While training, how can you get the final 13 channels model, since I have obtained 4 channels model which seems like a trained sdxl base model not a sdxl inpaint model?

[xjiao004]

[luxiaolili]

I realize it using sdxl inpaint model need a conv layer to change the in_channel, the train code for it see https://github.com/luxiaolili/IDM-VTON-train/

[yunyouke]

hello, in the inference code, the input channel need to be changed in 13? I donot see any code operation in the inferenc.py

[luxiaolili]

hello, in the inference code, the input channel need to be changed in 13? I donot see any code operation in the inferenc.py in the training script, unet.config['in_channels'] = 13 ,update the config.json to 13 channel, so you load the model with 13 channel