Transfer clothed character poses from mannequin renders with Qwen Pose
Apply an existing textureless 3D mannequin pose to a clothed character image through a repeatable local ComfyUI edit, with PNG and JSON artifacts.
npx skills add agentskillexchange/skills --skill transfer-clothed-character-poses-from-mannequin-renders-with-qwen-pose
Use this bounded Qwen-Image 2.1 workflow to transfer a pose from an existing textureless 3D mannequin render to an authorized, clothed character image. Invoke it when an agent needs a repeatable local character-pose variation with explicit input files and a reviewable image artifact, rather than manually reconnecting a ComfyUI graph for each request.
What this skill actually does
The agent checks the two input paths and ComfyUI prerequisites, runs the upstream Route 3 script, and returns the generated PNG with its JSON metadata sidecar. From the repository root, the documented invocation is python3 scripts/run_pose_transfer.py --char /absolute/character.png --pose-ref /absolute/doll_render.png --out outputs/pose_transfer.png. The script uploads both images to the configured ComfyUI server, applies the bundled dual-image edit workflow, waits for the result, and downloads the output. Check the printed JSON, the sidecar’s ok value, and the actual image before reporting success. The implementation’s ok check establishes a nonempty output file; visual review must assess pose fidelity, identity and clothing preservation.
Prerequisites are Python 3.10 or newer, a running ComfyUI installation, the Qwen-Image 2.1 base models, the VNCCS PoseStudio LoRA, and the repository’s bundled workflow and helper scripts. Follow the route-specific installation guide and configure the ComfyUI endpoint before invoking the script. Inputs are sent to that endpoint, so use an operator-approved server. This route reuses an existing mannequin render; it does not create new mannequin poses. Ordinary photographs and illustrations are unsupported pose references because upstream reports reference-person leakage and redrawing. Camera framing may change, and identity preservation is an upstream claim to verify on each result.
The catalog scope is non-explicit, clothed character pose transfer through Route 3. It excludes the repository’s NSFW text-to-image recipes, undressing edits, explicit prompts and samples, browser-driven pose creation, general ComfyUI administration, and arbitrary model or workflow management. The upstream project includes adult-oriented material; this entry describes only its independently documented pose-transfer path.
Framework fit is Multi-Framework: upstream names Claude Code, Codex, Cursor and OpenClaw, and this route is an ordinary Python CLI without a required agent SDK or MCP server. Category: Image & Creative Automation. Metadata and invocation shape were checked against upstream documentation and source; runtime behavior and security have not been tested by this intake.