Skip to content
Agientry
← Back to Videos

PixelUMM: Pixel-Space Image and Video Model

Loading video
Loading video

PixelUMM removes vision encoders and VAEs, using 16×16 image patches and four-frame video tubelets in a Qwen3-8B backbone for unified understanding and generation.

NVIDIAPixelUMMMultimodalEncoder-FreeImageGenerationVideoGenerationVideo Generation
Category: research
Author: @CongWei1230
Date: 2026-10-01T00:00:00
Duration: 28.966s
Source: https://x.com/CongWei1230/status/2105808995910816159
Reference: https://arxiv.org/abs/2609.38597

Related Videos

NVIDIA PixelUMM: encoder-free unified understanding and generation in raw pixel space
0:30

NVIDIA PixelUMM: encoder-free unified understanding and generation in raw pixel space

@robotsdigest · 392 views · 2026-10-04
NVIDIAPixelUMM
A Multi-Character Audio-Video Generation Pipeline
6:43

A Multi-Character Audio-Video Generation Pipeline

@nakazakifam · 88 views · 2026-10-31
Video GenerationMultimodal
Freefield: 117 Open Generative Apps in ComfyUI
1:56

Freefield: 117 Open Generative Apps in ComfyUI

@aiaicreate · 104 views · 2026-10-28
ComfyUIHiggsfield
RoboRender: Turning Simulated Trajectories into Photorealistic Video for Policy Training
2:02

RoboRender: Turning Simulated Trajectories into Photorealistic Video for Policy Training

@RavenHuang4 · 750 views · 2026-10-08
Sim-to-RealVideo Generation
Mistral Large 4: a 1T-parameter natively multimodal model
0:17

Mistral Large 4: a 1T-parameter natively multimodal model

@MistralAI · 513 views · 2026-10-06
MistralLMM
LongCat-Video: Meituan open-sources a 13.6B video model
0:30

LongCat-Video: Meituan open-sources a 13.6B video model

@denziideng · 405 views · 2026-10-06
VideoGenerationText-to-Video

As an Amazon Associate, we earn from qualifying purchases.