Self-Supervised Flow Matching for Scalable Multi-Modal Synthesis

Strong semantic representations improve the convergence and generation quality of diffusion and flow models. Existing approaches largely rely on external models, which require separate training, operate on misaligned objectives, and exhibit unexpected scaling behavior. We argue that this dependence arises from the model's training objective, which poses a denoising task with little incentive to learn semantic representations. We introduce Self-Flow: a self-supervised flow matching paradigm that integrates representation learning within the generative framework. Our key mechanism, Dual-Timestep Scheduling, applies heterogeneous noise levels across tokens, creating an information asymmetry that forces the model to infer missing information from corrupted inputs. This drives learning strong representations alongside generative capabilities without external supervision. Our method generalizes across modalities and enables multi-modal training while following expected scaling laws, achieving superior image, video, and audio generation.

Despite the dominance of diffusion and flow models in visual and multi-modal generation, they remain paradoxically dependent on external engines for global semantic understanding. This reliance suggests a fundamental limitation: standard local denoising objectives provide little incentive for models to develop a global structural understanding. To address this, we introduce Self-Flow, a self-supervised framework that unifies representation learning and generative modeling within a single pipeline. Our framework works by creating an information asymmetry across tokens, forcing the model to infer corrupted inputs from cleaner surrounding context. This process naturally drives the emergence of strong internal representations alongside generative capabilities. Because Self-Flow operates without external models, it generalizes seamlessly across diverse modalities including images, video, and audio, and follows expected scaling laws, offering a path toward world models that are both perceptually grounded and semantically rich.