news | Xingchen Wan

Feb 06, 2026	Two papers have been accepted to ICLR 2026: MASS (Optimizing Agents with Better Prompts and Topologies) and Visual Planning with Reinforcement Learning.
Oct 17, 2025	We present VISTA and Maestro, two self-improving multimodal generation agents for text-to-video and text-to-image generation, respectively. 🚨 Google just dropped the most advanced self-improving video AI ever built. It’s called VISTA, and it literally rewrites its own prompts to make every new generation better than the last. No retraining. No fine-tuning. Just pure test-time self-reflection. Here’s how it works:… pic.twitter.com/WyhX9uur9l— Louis Gleeson (@aigleeson) October 24, 2025</blockquote> </td> </tr>
May 19, 2025	We present Visual Planning, where we apply reinforcement learning post-training on pure-vision models to achieve state-of-the-art performance in visual reasoning tasks. 🚀Let’s Think Only with Images. No language and No verbal thought.🤔 Let’s think through a sequence of images💭, like how humans picture steps in their minds🎨. We propose Visual Planning, a novel reasoning paradigm that enables models to reason purely through images. pic.twitter.com/ly9JtuEC33— Yi Xu (@_yixu) May 19, 2025</blockquote> </td> </tr> </table> </div> </div>