@shop-os/modelkit 0.2.21 → 0.2.23

package/CHANGELOG.md

@@ -9,6 +9,30 @@ header in the release PR.
 
 ## [Unreleased]
 
+## [0.2.23] - Unreleased
+
+### Added
+- `billedAudio(row, requested?)` on `@shop-os/modelkit/catalog`, next to `outputAudio`: whether a video request is billed as having sound. `always` rows are true, `never` rows false; an `optional` row takes `requested`, then the row's `defaults.generateAudio`, then true. A row with no valid `constraints.audio` is true. Every fallback is true because pro-backend's base video price is the sound-on price and a silent rate applies only to a line item that says `with_audio: false`, so a guess can overcharge but never undercharge. Callers (cowork Studio and chat, vibe-api's pricing classifier) pass the same `requested` value they dispatch with, so the billed and the generated sound cannot drift apart.
+
+## [0.2.22] - Unreleased
+
+### Added
+- `constraints.prompt.maxChars` on every bundled row with a documented prompt limit, so Studio accepts prompts longer than its 5,000-character default where the model takes them. Each row keeps its reference syntax (`numbered`, or `caption` on the Gemini rows):
+  - `seedance-2.0`, `seedance-2.0-fast`, `seedance-2.5`: 30000. BytePlus caps the prompt at 10,000 tokens; 30,000 characters assumes English prompts at about 3 characters per token, with margin. A CJK prompt can reach the token limit sooner.
+  - `gpt-image-2`, `gpt-image-2.5-flare`, `gpt-image-2.5-sunburst`: 32000 (OpenAI images API: 32,000 characters for GPT image models).
+  - `flux-2-pro`: 32768 (Prodia schema `maxLength`).
+  - `minimax-h3`, `minimax-h3-max`, `minimax-h3-reference`, `minimax-h3-max-reference`: 50000 (fal schema `maxLength` 50,000).
+  - `gemini-3-pro-image-preview`, `gemini-3.1-flash-image-preview`, `gemini-3.1-flash-lite-image`, `gemini-omni-flash`, `gemini-omni-1.1-flash`: 50000. Their token budgets are far above this; 50,000 is cowork's global prompt ceiling.
+  - Unchanged: `kling-2.5`, `kling-3.0`, `kling-v3-omni-video` at 2500 and `veo-3.1` at 3000. No limit on `grok-imagine`, `muse-image-1.0`, `seedream-5-0-pro` or any other row without a documented one, so Studio keeps its 5,000 default there.
+
+### Live rows to edit after publish
+Apps read the pro-backend `model_catalog` rows, not this file. Edit an environment's rows only after the companion cowork + spacelab-mcp change (Studio's global prompt ceiling raised from 5,000 to 50,000) is deployed there. Before that, cowork's projection would report `maxPromptChars: 30000`, while its request schema and spacelab-mcp still refuse anything over 5,000. After pro-backend is on 0.2.22, in each environment:
+- `seedance-2.0`, `seedance-2.0-fast`, `seedance-2.5`: `constraints.prompt` → `{ "referenceSyntax": "numbered", "maxChars": 30000 }` (assumes English prompts).
+- `gpt-image-2`, `gpt-image-2.5-flare`, `gpt-image-2.5-sunburst`: `constraints.prompt` → `{ "referenceSyntax": "numbered", "maxChars": 32000 }`.
+- `flux-2-pro`: `constraints.prompt` → `{ "referenceSyntax": "numbered", "maxChars": 32768 }`.
+- `minimax-h3`, `minimax-h3-max`, `minimax-h3-reference`, `minimax-h3-max-reference`: `constraints.prompt` → `{ "referenceSyntax": "numbered", "maxChars": 50000 }`.
+- `gemini-3-pro-image-preview`, `gemini-3.1-flash-image-preview`, `gemini-3.1-flash-lite-image`, `gemini-omni-flash`, `gemini-omni-1.1-flash`: `constraints.prompt` → `{ "referenceSyntax": "caption", "maxChars": 50000 }`.
+
 ## [0.2.21] - Unreleased
 
 ### Added

package/dist/catalog.cjs

@@ -36,6 +36,7 @@ __export(catalog_entry_exports, {
   SurfaceSchema: () => SurfaceSchema,
   TEXT_MODELS: () => TEXT_MODELS,
   VIDEO_MODELS: () => VIDEO_MODELS,
+  billedAudio: () => billedAudio,
   compareCatalogOrder: () => compareCatalogOrder,
   isListed: () => isListed,
   isListedOn: () => isListedOn,
@@ -217,13 +218,16 @@ var GEMINI_RATIOS = [
   "5:4",
   "adaptive"
 ];
-function geminiImage(id, label, aliases, sizes, display, exposure, references) {
+var GEMINI_PROMPT_MAX_CHARS = 5e4;
+var GPT_IMAGE_PROMPT = { referenceSyntax: "numbered", maxChars: 32e3 };
+function geminiImage(id, label, aliases, sizes, display, exposure, references, promptMaxChars) {
+  const prompt = { referenceSyntax: "caption" };
+  if (promptMaxChars !== void 0) prompt.maxChars = promptMaxChars;
   const constraints = {
     aspectRatios: [...GEMINI_RATIOS],
     sizes,
     supports: ["safetySettings", "temperature"],
-    // google-image sends MediaRef.label as a text part beside each reference.
-    prompt: { referenceSyntax: "caption" }
+    prompt
   };
   if (references !== void 0) constraints.references = references;
   const model = {
@@ -255,7 +259,8 @@ var IMAGE_MODELS = [
       logo: "https://d31vvfpk2ppdyq.cloudfront.net/assets/image/192485d2-a76a-4657-9c06-930b2c70badd_compressed.png"
     },
     ["cowork", "space_builder"],
-    { images: 14 }
+    { images: 14 },
+    GEMINI_PROMPT_MAX_CHARS
   ),
   geminiImage(
     "gemini-3.1-flash-image-preview",
@@ -269,7 +274,8 @@ var IMAGE_MODELS = [
       logo: "https://d31vvfpk2ppdyq.cloudfront.net/assets/image/192485d2-a76a-4657-9c06-930b2c70badd_compressed.png"
     },
     ["cowork", "space_builder"],
-    { images: 14 }
+    { images: 14 },
+    GEMINI_PROMPT_MAX_CHARS
   ),
   geminiImage(
     "gemini-3.1-flash-lite-image",
@@ -282,7 +288,8 @@ var IMAGE_MODELS = [
       logo: "https://d31vvfpk2ppdyq.cloudfront.net/assets/image/192485d2-a76a-4657-9c06-930b2c70badd_compressed.png"
     },
     ["cowork", "space_builder"],
-    { images: 14 }
+    { images: 14 },
+    GEMINI_PROMPT_MAX_CHARS
   ),
   // trigger-workflows validates 1K only; pricing seeds 2K/4K rows so allow all three.
   geminiImage(
@@ -320,7 +327,8 @@ var IMAGE_MODELS = [
       sizes: ["1K", "2K", "4K"],
       qualityOptions: ["auto", "low", "medium", "high"],
       references: { images: 16 },
-      supports: ["quality", "count", "outputFormat"]
+      supports: ["quality", "count", "outputFormat"],
+      prompt: GPT_IMAGE_PROMPT
     },
     defaults: { size: "1K", aspectRatio: "adaptive", quality: "auto" },
     usage: { ...IMAGE_USAGE, slug: "gpt" },
@@ -355,7 +363,8 @@ var IMAGE_MODELS = [
       sizes: ["1K", "2K", "4K"],
       qualityOptions: ["auto", "low", "medium", "high", "xhigh", "max"],
       references: { images: 16 },
-      supports: ["quality", "count", "outputFormat"]
+      supports: ["quality", "count", "outputFormat"],
+      prompt: GPT_IMAGE_PROMPT
     },
     defaults: { size: "1K", aspectRatio: "adaptive", quality: "auto" },
     usage: { ...IMAGE_USAGE },
@@ -389,7 +398,8 @@ var IMAGE_MODELS = [
       sizes: ["1K", "2K", "4K"],
       qualityOptions: ["auto", "low", "medium", "high", "xhigh", "max"],
       references: { images: 16 },
-      supports: ["quality", "count", "outputFormat"]
+      supports: ["quality", "count", "outputFormat"],
+      prompt: GPT_IMAGE_PROMPT
     },
     defaults: { size: "1K", aspectRatio: "adaptive", quality: "auto" },
     usage: { ...IMAGE_USAGE },
@@ -512,7 +522,9 @@ var IMAGE_MODELS = [
       sizes: ["1K"],
       references: { images: 8 },
       // flux-2-pro's job type always returns image/png; only the seed is forwarded.
-      supports: ["seed"]
+      supports: ["seed"],
+      // Prodia's flux-2-pro schema gives the prompt a maxLength of 32,768.
+      prompt: { referenceSyntax: "numbered", maxChars: 32768 }
     },
     defaults: { size: "1K", aspectRatio: "1:1" },
     usage: { ...IMAGE_USAGE },
@@ -1314,6 +1326,9 @@ var H3_REFERENCE_RATIOS = ["adaptive", ...H3_RATIOS];
 var NEGATIVE_PROMPT = "blur, distort, and low quality";
 var FRAMES_EXCLUDE_REFERENCES = { framesExcludeReferences: true };
 var KLING_PROMPT = { referenceSyntax: "numbered", maxChars: 2500 };
+var SEEDANCE_PROMPT = { referenceSyntax: "numbered", maxChars: 3e4 };
+var H3_PROMPT = { referenceSyntax: "numbered", maxChars: 5e4 };
+var OMNI_PROMPT = { referenceSyntax: "caption", maxChars: 5e4 };
 var BYTEDANCE_LOGO = "https://d31vvfpk2ppdyq.cloudfront.net/assets/image/3bba5d81-4e57-4e71-911b-cdfa05acca2c_compressed.png";
 var GOOGLE_LOGO2 = "https://d31vvfpk2ppdyq.cloudfront.net/assets/image/192485d2-a76a-4657-9c06-930b2c70badd_compressed.png";
 var VIDEO_MODELS = [
@@ -1437,8 +1452,7 @@ var VIDEO_MODELS = [
       references: { images: 6, videos: 1, videoSeconds: { max: 10 } },
       sourceVideoSeconds: { max: 10 },
       supports: [],
-      // generateOmniVideo sends MediaRef.label as a text part beside each image.
-      prompt: { referenceSyntax: "caption" },
+      prompt: OMNI_PROMPT,
       audio: "always"
     },
     defaults: {},
@@ -1475,8 +1489,7 @@ var VIDEO_MODELS = [
       // First-to-last-frame interpolation: `<LAST_FRAME>` goes with `<FIRST_FRAME>`.
       supports: ["videoExtension", "lastFrame"],
       extension: { segmentSeconds: 10, maxTotalSeconds: 40 },
-      // generateOmniVideo sends MediaRef.label as a text part beside each image.
-      prompt: { referenceSyntax: "caption" },
+      prompt: OMNI_PROMPT,
       audio: "always"
     },
     defaults: {},
@@ -1515,6 +1528,7 @@ var VIDEO_MODELS = [
       ratioIgnoredWith: ["firstFrame", "sourceVideo"],
       // No seed: 2.x echoes it but does not honour it (two seeded runs differed, 2026-10-01).
       supports: ["generateAudio", "lastFrame", "watermark"],
+      prompt: SEEDANCE_PROMPT,
       audio: "optional"
     },
     defaults: {
@@ -1556,6 +1570,7 @@ var VIDEO_MODELS = [
       ratioIgnoredWith: ["firstFrame", "sourceVideo"],
       // No seed: 2.x echoes it but does not honour it (two seeded runs differed, 2026-10-01).
       supports: ["generateAudio", "lastFrame", "watermark"],
+      prompt: SEEDANCE_PROMPT,
       audio: "optional"
     },
     defaults: {
@@ -1597,6 +1612,7 @@ var VIDEO_MODELS = [
       ratioIgnoredWith: ["firstFrame", "sourceVideo"],
       // No seed: 2.x echoes it but does not honour it (two seeded runs differed, 2026-10-01).
       supports: ["generateAudio", "lastFrame", "watermark", "taskType"],
+      prompt: SEEDANCE_PROMPT,
       audio: "optional"
     },
     defaults: {
@@ -1758,6 +1774,7 @@ var VIDEO_MODELS = [
       // fal's image-to-video endpoint has no aspect_ratio; the frame sets it.
       ratioIgnoredWith: ["firstFrame"],
       supports: ["seed", "promptExpansion", "lastFrame", "safetyChecker"],
+      prompt: H3_PROMPT,
       audio: "always"
     },
     defaults: {
@@ -1791,6 +1808,7 @@ var VIDEO_MODELS = [
       // fal's image-to-video endpoint has no aspect_ratio; the frame sets it.
       ratioIgnoredWith: ["firstFrame"],
       supports: ["seed", "promptExpansion", "lastFrame", "safetyChecker"],
+      prompt: H3_PROMPT,
       audio: "always"
     },
     defaults: {
@@ -1832,6 +1850,7 @@ var VIDEO_MODELS = [
         videoSeconds: { min: 2, max: 15, total: 15 }
       },
       supports: ["seed", "promptExpansion", "safetyChecker"],
+      prompt: H3_PROMPT,
       audio: "always"
     },
     defaults: {
@@ -1875,6 +1894,7 @@ var VIDEO_MODELS = [
         videoSeconds: { min: 2, max: 15, total: 15 }
       },
       supports: ["seed", "promptExpansion", "safetyChecker"],
+      prompt: H3_PROMPT,
       audio: "always"
     },
     defaults: {
@@ -2051,6 +2071,17 @@ function outputAudio(row) {
   const view = ConstraintsReadView.safeParse(row.constraints);
   return view.success ? view.data.audio : void 0;
 }
+var DefaultsReadView = import_zod.z.looseObject({
+  generateAudio: import_zod.z.boolean().optional().catch(void 0)
+});
+function billedAudio(row, requested) {
+  const audio = outputAudio(row);
+  if (audio === "never") return false;
+  if (audio !== "optional") return true;
+  if (requested !== void 0) return requested;
+  const defaults = DefaultsReadView.safeParse(row.defaults);
+  return (defaults.success ? defaults.data.generateAudio : void 0) ?? true;
+}
 function promptConventions(row) {
   const view = ConstraintsReadView.safeParse(row.constraints);
   const prompt = view.success ? view.data.prompt : void 0;