Chippo/kale-img-1-lora

kale-img-1 (kxv) — style LoRA for MOSS-Image

A style adapter for MOSS-Image-Base that aims for a soft 3D children's illustration look: colorful, rounded subjects with gentle shading and a clean light background.

Trigger

Use kxv to call the learned style. For example:

kxv, a small red lighthouse on a rocky island, soft 3D children's illustration, plain white background

Generated samples

These are the nine standalone generated images from the additional sample set. Each image is available as its own file.

Sample	File
00014
00017
00025
00039
00046
00068
00071
00072
00077

Original vs. generated

Each source and generated image is uploaded as a separate PNG. Original files are listed alongside their generated counterparts for one-to-one comparison.

Subject	Original	Generated
Owl
Glacier
Helicopter
Pagoda
Parrot
Pancakes
Treehouse
Emperor penguin
Alarm clock

Use

Use this adapter with a MOSS-Image-Base workflow that supports LoRA adapters. For ComfyUI, place the file in ComfyUI/models/loras/ and load it with a compatible MOSS-Image LoRA loader. Loader support can vary with the MOSS-Image integration and checkpoint format.

A reasonable starting point for the base model is flow-matching Euler, 50 steps, CFG 3, and shift 5. Adjust the sampler settings and LoRA strength for your workflow and prompt.

Training details

The checkpoint metadata identifies kxv as the trigger token and records:

• Base model: MOSS-Image-Base
• Training toolkit: AI-Toolkit, version 0.13.39

License

This repository is released under Apache-2.0. See LICENSE. MOSS-Image-Base is also released under Apache-2.0; consult the upstream repository for its notices and terms.