search

ShivamShrirao / diffusers

🤗 Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch
https://huggingface.co/docs/diffusers
Apache License 2.0
1.89k stars 505 forks source link

Support for SD 2.1 in dreambooth (ShivamShrirao) #175

Open slives-lab opened 1 year ago

slives-lab commented 1 year ago

Support for SD 2.1 in dreambooth (ShivamShrirao) Is there an updated SD v2.1 model for dreambooth?

https://github.com/ShivamShrirao/diffusers/tree/main/examples/dreambooth

chris-aeviator commented 1 year ago

I was able to successfully use this repo (and the example/dreambooth scripts) with SD2.1 512 base. Just the AUTOMATIC1111 is a bit picky with the resulting models and needs an additional yaml config file

NeoAnthropocene commented 1 year ago

@chris-aeviator Hey, I wonder if those YAML files do the work in your case? Or is there any other YAML file sources that you are using with your fine-tuned SD v2.x models?

https://github.com/Stability-AI/stablediffusion/tree/main/configs/stable-diffusion

sebaxakerhtc commented 1 year ago

I have success with 2.1 768, but always have unusable result with 2.1 512(base) model

NeoAnthropocene commented 1 year ago

I have success with 2.1 768, but always have unusable result with 2.1 512(base) model

Thanks for the feedback! I wonder if there is an update on the repo, because whatever I tried; I can't get reasonable results with the SD 2.1 -768.

If you don't mind, would you please share your training settings with us?

sebaxakerhtc commented 1 year ago

would you please share your training settings with us?

Ok. Give me a half an hour - i will repeat the steps to train 2.1 768 model If matters - i use Automatic1111 with dreambooth extension based on this repo

Upd: So... It works again. Quadro RTX A4000 Scheduler: euler-ancestral

Screenshots of settings:

I have these files

v2-1_768-ema-pruned.ckpt
v2-1_512-ema-pruned.ckpt