ernie-research (Jiahao Hu, Jingdong Du, Yichen Zhang et al.) · Unified Multimodal Model

HiUni: Harmonizing Understanding and Generation in Shared Visual Space

HiUni is a unified multimodal model that reuses one language-aligned ViT and dense Transformer backbone for both visual understanding and image generation, confining task-specific computation to the output side. It adds a Pixel Diffusion Head for pixel-space optimisation and a lightweight HiGate pathway that supplies fine spatial detail near the output. The authors report average understanding scores similar to the pretrained VLM, with competitive results on image generation and editing. No licence or code availability is stated on the page.

Open the source ↗

Similar Vision & Image releases

  • PixelUMM

    NVIDIA and University of Waterloo · Encoder-Free Unified Model · Week 41, 2026

  • InternLumina-U2

    Shanghai AI Laboratory · Omni-Visual Diffusion LLM · Week 36, 2026

  • MODUS

    EPFL / Apple / U. Copenhagen / CUHK / U. Geneva / Lambda · Visual Understanding · Week 31, 2026

  • Tuna-2

    Tuna AI · Visual Understanding · Week 18, 2026

  • Vision Banana

    Google DeepMind · Visual Understanding · Week 17, 2026

  • LongCat-Image

    Meituan · Text-to-Image · Week 50, 2025

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.