XNews

Black Forest Labs Releases FLUX 3 Multimodal Model

Black Forest Labs Releases FLUX 3 Multimodal Model

Black Forest Labs released FLUX 3, positioning it as a 'real world model' rather than just a text-to-video or text-to-image system. It integrates image, video, and audio into a single multimodal flow architecture to learn how the world operates, with the same backbone used for content generation and embodied/action prediction. The rationale: image provides spatial structure, video adds time and physics, audio contributes causal acoustic cues, and language supplies goals and instructions; training them together yields mutual constraints closer to a world representation. Three product lines were announced: Video (early access) supports up to ~20 seconds with native audio, text/image/video reference, continuation, keyframes, multilingual dialogue, and chaining for longer films. Image (coming weeks) offers synthesis and editing with complex prompts, multilingual text, and improved style/resolution. Action (partner channel) enables native action prediction or fine-tuning the video backbone as a dynamics model, already used in FLUX-mimic for robotics and Audi scenarios. The model will be open-sourced as Flux 3 Dev; early access is available now, API access opens in coming weeks, followed by the open-source release. Features include text-to-video, image-to-video, video reference generation, video and audio extension, keyframe control, multilingual dialogue, diverse visual styles and aspect ratios, smart clip chaining, text and animation generation, and image output.

Key insights

FLUX 3's Unified Multimodal Architecture Aims to Learn World Dynamics, Not Just Generate Media

Black Forest Labs explicitly frames FLUX 3 as a 'real world model' that learns from the interplay of image, video, audio, and language. By training these modalities together, the model captures physical constraints like motion matching mass and sound matching impact. This goes beyond typical generative models and could improve performance in tasks requiring causal understanding, such as robotics and simulation.

Open-Source Release of FLUX 3 Will Democratize Advanced Multimodal Generation

The planned open-source release of Flux 3 Dev is expected to significantly impact the AI ecosystem. Previous FLUX models have been widely adopted; a full multimodal model with video, audio, and action capabilities available under open license could accelerate innovation in content creation, robotics, and research, similar to how Stable Diffusion spurred image generation.

Source posts