Skip to content

Back to the catalog

SenseNova-U1.5: Towards Native Unified Visual Intelligence
arxiv.orgpaper

SenseNova-U1.5: Towards Native Unified Visual Intelligence

SenseNova-U1.5 is an 8B-parameter Mixture-of-Transformers native unified multimodal model that understands, reasons about, and generates visual content within an encoder-free and VAE-free architecture. It uses spatially coherent patch reconstruction, native 4K resolution training, and post-training with specialized experts (aesthetics, bilingual text rendering, infographics, image editing) consolidated via multi-expert on-policy distillation. The paper reports advances in image fidelity, text rendering, complex composition, and multi-reference editing, and plans to open-source training code including SFT, RL, and on-policy distillation.

TL;DR An 8B encoder-free, VAE-free unified multimodal model for visual understanding, reasoning, and generation, with multi-expert distillation for editing and text rendering.
Open original ↗

Technologies named here

  • SenseNova-U1.5
  • Mixture-of-Transformers

Techniques named here

  • Encoder-free and VAE-free architecture
  • Spatially coherent patch reconstruction
  • Structural prompt enhancement
  • Multi-expert on-policy distillation

Related in multimodal