Sayak Paul

Hi there 👋 I am Sayak Paul (সায়ক পাল). I work on 🧨 diffusion models at Hugging Face. Know more about me from here. I maintain a Google Doc answering some FAQs at length. You can check it out here.

My external articles and other publishing engagements are listed here. Decks from my speaking engagements are listed here. A detailed account of the things that are not directly available from the top navbar (interviews, talks etc.) can be found here.

The structure of this website is inspired by Omar’s site.

News

• From Statics to Dynamics: Physics-Aware Image Editing with Latent Transition Priors with Zhao et al. was accepted to ICML’26 🎉
• New work – Faster Diffusion on Blackwell: MXFP8 and NVFP4 with Diffusers and TorchAO.
• Gave a talk on “torch.compile and Diffusers: A Hands-On Guide to Peak Performance” at PyTorch Conference EU 2026 (Paris). Slides are here. Recording is here.
• Organizing a tutorial on Post-Training Diffusion Models: Enhancing Capabilities, Control, and Alignment at ECCV’26. Check out the details here.
• Published a kernel (and its development journey) for optimizing LTX2.3: sayakpaul/ltx2-simple-optims.

To know more about my projects, please refer to my GitHub profile. For an up-to-date list of my publications, refer to my Google Scholar page.

Apart from the blogs here, I try to contribute to other platforms in the form of writing. Please refer here for more details.