OpenSenseNova/SenseNova-U1

SenseNova-U series: Native Unified Paradigm with NEO-unify from the First Principles

What it solves

SenseNova-U1 is designed to move beyond simple modality integration (using adapters to connect separate models) toward true unification. It solves the problem of fragmented multimodal AI by creating a single, monolithic architecture that can natively handle visual understanding, reasoning, and generation within one model.

How it works

The project utilizes a novel architecture called NEO-unify, which eliminates the need for separate Visual Encoders (VE) and Variational Auto-Encoders (VAE). Instead, it treats pixel and word information as a unified compound, allowing the model to reason across modalities using native Mixture-of-Experts (MoTs). This end-to-end approach enables the model to perform tasks like text-to-image generation, image editing, and visual question answering (VQA) using the same underlying weights.

Who it’s for

This project is for developers and researchers working with multimodal AI, specifically those needing high-fidelity image generation, complex layout design (like infographics), and interleaved image-text content creation.

Highlights

  • Native Unification: Combines understanding and generation in one model rather than using separate adapters.
  • Interleaved Generation: Capable of producing coherent text and images in a single flow for guides or diaries.
  • High-Density Rendering: Strong performance in creating structured layouts for posters, resumes, and infographics.
  • Advanced Image Editing: Supports local, text-based, and multi-reference edits while preserving subject identity.
  • 4K Generation: Native support for high-resolution 4K image output.
  • Efficient Inference: Offers GGUF quantized checkpoints and LoRA-distilled versions for faster performance on lower-VRAM hardware.

Related

  • Project
    OpenSenseNova/SenseNova-Vision

    SenseNova-Vision is a unified multimodal model that treats diverse computer vision tasks as a generation problem, enabling a single model to perform detection, segmentation, and geometric prediction via text and image outputs.

  • Project
    Ma-Zhuang/OmniNWM

    OmniNWM is a research codebase that builds a panoramic, multi‑modal world model for autonomous‑driving simulation. It jointly generates RGB, semantic, depth, and 3‑D occupancy maps from a vehicle trajectory, uses normalized Plücker ray‑maps for precise control, and provides occupancy‑based dense rewards for closed‑loop policy evaluation. The repo includes installation steps (including a manual patch to `transformers`), data preparation instructions for nuScenes, pretrained checkpoint download links, and scripts for inference, out‑of‑distribution testing, and staged training.

  • Dispatch
    Qwen 3.5‑397B‑A17B release: hybrid linear‑attention MoE model with 1 M token context and state‑of‑the‑art multimodal performance

    Qwen 3.5‑397B‑A17B is a 397 billion‑parameter multimodal model that activates only 17 billion parameters per token, delivering state‑of‑the‑art performance on language, coding, reasoning and vision tasks while being up to 19× faster than its predecessor.

  • Dispatch
    Qwen-Image-2.1 Release: 7B Unified Text-to-Image and Editing Model with Native Transparency

    Qwen-Image-2.1 is a 7B-parameter open‑source model that unifies text‑to‑image generation and editing, adds native RGBA support, and delivers faster inference through mixed‑granularity attention.

  • Project
    boogu-project/Boogu-Image

    An open-source unified image generation and editing model family that provides high-quality text-to-image generation and precise image editing with strong bilingual text rendering.