search

invoke-ai / InvokeAI

Invoke is a leading creative engine for Stable Diffusion models, empowering professionals, artists, and enthusiasts to generate and create visual media using the latest AI-driven technologies. The solution offers an industry leading WebUI, and serves as the foundation for multiple commercial products.
https://invoke-ai.github.io/InvokeAI/
Apache License 2.0
23.8k stars 2.45k forks source link

[enhancement]: SGMUniform scheduler for Hyper loras needs support! #6909

Open rollingcookies opened 2 months ago

rollingcookies commented 2 months ago

Is there an existing issue for this?

Contact Details

discord: @digital_groove

What should this feature add?

SGMUniform scheduler is needed to support Hyper loras! Please add support for this scheduler in Invoke interface.

I also opened a problem in the Huggingface Diffusers library where I described the problem to add support for SGMUniform scheduler in the library: https://github.com/huggingface/diffusers/issues/9397

And here they solved this problem and it looks like SGMUniform scheduler is now present in the Diffusers library (please check if everything is ok): https://github.com/huggingface/diffusers/issues/9416

Alternatives

No response

Additional Content

No response

Vargol commented 2 months ago

There's a crap way to get SGMUniform running on InvokeAI, as in diffusers it's simply a matter of changing the schedulers timestep spacing to trailing. So if your model is in diffusers format you can edit the schedulers config file. The value is then inherited by any scheduler that supports that setting.

Unfortunately this means editing the schedulers config file back and forth if not using Hyper all the time or creating a copy of the and editing the scheduler config of the copy so you're got a new model. You can use symlinks to save disk space. it would be nice if the dev's added an better way of doing it.

Note some models already have the config set correctly if they've been imported from diffusers versions e.g. RunDiffusion/Juggernaut-XL-v9 so it's worth checking first.