Рет қаралды 31,441
We explain the new Offset Noise discovery which allows latent diffusion model trainers to get vastly improved results by changing a single line of code. We also compare images generated by offset noise models to pre-offset noise images and Midjourney images. This is probably the first time since the release of the v4 model in December 2022 that the stable diffusion community has achieved parity with Midjourney.
Thanks so much to Nicholas Guttenberg for the excellent research. That article should by all rights be a research paper.
======= Links =======
Offset Noise: www.crosslabs....
Cacoe's Server: / discord
Old version of the Illuminate model: huggingface.co...
An Image Made of Sin Waves: • Fourier Image Decompos...
======= Music =======
Music from freetousemusic...
‘Onion’ by LuKremBo: • (no copyright music) l...
From KZbin Audio Library:
Escapism Yung Logos