Using stable horde to generate dreambooth models

[Tophness]

Google Dreambooth has come to SD but it currently requires 24gb of vram, making it only accessible to people who pay for a server. As SD itself runs on even 2gb cards now, wouldn't this be the perfect additional use for a server farm like this?

[db0]

Sure. The horde can support multiple models (check the KoboldAI horde). But I don't have the necessary AI understanding (or time) to implement the worker code for this myself. If there is an existing worker implementation in python however, I can modify it with bridge code and allow it to join the horde.

As always, PRs are welcome

[0xdevalias]

Not sure if the implementation/etc would be compatible, but here's another distributed StableDiffusion training project a friend recently linked me to:

• https://github.com/chavinlo/distributed-diffusion
  • Which seems to build on top of: https://github.com/learning-at-home/hivemind

  • Decentralized deep learning in PyTorch. Built to train models on thousands of volunteers across the world.

There's also some examples of dreambooth training from the diffusers repo:

• https://github.com/huggingface/diffusers/tree/main/examples
  • https://github.com/huggingface/diffusers/tree/main/examples/dreambooth

[tazlin]

This issue dates back to a different era of the horde. Closing as stale.