Skip to content

feat: time-domain 30fps blink review - #14

Draft
nkcbuilds wants to merge 1 commit into
feat/eye-transition-no-ghostfrom
feat/timed-blink-qa
Draft

nkcbuilds wants to merge 1 commit into
feat/eye-transition-no-ghostfrom
feat/timed-blink-qa

Conversation

@nkcbuilds

Copy link
Copy Markdown
Owner

Goal

Close the gap between PuppetControl's 25-frame non-uniform parameter sweep and what an actual 30 FPS video displays. Provide real timed-playback evidence, rather than continuing to add unverified global fade/opacity variants.

Draft stacked PR on feat/eye-transition-no-ghost (#13), which is itself built on #12/#11/#10/#9. Do not merge directly into main.

Changes

  • New character qa blink <imported-project> --timed [--region x,y,w,h] retains full source/model/provenance validation and does not change artwork or approval.
  • Exports 36 actual PNG frames at 30 FPS plus a reproducible 1.2-second H.264 MP4 and 6×6 full/eye contact sheets under a new ignored QA directory. Frame n samples at n/30 exactly. Synthetic schedule: 250 ms open lead-in, 140 ms close, 65 ms hold, 240 ms reopen and open tail.
  • Reuses exact identical control-state PNGs instead of rerendering them: 13 genuine Rev2D renders + 23 copies in the verified sample.
  • Pixel QA reports candidate abrupt changes, changed-control stalls, frame values and timing; none are automatic quality approval.
  • Added pure deterministic 30fps timeline tests, complete docs/EYE_TIMING.md, README/agent-skill guidance, and a tiny MIT-licensed Hana 30fps sample MP4 attributed in third-party notices.

Actual visual findings

  • Rendered from independently verified, unapproved stepped-eye and short-fade Hana experimental rigs. Both produced valid 30fps, 512x640, H.264, 1.2s, 36-frame MP4s verified by FFprobe.
  • SHA-256 identity across all 36 rendered PNGs of the two modes for this exact synthetic blink schedule: different fade windows occur between frame timestamps. The two modes are NOT equivalent at other timing/FPS; their earlier threshold still comparisons remain different.
  • Both time-domain reports flagged 5 possible abrupt changes and 5 possible changed-control stalls. Human inspection of the 6x6 contact sheet still shows a stylized/stiff eye closure, not polished Live2D-like artwork.
  • Additional continuous-mesh/late-overlay experiments tried at 0.11–0.20, 0.035–0.105, 0.065–0.15 with warp scaled to 80%, and 0.01–0.055. All failed visual QA due to pinched lash corners or shifted translucent contour ghosting. Those experiments remain ignored/local and were not added as production options.

Validation

  • npm run verify: 220 passed tests across 38 files, TypeScript check/build passed.
  • git diff --check clean; only MIT-licensed attributed tiny video is committed, no user art, private recordings, or generated rig snapshot.
  • Maintains default character qa blink 25-frame threshold inspection unchanged.

Remaining blockers

  • Better curved intermediate eyelid drawings or an actual eyelid-specific deformable shape that correctly occludes iris/skin through all parameter values. Global fades and hard switches alone cannot solve source-art inconsistency.
  • Wider character-style QA, 15/60fps playback, human review, and true 10–15s imported-art video still pending. Original experiment stays approved:false.

@coderabbitai

coderabbitai Bot commented Nov 7, 2026

Copy link
Copy Markdown

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

  • Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

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