SenseNova U1Pro Hands-On: SenseTime's Image Model Aims at Real Delivery, Not One-Off Generations
智东西 · wechat · 2026-09-21
Zhidongxi put SenseTime's newly launched SenseNova U1Pro (now on the Xiaohuanxiong app with an enterprise API) through a multi-round review testing whether an image model can complete actual work tasks rather than just generate pretty pictures.
- Single-image creation: It produced a rap festival poster, a xianxia-style character sheet for an AI comic, and a photoreal collectible toy concept shot with convincing materials and packaging.
- Full task workflows: It generated article covers and illustrations that accurately reflected paragraph content, and produced a cohesive set of campus-recruiting materials (posters, roll-ups, brochures, job charts) with unified styling but scenario-appropriate layouts.
- Editing: Outfit/background swaps preserved the subject's identity; multi-reference fusion combined a photo, two cartoon avatars, and a Polaroid frame; localized edits (room renovation, weather change, text tweaks) touched only the intended regions without degrading the rest.
Verdict: U1Pro's edge is task-oriented generation—understanding context, organizing deliverable sets, and controllable revision—though outputs still need human verification.
More from Multimodal
- Running MiniMax H3 on an RTX 3090: Sage Attention + Triton cut 10s video gen from 18 to 10 minutes — Fun-Helicopter21 · 2026-09-22
- RTX 4090 H3 generation temps: 72°C core max, 83°C hotspot, 78°C VRAM — taurine_bitch · 2026-09-22
- SenseNova U1.5 Lite beats FLUX.2-klein-9b on multi-reference image fusion in 5-task test — Kakash1i · 2026-09-22
- One-line input to finished episode: an agent pipeline built on CREAO — socialwithaayan · 2026-09-22
- AI virtual singer YURI's SURREAL closes Shanghai show on a 100-meter screen — hq4ai · 2026-09-22
- Chromovolume: visualizing human movement by stacking time into a hypnotic 3D volume — CurieuxExplorer · 2026-09-22