Skip to content

feat: generative card grammar, new motion system and trending suggestions - #4

Merged
nikolmedo merged 1 commit into
mainfrom
feat/generative-card-grammar
Sep 24, 2026
Merged

nikolmedo merged 1 commit into
mainfrom
feat/generative-card-grammar

Conversation

@nikolmedo

Copy link
Copy Markdown
Owner
  • Add an enum-only card grammar to the Scene contract (surface, corner, header, pattern, size, per-kind variant, typeScale), mirrored and drift-checked on server and client, with per-kind renderer variants and a 12-column module grid driven by a single --cols source.
  • Seed a deterministic client fallback for modules the model leaves plain, and seed SVG card patterns from the module's scene index.
  • Wire mood and density into CSS, drop the anti-variety prompt rules and raise the answer temperature to 0.6.
  • Replace blur entrances with five clip-path/transform families, make reveal "count" drive a decoding count-up, cap the cascade, fix the holding window, crossfade the skeleton and FLIP span changes.
  • Pause and slim the ambient canvas and stop per-frame React renders in the count-up.
  • Generate landing suggestions from Google Trends through the fast model, with validated output, a TTL cache, single-flight and a failure cooldown, falling back to the static examples.

…ions

- Add an enum-only card grammar to the Scene contract (surface, corner,
  header, pattern, size, per-kind variant, typeScale), mirrored and
  drift-checked on server and client, with per-kind renderer variants
  and a 12-column module grid driven by a single --cols source.
- Seed a deterministic client fallback for modules the model leaves
  plain, and seed SVG card patterns from the module's scene index.
- Wire mood and density into CSS, drop the anti-variety prompt rules and
  raise the answer temperature to 0.6.
- Replace blur entrances with five clip-path/transform families, make
  reveal "count" drive a decoding count-up, cap the cascade, fix the
  holding window, crossfade the skeleton and FLIP span changes.
- Pause and slim the ambient canvas and stop per-frame React renders in
  the count-up.
- Generate landing suggestions from Google Trends through the fast
  model, with validated output, a TTL cache, single-flight and a
  failure cooldown, falling back to the static examples.
@nikolmedo nikolmedo self-assigned this Sep 24, 2026
@nikolmedo
nikolmedo merged commit cb79258 into main Sep 24, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant