sdxl fine tunes - Githubissues

matt-bornstein commented 1 year ago

This is a new tab to run a bunch of SDXL fine tunes on a single prompt.

It pulls models from https://replicate.com/collections/sdxl-fine-tunes.

The hard-ish part is deciding how to prompt each model. I hard-coded templates for a bunch of them in lib/sdxlfinetunes.js. If the app encounters a model not in that set, it uses llama2 to guess at the format from a single example. The logic for this is in pages/api/predictions/sdxlfinetunes.js.

This could be improved in various ways:

Store the templates in supabase instead of hard coded
Add more templates for the rest of the models
Pull more example prompts for each model before guessing (few shot vs single shot)
Improve llama2 prompt in some other way

I'm sure I broke a bunch of conventions etc, so happy to change anything

vercel[bot] commented 1 year ago

@matt-bornstein is attempting to deploy a commit to the Replicate Team on Vercel.

A member of the Team first needs to authorize it.

vercel[bot] commented 11 months ago

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
zoo	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Oct 19, 2023 5:58pm

replicate / zoo

sdxl fine tunes #53