MOSS Manga Colorist | Cinematic Manga Colorization

作者：moonrailgun

MOSS Manga Colorist v1 is a MOSS-Image Base LoRA for black-and-white manga colorization, with cinematic lighting, warm skin tones and anime-style rendering.

Use the MOSS reference-image workflow in anima_lora. A generic text-to-image LoRA loader does not reproduce this workflow. Civitai on-site generation and a standalone ComfyUI workflow have not been validated.

Quick start

• Install anima_lora and download its MOSS Base model pack through the GUI or with make download-model ARGS=moss-base.

• Download the LoRA and optional Usage ZIP below. Put the weight file and the extracted ZIP contents in a manga-colorist folder inside your anima_lora directory.

• Run this from your anima_lora directory, replacing the input path with your page or folder: uv run --no-sync python scripts/moss/generate.py --config manga-colorist/inference.toml --ref_image /path/to/your/page.png --out_dir output/my-colorized-pages

Example settings: prompt comic, strength 1.0, 30 sampling steps, CFG 3, seed 7, ER-SDE. Keep the other inference settings in the included inference.toml.

Base model: OpenMOSS-Team/MOSS-Image-Base. The LoRA requires the base model; it is not a standalone checkpoint.

Examples and limitations

The gallery contains five example pages and their comparisons. Comparison columns: Original | Base Model (No LoRA) | Trained LoRA (First Pass). The base and LoRA examples use different inference workflows and settings, so these are practical workflow comparisons, not a controlled LoRA on/off test.

This is stylized colorization with re-rendering. Faces, clothing, linework, lettering and other details may change; canonical character colors and exact artwork preservation are not guaranteed.

Hugging Face project and original example images

License and credits

LoRA weights and original release materials: Apache License 2.0. Base model by OpenMOSS; tooling by sorryhyun/anima_lora. Example artwork remains the property of its respective creators and publishers. See LICENSE and NOTICE in the Usage ZIP for license text and artwork credits.

中文说明

基于 MOSS-Image Base 的漫画上色 LoRA，侧重电影感光照与动画式渲染。请使用 anima_lora 的 MOSS 参考图推理流程；下载下方权重和使用说明包，按 USAGE.txt 操作即可。常用设置：提示词 comic、权重 1.0、30 步、CFG 3。

模型会重绘局部细节，不保证完整保留线条、对白或官方配色。示例中的基础模型与 LoRA 使用不同推理流程，三列对比不属于严格的 LoRA 开关实验。