Spaces:

multimodalart

/

dreambooth-training

like 498

Running

Hello.
Generated 2.1-768 model.ckpts always have the same SHA256 no matter what pictures were used as input. They dont include the changes/new token.

Using the diffusers-to-ckpt script helps as a workaround.

SD 1.5 model.ckpt is always working fine.

PS: For the 2.1-768 models, i use the 2.1-v-inference.yaml delivered with easydiffusion.
I simple copy it into the models folder and rename it according to the cpkt I am using.

For 1.5 models i dont use a yaml file.

Am I doing it right?