MOSS-Image-CharRef

MOSS-Image-CharRef

Draws a character from one reference picture in new scenes, as a LoRA for MOSS-Image Base in ComfyUI, with no per-character training. Made for anime and illustrated characters, on single images and comic pages.

Made by 7err4 on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: MOSS-Image CharRef is a LoRA for the MOSS-Image Base model in ComfyUI. Feed it a single character image plus a text prompt, and it renders that same figure in a scene you describe, with matching face, hair, clothes and drawing style. You do not train a separate LoRA for each character.

Who it suits: It suits artists, comic makers and tinkerers who work with anime or illustrated characters in ComfyUI and want the same character across many scenes.

Industry | Who uses it | How they use it | Example
Comics and manga | Comic artists | Keep one character consistent across panels and whole pages | Drawing the same hero in a library and on a rooftop
Illustration | Digital artists | Place a reference character into new scenes by prompt | Changing the outfit through the prompt text
Game and app art | Indie developers | Make varied scenes of an original character | Generating a cast member in a kitchen scene
Hobby art | Hobbyists | Try character scenes without training anything | Testing a favorite design in several settings

You can skip it if you work with photorealistic or 3D faces, since the card says those do not hold reliably.

Where to use it:

• ComfyUI version 0.42 or newer
• The Fumika app, which uses it as its reference picture option
• Windows, macOS or Linux with Python, to download the files

Use case | Who | What they get back
Single character image | Artists | A new scene that keeps the face, hair, outfit and style of the reference
Comic page | Comic makers | Pages built from a MOSS JSON layout with panels and speech bubbles
Outfit change | Illustrators | The same character in different clothing named in the prompt

How to set it up: You need ComfyUI, the MOSS-Image Base model and one extra node.

• ComfyUI 0.42 or newer
• The MOSS-Image Base model, which this LoRA builds on
• The ComfyUI-MOSS-Image-Ostris-Edit node, which supplies reference latents and the KV cache
• Python and huggingface_hub to download the files

1. Update ComfyUI and install the ComfyUI-MOSS-Image-Ostris-Edit node.
2. Download this repository's files to your computer.
3. Put the LoRA file where ComfyUI looks for LoRA files.

pip install -U huggingface_hub
hf download 7err4/MOSS-Image-CharRef

How to use it: Build the workflow in order. Patch the model with the KV cache on, then load this LoRA at strength 1.0. Encode your reference as image1 with a prompt. Sample with the card's settings of 50 steps and cfg 3. The prompt must start with the card's prefix, then the scene.

Same character and art style as Picture 1.

Tool | What it does
MOSSImageOstrisEditModelPatch | Patches the model with the KV cache turned on
LoraLoaderModelOnly | Loads this LoRA at strength 1.0
TextEncodeMOSSImageOstrisEdit | Encodes your reference image and the prompt, and also the empty negative prompt

Example: You load a clean full-body picture of your character as image1 and write the prefix followed by "reading in a library, long black hair, red jacket". The result is a new image of that character in a library, keeping the look of the reference.

For newcomers and experts: If you are new to this, a LoRA is a small add-on file that teaches an image model one skill. Here the skill is copying a character from a reference picture. A first small try is one clean reference and one short scene prompt.

For people who know the field, the card describes training with ai-toolkit in edit mode at rank 32 for 1000 steps on about 1330 reference and target pairs of anime art. Sampling uses euler with sgm_uniform and skip layer guidance on a single layer, 22. The tradeoff is anime strength at the cost of photoreal faces.

Reference picture goes in as image1
Prompt starts with the same-character prefix
Model patch and LoRA guide the sampler
A new scene of the character comes out

Good to know: Appearance words in the prompt, such as hair style, eye colour and key outfit pieces, noticeably improve how closely the character matches.

• Limits: fine hair details can drift, and several characters in one scene only take the identity of the first reference reliably
• Limits: small text and hands stay weak as in MOSS-Image Base, and a reference taken from a comic panel brings its props along
• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the reference pictures you load and the downloaded model files
• Not for: intimate images of a real person without consent, sexual depictions of minors, facial recognition databases, or deceptive manipulation

On Hugging Face, 13 November 2026

• Library: comfyui
• Downloads in the last 30 days: 0
• Likes: 51
• License: apache-2.0
• Based on: OpenMOSS-Team/MOSS-Image-Base
• Tags: comfyui, lora, moss-image, character-reference, image-editing, anime

Links

• Model: https://huggingface.co/7err4/MOSS-Image-CharRef
• Files: https://huggingface.co/7err4/MOSS-Image-CharRef/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download 7err4/MOSS-Image-CharRef

Then follow the usage in its model card. Nothing in this kit runs by itself.