Turn your cat into a tiny felt actor!
Felt Cat Studio transforms a cat photo into an expressive, needle-felted chibi character. Choose an acting preset and a character role from English dropdown menus—no detailed prompt writing required.
IMPORTANT: HOW TO OPEN THIS FILE
This download is a ComfyUI workflow JSON, even if Civitai labels the file as “Config.”
Import the JSON into ComfyUI, or drag it onto the ComfyUI canvas. Do not place it in the checkpoints folder.
Model weights are not included. Install the required models listed below and select them in the corresponding loader nodes.
FEATURES
• Soft needle-felted wool texture and chibi proportions
• Six acting presets combining expressions and body gestures
• Five character roles, including a custom option
• English controls and dropdown menus
• Optional clothing and character details
• Camera and framing instructions tailored to each acting preset
• Still images for sharing or use in a separate image-to-video workflow
ACTING PRESETS
• Secretly smug
• Playing innocent
• Politely unimpressed
• Holding back anger
• Suddenly stunned
• Pretending to understand
CHARACTER ROLES
• Everyday
• Antique appraiser
• Wedding host
• Space steward
• Custom
QUICK START
1. Import Felt_Cat_v03_2_CIVITAI_EN_DROPDOWNS.json into ComfyUI.
2. Select your installed diffusion model, text encoder, and VAE in the loader nodes.
3. Upload a clear photo of one cat in node #1.
4. Choose an acting preset from the dropdown in node #30.
5. Choose a character role from the dropdown in node #40.
6. Optionally enter clothing or character details in node #46. Select Custom if you want to describe your own role.
7. Run the workflow and save the output from node #18.
Change the random seed to try another variation. You do not need to edit the internal preset library.
REQUIRED MODELS
The tested configuration uses:
Diffusion model:
flux-2-klein-4b-fp8.safetensors
Folder: ComfyUI/models/diffusion_models/
Text encoder:
qwen_3_4b_fp4_flux2.safetensors
Folder: ComfyUI/models/text_encoders/
CLIPLoader type: flux2
VAE:
flux2-vae.safetensors
Folder: ComfyUI/models/vae/
Use the FLUX.2 Klein 4B distilled model intended for 4-step generation. If your compatible model files have different names, select them manually in the loaders.
Different model variants or text encoder precision may produce different results.
COMFYUI REQUIREMENTS
Use a recent ComfyUI installation with a matching, up-to-date frontend.
The dropdowns use CustomCombo, an experimental ComfyUI core node. No third-party custom-node package is intentionally required.
If CustomCombo is missing or the dropdowns do not display correctly, check your ComfyUI and frontend versions.
DEFAULT SETTINGS
• Sampler: Euler
• Scheduler: Simple
• Steps: 4
• CFG: 1
• Output size: 640 × 960
TESTING
Two local generation tests passed on October 29, 2026:
• Holding back anger + Everyday
• Pretending to understand + Antique appraiser
Both tests used the workflow’s actual backend nodes, including CustomCombo. The outputs showed changes in expression, gesture, clothing, and role details.
Frontend dropdown interaction was not separately retested. These local tests do not guarantee compatibility with every installation. This resource is intended for download and local ComfyUI use; Civitai-hosted generation is not claimed.
TIPS AND LIMITATIONS
Use a clear, well-lit photo with the cat’s face and markings visible.
The workflow aims to retain recognizable coat colors, facial markings, and eye color while transforming the cat into a stylized felt character. Likeness, anatomy, and expression can vary between generations.
Camera instructions guide composition but do not guarantee exact framing. In the two tested outputs, framing remained wider than the requested close-ups.
This workflow generates still images only. Video generation and audio are not included.
TROUBLESHOOTING
If generation fails, please share:
• The full error message or traceback
• Your ComfyUI version
• Your GPU and available VRAM
• The diffusion model, text encoder, and VAE filenames
• The workflow version you downloaded
A node name such as “SamplerCustomAdvanced” alone is not enough to identify the cause.
This resource is an authored ComfyUI workflow, not a trained LoRA or checkpoint. Use of the underlying models remains subject to their respective licenses.