Convert diffusers to original MS format

[kabachuha]

• Convert the In/Out blocks
• Convert the Text encoder
• Convert Unet ResBlocks
• Testing

closes #29

[kabachuha]

ModelStructure reference

Diffusers:

message.txt

ModelScope:

model_info.txt

[kabachuha]

The conversion works!

python convert_diffusers_to_original_ms_text_to_video.py --model_path animov --checkpoint_path text2video_pytorch_model.pth --clip_checkpoint_path open_clip_pytorch_model.bin 
Initializing the conversion map
Converting the UNET
Saving UNET
Saving CLIP
Operation successfull

The only thing left is to load the model and see if it works

[kabachuha]

Fixed all the keys mismatch

[kabachuha]

All complete

alUntitled.mp4

[ExponentialML]

This was on my todo list, but you knocked it out of the park. Thank you very much!

[sanatorios]

This is amazing but I'm not really sure about the needed process. Would love to convert strangeman3107's animov-512x fine-tuned model into MS format as well. How it could be possiblle?

[howardgriffin]

Hi, I have noticed that the newest version(v3.0.0) of 'convert_diffusers_to_original_ms_text_to_video.py' doesn't save clip_checkpoint.
What's more, this script doesn't save the configuration.json as https://github.com/kabachuha/sd-webui-text2video mentioned.
So，I cann't test my own trained model with sd_webui. How to solve my problem? @kabachuha @ExponentialML