ComfyUI Workflow: ControlNet OpenPose (2-Pass)

This is the official 2-pass OpenPose example from the ComfyUI examples repository. The first pass uses AnythingV3 to establish a base composition from the pose reference. The second pass refines the same composition with AOM3A3 (Abyss Orange Mix), keeping the OpenPose conditioning so the pose stays locked while the style and detail improve.

OpenPose ControlNet takes a pose image (the body skeleton) and uses it as a structural guide for generation. The pose image in this workflow is a stick-figure render of a person in a worship pose, included in the repo as pose_worship.png.

Key benefits

• 2-pass setup: the base model lays the composition, the refiner model polishes the style. Both passes share the same pose conditioning.
• Official ComfyUI example, maintained alongside the core.
• Drag-and-drop the example PNG into ComfyUI to load the workflow. The JSON is embedded in the PNG.
• No preprocessor node required, the pose image loads directly as a conditioning image.
• Pairs cleanly with the T2I-Adapter (lighter weight, similar effect for some use cases).

How it works

1. Download the workflow PNG from the linked GitHub URL and drag it into ComfyUI. The workflow loads automatically.
2. Get the model files: AnythingV3 and AOM3A3 checkpoints go in models/checkpoints. OpenPose controlnet weights go in models/controlnet.
3. Optional: swap the pose image. Replace pose_worship.png with your own pose reference. The workflow is pose-agnostic.
4. Queue the prompt. The first pass runs the base generation, the second pass refines. Total time depends on resolution and the refiner model.
5. Iterate on the conditioning strength. Lower the controlnet strength for more creative freedom, raise it for stricter pose adherence.

OpenPose ControlNet is the right tool when you have a specific pose in mind and you want every generation to follow it. Character sheets, action sequences, and reference-heavy illustration work all benefit.