SenseTime Open-Sources SenseNova U1 Lite Multimodal Model
SenseTime has open-sourced the SenseNova U1 Lite Series, a multimodal model built on its NEO-Unify architecture. The release describes compact 8B and A3B versions that combine multimodal understanding and generation, while generating images natively in a single system without a separate visual encoder or VAE.
SenseTime said the model supports interleaved text-and-image generation in one flow and is designed for dense visual communication tasks including knowledge illustrations, posters, presentations and comics. The company described the open-source models as aiming for commercial-grade performance and cost efficiency.
From the sources (2 posts)
@testingcatalogSenseTime open-sourced SenseNova-U1, a multimodal image generation model built on NEO-Unify! This architecture drops the visual encoder and VAE entirely. It generates images natively as one system that can handle understanding, reasoning,
@adinayakupSenseTime is back to open source👀 SenseNova U1 🔥 a unified multimodal model (8B/A3B, Apache 2.0) with no visual encoder/ VAE, just end-to-end pixel word modeling