
Radamés Ajna
2.5K posts

Radamés Ajna
@radamar
ML Engineering @adobe prev: @huggingface



My favorite editing model, FLUX.2 [klein] 9B, just got 2x faster: Meet FLUX.2 [klein] 9B-KV 😍💨 > Using KV-Cache Optimization to reduce computation & speed up inference by up to 2.5 times for multi-reference editing love how well it edits "around" the bullets





FLUX.2 is out! 🚨 FLUX.2-dev bing state-of-the-art to open weight, with a chonky 32B parameter model, however, it can run on low-end cards with quantization and the new remote text encoder Read more on the 🧨 diffusers welcomes FLUX-2 blog huggingface.co/blog/flux-2


i sit in parks…


This is how giraffes make their necks longer

Testing nano-banana with recursive editing from simple prompts "Show this scene one second in the future"


Another gpt-image feedback loop: "Generate the same photo 5 seconds into the future." Starting with distracted boyfriend meme.


Had a great conversation with @l2k on Gradient Dissent about what's actually happening in post-training right now. We dug into why LMSYS is setting the industry back - researchers are literally being forced to make models hallucinate more to climb the leaderboard...





