Dreambooth can be trained on a 10gb GPU, even on windows with WSL + Docker.

[ZeroCool22]

https://github.com/smy20011/efficient-dreambooth

Some chance to get it on the WEBUI?

@AUTOMATIC1111

[bmaltais]

Doesn't it produce a diffuser model instead of a CKPT? Webui can't use diffusers models.

[bmaltais]

Just saw someone with the handle Josh is working on converting diffusers model back to ckpt... So this might not be an issue much longer.

[ZeroCool22]

Just saw someone with the handle Josh is working on converting diffusers model back to ckpt... So this might not be an issue much longer.

OMG, that would be awesome.

Hey, you have knowledge with WSL and Docker?

[bmaltais]

It is now reality. I am able to create a dreambooth model in colab, convert it to ckpt and use it in webui!: https://gist.github.com/jachiam/8a5c0b607e38fcc585168b90c686eb05

[bmaltais]

Also, I am testing adding more than one subject to the same model. So far it worked for a man and a woman... I am now attempting to add a 2nd men and see how that goes!

[bmaltais]

So far testing does not look like a great success… but I want to test another technique this morning that “might” be interesting.