Problem when loading SANA's VAE

[frutiemax92]

Downloaded the VAE from here: https://huggingface.co/mit-han-lab/dc-ae-f32c32-sana-1.0-diffusers/tree/main

I'm getting a black screen for now, but there seems to be many missing layers... This is on Windows 11.

got prompt
Missing VAE keys ['encoder.project_in.weight', 'encoder.project_in.bias', 'encoder.stages.0.0.conv1.conv.weight', 'encoder.stages.0.0.conv1.conv.bias', 'encoder.stages.0.0.conv2.conv.weight', 'encoder.stages.0.0.conv2.norm.weight', 'encoder.stages.0.0.conv2.norm.bias', 'encoder.stages.0.1.conv1.conv.weight', 'encoder.stages.0.1.conv1.conv.bias', 'encoder.stages.0.1.conv2.conv.weight', 'encoder.stages.0.1.conv2.norm.weight', 'encoder.stages.0.1.conv2.norm.bias', 'encoder.stages.0.2.main.weight', 'encoder.stages.0.2.main.bias', 'encoder.stages.1.0.conv1.conv.weight', 'encoder.stages.1.0.conv1.conv.bias', 'encoder.stages.1.0.conv2.conv.weight', 'encoder.stages.1.0.conv2.norm.weight', 'encoder.stages.1.0.conv2.norm.bias', 'encoder.stages.1.1.conv1.conv.weight', 'encoder.stages.1.1.conv1.conv.bias', 'encoder.stages.1.1.conv2.conv.weight', 'encoder.stages.1.1.conv2.norm.weight', 'encoder.stages.1.1.conv2.norm.bias', 'encoder.stages.1.

[city96]

Hmm, looks like the repo with the separate VAE was (re)moved, and the one you link to is in the reference format (which doesn't have any conversion logic in place at the moment).

Could you check if this one works? I'll update the link in the readme if yes. https://huggingface.co/Efficient-Large-Model/Sana_1600M_1024px_diffusers/blob/main/vae/diffusion_pytorch_model.safetensors

[frutiemax92]

This works with the model you provided.

[city96]

Thanks, updated the link in the readme.