# Anima example workflows
Examples for the 3D pose and RGB regional adapters.
## 3D pose example workflow
Files:
- [3dpose_example.json](3dpose_example.json): importable ComfyUI workflow.
- [3dpose_example.png](3dpose_example.png): sample RGB 3D-mesh pose condition (768 × 1152).
- [3dpose_example_workflow.png](3dpose_example_workflow.png): workflow screenshot showing the condition, node connections and generated preview.
### Use
1. Install [ComfyUI-Anima-LLLite](https://github.com/kohya-ss/ComfyUI-Anima-LLLite) in an Anima-compatible ComfyUI setup.
2. Import `3dpose_example.json` and upload `3dpose_example.png` in **Load Image**.
3. Select the compatible models in the loader nodes: `anima-base-v1.0.safetensors`, `qwen_3_06b_base.safetensors`, `qwen_image_vae.safetensors`, and [anima_3dpose_6200.safetensors](../anima_3dpose_6200.safetensors). Obtain the base components from [circlestone-labs/Anima](https://huggingface.co/circlestone-labs/Anima).
4. Generate using the workflow's existing prompt and sampler settings. The node is **Apply Anima ControlNet-LLLite (sd-scripts)** (`AnimaLLLiteApply_sdscripts`).
The supplied settings are **768 × 1152**, batch 1, **30 steps**, **CFG 4**, **er_sde / simple**, denoise 1.0, LLLite strength **1.0**, active window **0.0–1.0**, and preserve wrapper enabled. The saved seed is `85793728931883`, with **randomize** enabled; choose fixed seed handling for repeatable comparisons.
The shared JSON uses portable published model filenames and the included sample filename. The original screenshot shows model dropdowns with subfolder prefixes; adjust the dropdowns to your own installed model filenames. Node links, layout, prompts and sampling settings are preserved. PNG workflow/text/EXIF metadata is removed while visible pixels are unchanged; import the separate JSON to load the workflow.
This is a user-supplied standalone 3D-pose example. It was inspected and packaged here; no new generation or quantitative evaluation was performed for this upload.
### Preview
[Back to the model card](../README.md)
## Regional control example workflow
Files:
- [regional_example.json](regional_example.json): importable ComfyUI workflow.
- [regional_example.png](regional_example.png): sample RGB instance-color region map (768 × 1152).
- [regional_example_workflow.png](regional_example_workflow.png): workflow screenshot showing the condition, node connections and generated preview.
### Use
1. Install [ComfyUI-Anima-LLLite](https://github.com/kohya-ss/ComfyUI-Anima-LLLite) in an Anima-compatible ComfyUI setup.
2. Import `regional_example.json` and upload `regional_example.png` in **Load Image**.
3. Select the compatible models in the loader nodes: `anima-base-v1.0.safetensors`, `qwen_3_06b_base.safetensors`, `qwen_image_vae.safetensors`, and [anima_regional_4335.safetensors](../anima_regional_4335.safetensors). Obtain the base components from [circlestone-labs/Anima](https://huggingface.co/circlestone-labs/Anima).
4. Generate using the workflow's existing prompt and sampler settings. The node is **Apply Anima ControlNet-LLLite (sd-scripts)** (`AnimaLLLiteApply_sdscripts`).
The sample region map and output canvas are **768 × 1152**, batch 1, **30 steps**, **CFG 4**, **er_sde / simple**, denoise 1.0, LLLite strength **1.0**, active window **0.0–1.0**, and preserve wrapper enabled. The saved seed is `965383411713769`, with **randomize** enabled; choose fixed seed handling for repeatable comparisons.
Colors are region identifiers, not fixed semantic classes; black marks unassigned space. The sample map was resized to 768 × 1152 with nearest-neighbor resampling to preserve the flat region colors and boundaries. The control node resizes the supplied region map to the output canvas.
The shared JSON uses portable published model filenames and the included sample filename. The original screenshot shows model dropdowns with subfolder prefixes; adjust the dropdowns to your own installed model filenames. Node links, layout, prompts and sampling settings are preserved. PNG workflow/text/EXIF metadata is removed. The workflow screenshot is unchanged; the sample map is resized as noted above. Import the separate JSON to load the workflow.
This is a user-supplied standalone regional-control example. It was inspected and packaged here; no new generation or quantitative evaluation was performed for this upload.
### Preview
[Back to the model card](../README.md)
## Advanced workflow: 3D pose + regional control
Files: [advanced_region_3dpose_example.json](advanced_region_3dpose_example.json) and [workflow screenshot](advanced_region_3dpose_example_workflow.png).
Install [ComfyUI_toyxyz_test_nodes](https://github.com/toyxyz/ComfyUI_toyxyz_test_nodes) for **3D Pose Editor** (`ToyxyzPose3DEditor`), **ComfyCouple Region painter** (`ComfyCoupleRegionPainter`), and **ComfyCouple Mask** (`ComfyCoupleMask`). Install [ComfyUI-Anima-LLLite](https://github.com/kohya-ss/ComfyUI-Anima-LLLite) for the two **Apply Anima ControlNet-LLLite (sd-scripts)** nodes. Import the JSON and select your installed **Anima Base 1.0**, `qwen_3_06b_base.safetensors`, `qwen_image_vae.safetensors`, `anima_3dpose_6200.safetensors`, and `anima_regional_4335.safetensors`.
The saved workflow includes two editable bodies, their pose/camera/lighting state, painter masks, and per-person prompts. It applies the 3D pose adapter first, then the regional adapter, then ComfyCouple Mask regional prompting. Both adapters use strength **1.0**, start **0.0**, and `preserve_wrapper = true`. The pose adapter ends at **1.0**; the regional adapter ends at **0.3**.
Keep the regional ControlNet's **`end_percent` low**, using **0.3** as the supplied example, to release regional guidance earlier. Alternatively, **paint additional mask areas in ComfyCouple Region painter** to expand or refine each character's region, including areas for clothing outside the body silhouette. Choose the corresponding character layer when editing its mask and prompt.
Saved sampling settings: **768 × 1152**, batch **1**, **30 steps**, **CFG 4**, **er_sde / simple**, denoise **1.0**, seed `43365297866659` with **randomize** enabled. The editor and painter are saved at **768 × 1024**; these original settings are preserved. Align their dimensions with the output canvas if you want matching aspect ratios. Choose fixed seed handling for repeatable comparisons.
The published JSON uses portable model filenames. Pose state, embedded masks, prompts, node positions, connections, and generation settings are preserved. The screenshot shows the original model subfolders; select the equivalent installed files. Screenshot metadata is removed without changing its visible pixels. Import the separate JSON to load the workflow.
This is a user-provided combined-control example; no new generation or quantitative evaluation was performed for this upload.
[Back to the model card](../README.md)