Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Very cool. Will give this idea a spin soon. I'm a bit of a scientist myself too :)

Here's something interesting I did few days ago.

- Generated images using mixture of different styles of prompts with SDXL Base Model ( using Diffusers )

- Trained a LoRA with them

- Generated again with this LoRA + Prompts used to generate the training set.

Ended up with results with enhanced effects - glitchier, weirder, high def.

Results => https://imgur.com/gallery/vUobKPK

I’m gonna train another LoRA with these generations and repeat the process obviously!

This is a pretty neat way to bypass the 77 token limit in Diffusers and develop tons of more styles now that I think about it.

You can play around with the LoRA at https://replicate.com/galleri5/nammeh ( GitHub account needed )

Will publish it to CivitAI soon.



Consider applying for YC's Winter 2026 batch! Applications are open till Nov 10

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: