How can I use the native mmagic implementation for stable-diffusion instead of 'from_pretrained=stable_diffusion_v15_url' so that I can debug line by line through VScode?

[AyanKumarBhunia]

Can someone help me with the config description with the pre-trained weights link?

[Z-Fran]

UNet2DConditionModel and AutoencoderKL are from diffusers. So you need to debug in diffusers. I think you can add breakpoints in magic https://github.com/open-mmlab/mmagic/blob/main/mmagic/models/editors/stable_diffusion/stable_diffusion.py#L119 , and execute step by step until step into diffusers. For example, if you want to debug in UNet2DConditionModel , you can add a breakpoints in https://github.com/open-mmlab/mmagic/blob/main/mmagic/models/editors/stable_diffusion/stable_diffusion.py#L250