🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

SANA-Sprint One-Step Diffusion with Continuous-Time Consistency Distillation

The best few-step sampling model across the speed-memory frontier? 😱 Introducing SANA-Sprint in collaboration with the great SANA team! Beyond the results, perhaps more importantly, the work is about the recipe of SANA-Sprint. Code & model will be open ❤️ Let's go ⬇️

👀SANA-1.5 4.8B checkpoint is released!🎉🥳 Much better than SANA-1.0 1.6B. We also release more training code e.g. FSDP support / webdataset loader / multi-scale image sampler. One-click start training and feel free to try! github.com/NVlabs/Sana?ta… We will release Sprint soon😎

I will attend GTC 2025 at NVIDIA Santa Clara from March 17 to March 21, welcome coffee chat ! ☕️

🚀 SANA 1.5 Update: Inference Scaling Now Open-Source! 🎉 📈 Breakthrough on GenEval benchmark: • SANA 1.5 + Inference Scaling: 0.81 → 0.96 (!!) 🎯 • SD 1.5 + Inference Scaling: 0.42 → 0.87 ⬆️ 💫 The secret sauce: 1. Generate n candidates 🎨 2. Pick top k with NVILA

It is interesting to see the idea of NoPE (no Position Encoding) used in Llama-4! By the way, our work SANA was also inspired by Meta's original NoPE paper and is the first work to remove positional encoding in diffusion transformers (DiT) 😊

I tried WriteHere's Deep Research report feature, and the results were truly impressive! 🤩 Congrats Yimeng Chen !

I'll be attending ICLR from April 24th to 28th in Singapore—hope to see you there! I'd be delighted to grab coffee and chat about topics such as #GenerativeAI and #EfficientAI

Deep compression in vae is hard, but this paper beautifully explains how to achieve this

I gave an oral presentation on our image generation foundation model, SANA, at ICLR 2025 Singapore. Welcome to follow! In addition, we are developing SANA Video model, stay tuned! ICLR oral replay: youtube.com/watch?v=rrKFyY…

I know you have always secretly craved a cool distillation script that actually gets results. That time has come 🤯 In collaboration w/ Junsong_Chen & Shuchen Xue, we present a Diffusers-compatible training script for SANA Sprint 🏃 Links ⬇️

🚀 Fast-dLLM: 27.6× Faster Diffusion LLMs with KV Cache & Parallel Decoding 💥 Key Features🌟 - Block-Wise KV Cache Reuses 90%+ attention activations via bidirectional caching (prefix/suffix), enabling 8.1×–27.6× throughput gains with <2% accuracy loss 🔄 -

🚀The code for Fast-dLLM is now open-source! 💥 Fast-dLLM achieves a 27.6× end-to-end speedup on 1024-token sequences with less than 2% accuracy drop. Check out the code here: github.com/NVlabs/Fast-dL…

I will be attending CVPR from June 11-15. Welcome to meet me for coffee! I will also share our team's research on Efficient Image Generation (including SANA, DC-AE, VILA-U, HART) in two workshops on Jun 12. 我将在 6月11-15 参加 CVPR, 欢迎约咖啡！👀 coop-intelligence.github.io

Thank you Felix for hosting this excellet workshop!