qwen21-nsfw-pose-skill/CHANGELOG.md

Changelog

Notable changes to this package. Format follows Keep a Changelog; versions are git tags. The package ships as a skill, not a library, so versions mark documented, reproducible states of the scripts + docs rather than an API contract.

[Unreleased]

Added

• VNCCS_LORA environment variable: choose the VNCCS PoseStudio LoRA file for routes ③ and ④ (run_pose_transfer.py, run_vnccs_ui.py) and for the doctor.py LoRA check. The default is still VNCCS_QI2_PoseStudioV1.1.safetensors. Route ③'s --lora flag still takes precedence.

1.1.0 — 2026-10-20

Since 1.0.0: a new browser-free route, a self-check script, macOS and Windows support, and a rewritten README showcase. Routes are now ①②③④ with no gaps.

Added

• Route ③ — scripted pose transfer (scripts/run_pose_transfer.py): change a character's pose from a textureless 3D mannequin render, no browser involved. Dual-image edit + VNCCS PoseStudio LoRA @1.0. Route ④ saves suitable mannequin renders (doll_*.png) automatically.
• scripts/doctor.py — read-only environment self-check: ComfyUI connectivity, base models, LoRAs, VNCCS nodes, pose library and browser backend, each with fix instructions.
• scripts/make_custom_pose.py — build custom poses from poses already in the library (#4).
• English README (README.en.md), with a language switcher on both READMEs.
• macOS (Apple Silicon) support: documented guidance, plus a macos-latest CI job that runs the offline package checks on Apple's runner.
• Windows compatibility for doctor.py output, the media lock and the pose index (#9).
• Hardware requirements section (issue #5) and an acknowledgements section.

Changed

• Licensed original code and docs under Apache-2.0 (third-party assets keep their own terms).
• README showcase rewritten: full-width animated cover card for the NSFW author originals, click-to-expand instead of inline, plus a Route ④ demo GIF at the top.
• Setup guide corrected for the edit LoRA, VNCCS and the pose library (#2).
• --list-poses hint, usage text and setup wording polished; dead code dropped.
• Photo/illustration pose references are now called out explicitly as unsupported (the model composites or redraws the reference person) — use a 3D mannequin render instead.

Fixed

• Result naming for full pose keys (#1).
• Five pose previews replaced; validate_package.py now fails on empty preview files.
• Placeholder pose renamed.
• validate_package.py also validates README.en.md images.

1.0.0 — 2026-10-13

Initial stable release: Routes ① T2I and ② Edit, the author recipe and prompt structure, boldness tiers L0–L3, the 136-pose VNCCS library, install guide and the author-case reproduction.