Skip to content

feat: bounded PNG animation quality probes - #8

Draft
nkcbuilds wants to merge 2 commits into
mainfrom
feat/pixel-visual-qa
Draft

nkcbuilds wants to merge 2 commits into
mainfrom
feat/pixel-visual-qa

Conversation

@nkcbuilds

@nkcbuilds nkcbuilds commented Nov 6, 2026 •

Copy link
Copy Markdown
Owner

Scope

This draft PR adds local, bounded PNG visual-QA triage for PuppetControl's 2D animation pipeline. The report never auto-approves, blocks or certifies visual quality.

  • Compare two same-resolution workspace-local PNGs, full frame or an explicit ROI.
  • Report mean per-channel RGB difference, changed-pixel fraction and opaque-border transparency regression.
  • Inspect 2–60 ordered PNG frames via character qa sequence, emitting adjacent-pair metrics, loop first/last comparison and review candidates for unusually large changes.
  • Respect declared editorial cuts when analyzing unexpected pair deltas.
  • Optionally supply --control-values sampled from the rig; flag cases where the control changed but the selected region's pixels did not.
  • Strict bounds for file size, dimensions, frame count, pixel buffers, ROI, control arrays and workspace-contained paths. FFmpeg subprocesses use fixed argv with no shell. Includes TS unit tests, CLI help and agent skill updates.

Real visual evidence

A 512×640 / 30fps / 13-frame / 0.433333s local Hana blink sequence was rendered through the pinned Rev2D backend, with sample EyeLOpen/EyeROpen stepping 1→0→1. The frames, MP4, contact sheet and JSON QA reports remain in ignored outputs/qa/temporal-blink-20261107/ (they are not checked in or claimed to be acceptable production artwork).

  • Region (175,220)-(345,295): median adjacent RGB absolute error 10.7044/255.
  • First and last frames match exactly, including full-frame RGB (zero changed pixels).
  • No candidate jumps under the generic outlier threshold; that does not imply a good blink.
  • Changed-control motion stalls flagged at 3→4 and 8→9, both pixel-identical with control delta 0.166667. This is consistent with upstream Rev2D's explicitly held half-eye sprite over EyeOpen 0.3–0.6 (not a surprising renderer malfunction).
  • The contact sheet was visually inspected: intermediate 0.667 and 0.167 eye-open states contain duplicate/ghosted lash contours during sprite/mesh fades. These artifacts remain unresolved and the character stays unapproved.

Validation

  • npm run verify: 135 tests passed, typecheck and TypeScript build succeeded.
  • git diff --check: clean.
  • A real local CLI sequence command returned the expected two motion-stall candidates and no false claims of a quality pass.
  • No third-party art, secrets, or generated user assets committed.

Remaining / review

  1. Fix actual sprite/mesh transitional ghosting; visual QA cannot fix artwork or blending.
  2. Integrate a first-class contact sheet and a diverse licensed-art benchmark, plus region-specific seam analysis.
  3. PR feat: guarded local drawn eye and mouth variants #7 (drawn-variant workflow) is independent but overlaps CLI/docs. Reconcile after review/merge order; do not merge unreviewed changes automatically.

@coderabbitai

coderabbitai Bot commented Nov 6, 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