A practical cache review checklist for AI image generator pages

I am documenting a repeatable CyberPanel/OpenLiteSpeed review for browser-based image tools and would appreciate feedback on the cache boundaries.

The page combines a mostly static interface with generation requests and optional reference-image uploads. My current checklist is:

1. Keep generation/API responses uncached.
2. Cache versioned CSS, JavaScript, and style thumbnails with long immutable headers.
3. Exclude authenticated, upload, and result routes from LSCache.
4. Verify that query parameters used by the UI do not fragment the public landing-page cache unnecessarily.
5. Test one cold and one warm load from a normal browser before changing TTLs.

For a concrete page shape, I am testing this anime generator landing page. The generation workflow stays in the browser while the public explanation and style examples are cache-friendly.

Would you recommend a separate cache rule for uploaded reference previews, or should those always be treated as private/no-store assets?