Yong-Xin Wang

High-fidelity image synthesis does not require paired text from day one: pre-training visual priors on uncaptioned images before multimodal alignment beats conventional joint training pipelines to establish a new open-source DiT benchmark.

MLLMs struggle with visual prompts, showing a 17.8-point accuracy drop when questions are embedded in images, revealing a critical semantic gap in multimodal reasoning.

Continuous latent visual representations can drastically improve multimodal reasoning efficiency and performance, outpacing traditional discrete output methods.