Daily papers

Thursday, 19 November 2026 · 6 min read

Jeveration generates images and dense predictions in one fast forward pass

Jeveration uses a frozen vision-language model with shared random states to generate 512×512 images and dense predictions in a single forward pass, achieving up to 122× speedup over prior unified models while maintaining competitive quality.

Based on the paper Jeveration: System One Models Now Generate by Yuhe Liu, Qi Li, Shihua Zhang, Xinchao Wang

Challenges in fast, coherent image generation with unified models

Current image generation methods mainly rely on autoregressive or diffusion models that generate images sequentially, requiring many model evaluations and resulting in high latency.

System One models like Jev produce decisions in parallel without iterative reasoning, offering fast inference but have not been shown to generate full images coherently in one pass.

Generating a whole image in parallel is difficult because a single prompt admits many valid images, and independent predictions at each image code position can lead to incoherent images where parts do not fit together.

Extending such parallel generation to multiple tasks like text-to-image, image-to-image, and dense prediction requires a unified architecture that can handle different inputs and outputs efficiently.

Sign in to read the whole paper

You have read the opening. The rest is free with a Bezen account, and no card is needed.

In the rest of this paper

  • How the method works, step by step, with a flow chart
  • The key formula from the paper, explained symbol by symbol
  • Three figures from the paper, and how to read them
  • How it was tested, what was found, and where it falls short
Jeveration generates images and dense predictions in one fast forward pass · Bezen