license: other
library_name: anima-edit
tags:
- anima
- controlnet
- controlnet-lllite
- anime
- image-restoration
- image-to-image
- denoise
- deblur
- tile
pipeline_tag: image-to-image
base_model: circlestone-labs/Anima
Anima Tile & Repair ControlNet-LLLite
This repository contains the first public release of Anima Tile & Repair ControlNet-LLLite, a lightweight ControlNet-LLLite model for the Anima image model family.
The model is intended for anime image repair workflows. It is designed to provide tile/repair-style guidance for improving anime images affected by visible noise, blur, compression artifacts, or low-detail degradation, while keeping the original composition and character structure stable.
This is a v1 release. Future updates are planned.
Files
| File | Description |
|---|---|
anima_tiled_lllite_v1.safetensors |
Recommended v1 checkpoint. |
anima_tiled_lllite_v1-000002.safetensors |
Earlier training checkpoint. |
anima_tiled_lllite_v1-000004.safetensors |
Earlier training checkpoint. |
anima_tiled_lllite_v1-000006.safetensors |
Earlier training checkpoint. |
anima_tiled_lllite_v1-000008.safetensors |
Later training checkpoint before final export. |
Intended Use
Use this model as a ControlNet-LLLite guidance module together with an Anima-compatible generation or restoration workflow.
Typical use cases:
- repairing noisy anime images;
- reducing blur and softness;
- improving tiled/local detail consistency;
- restoring degraded anime-style images while preserving layout and character structure.
Usage Notes
The anima-edit project documentation describes ControlNet-LLLite as a lightweight alternative to ControlNet. Relevant inference options include:
--control_net_lllite_models: path to the ControlNet-LLLite model file;--control_net_multipliers: guidance strength / multiplier;--control_net_ratios: step ratio during which ControlNet-LLLite is applied.
ControlNet and ControlNet-LLLite should not be used at the same time in the same workflow.
Example path usage:
--control_net_lllite_models /path/to/anima_tiled_lllite_v1.safetensors \
--control_net_multipliers 1.0 \
--control_net_ratios 1.0
The exact strength depends on the base Anima checkpoint, sampler, resolution, denoise strength, and the severity of the degradation. For a first test, start with a moderate multiplier and adjust upward only if the repair guidance is too weak.
Compatibility
This release is intended for Anima-based anime image restoration workflows and was prepared in the anima-edit ecosystem.
Anima uses a DiT-style architecture with a Qwen3 text encoder / adapter stack and Qwen-Image VAE components. This ControlNet-LLLite checkpoint is not a standalone image model; it must be loaded as an auxiliary guidance model in a compatible workflow.
Limitations
- This is an early v1 release.
- It is focused on anime image repair, not general photographic restoration.
- Strong guidance may alter fine details or over-sharpen images.
- Results depend heavily on the base Anima model and the surrounding inference pipeline.
Release Notes
v1
Initial release of Anima Tile & Repair ControlNet-LLLite for anime denoising, deblurring, and tiled repair guidance.
Future versions may include improved robustness, more training data, and updated checkpoints.