Exploring MLLM-Diffusion Information Transfer with MetaCanvas

TL;DR

MetaCanvas enables multimodal large language models to directly reason and plan in spatial and spatiotemporal latent spaces, significantly improving control and precision in visual generation tasks by tightly integrating with diffusion models.

Contribution

The paper introduces MetaCanvas, a lightweight framework allowing MLLMs to operate in latent spaces for better visual generation control, bridging the gap between understanding and generation.

Findings

MetaCanvas outperforms global-conditioning baselines across six tasks.

Empirical evaluation on three diffusion backbones demonstrates versatility.

Treating MLLMs as latent-space planners enhances generation precision.

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.