Trained remotely on Modal using Qwen/Qwen-Image-2512 as the base model. This repository contains the LoRA adapter only and does not redistribute the base model.
Description
A LoRA model fine-tuned to generate watercolor anime illustrations with soft ink linework, pale paper texture, transparent washes, muted palettes, and quiet storybook-like scene mood.
The model is best for rainy streets, bamboo landscapes, moonlit rivers, quiet train scenes, rice fields, and gentle character illustrations.
Trigger words
Use watercolor anime to activate the LoRA style/concept.
Usage
I recommend using this LoRA with Qwen Image 2512 in ComfyUI, Diffusers, Draw Things, or another compatible LoRA loader.
Recommended starting settings:
LoRA weight: 0.80-0.95
Steps: 25-32
Guidance: 3.0-3.3
Shift: 2.83
Sampler: FlowMatch if available, otherwise UniPC Trailing
Resolution: 1024 x 1024
Preview images use the images/preview_01.png through images/preview_06.png naming pattern.
Example prompt:
watercolor anime of a quiet bamboo path after rain, pale paper texture, transparent watercolor washes, soft ink linework, muted colors
Training Data
This model was trained on 120 GPT-generated watercolor anime images.
