Anima Regional Canvas

<img width="1712" height="745" alt="Clip_7" src="https://github.com/user-attachments/assets/73c40c9b-d6b1-4eab-a3e7-0baf36a4f8af" />

Update

Fixed a canvas persistence bug where painted regions could disappear after switching ComfyUI tabs or when the canvas was rebuilt. The canvas now preserves and restores painted mask data more safely.

An ANIMA-focused custom node for Anima-LLLite Regional ControlNet workflows.

It is designed for ANIMA workflows using the Anima base model, Anima-LLLite, and the Anima-LLLite Regional ControlNet model. The node lets you paint color-coded regions directly inside ComfyUI, outputs the color mask image for Apply Anima ControlNet-LLLite, and generates masked conditioning from matching region prompts.

Requirements

• Node: kohya-ss/ComfyUI-Anima-LLLite
• Anima base model: circlestone-labs/Anima Other Anima fork models
• Model: anima-lllite-regional-exp-v3.safetensors
• Model repository: Sen-sou/Anima-LLLite-Regional-Controlnet This node does not include the Anima-LLLite node or the regional ControlNet model. Install them separately

Install

Search for "anima regional" from ComfyUI Manager. <img width="1033" height="250" alt="1782109442-qZH5iIbnRBG4mx23WP7yMvTz" src="https://github.com/user-attachments/assets/64d29b98-74a1-4c47-90fe-010ed2f4ad29" />

Clone this repository into ComfyUI's custom_nodes folder:

cd D:\Codex\ComfyUI\custom_nodes
git clone https://github.com/ukr8b3g-cmyk/Anima_Regional_Canvas.git

Restart ComfyUI after installation.

Example workflows are included in workflows/:

• Anima_Regional_Canvas_Test.json
• Anima_Regional_Inpaint_Canvas_Test.json

Usage

1. Add one of the canvas nodes:
   • Anima Regional Canvas: normal regional generation.
   • Anima Regional Inpaint Canvas: regional inpaint generation from an optional input image.
2. Check the canvas size shown in the Canvas width x height info badge.
   • The default size is 1024 x 1024.
   • Enter a width and height in the number boxes, then press Enter or move focus away to apply the resize.
   • Load Canvas and a connected image update the canvas size. Sizes are normalized down to multiples of 8 to match the latent output.
   • For ANIMA workflows, a larger size is recommended. Smaller sizes work, but they may be too low-resolution for detailed ANIMA output.
3. Enter prompts:
   • QUALITY: quality/style tags, for example masterpiece, absurdres, score_7, anime style
   • SCENE: count, character names, background, and situation, for example 2girls, cirno, reimu, cafe
   • RED / BLUE / YELLOW / GREEN / MAGENTA: prompt for each painted region
   • NEGATIVE: negative prompt
4. Paint regions on the canvas with the color buttons.
   • Save Canvas: save the painted canvas as PNG.
   • Load Canvas: load a saved canvas PNG or image back into the canvas.
   • Brush size can be changed with the Brush number box or slider.
   • Windows shortcut: Alt + right-drag on the canvas. Move left/right to change brush size.
   • Mac shortcut: Control + Option + left-drag on the canvas. Move left/right to change brush size.
   • Moving up/down during the shortcut adjusts brush opacity.
   • The brush circle preview shows the current brush size on the canvas.
5. Connect IMAGE to Apply Anima ControlNet-LLLite image.
6. Connect POSITIVE, NEGATIVE, and LATENT to KSampler.
7. For a mask overlay preview, use ComfyUI core Blend Images:
   • generated image -> Blend Images image1
   • MASK_PREVIEW -> Blend Images image2
   • Blend Images -> Preview Image
8. Save the final image with Save WEBP Meta if metadata output is needed.

Node Variants

Anima Regional Canvas

Use this for normal generation.

• IMAGE outputs the painted color mask for Apply Anima ControlNet-LLLite image.
• LATENT outputs an empty latent using the canvas size.
• Paint red, blue, yellow, green, or magenta regions and enter matching region prompts.

Anima Regional Inpaint Canvas

Use this for inpaint generation.

• Connect an input image to image when you want to inpaint over an existing image.
• Connect vae when using the node's INPAINT_LATENT output.
• The connected image is shown on the canvas automatically when available.
• Paint only the areas that should be controlled or repainted.
• White/unpainted areas are treated as the keep/base area.
• grow_mask_by expands the inpaint mask slightly to reduce hard edges.
• If no image and vae are connected, the node falls back to an empty latent, similar to the normal node.

Design

• Apply Anima ControlNet-LLLite stays separate.
• KSampler, VAE Decode, and Save WEBP Meta stay separate.
• External custom nodes are not imported or called.
• This implementation is independently designed, inspired by regional conditioning workflows, and optimized for this canvas-based node. It does not reuse external custom-node code.
• Regional control uses ComfyUI's standard masked conditioning: only painted colors with non-empty prompts are encoded.
• QUALITY is for quality/style tags.
• SCENE is for count, subject names, background, and situation, for example 2girls, cirno, reimu, cafe.
• RED, BLUE, YELLOW, GREEN, and MAGENTA are region prompts.

Outputs

Anima Regional Canvas

• IMAGE: color mask image for Apply Anima ControlNet-LLLite image
• MODEL: passthrough model
• POSITIVE: masked conditioning for KSampler positive
• NEGATIVE: conditioning for KSampler negative
• LATENT: empty latent using the canvas size
• METADATA: prompt metadata string
• MASK_PREVIEW: preview-only image

Anima Regional Inpaint Canvas

• IMAGE: color mask image for Apply Anima ControlNet-LLLite image
• MODEL: passthrough model
• POSITIVE: masked conditioning for KSampler positive
• NEGATIVE: conditioning for KSampler negative
• INPAINT_LATENT: inpaint latent when image and vae are connected; otherwise empty latent
• INPAINT_MASK: inpaint mask generated from painted regions
• METADATA: prompt metadata string

Compatibility

Verified in this workspace:

• Python 3.13.11
• PyTorch 2.12.1+cu130
• CUDA build 13.0
• Pillow 12.2.0
• NumPy 2.4.4

Inferred minimum:

• Python: ComfyUI-supported Python, practically 3.10+.
• PyTorch: ComfyUI-supported PyTorch. This node uses only basic tensor ops and should not require a specific CUDA build.
• CUDA: no direct dependency. CPU or any CUDA build that your ComfyUI/PyTorch already supports is acceptable.
• Pillow/NumPy: no special version pin; ComfyUI's installed versions are sufficient.

The node avoids hard version pins and only lazily uses ComfyUI core helpers when available.

Standard Connection

Anima Regional Canvas IMAGE -> Apply Anima ControlNet-LLLite image
Apply Anima ControlNet-LLLite MODEL -> KSampler model
Anima Regional Canvas POSITIVE -> KSampler positive
Anima Regional Canvas NEGATIVE -> KSampler negative
Anima Regional Canvas LATENT -> KSampler latent_image
KSampler LATENT -> VAE Decode -> Save Image

Inpaint Connection

Load Image IMAGE -> Anima Regional Inpaint Canvas image
Load VAE VAE -> Anima Regional Inpaint Canvas vae
Anima Regional Inpaint Canvas IMAGE -> Apply Anima ControlNet-LLLite image
Apply Anima ControlNet-LLLite MODEL -> KSampler model
Anima Regional Inpaint Canvas POSITIVE -> KSampler positive
Anima Regional Inpaint Canvas NEGATIVE -> KSampler negative
Anima Regional Inpaint Canvas INPAINT_LATENT -> KSampler latent_image
KSampler LATENT -> VAE Decode -> Save Image

Standard Connection Chart

flowchart LR
  Model["Load Diffusion Model"] --> Canvas["Anima Regional Canvas"]
  Clip["Load CLIP"] --> Canvas
  VAE["Load VAE"] --> Decode["VAE Decode"]
  Canvas -- IMAGE --> LLLite["Apply Anima ControlNet-LLLite"]
  Canvas -- MODEL --> LLLite
  LLLite -- MODEL --> KSampler
  Canvas -- POSITIVE --> KSampler
  Canvas -- NEGATIVE --> KSampler
  Canvas -- LATENT --> KSampler
  KSampler -- LATENT --> Decode
  Decode -- IMAGE --> Save["Save Image"]
  Canvas -- MASK_PREVIEW --> Preview["Preview Image optional"]
  classDef canvasNode fill:#0f4f3a,stroke:#69d89b,color:#ffffff
  classDef llliteNode fill:#50321a,stroke:#ffb45c,color:#ffffff
  class Canvas canvasNode
  class LLLite llliteNode

Inpaint Connection Chart

flowchart LR
  Model["Load Diffusion Model"] --> Canvas["Anima Regional Inpaint Canvas"]
  Clip["Load CLIP"] --> Canvas
  Image["Load Image"] --> Canvas
  VAE["Load VAE"] --> Canvas
  VAE --> Decode["VAE Decode"]
  Canvas -- IMAGE --> LLLite["Apply Anima ControlNet-LLLite"]
  Canvas -- MODEL --> LLLite
  LLLite -- MODEL --> KSampler
  Canvas -- POSITIVE --> KSampler
  Canvas -- NEGATIVE --> KSampler
  Canvas -- INPAINT_LATENT --> KSampler
  KSampler -- LATENT --> Decode
  Decode -- IMAGE --> Save["Save Image"]
  Canvas -- INPAINT_MASK --> MaskToImage["Convert Mask to Image optional"]
  MaskToImage --> MaskPreview["Preview Image optional"]
  classDef canvasNode fill:#0f4f3a,stroke:#69d89b,color:#ffffff
  classDef llliteNode fill:#50321a,stroke:#ffb45c,color:#ffffff
  class Canvas canvasNode
  class LLLite llliteNode

UI Prompt Fields

• QUALITY: quality and style tags, for example masterpiece, absurdres, score_7, anime style.
• SCENE: global composition, count, subject names, pose, background, and situation, for example 1girl, full body, standing with arms out, outdoor, blue sky, green field.
• RED / BLUE / YELLOW / GREEN / MAGENTA: prompt for each painted region.
• NEGATIVE: negative prompt.

Common prompt rule:

• Put the overall scene in SCENE.
• Put region-specific details in the matching color prompt.
• Leave unused color prompts empty.
• White/unpainted areas use the default QUALITY + SCENE conditioning.

Example:

SCENE:
1girl, full body, standing with arms out, outdoor, blue sky, green field

YELLOW:
long blonde twin tails, large fluffy hair

MAGENTA:
pink one-piece dress

BLUE:
blue eyes

GREEN:
green grass field

Colors

• RED
• BLUE
• YELLOW
• GREEN
• MAGENTA
• white background uses the default QUALITY + SCENE conditioning

Acknowledgements

• kohya-ss/ComfyUI-Anima-LLLite for the Anima-LLLite ComfyUI node.
• Sen-sou/Anima-LLLite-Regional-Controlnet for the regional ControlNet model.
• This project was inspired by Sen-sou's Anima-LLLite-Regional-Controlnet, but the code in this repository is original, independently developed, and not copied or plagiarized.
• ComfyUI and its community.

License

MIT License. See LICENSE.