Edit anime images using plain-English instructions, directly inside ComfyUI.
Anima Smart Edit 1.0 combines Anima Base V1.0, Anima Edit V2 and Cosmos Reference in a fully local, two-pass workflow. Load an image, describe the desired change, and refine the result at a higher resolution.
What can you do?
Change clothing, hairstyles and colors through text instructions.
Adjust lighting, atmosphere and time of day.
Rework image details while guiding the model to retain the original subject and composition.
Preservation depends on the image, prompt and settings. Exact identity, small details and untouched pixels are not guaranteed.
Workflow features
Organized sections with setup notes and credits.
Original, half-original or manual first-pass resolution.
First sampling pass for editing, followed by neural upscaling and a second pass for refinement.
Final dimensions equal to 2× the first-pass width and height.
Previews for comparing the first pass and refined result.
Fully local inference, without an external image-generation API.
Quick start
Follow
README_INSTALL_EN.mdin the ZIP to install the required nodes and download the models.Import the workflow and select your local model files.
Upload your image. Start with the saved Half original resolution setting.
Write your edit request in English and queue the workflow.
Download the result from the final preview.
Example:
“Change her hair to pastel pink. Preserve her eye color, outfit, pose, background and lighting.”
The supplied settings are 25 steps / CFG 4 / denoise 1.0 for editing and 8 steps / CFG 1 / denoise 0.25 for refinement. Lower first-pass denoise may preserve more structure but weaken the edit.
Requirements
Anima Base V1.0, the matching text encoder and VAE, Anima Edit V2, and the RealESRGAN anime upscaler. Required node packages are Cosmos Reference, rgthree and the included resolution helper. Model weights are downloaded separately; links and folder locations are in the guide.
Hardware
Tested locally, 5 to 15 seconds per image on an RTX 5090, 64 GB system RAM (depends on the image resolution). This is the tested configuration, not a minimum requirement but I would recommend at least 8-10GB VRAM for a smooth Anima sampling, use more aggressive quantizations if needed (8GB VRAM). Memory use and generation time depend on resolution and model configuration.
Body Proportions
This workflow is actually very bad at altering body proportions such as height, fat percentage, breasts size and other "proportion altering" editings. This Krea2 Editing Workflow can work together with Krea2 Anime models such as this, some LoRAs and good prompting to alter body proportions.
Credits
CircleStone Labs and Comfy Org — Anima, built on NVIDIA Cosmos.
Mirumo0u0, pamparamm and levzzz5154 — Cosmos Reference and its credited contributions.
rgthree, Real-ESRGAN contributors, and the ComfyUI community.
Feedback and example results are welcome!