English

Exploring MLLM-Diffusion Information Transfer with MetaCanvas

Computer Vision and Pattern Recognition 2025-12-15 v1 Artificial Intelligence Machine Learning

Abstract

Multimodal learning has rapidly advanced visual understanding, largely via multimodal large language models (MLLMs) that use powerful LLMs as cognitive cores. In visual generation, however, these powerful core models are typically reduced to global text encoders for diffusion models, leaving most of their reasoning and planning ability unused. This creates a gap: current multimodal LLMs can parse complex layouts, attributes, and knowledge-intensive scenes, yet struggle to generate images or videos with equally precise and structured control. We propose MetaCanvas, a lightweight framework that lets MLLMs reason and plan directly in spatial and spatiotemporal latent spaces and interface tightly with diffusion generators. We empirically implement MetaCanvas on three different diffusion backbones and evaluate it across six tasks, including text-to-image generation, text/image-to-video generation, image/video editing, and in-context video generation, each requiring precise layouts, robust attribute binding, and reasoning-intensive control. MetaCanvas consistently outperforms global-conditioning baselines, suggesting that treating MLLMs as latent-space planners is a promising direction for narrowing the gap between multimodal understanding and generation.

Keywords

Cite

@article{arxiv.2512.11464,
  title  = {Exploring MLLM-Diffusion Information Transfer with MetaCanvas},
  author = {Han Lin and Xichen Pan and Ziqi Huang and Ji Hou and Jialiang Wang and Weifeng Chen and Zecheng He and Felix Juefei-Xu and Junzhe Sun and Zhipeng Fan and Ali Thabet and Mohit Bansal and Chu Wang},
  journal= {arXiv preprint arXiv:2512.11464},
  year   = {2025}
}

Comments

Project page: https://metacanvas.github.io

R2 v1 2026-07-01T08:22:05.765Z