at://
/
app.bsky.feed.post
/
3mnekb3ow7226
sign in
All
4
Record
2
Post
1
PostEmbed
1
Post
by @danabra.mov
PostEmbed
by @danabra.mov
Record
by @jimpick.com
Record
by @atsui.org
+ new component
Post
There's nothing more satisfying than watching the right noise do its magic with diffusion models! A few interesting takeaways I had from this work š§µ
14d
Shyamgopal Karthik
#CVPR2026 paper: It's Never Too Late: Noise Optimization for Collapse Recovery in Trained Diffusion Models Text-to-image models often collapse to near-identical samples. Our fix: optimize the noise. Start from pink š©·, not white noise. š akoepke.github.io/divgen/index... 1/6
14d
Video
A. Sophia Koepke
