Plan to support model loaded directly from huggingface/diffusers?

[nguyendothanhtruc]

Hi, thank you for your amazing work!.
May I ask how to load a pretrained model directly from huggingface/diffusers? I also use the convert script of diffusers to convert from diffusers format to safetensors or ckpt. However, I encouter the key error as bellow:

key cond_stage_model.transformer.text_model.embeddings.position_ids not found!

Can your work only load checkpoint downloaded on civital?.
Hope you can reply and provide me some advices. Thank you a lot!

[tyxsspa]

Maybe that error is caused by a mis-matched transformer version? (try transformers==4.30.2)
Yes in demo.py you can load any SD1.5 ckpt or lora models from civitai without problems.