---
license: apache-2.0
base_model:
- lllyasviel/FramePackI2V_HY
pipeline_tag: text-to-image
---

# Animetic Light

[<img src="./assets/images/main_image.webp" alt="main image" width="400">](./assets/images/main_image.webp)

[![README - Japanese](https://img.shields.io/badge/README-Japanese-blue.svg?logo=markdown)](README_ja.md)
[![README - English](https://img.shields.io/badge/README-English-blue.svg?logo=markdown)](README.md)

An experimental model for background generation and relighting targeting anime-style images.
This is a LoRA compatible with FramePack's 1-frame inference.

For photographic relighting, IC-Light V2 is recommended.

- [IC-Light V2 (Flux-based IC-Light models) · lllyasviel IC-Light · Discussion #98](https://github.com/lllyasviel/IC-Light/discussions/98)
- [IC-Light V2-Vary · lllyasviel IC-Light · Discussion #109](https://github.com/lllyasviel/IC-Light/discussions/109)

IC-Light V2-Vary is available on Hugging Face Spaces, while IC-Light V2 can be used via API on platforms like fal.ai.

- [fal-ai/iclight-v2](https://fal.ai/models/fal-ai/iclight-v2)

## Features

- Generates backgrounds based on prompts and performs relighting while preserving the character region.

## Generation Examples

[<img src="./assets/images/sample.webp" alt="animatic light sample" width="500">](./assets/images/sample.webp)

## How to Use

- The recommended image resolution is approximately 1024x1024 total pixels.

- Prepare an image with a simple, flat background.
Even with an image that already has a background, roughly filling that background area with a single color may still work.

[<img src="./assets/images/erased_bg_sample.webp" alt="erase bg sample" width="300">](./assets/images/erased_bg_sample.webp)

- Prompts are recommended to be created with Gemini Flash 2.5.
A prompt length of 700 characters or more (including spaces and line breaks) is recommended.
Prompts for Gemini Flash 2.5 to help with prompt generation are available [here](./prompt.txt).
By inputting an image and a brief description, a complete prompt will be output.
  - Example 1: <Image of a dragon\> "A dragon flying while breathing blue fire. The background is a volcano."
  - Example 2: <Image of a dragon\> "Volcano"
- Add the following tags to the beginning of your prompt:
`illustration style`, (optional: `ambient lighting`, `dim lighting`)
- Note: Unsupported image styles (e.g., figure-like) may result in noise or residual backgrounds.

### ComfyUI Workflow

Workflow file is [here](./workflows/workflow_animetic_light.json)  
Sample image is [here](./sample_images/dragon.webp)  
Please use [xhiroga/ComfyUI-FramePackWrapper_PlusOne](https://github.com/xhiroga/ComfyUI-FramePackWrapper_PlusOne).  
For model installation instructions, please refer to the documentation for your respective inference environment.  
Also, please place `animetic_light.safetensors` in the `ComfyUI/models/loras` directory.

## Resources

### FramePack

- [GitHub - lllyasviel/FramePack: Lets make video diffusion practical!](https://github.com/lllyasviel/FramePack)

### Inference Environments

- [GitHub - xhiroga/ComfyUI-FramePackWrapper_PlusOne](https://github.com/xhiroga/ComfyUI-FramePackWrapper_PlusOne)
- [GitHub - kijai/ComfyUI-FramePackWrapper](https://github.com/kijai/ComfyUI-FramePackWrapper)
- [GitHub - git-ai-code/FramePack-eichi](https://github.com/git-ai-code/FramePack-eichi)

### Training Environments

- [GitHub - kohya-ss/musubi-tuner](https://github.com/kohya-ss/musubi-tuner)
  - [musubi-tuner/docs/framepack_1f.md at main · kohya-ss/musubi-tuner](https://github.com/kohya-ss/musubi-tuner/blob/main/docs/framepack_1f.md)

---
[![x - ippanorc](https://img.shields.io/badge/ippanorc-7e913b.svg?logo=x)](https://x.com/ippanorc)