README.md

metadata

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: