Anima Character Sheet Designer for ComfyUI
Create character reference sheets with Anima — using text prompts or local image tagging.
Choose up to 13 views, generate each view independently, and assemble the results into one character sheet.
This project builds on my MOSS-Image, H3, and Qwen-Image 2.1 Character Sheet Designers.
The familiar interface includes illustrated view selection, clothing details, part-specific prompts, styles, and automatic/manual sheet layouts. The Anima version also lets you inspect and edit the internal base prompt for every view.
Anima Character Sheet Designer v1.0.1 for ComfyUI
Create character sheets with up to 13 views from a text prompt or tags extracted from an image.
After cloning or installing, turn on Use Image Input and select PixAI FP16 or WD EVA02. The first time you select a Tagger that isn’t installed, ComfyUI automatically downloads about 1–1.3 GB for that model. This may take a few minutes, depending on your connection. Verified files are reused on later runs.
Important: Both Input Modes Use Text-to-Image Generation
Unlike the H3 and Qwen versions, this workflow does not directly use the input image as a visual reference for generation.
There are two input methods, selected with Use Image Input:
1. Text Input — OFF / false
Connect your character description to Text input.
Your text is used to generate the selected views. Image analysis and Tagger model weights are not needed in this mode.
2. Image Input — ON / true
Connect a standard Load Image node to Image input.
The selected Tagger analyzes the image locally and converts its appearance into tags:
Input Image → PixAI / WD Tagger → Appearance Tags → New Anima Generations → Assembled Sheet
This is similar to the MOSS version’s image-to-description approach, but uses appearance tags instead of a generated text description.
It is not image editing or reference-image conditioning. Faces, clothing, colors, accessories, proportions, and unseen details may change. Exact character or image preservation is not guaranteed.
A crop focused on the intended character can help reduce unrelated background or other-person tags.
PixAI or WD: Choose Your Tagger
PixAI Tagger v1.0 FP16 is the recommended default, offering newer data and tag coverage. The official PixAI v1.0 model card lists a May 2026 cutoff.
WD EVA02 Large v3 is also supported for users who prefer the established WD tagging approach or its results.
You only need the Tagger you select.
You can either:
Download from the node: open the input/model subgraph, find Anima Image Tagger, select a model, and click Download selected model.
Install manually: follow the download links and folder instructions in the GitHub README.
The download button retrieves only the selected model. Progress and model status appear inside the Tagger node. Model ready means the selected files are installed.
The PixAI implementation uses an unofficial FP16 ONNX export by Mexes of the official PixAI v1.0 model.
Editable Prompts for Every View
Open Prompts → Edit view prompts to inspect or replace a view’s internal base prompt.
These prompts control framing, orientation, and what the view depicts. For example:
full body, standing, front view, neutral pose, complete subject visible
Select a view, edit its prompt, and press Apply. Reset this prompt restores the default for that view.
Keep phrases short and separate them with comma + space.
Part Prompts Are Additional Appearance Instructions
Part prompts serve a different purpose: they add appearance changes to the relevant views.
For example:
Footwear:
black bootsHands / gloves:
fingerless gloves
They supplement the view prompt rather than replacing it.
Quality tags and negative prompts remain under your control. The node does not automatically add a quality prefix; edit the normal negative prompt inside the rendering subgraph.
Views, Styles, and Settings
Available presets include:
Standard — 5 views
Basic — 4 views
Turnaround — 3 views
Single view
7 views — hands and feet
Head — 6 views
Clothing — front/back
All — 13 views
You can also select views individually.
Auto / Manual output size controls the finished sheet. Per-view rendering resolution is configured separately in Settings.
The layout guide is a composition preview; it is not sent to Anima.
Optional styles add a rendering direction. Choose None to leave the style unchanged by the node.
Required Generation Models
The supplied workflow uses:
ComponentDownloadWAI-ANIMA v1.0 — base 1.0waiANIMA_v10Base10.safetensorsQwen3 0.6B — Text Encoderqwen_3_06b_base.safetensorsQwen Image VAEqwen_image_vae.safetensorsAnima Turbo LoRA v0.2Anima Turbo LoRA
Template starting settings:
Euler / simple · 10 steps · CFG 1.5 · LoRA strength 0.7
Set the CLIP loader type to stable_diffusion.
These are the supplied workflow’s starting settings, not guaranteed optimal values for every model or character.
Installation
In your ComfyUI custom_nodes directory:
git clone https://github.com/ukr8b3g-cmyk/Anima-Character-Sheet-Designer.gitInstall the required Python dependencies in ComfyUI’s Python environment. Image tagging requires ONNX Runtime; see the README for CPU/CUDA setup.
The supplied workflow also uses Save WEBP Meta.
Restart ComfyUI and reload your browser after installation or updates. A recent ComfyUI with native Anima and subgraph support is required.
Tagger Model Installation Paths
Download the files for your selected Tagger and place them in the following folders. Keep the original filenames. You only need one Tagger model.
PixAI v1.0 FP16 — recommended/default
ComfyUI/
└─ models/
└─ anima_tagger/
└─ pixai_v1_fp16/
├─ model_fp16.onnx
├─ selected_tags.csv
└─ preprocess.jsonWD EVA02 Large v3 — alternative
ComfyUI/
└─ models/
└─ anima_tagger/
└─ wd_eva02_large_v3/
├─ model.onnx
└─ selected_tags.csvCreate these folders if they do not exist. Keep each model’s selected_tags.csv together with its corresponding weights.
Alternatively, open Anima Image Tagger inside the workflow’s input/model subgraph, select a Tagger, and click Download selected model to install its files automatically.
After manual installation, click Check model files. Model ready confirms that the selected files have passed verification.
Limitations
Independent generation can introduce differences between views. Left/right orientation, garment separation, and small details may vary. Clothing-only views may still contain a person or mannequin, and complex 13-view sheets require visual review.
Image tagging provides a starting description, not a guarantee of faithful reproduction.
GitHub & Workflow
Remember: image input is converted into tags for new T2I generation. The original image is not directly used to edit or reproduce the character.