With image generation models, it’s hard to describe what you want with only a text prompt. Collage Diffusion makes this a lot easier—drag and resize layers like in Photoshop, and get a well-harmonized image as a result. Demo at collagediffusion.stanford.edu, code is open source! 1/5

Excited about models that are sub-quadratic in sequence length and model dimension? Our Monarch Mixer paper is now on arXiv -- and super excited to present it as an oral at #NeurIPS2023! Let's dive in to what's new with the paper and the new goodies from this release: Monarch

Excited to be sharing this at SIGGRAPH Asia at the end of the year! Check out more at sxzhang25.github.io/publications/m…

Excited to announce ARES: An Automatic Evaluation Framework for Retrieval-Augmented Generation (RAG)! ARES allows you to evaluate your RAG approaches with minimal human annotations while providing statistical guarantees for LLM-based scoring.

My TED Talk from October 2023 is live! ted.com/talks/joon_sun… I talked about generative agents, and what their power to simulate ourselves and our societies will mean for us in the future. Here is the message I ended with: 🧵

(1/7) Happy mother’s day! We think what the mothers of America really want is a Flash Attention implementation that’s just 100 lines of code and 30% faster, and we’re happy to provide. We're excited to introduce ThunderKittens (TK), a simple DSL embedded within CUDA that makes

Really excited to share "Iterative Motion Editing with Natural Language", which will be presented at #SIGGRAPH2024 (1/3) Website: purvigoel.github.io/iterative-moti… Paper: arxiv.org/abs/2312.11538 Video: youtube.com/watch?v=ySSGd5…

NVIDIAGameDev Bots obviously not human? Eat GPUs for breakfast? Come to our talk on Thursday to find out how a small ML transformer can generate human-like movement for two teams of CSGO bots in 0.5ms per frame (amortized) on one CPU core! Paper has details davidbdurst.com/mlmove/

I am presenting Block and Detail: Scaffolding Sketch-to-Image Generation at ACM UIST today, 3-4PM EST in the New Vizualizations paper session! Paper: arxiv.org/abs/2402.18116 Website: miatang13.github.io/BlockAndDetail/ Work with Lu Yuan miatang Maneesh Agrawala Kayvon Fatahalian

Some thoughts on the difficulty of animation timing, particularly in the context of learned motion inbetweening. medium.com/@goel.purvi/th…

What is Halfmoon? 👀

So thrilled to share “Generative Motion Infilling From Imprecisely Timed Keyframes”, to be presented at #Eurographics2025 (1/4) Website: purvigoel.github.io/imprecise-timi… Paper: arxiv.org/abs/2503.01016 Video: youtube.com/watch?v=izbbt1…

Excited to come out of stealth at Reve! Today's text-to-image/video models, in contrast to LLMs, lack logic. Images seem plausible initially but fall apart under scrutiny: painting techniques don't match, props don't carry meaning, and compositions lack intention. (1/4)

Tabracadabra 🎉 is a system that brings tab-to-autocomplete literally *anywhere* there’s a textbox. Instead of relying on a single codebase for context, Tabracadabra uses a General User Model to autocomplete with everything you see or do on your computer.

Excited to present Axes-and-Tags this week at #ICCV2025! We put a lot of thought into enabling people to use AI to idea the in a productive way. Check out the paper or live demo below!