Skip to content
Tagged articles

pixel-space

2 articles · Page 1 of 1
Machine Heart
Machine Heart
Oct 31, 2026 · Artificial Intelligence

NVIDIA's PixelUMM: Encoder-Free Unified Image & Video Understanding in Pixel Space

NVIDIA and University of Waterloo introduce PixelUMM, an encoder-free unified multimodal model that processes images and videos directly in pixel space using patchified representations and a Mixture-of-Transformers backbone, achieving competitive performance across understanding and generation benchmarks without VAE or ViT.

Mixture-of-TransformersNVIDIAPixelUMM
0 likes · 15 min read
NVIDIA's PixelUMM: Encoder-Free Unified Image & Video Understanding in Pixel Space
Machine Heart
Machine Heart
Jun 22, 2026 · Artificial Intelligence

Why Dropping VAE and Private Data Boosts Text-to-Image Generation Performance

MiniT2I, a minimalist pixel-space text-to-image model that discards VAE, AdaLN, and private data, achieves 0.87 GenEval and 84.2 DPG-Bench scores with only 258 M parameters, demonstrating that a stripped-down architecture and public data can outperform larger, more complex systems.

AI researchMiniT2ITransformer
0 likes · 8 min read
Why Dropping VAE and Private Data Boosts Text-to-Image Generation Performance