Initializing dashboard...
FLUX 3 is Black Forest Labs' new frontier model — the first to jointly learn from images, video, and audio within a single unified architecture. Rather than modeling each modality in isolation, FLUX 3 builds a shared representation of the physical world: how objects move, how physics constrains motion, and how events produce sound. The result is a model that generates highly realistic, coherent video with native audio — up to 20 seconds in a single pass.
Key breakthroughs that make FLUX 3 a new class of generative model
FLUX 3 is trained jointly across video, images, and audio using the Self-Flow architecture. Each modality reinforces the others: sound constrains motion, motion constrains appearance, and language ties everything to intent. This cross-modal learning produces a world model — not just a generator — capable of physically plausible, temporally coherent output that isolated models cannot match.
Every video generated by FLUX 3 comes with natively synthesized audio. The model learns the causal relationship between visual events and their acoustic consequences — impacts sound like impacts, water sounds like water, and dialogue syncs with lip motion. No post-processing or separate audio model is required.
FLUX 3 can produce highly diverse, coherent video clips up to 20 seconds long from a single inference pass. Capabilities include text-to-video, image-to-video animation, video-to-video style transfer, keyframe-controlled transitions, and generative video-audio continuation — all with multilingual support and a broad range of visual styles from cinematic to candid.
Everything you need to know about FLUX 3
FLUX 3 is Black Forest Labs' new multimodal foundation model, announced in July 2026. It jointly learns from images, video, and audio to build a unified world representation. Unlike previous FLUX models focused on image generation, FLUX 3 is a video-first model with native audio output and strong physical understanding.
FLUX 3 is built on the Self-Flow architecture, which aligns multimodal generation and understanding within a single backbone. This means the model doesn't just generate video — it learns how the world works across sight and sound simultaneously. The result is superior temporal coherence, physics-aware motion, and native audio that other models produce separately via post-processing.
Yes. Audio is a native output of every FLUX 3 video generation. The model has learned the relationship between physical events and sounds during training, so generated audio matches the visual content without needing a separate TTS or audio generation step.
FLUX 3 can generate videos up to 20 seconds in a single pass. For longer productions, individual clips can be chained together using FLUX 3's agentic multi-shot sequencing capability, with visual references to maintain character and scene consistency across shots.
FLUX 3 supports text-to-video, image-to-video (animation from a starting frame or using images as visual references), video-to-video generation from a reference clip, keyframe-to-video controlled transitions, and generative video-audio continuation from input video and audio. It also supports a wide range of aspect ratios and visual styles.
FLUX 3 Video is available in Early Access as of July 23, 2026. Access to FLUX 3 Image and the open-weight FLUX 3 Dev will follow in subsequent weeks. Our platform will integrate FLUX 3 as soon as the API is publicly available.
FLUX 3 supports 720p output with a wide variety of aspect ratios including 16:9, 9:16, 1:1, 4:3, 3:4, and 21:9. Higher resolutions are expected to be supported as the model exits early access.
Self-Flow is Black Forest Labs' research approach for efficiently aligning multimodal generation and understanding within the same underlying flow-matching model. It allows FLUX 3 to train jointly on video, images, and audio without dedicated per-modality decoders, enabling strong cross-modal generalization. Early results show Self-Flow outperforms standard flow matching across all modalities.
According to Black Forest Labs' early evaluations, FLUX 3 was preferred over Runway Gen-4.5 in 77% of comparisons and over Luma Ray 3.2 in 93% of comparisons. It was competitive with Kling v3 Pro (preferred in 60% of comparisons) and Seedance 2.0 (52%). These are preliminary results and improvements are ongoing during the early access phase.
Yes. Black Forest Labs has partnered with mimic robotics to develop FLUX-mimic, a video-action model that uses the FLUX 3 backbone for robot dexterous manipulation tasks. The pretrained video backbone serves as a dynamics-aware foundation that specialized action models can be efficiently finetuned from with limited task-specific data.
