raulc0399/pixart-alpha-hed-controlnet

These are the weights for the Pixart HED Controlnet, converted to be used with the pixart alpha controlnet pipeline.

To use:

git clone https://github.com/raulc0399/PixArt-alpha.git .
git checkout master_train_controlnet_diffusers
pip install -r requirements.txt
python controlnet/pipeline/run_controlnet_pipeline.py

(make sure to change the input_image_path and prompt)

the weights were created using the controlnet/convert_pixart_alpha_controlnet_to_diffusers.py from the github repo

python controlnet/convert_pixart_alpha_controlnet_to_diffusers.py --orig_ckpt_path=<full path to>/PixArt-XL-2-1024-ControlNet.pth --dump_path=./<output folder>/hed-controlnet

the training script has been implemented in controlnet/train_pixart_controlnet_hf.py to start it with the toy-dataset "fusing/fill50k" run the file ./train_controlnet_hf_diffusers.sh

more info can be found in this PR https://github.com/PixArt-alpha/PixArt-alpha/pull/164