What Is FLUX 3?

Black Forest Labs announced FLUX 3 on July 23, 2026. It is a multimodal frontier model family trained jointly across images, video, and audio, with a path toward scalable action prediction. For creators, the important part is not the label. It is the set of inputs and controls that FLUX 3 Video is designed to bring into one generation workflow.

This guide separates the announced capabilities from the access that is actually available today.

FLUX 3 Video in one paragraph

FLUX 3 Video accepts text, image, and video inputs. It can generate clips up to 20 seconds long with native audio, continue an existing video, create transitions between keyframes, synchronize multilingual dialogue with lip movement, render typography, and connect multiple clips.

Those are announced capabilities, not results from an independent AI Flux3 benchmark. Public API access is not yet available for us to test them directly.

Text, image, and video can all be inputs

The three input types support different creative starting points:

• Text input describes a shot from scratch.
• Image input supplies a visual reference or starting frame.
• Video input supplies existing motion that can be continued.

This matters because a production brief is rarely just a sentence. A creator may already have a product image, storyboard frame, or opening clip. FLUX 3 Video is designed to accept those assets without reducing the whole brief to text.

Native audio is part of the generation

The announced maximum is a 20-second video with native audio. “Native” means sound is part of the generated result rather than a separate track that the user must create afterward.

That opens a more useful prompt structure: one section for the image and motion, another for dialogue, ambience, and specific sound events. It also makes the announced multilingual lip-sync support especially relevant. Until public access arrives, however, timing accuracy and language quality remain untested by AI Flux3.

Continuation, keyframes, and multiple clips

FLUX 3 Video is intended to support three controls that matter when a single prompt is not enough:

1. Video continuation extends an existing clip.
2. Keyframe transitions describe motion between selected visual states.
3. Multi-clip workflows connect more than one generated clip.

Together, these features suggest a workflow built around shots and transitions instead of one long request. Their practical limits are still unknown because the model is currently restricted to initial API partners.

Typography and multilingual dialogue

Black Forest Labs also lists text rendering and multilingual dialogue lip sync among FLUX 3 Video’s capabilities. Both are valuable claims for product videos, explainers, title cards, and dialogue-led scenes.

They are also easy to overstate. AI Flux3 will treat them as claims to test when access opens, not as guaranteed outcomes today. Legibility, spoken-language coverage, and consistency across a full 20-second clip all require real outputs before they can be judged.

The robotics direction

The FLUX 3 model family is described as extending toward scalable action prediction, a direction relevant to robotics as well as media generation. The launch discussion also refers to this robotics-oriented work as FLUX-mimic. Our current fact baseline does not include implementation details, so it would be premature to infer specific robotics capabilities from the name.

Image and Dev releases

FLUX 3 Image is expected within weeks of the July 23 announcement. An open FLUX 3 Dev edition is planned for later in 2026.

“Planned” is the important word. No public release date or local hardware requirement is part of the information available to us, so claims about speed, memory use, fine-tuning, or local deployment would be speculation.

Can you use FLUX 3 Video now?

Only initial partners have early API access. As of July 24, 2026, neither fal.ai nor Replicate offers a FLUX 3 endpoint. AI Flux3 therefore does not claim to generate with FLUX 3 today.

The Studio is being prepared so FLUX 3 can be added when a third-party API becomes available. Until then, the FLUX 3 Early Access option is a waitlist, not a disguised connection to another model.

That distinction is the simplest way to understand the current release: the model has been announced, its capabilities are documented, and broad API access has not arrived yet.

FLUX is a trademark of Black Forest Labs Inc. AI Flux3 is an independent project and is not affiliated with Black Forest Labs.