Zihao Liu

Existing text-to-image benchmarks miss the mark on real-world artistic creation, but Qwen-Image-Bench finally provides a creator-centric evaluation that reliably distinguishes state-of-the-art models.

Achieve more human-like and consistent user simulation in Chinese across multiple domains by iteratively refining user profiles with real dialogue data and aligning behavior with a rubric-guided reward model.

Smooth, stable robotic manipulation of dynamic space targets is now possible thanks to a data-driven approach that exploits inter-frame correlations.