@kolbo/mcp 1.77.2 → 1.78.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@kolbo/mcp",
-  "version": "1.77.2",
+  "version": "1.78.0",
   "description": "Kolbo AI MCP Server - Generate images, videos, music, speech, and sound effects from Claude Code",
   "main": "src/index.js",
   "bin": {

package/src/tools/_shared.js

@@ -611,6 +611,7 @@ async function uiGenerating(p) {
       ? { failed_submissions: p.failed_submissions } : {}),
     ...(p.warning ? { _warning: p.warning } : {}),
     _widget_note: 'A live Kolbo widget is rendering this generation for the user (progress + final result + action buttons). Tell the user it is generating and the card above will update — do NOT poll in a loop. If you need the output URLs (e.g. for a follow-up edit or a report), call get_generation_status ONCE with wait=true — it blocks until done. Tracking several generations? Pass ALL their ids in generation_ids in that one call.',
+    _paid_note: 'This generation is RUNNING and the user is paying for it. If you now realize the tool, model, or parameters were wrong, call cancel_generation with this generation_id FIRST, then start the replacement — never leave a wrong generation running alongside its retry (the user gets two cards and two charges).',
   }, null, 2);
   return uiResult(UI.generation, text, structured);
 }

package/src/tools/generate.js

@@ -541,7 +541,7 @@ function registerGenerateTools(server, client, options = {}) {
   // retired textToVideoGeneration path and was stale.
   server.tool(
     'generate_video',
-    'Generate a video from a text prompt using Kolbo AI. For SEVERAL different videos, pass all their prompts in `prompts` in ONE call (one combined widget) — never a series of separate calls. For animating an existing still image into motion, use generate_video_from_image instead. For a coordinated multi-scene video campaign, use generate_creative_director with workflow_type="video". Supports reference images (for style/composition guidance) and Visual DNA for character consistency. Returns the final video URL when complete.',
+    'Generate a video from a text prompt using Kolbo AI. For SEVERAL different videos, pass all their prompts in `prompts` in ONE call (one combined widget) — never a series of separate calls. For animating an existing still image into motion, use generate_video_from_image instead. For a coordinated multi-scene video campaign, use generate_creative_director with workflow_type="video". Supports reference images (for style/composition guidance) and Visual DNA for character consistency. ROUTE BEFORE CALLING: when reference images anchor IDENTITY (specific characters, a specific product, a location that must match) — especially 2+ of them — that is generate_elements, not this tool; reference_images here are loose style/composition hints. Decide the right tool FIRST: a mis-routed call still starts a PAID generation, and switching tools afterwards without cancel_generation leaves the user paying for both. Returns the final video URL when complete.',
     {
       prompt: z.string().optional().describe('Text description of the video to generate. Required unless `prompts` is provided.'),
       prompts: promptsField('videos'),
@@ -1581,7 +1581,7 @@ function registerGenerateTools(server, client, options = {}) {
   // ─── generate_video_from_video ─────────────────────────────
   server.tool(
     'generate_video_from_video',
-    'Restyle / transform an existing video (video-to-video). Use for style transfer, scene restyling, subject swap, motion transfer, character replacement, or burning in styled subtitles (VEED Subtitles). Source video can be a URL or absolute local path. `prompt` is OPTIONAL: most models need it, but prompt-less models (VEED Subtitles, Act Two, Wan Animate, Kling Motion Control) ignore it. For VEED Subtitles, pass a `preset` style and optional `source_language` / `translation_language` instead of a prompt. IMPORTANT: different models support different extra inputs — call list_models type="video_to_video" and read max_images / max_videos / max_elements on the chosen model before generating. Pass reference_images for models with max_images > 0 (e.g. Kling O1/O3, Aleph, WAN VACE), reference_videos for models with max_videos > 1 (e.g. WAN 2.6 reference-to-video accepts up to 3), and elements for models with max_elements > 0. For animating a still image use generate_video_from_image instead. For text-only → video use generate_video.',
+    'Restyle / transform an existing video (video-to-video). Use for style transfer, scene restyling, subject swap, motion transfer, character replacement, or burning in styled subtitles (VEED Subtitles). Source video can be a URL or absolute local path. `prompt` is OPTIONAL: most models need it, but prompt-less models (VEED Subtitles, Act Two, Wan Animate, Kling Motion Control) ignore it. For VEED Subtitles, pass a `preset` style and optional `source_language` / `translation_language` instead of a prompt. IMPORTANT: different models support different extra inputs — call list_models type="video_to_video" and read max_images / max_videos / max_elements on the chosen model before generating. Pass reference_images for models with max_images > 0 (e.g. Kling O1/O3, Aleph, WAN VACE), reference_videos for models with max_videos > 1 (e.g. WAN 2.6 reference-to-video accepts up to 3), and elements for models with max_elements > 0. REPAIR / RETIME (not restyling, no prompt needed — these keep the footage and fix or retime it): model "topaz/deblur/video" removes lens, motion and compression blur; "topaz/colorize/video" colorizes black-and-white footage; "topaz/interpolate/video" is SLOW MOTION and frame-rate conversion (see slowdown_factor / target_fps); "topaz/sdr-to-hdr/video" masters SDR footage to HDR (see output_format). Reach for these when the user says blurry, shaky-detail, black-and-white, slow motion, smoother frame rate, or HDR — a restyle model would repaint the video instead of repairing it. For animating a still image use generate_video_from_image instead. For text-only → video use generate_video.',
     {
       source_video: z.string().describe('URL or absolute local path to the primary source video to restyle. **Source duration must fall within `min_video_duration`-`max_video_duration` from list_models for the chosen model** — videos outside that range are rejected (or silently truncated by some upstream providers). For models that use reference_videos as their primary input (e.g. WAN 2.6 reference-to-video), pass the first reference video here and also include it in reference_videos.'),
       prompt: z.string().optional().describe('Text description of the desired restyle / transformation. Required by most video-to-video models; omit for prompt-less models (VEED Subtitles, Act Two, Wan Animate, Kling Motion Control).'),
@@ -1898,7 +1898,7 @@ function registerGenerateTools(server, client, options = {}) {
 
       // ── upscale ────────────────────────────────────────────
       scale: z.number().optional()
-        .describe('Upscale factor: 1, 2, 4 or 8. Used with operation="upscale". Default: 2.'),
+        .describe('Upscale factor: 1, 2 or 4. Used with operation="upscale". Default: 2. There is NO 8 — every Topaz split endpoint declares upscale_factor max 4 in its OpenAPI, so an 8 is a guaranteed provider error. Output size is otherwise unbounded: 4x on a 4K source is a 16K result and is priced pro-rata, not refused.'),
 
       enhancement_model: z.string().optional()
         .describe('Topaz engine. With operation="upscale" on model "topaz/upscale/image" it selects the engine family: "Standard V2" / "High Fidelity V3" / "CGI" / "Text Refine" (faithful), "Wonder 3.5" (rebuilds natural detail), "Bloom 2" (reinvents detail — most expensive), "Transparent" (keeps the alpha channel). With operation="enhance" it selects the correction engine for the chosen model. Omit for the model default. Call list_models to see the engines a model offers.'),