TheLastBen / fast-stable-diffusion

fast-stable-diffusion + DreamBooth
MIT License
7.52k stars 1.31k forks source link

Is there a way to implement kohya_ss for lora training? #1647

Open kelvin-zhao opened 1 year ago

kelvin-zhao commented 1 year ago

Hi Ben,

Wondering if there's a way to implement https://github.com/bmaltais/kohya_ss as a Google Collab? The versatility of Lora models seem to make it much more flexible to use with different models, not mentioning the small size and quick training!

TheLastBen commented 1 year ago

Hi, this is on my todo list, it will be available in the near future.

kelvin-zhao commented 1 year ago

That'll be awesome! I've been trying to get kohya ss gui to work for me (using mac), but I just can't find any solutions so far...

Omenizer commented 1 year ago

Hi, this is on my todo list, it will be available in the near future.

Yess πŸ™πŸΌπŸ˜‹πŸ˜πŸ†πŸ‘

cerarslan commented 1 year ago

That'll be awesome! I've been trying to get kohya ss gui to work for me (using mac), but I just can't find any solutions so far...

Use additional network extension then upload your lora model to additional-network/-extension/model. Normal stable diffusion webui folder(model/lora) doesn't work.

kelvinz commented 1 year ago

That'll be awesome! I've been trying to get kohya ss gui to work for me (using mac), but I just can't find any solutions so far...

Use additional network extension then upload your lora model to additional-network/-extension/model. Normal stable diffusion webui folder(model/lora) doesn't work.

The issue isn't to use Lora, but to train Lora though.

cerarslan commented 1 year ago

Ohh, so sorry mate. That is just miss understand them πŸ™

Omenizer commented 1 year ago

@TheLastBen it would be awesome to be able to export LORA from a checkpoint. There's an option in the Kohya tools, but it doesn't work on colab. We have dreambooth training down thanks to your work, and if we could just make a Lora from a good checkpoint that would be awesome!