This dataset contains a 25k subset prepared for SDXL ControlNet training with synthetic lineart and scribble conditioning maps.
The original training target images were encoded into SDXL VAE latents using:
coco_lineart_sdxl_latents/ — Hugging Face dataset with lineart conditioning images
coco_scribble_sdxl_latents/ — Hugging Face… See the full description on the dataset page:
https://huggingface.co/datasets/Stoland/coco-sdxl-lineart-scribble-latents-25k.