Mixture of States: Routing Token-Level Dynamics for Multimodal Generation

TL;DR

MoS introduces a token-wise routing mechanism for multimodal diffusion models, enabling efficient and flexible fusion of modalities that achieves state-of-the-art results with fewer parameters.

Contribution

The paper presents MoS, a novel token-level fusion paradigm with a learnable router for multimodal diffusion models, improving efficiency and performance.

Findings

Achieves state-of-the-art results in text-to-image generation and editing.

Models with 3B-5B parameters match or surpass larger counterparts.

Efficient routing with minimal computational overhead.

Abstract

We introduce MoS (Mixture of States), a novel fusion paradigm for multimodal diffusion models that merges modalities using flexible, state-based interactions. The core of MoS is a learnable, token-wise router that creates denoising timestep- and input-dependent interactions between modalities' hidden states, precisely aligning token-level features with the diffusion trajectory. This router sparsely selects the top-$k$ hidden states and is trained with an $\epsilon$-greedy strategy, efficiently selecting contextual features with minimal learnable parameters and negligible computational overhead. We validate our design with text-to-image generation (MoS-Image) and editing (MoS-Editing), which achieve state-of-the-art results. With only 3B to 5B parameters, our models match or surpass counterparts up to $4\times$ larger. These findings establish MoS as a flexible and compute-efficient paradigm for scaling multimodal diffusion models.