remix-from-library

Remix from library

Take something the user already has in their Tolstoy library and spin new variations — using the original as a visual reference so the new creative stays on-brand.

The workflow

1. Find the source asset.

   • Named or described → search_assets with that query.
   • "My recent / latest" → list_assets.
   • "My favorites / starred" → list_assets with favoritedOnly: true.
   • Be honest: the Library MCP does NOT expose performance metrics — you cannot know which asset is actually "best-performing." If the user says "my best ad," surface the most likely candidates (favorited / recent / matching their description) and confirm which one before remixing. Don't guess silently.

2. Get the reference image URL. The reference must be an image:

   • Image asset → its mediaUrl.
   • Video asset → its thumbnailUrl. Call get_asset for full detail if the search/list result is thin.

3. Generate variations. Call generate_studio_content with that URL in referenceImageUrls, assetType for what they want (image or video), an aspectRatio for the destination, and a prompt describing the variation intent (e.g. "3 variations with different backgrounds, same product and brand feel"). Reference-to-output keeps the product/identity while letting composition vary.

4. Iterate with iterate_studio_content (same chatId) for refinements.

Example

"Clone my best serum ad and give me 3 fresh versions"

1. list_assets { favoritedOnly: true } (or search_assets "serum ad") → show candidates, confirm which one.
2. Take the chosen asset's thumbnailUrl (video) or mediaUrl (image).
3. generate_studio_content { assetType: "video", aspectRatio: "9:16", prompt: "3 fresh variations of this serum ad — new hooks and backgrounds, same product and brand feel", referenceImageUrls: ["<asset url>"] }

Notes

• referenceImageUrls must be public HTTPS image URLs (library CDN URLs qualify).
• This creates new content; it does not modify or replace the original asset.