Exploring MLLM-Diffusion Information Transfer with MetaCanvas

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Lin, Han, Pan, Xichen, Huang, Ziqi, Hou, Ji, Wang, Jialiang, Chen, Weifeng, He, Zecheng, Juefei-Xu, Felix, Sun, Junzhe, Fan, Zhipeng, Thabet, Ali, Bansal, Mohit, Wang, Chu
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911315787776000
author Lin, Han
Pan, Xichen
Huang, Ziqi
Hou, Ji
Wang, Jialiang
Chen, Weifeng
He, Zecheng
Juefei-Xu, Felix
Sun, Junzhe
Fan, Zhipeng
Thabet, Ali
Bansal, Mohit
Wang, Chu
author_facet Lin, Han
Pan, Xichen
Huang, Ziqi
Hou, Ji
Wang, Jialiang
Chen, Weifeng
He, Zecheng
Juefei-Xu, Felix
Sun, Junzhe
Fan, Zhipeng
Thabet, Ali
Bansal, Mohit
Wang, Chu
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2512_11464
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Exploring MLLM-Diffusion Information Transfer with MetaCanvas
Lin, Han
Pan, Xichen
Huang, Ziqi
Hou, Ji
Wang, Jialiang
Chen, Weifeng
He, Zecheng
Juefei-Xu, Felix
Sun, Junzhe
Fan, Zhipeng
Thabet, Ali
Bansal, Mohit
Wang, Chu
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
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.
title Exploring MLLM-Diffusion Information Transfer with MetaCanvas
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2512.11464