RahimEntezari

I am proud member of Stable Diffusion 3, and we won best paper award at #ICML2024

@bneyshabur Congratulation Behnam! Wish you and the team luck and Looking forward to hearing more!

I've left Anthropic to start something new. 🧵

This is why Iranian regime shutdown internet, to kill thousands of people. Please spread the world

I still get a lot of questions about doing a PhD in ML so I'm resurfacing a three threads I wrote about this including some fun personal stories 🧵

@DongWan06935465 Consistency was much better than I have seen with Wan, but what happened to the cheeze in the frames before last? my 5x5 was the basic version

@rahiment @shlomifruchter @GoogleDeepMind @thwiedemer @priyankjaini Just tried sora2. It seems to work.

Quite an a-maze-ing discovery: Veo 3 demonstrates emergent visual reasoning, like finding finding the path to the cheese.

@shlomifruchter @GoogleDeepMind @thwiedemer @priyankjaini Tried Wan2.5 an old Sora, both failed too in 5x5 maze. not success at all. waiting for sora2

@shlomifruchter @GoogleDeepMind @thwiedemer @priyankjaini These are actually amazing properties! I just tried the maze, 5x5 with Wan model and it totally failed, not a single solution for 100 generations x different CFGs. I wonder if Maze was part of training set though

🎬 Introducing Stable Cinemetrics, to be presented at NeurIPS 2025. We present the first taxonomy of professional controls to systematically study and control video generative models through the lens of filmmaking. Interactive webpage with paper link: stable-cinemetrics.github.io 🧵

@shlomifruchter We also included some evals on Veo3 in Table 13.

How ready are T2V for professional filmmaking? In our Nuerips paper, we showed that how far we are from making cinematic ready generations, in Events, Camera, Setup and lighting.

@hamidInventions @jampani_varun @hmrishavbandy @yizhe_song 2 vs 12B params

@jampani_varun @hmrishavbandy @rahiment @yizhe_song Looks decent but it is still costlier than Flux Schnell i guess

Introducing Stable Diffusion 3.5 Flash ⚡, a state of the art distilled image generation model that is best in class for all compute budgets. 🔗Project page: hmrishavbandy.github.io/sd35flash/ Authors: @hmrishavbandy, @rahiment, Jim Scott, Reshinth Adithyan, @yizhe_song, @jampani_varun

T1/11 Text-to-image AI is amazing, but what if you don't speak English? Most are English-centric, creating barriers for billions. We introduce NeoBabel, a new open model for visual generation that natively understands 6 languages, breaking down the walls of digital inequity. 🌍🗣️

Here is another uncut video of real-time interactions with @GoogleDeepMind 's Gemini Robotics!

CVPR2025: Are you passionate about Large Vision models? How about efficiency? - We are looking for reviewers for eLVM workshop - Also eager to read your submissions! workshop page: sites.google.com/view/elvm/home reviewer enrolment: bit.ly/3Flhjcb #CVPR2025

@DongWan06935465 Congratulations!

🚀 Excited to share our paper was accepted at #ICLR2025! 🎉 ✨Pruning models? No data, no fine-tuning! Just fold it! ✨Introducing Model Folding: a groundbreaking data-free model compression technique. 📖 Read more: openreview.net/forum?id=W2Wkp…

Our agents are working 24/7, and more registration codes are rolling out! Try out our hallucination detection agents for free now! Request your registration code here: platform.relai.ai/auth/waitlist

Hi AI world ✋ We are committed to make AI reliability accessible and achievable for everyone! First up: RELAI agents for hallucination detection Chat with popular LLMs, Verify with RELAI. Try it for free today: relai.ai #AI #LLM #ReliableAI #Hallucination

Interesting to see RELAI looking into one the most important problems of GenAI: reliability!

SD3.5 medium is out this time! Great to use with consumer hardware for finetuning and a lot of research on diffusion models!

Our work on “Rephrasing natural text data with different languages and quality levels for Large Language Model pre-training” is finally out: arxiv.org/abs/2410.20796