SenseNova Unveils U1.5 Lite

SenseNova has unveiled the U1.5-Lite-Preview image generation and editing model. On official benchmarks, Qwen-Image-Bench improved from 47.14 to 55.20 points, ImgEdit-Bench from 3.90 to 4.37 points, and GEdit-Bench-en from 7.47 to 8.17 points.

Key updates include the following.

• Native 4K generation and improved texture, material, and lighting representation
• Enhanced Chinese and English text rendering in dense layouts such as posters and infographics
• Support for long, layered prompts including timelines, colors, composition, and prohibited elements
• Local editing that modifies only specific regions while minimizing changes to the rest of the image
• Support for multi-reference composition, style transfer, and iterative editing workflows that continue without regeneration

However, as this is still a preview stage, short prompt understanding, small text rendering, facial detail, and aesthetic consistency remain limitations. Common-sense understanding of complex scenes and editing stability also need further improvement.

The code and model can be found on GitHub and Hugging Face.