Release v1.2.2

Changes

• fix: match registered node ID casing (ComfyUI-WanAnimatePreprocessV2) (aef268a)
• chore: bump v1.2.2 (bf7af28)

Full Changelog: v1.2.1...v1.2.2

v1.2.1 — Release-management bump

v1.2.1 — Release-management bump

This is a release-management-only version bump. There are no functional code changes versus v1.2.0.

Why

• Backfills proper GitHub Releases for v1.1.0, v1.1.1, and v1.2.0 (previously only tags existed).
• Establishes a clean cadence so future versions are bumped (never recreated) per pyproject.toml.

Changes

• pyproject.toml: version 1.2.0 → 1.2.1.

All v1.2.0 features (MediaPipe FaceMesh integration, iris/lip outputs, graceful fallback) are unchanged.

v1.2.0 — MediaPipe FaceMesh iris & lip tracking

v1.2.0 — MediaPipe FaceMesh integration for high-fidelity iris & lip tracking

Highlights

• MediaPipe FaceMesh integration (refine_landmarks=True, 478 landmarks incl. 10 iris points) replaces the OpenCV gradient-voting pupil finder for dramatically improved iris/gaze and lip tracking on Wan 2.2 Animate.
• Verified MP_TO_DLIB68 mapping slices the 478 MediaPipe vertices to the standard 68-point dlib layout, dropped into face_kps[1:69] so Wan 2.2's pose encoder, limbSeq, and existing visualization stay byte-compatible.
• Iris centres now come directly from MediaPipe indices 468 (right) and 473 (left).
• Lip-sync metric — inner-lip Mouth Aspect Ratio (|13–14| / |78–308|) exposed for Wan 2.2 audio-driven lip-sync.

New PoseAndFaceDetectionV2 outputs

• right_pupil_xy (STRING JSON) — per-frame [x, y] in full-image pixel space.
• left_pupil_xy (STRING JSON) — per-frame [x, y] in full-image pixel space.
• lip_openness_ratio (FLOAT) — clip-mean inner-lip MAR; per-frame values also embedded in iris_data JSON.

Reliability

• Graceful fallback: if mediapipe is missing, fails to load, or fails on any individual frame, the node transparently falls back to the legacy ViTPose + _find_pupil_center pipeline. No crashes.
• pose_utils/pose2d_utils.py is intentionally untouched — keypoint_list, limbSeq, and the retarget_pose math are byte-for-byte unchanged, so existing Wan 2.2 controlnet workflows continue to work.

Install

pip install mediapipe

(also added to requirements.txt and pyproject.toml).

Backwards compatibility

The first 7 outputs of PoseAndFaceDetectionV2 are preserved in their original order; older workflows wired to only those sockets keep working unchanged.

v1.1.1 — CI hotfix

v1.1.1 — CI hotfix

Fixes

• CI: resolve ModuleNotFoundError: tomllib on Python 3.10 in the publish workflow by falling back to tomli.

No runtime changes for end users.

v1.1.0 — Comfy Registry compliance

v1.1.0 — Comfy Registry compliance & modernized publish pipeline

Highlights

• Modernized CI publish pipeline aligned with Comfy Registry conventions.
• Project metadata cleanup in pyproject.toml (publisher id, display name, icon).
• V2 node naming finalized (OnnxDetectionModelLoaderV2, PoseAndFaceDetectionV2, DrawViTPoseV2, category WanAnimatePreprocess_V2).
• Step parameter added to threshold inputs for 2-decimal precision.

Compatibility

• Requires Python ≥ 3.10.
• Drop-in replacement for v1.0.x workflows.