---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: control_images
    list: image
  - name: control_mask
    dtype: image
  - name: target_image
    dtype: image
  - name: prompt
    dtype: string
  splits:
  - name: train
    num_bytes: 105550701
    num_examples: 35
  - name: test
    num_bytes: 6884231
    num_examples: 12
  download_size: 107056437
  dataset_size: 112434932
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
license: apache-2.0
task_categories:
- image-to-image
language:
- en
pretty_name: char-compos
tags:
- image
- images-compose
size_categories:
- n<1K
---


- Training code: https://github.com/tsiendragon/qwen-image-finetune
- mask is used for add different weights to latent
- model generated results:

![image/png](https://cdn-uploads.huggingface.co/production/uploads/641af68ea5f876fe30c38508/K59UaUUqqFEtKhfQ576lD.png)