arXiv:2512.15420v1 Announce Type: new 
Abstract: Any-to-any generation seeks to translate between arbitrary subsets of modalities, enabling flexible cross-modal synthesis. Despite recent success, existing flow-based approaches are challenged by their inefficiency, as they require large-scale datasets often with restrictive pairing constraints, incur high computational cost from modeling joint distribution, and rely on complex multi-stage training. We propose FlowBind, an efficient framework for any-to-any generation. Our approach is distinguished by its simplicity: it learns a shared latent space capturing cross-modal information, with modality-specific invertible flows bridging this latent to each modality. Both components are optimized jointly under a single flow-matching objective, and at inference the invertible flows act as encoders and decoders for direct translation across modalities. By factorizing interactions through the shared latent, FlowBind naturally leverages arbitrary subsets of modalities for training, and achieves competitive generation quality while substantially reducing data requirements and computational cost. Experiments on text, image, and audio demonstrate that FlowBind attains comparable quality while requiring up to 6x fewer parameters and training 10x faster than prior methods. The project page with code is available at https://yeonwoo378.github.io/official_flowbind.

تم تقديم FlowBind كإطار فعال للتوليد من نوع 'any-to-any'، حيث يتناول التحديات التي تواجه الأساليب القائمة على التدفقات الحالية التي تتطلب مجموعات بيانات كبيرة وتتحمل تكاليف حسابية مرتفعة. يستخدم هذا الإطار مساحة كامنة مشتركة لالتقاط المعلومات بين الأنماط، مما يحسن كل من التدفقات القابلة للعكس المحددة لكل نمط وهدف مطابقة التدفقات لترجمة مباشرة بين الأنماط.

FlowBind se ha presentado como un marco eficiente para la generación de tipo 'any-to-any', abordando los desafíos de los enfoques basados en flujos existentes que requieren grandes conjuntos de datos y generan altos costos computacionales. Este marco utiliza un espacio latente compartido para capturar información intermodal, optimizando tanto los flujos invertibles específicos de modalidad como el objetivo de emparejamiento de flujos para una traducción directa entre modalidades.

FlowBind a été introduit comme un cadre efficace pour la génération de type 'any-to-any', répondant aux défis des approches basées sur les flux existants qui nécessitent de grands ensembles de données et entraînent des coûts computationnels élevés. Ce cadre utilise un espace latent partagé pour capturer des informations intermodales, optimisant à la fois les flux inversibles spécifiques aux modalités et l'objectif d'appariement des flux pour une traduction directe entre les modalités.

FlowBind has been introduced as an efficient framework for any-to-any generation, addressing challenges in existing flow-based approaches that require large datasets and incur high computational costs. This framework utilizes a shared latent space to capture cross-modal information, optimizing both modality-specific invertible flows and the flow-matching objective for direct translation across modalities.

FlowBind: Efficient Any-to-Any Generation with Bidirectional Flows

Was this article worth reading? Share it

Magicley AI

Metaflow AI

AI Flow Chat

Summiflow

Orbit Flows

MixflowAI

Ready to build your own newsroom?