Twitxy Saved

Cong Wei

Verified account This account is verified. Learn more

@CongWei1230

@xAI: @grok multimodal | CS PhD @UWaterloo | MS & B.CS @UofT | Prev Intern @nvidia @AIatMeta @VectorInst
congwei1230.github.io Joined July 2023
491 Following 1.2K Followers 235 Posts

Latest posts 5

Cong Wei Verified account @CongWei1230
Congrats on the releas @an_epsilon0 ! Great to see this pixel space video gen direction gaining momentum alongside PixelUMM!
Shuo Chen Verified account @an_epsilon0 4d
“Real-time” video models are finally real-time!

Introducing PixVid: lifting a latent video model to generate one frame at a time, in pixel space. No VAE, no temporal compression.

The payoff? Fast high-res generation, precise editing, and ultra-low control-to-screen latency!
View quoted post on X
Cong Wei Verified account @CongWei1230
Thanks for sharing! We open-sourced code and models.

🤗Huggingface Paper: huggingface.co/papers/2609.38…
💻 Code: github.com/nv-tlabs/Pixel…
🤗 Model: huggingface.co/nvidia/PixelUMM
huggingface.co Paper page - PixelUMM: Encoder-Free Unified Image and Video Understanding and Generation Join the discussion on this paper page
Robots Digest 🤖 Verified account @robotsdigest Oct 4
NVIDIA researchers just released PixelUMM, a unified multimodal model that fundamentally changes how models process visual data.
Groundbreaking highlights:
Entirely removes VAEs and ViTs from the architecture
Operates directly in raw pixel space for both image and video tasks
Handles both visual understanding and generation within a single encoder-free model
View quoted post on X