 && film.llmJob?.status !== "running") return null;
@@ -1578,6 +1623,8 @@ async function produce(filmId, ctx) {
   if (!data.film) return { kind: "text", text: `Film not found: ${data.error ?? filmId}` };
   const planning = await resumePlanning(data.film, ctx);
   if (planning) return planning;
+  const uncast = await uncastRolesOf(ctx, filmId);
+  if (uncast.length) return { kind: "text", text: `Nothing was charged. ${castNeededText(filmId, uncast)}` };
   const pending = shots(data.film).filter((sh) => sh.status === "idle" || sh.status === "failed" && !sh.frameUrl);
   let fired = 0;
   for (const sh of pending.slice(0, 6)) {
@@ -1651,19 +1698,27 @@ It is saved in the Library folder "Film: ${film.title}".`
   if (shots(film).length > 0 && shots(film).every((sh) => sh.status === "idle")) {
     return { kind: "text", text: plannedText(film, ctx) };
   }
-  const idleFrames = all.filter((sh) => sh.status === "idle");
-  for (const sh of idleFrames.slice(0, 6)) {
-    await filmApi(ctx, "POST", `/api/films/${filmId}/shots/${sh.id}/generate`, { kind: "frame" });
-  }
-  const framesDone = all.every((sh) => ["frame-ready", "clip-generating", "clip-ready"].includes(sh.status));
-  if (framesDone && all.some((sh) => sh.status === "frame-ready" && needsSave(sh.frameUrl))) {
+  let started = 0;
+  const resent = /* @__PURE__ */ new Set();
+  const dispatch = async (sh, kind) => {
+    const r = await filmApi(ctx, "POST", `/api/films/${filmId}/shots/${sh.id}/generate`, { kind });
+    if (r.status !== 200) return;
+    started += 1;
+    if (sh.status === "failed") resent.add(sh.id);
+  };
+  const pendingFrames = all.filter((sh) => sh.status === "idle" || sh.status === "failed" && !sh.frameUrl && !isRefused(sh));
+  if (pendingFrames.length > 0) {
+    const uncast = await uncastRolesOf(ctx, filmId);
+    if (uncast.length) return { kind: "text", text: castNeededText(filmId, uncast) };
+  }
+  for (const sh of pendingFrames.slice(0, 6)) await dispatch(sh, "frame");
+  const framesDone = all.every(frameDone);
+  if (all.some((sh) => frameDone(sh) && needsSave(sh.frameUrl))) {
     await filmApi(ctx, "POST", `/api/films/${filmId}/save`, { kind: "frame" });
   }
   const inFlight = count(film, (sh) => sh.status === "clip-generating");
-  const clipQueue = all.filter((sh) => (sh.status === "frame-ready" || sh.status === "failed" && Boolean(sh.frameUrl)) && framesDone);
-  for (const sh of clipQueue.slice(0, Math.max(0, 4 - inFlight))) {
-    await filmApi(ctx, "POST", `/api/films/${filmId}/shots/${sh.id}/generate`, { kind: "clip" });
-  }
+  const clipQueue = framesDone ? all.filter((sh) => sh.status === "frame-ready" || sh.status === "failed" && frameDone(sh) && !isRefused(sh)) : [];
+  for (const sh of clipQueue.slice(0, Math.max(0, 4 - inFlight))) await dispatch(sh, "clip");
   const clipsDone = all.length > 0 && all.every((sh) => sh.status === "clip-ready");
   if (clipsDone) {
     if (all.some((sh) => needsSave(sh.videoUrl))) {
@@ -1686,30 +1741,53 @@ It is saved in the Library folder "Film: ${film.title}".`
         return { kind: "text", text: `Score still composing. Poll again in ~20s with {"film_id": "${filmId}", "action": "status", "music_job_id": "${musicJobId}"}.` };
       }
       const body = soundtrack === "scene-synced" ? buildScorePlanBody(film) : buildFlatMusicBody(film, all);
-      const dispatch = await ctx.client.run("text-to-music", body);
-      return { kind: "text", text: `All ${all.length} clips ready \u2014 the ${soundtrack} score is composing (charged like any music generation). Poll again in ~20s with {"film_id": "${filmId}", "action": "status", "music_job_id": "${dispatch.jobId}"}.` };
+      const dispatch2 = await ctx.client.run("text-to-music", body);
+      return { kind: "text", text: `All ${all.length} clips ready \u2014 the ${soundtrack} score is composing (charged like any music generation). Poll again in ~20s with {"film_id": "${filmId}", "action": "status", "music_job_id": "${dispatch2.jobId}"}.` };
     }
     await filmApi(ctx, "POST", `/api/films/${filmId}/export`, {});
     return { kind: "text", text: `All clips ready \u2014 export started (${progressLine(film)}). Poll again in ~60s with action "status" for the finished film URL.` };
   }
   const failed = all.filter((sh) => sh.status === "failed");
+  const refused = failed.filter(isRefused);
+  const retrying = failed.filter((sh) => !isRefused(sh));
+  const sentAgain = retrying.filter((sh) => resent.has(sh.id)).length;
+  const paused = refused.length > 0 && started === 0 && !all.some((sh) => sh.status === "frame-generating" || sh.status === "clip-generating");
+  const labels = new Map(film.scenes.flatMap((s, si) => s.shots.map((sh, i) => [sh.id, `${si + 1}.${i + 1}`])));
+  const reason = (sh) => {
+    const text = (sh.errorMessage ?? "no reason given").replace(/\s*Retry (?:the shot|it)\.\s*$/i, "");
+    return text.length > 300 ? `${text.slice(0, 297)}...` : text;
+  };
+  const one = refused.length === 1;
   return {
     kind: "text",
     text: [
-      `In progress: ${progressLine(film)}.`,
-      failed.length > 0 ? `${failed.length} shot(s) failed and were re-queued (${failed[0]?.errorMessage?.slice(0, 100) ?? ""}).` : "",
-      `Poll again in ~30s with {"film_id": "${filmId}", "action": "status"}.`
+      `${paused ? "Paused" : "In progress"}: ${progressLine(film)}.`,
+      refused.length > 0 ? `${one ? "Shot" : "Shots"} ${refused.map((sh) => labels.get(sh.id)).join(", ")} need${one ? "s" : ""} a new frame or prompt. The provider refused ${one ? "it" : "them"} (${reason(refused[0])}) and usually refuses the same input again, so status does not resend ${one ? "it" : "them"}: every retry is a charged render. Change ${one ? "it" : "them"} and render again at ${ctx.baseUrl}/film-studio/${filmId}.` : "",
+      retrying.length > 0 ? `${retrying.length} shot(s) failed (${reason(retrying[0])}): ${sentAgain === retrying.length ? "sent again" : sentAgain > 0 ? `${sentAgain} sent again, the rest on a later poll` : "sent again on a later poll"}.` : "",
+      paused ? `Nothing else is rendering. Once the shot${one ? "" : "s"} render${one ? "s" : ""} again, call pixeldojo_film with {"film_id": "${filmId}", "action": "status"} to carry on.` : `Poll again in ~30s with {"film_id": "${filmId}", "action": "status"}.`
     ].filter(Boolean).join("\n")
   };
 }
 async function run8(input, ctx) {
-  const action = input.action === "plan" && input.film_id ? "status" : input.action;
+  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 need the film_id returned by the plan step." };
+  if (!input.film_id) return { kind: "text", text: "produce/status/cast need the film_id returned by the plan step." };
+  if (action === "cast") {
+    if (!input.cast?.length) return { kind: "text", text: "Pass cast: [{ name, prompt | image_urls | character_id }] to cast characters." };
+    const { lines } = await applyCast(ctx, input.film_id, input.cast);
+    const uncast = await uncastRolesOf(ctx, input.film_id);
+    return {
+      kind: "text",
+      text: [
+        ...lines,
+        uncast.length ? castNeededText(input.film_id, uncast) : `Everyone is cast. Call pixeldojo_film with {"film_id": "${input.film_id}", "action": "produce"} to start production.`
+      ].join("\n")
+    };
+  }
   if (action === "produce") return produce(input.film_id, 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, count, PLANNER_POLL_MS, PLANNER_MAX_POLLS, isMusicVideo, needsSlice, stillPlanningText, LOOK_MUSIC, ARC, filmSkill;
+var STORY_LOOKS, MUSIC_VIDEO_LOOKS, LOOKS, MUSIC_CLIP_ENGINE_ID, InputSchema8, InputJsonSchema8, shots, 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();
@@ -1727,8 +1805,16 @@ var init_film = __esm({
       lyrics: z.string().max(1e4).optional().describe("Music video mode: the song lyrics, so shots land on the right lines. Optional."),
       aspect: z.enum(["16:9", "9:16"]).default("16:9"),
       length: z.enum(["30s", "60s", "2-3min", "5min"]).default("60s").describe("Target length; drives the shot count (~6/12/24/60 shots)."),
-      film_id: z.string().max(40).optional().describe("Existing film id (from a previous call) for produce/status."),
-      action: z.enum(["plan", "produce", "status"]).default("plan").describe("'plan' creates + storyboards (free). 'produce' starts charged generation. 'status' advances production and reports progress; poll it every ~30s until it returns the finished film."),
+      film_id: z.string().max(40).optional().describe("Existing film id (from a previous call) for produce/status/cast."),
+      action: z.enum(["plan", "produce", "status", "cast"]).default("plan").describe("'plan' creates + storyboards (free). 'cast' casts characters on an existing film. 'produce' starts charged generation. 'status' advances production and reports progress; poll it every ~30s until it returns the finished film."),
+      cast: z.array(
+        z.object({
+          name: z.string().trim().min(1).max(60),
+          image_urls: z.array(z.string().url().startsWith("https://").max(2048)).min(1).max(8).optional(),
+          prompt: z.string().trim().min(1).max(600).optional(),
+          character_id: z.string().max(40).optional()
+        })
+      ).max(8).optional().describe("Cast members, one per named character. Each gives image_urls (your own photos), a prompt (a portrait is generated, charged), or a character_id (a Character Studio character); with none of these the Character Studio character of that name is used."),
       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.")
     });
@@ -1756,8 +1842,26 @@ var init_film = __esm({
         film_id: { type: "string", description: 'The film_id returned by the plan step. Required for produce and status; with it, action "plan" is treated as "status".' },
         action: {
           type: "string",
-          enum: ["plan", "produce", "status"],
-          description: "plan (default, free): writes the script and plans shots, returns film_id and a cost estimate. produce (charged): starts rendering frames. status: advances production one step and reports progress; call it every ~30s until it returns the finished film URL."
+          enum: ["plan", "produce", "status", "cast"],
+          description: "plan (default, free): writes the script and plans shots, returns film_id, a cost estimate, and any characters still to cast. cast: casts characters on an existing film (pass film_id and cast). produce (charged): starts rendering frames; refused until every named character is cast. status: advances production one step and reports progress; call it every ~30s until it returns the finished film URL."
+        },
+        cast: {
+          type: "array",
+          maxItems: 8,
+          description: 'Cast members, one per named character. Every named character must be cast before frames render, so faces, ages and species hold in every shot. Pass them on the plan call (the script then stars them by name) or later with action "cast" using the names the plan reports. Each entry: name, plus ONE of image_urls (your own photos of an adult, made into a character), prompt (a portrait is generated from it, charged like any image), or character_id (a Character Studio character). An entry with only a name uses your Character Studio character of that name.',
+          items: {
+            type: "object",
+            properties: {
+              name: { type: "string", description: 'The character name as the film uses it (e.g. "Harlan").' },
+              image_urls: { type: "array", items: { type: "string" }, description: "Your own https photos of the character (1-8). PixelDojo uploads work; they are copied to permanent storage." },
+              prompt: {
+                type: "string",
+                description: 'A description to generate the portrait from: species if not human, gender, adult age, look and outfit (e.g. "man in his late 50s, grey mustache, sunburnt, red windbreaker").'
+              },
+              character_id: { type: "string", description: "A Character Studio character id." }
+            },
+            required: ["name"]
+          }
         },
         soundtrack: {
           type: "string",
@@ -1772,6 +1876,10 @@ var init_film = __esm({
     };
     shots = (f) => f.scenes.flatMap((s) => s.shots);
     count = (f, pred) => shots(f).filter(pred).length;
+    frameDone = (sh) => Boolean(sh.frameUrl) && sh.status !== "idle" && sh.status !== "frame-generating";
+    RETRY_CLEARS = /automatic audio check flagged|output audio\b[\s\S]*?sensitive information|temporarily unavailable/i;
+    REFUSED = /copyright|infringement|content filter|safety filter|moderation|blocked|flagged|sensitive|real (?:person|human)|nsfw|content policy|policy violation|inappropriate/i;
+    isRefused = (sh) => sh.status === "failed" && Boolean(sh.errorMessage) && !RETRY_CLEARS.test(sh.errorMessage) && REFUSED.test(sh.errorMessage);
     PLANNER_POLL_MS = 3e3;
     PLANNER_MAX_POLLS = 8;
     isMusicVideo = (f) => f.musicVideoMode === true;
@@ -1797,8 +1905,8 @@ var init_film = __esm({
     };
     filmSkill = {
       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. Phased: plan is free; produce starts charged generation; status advances production (poll every ~30s).",
-      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. Phased: plan is free; produce starts charged generation; status advances production (poll every ~30s).",
+      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. Phased: plan is free; cast; produce starts charged generation; status advances production (poll every ~30s).",
+      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. Phased: plan is free; cast; produce starts charged generation; status advances production (poll every ~30s).",
       inputZod: InputSchema8,
       inputJsonSchema: InputJsonSchema8,
       allowedModels: [],
@@ -2232,7 +2340,7 @@ async function produce2(id, ctx) {
     if (saved.status !== 200 || !url) {
       return { kind: "text", text: `The export is ready but could not be saved: ${saved.data.error ?? saved.status}. The draft (kept 24h): ${exp.url}. ${next(id, "produce", "~30s")}` };
     }
-    return { kind: "text", text: `"${ep.title}" is finished and saved to My Media: ${url}` };
+    return { kind: "text", text: `"${ep.title}" is finished and saved to your Library: ${url}` };
   }
   const started = await episodeApi(ctx, "POST", `/${id}/export`, {});
   if (started.status !== 202) {
@@ -2303,7 +2411,7 @@ var init_episode = __esm({
     stillPlanning = (id) => `The script is still being written (episode_id: ${id}). Nothing has been charged. ${next(id, "status", "~30s")}`;
     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 My Media. 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.",
+      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,
       allowedModels: [],
@@ -2853,7 +2961,7 @@ var init_library = __esm({
     };
     librarySkill = {
       name: "pixeldojo_library",
-      description: "Search your media library: saved My Media items (permanent) and recent API/MCP generations (deleted 1 hour after they finish), newest first. Returns asset URLs to reuse as reference inputs.",
+      description: "Search your Library: saved media (permanent) and recent API/MCP generations (deleted 1 hour after they finish), newest first. Returns asset URLs to reuse as reference inputs.",
       inputZod: InputSchema16,
       inputJsonSchema: InputJsonSchema16,
       allowedModels: [],
@@ -3275,7 +3383,7 @@ function makeRedactedLogger(target = console) {
 }
 
 // src/server.ts
-var PACKAGE_VERSION = "0.16.2" ;
+var PACKAGE_VERSION = "0.17.0" ;
 var SERVER_INSTRUCTIONS = "PixelDojo generates images, video, audio and short films across 100+ models, billed in credits from the connected PixelDojo 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, so download anything the user wants to keep. Finished films are the exception and stay in the account library.";
 function buildServer(options) {
   if (!options.apiKey) {