FLUX 3 Video: Multimodal Video Generation with Native Audio on Models and API | RunComfy

Black Forest Labs / FLUX 3 Video#

FLUX 3 is Black Forest Labs' next unified model, trained jointly on images, video, and audio so motion and sound reason from the same foundation. This page focuses on its video side and previews what FLUX 3 Video will feel like in production.

Note: FLUX 3 Video is coming soon. This page previews the experience while Black Forest Labs finalizes the public model and API.

FLUX 3 Video Highlights#

Why creators reach for the FLUX 3 Video experience:

• Skip the edit suite: a written brief stands in for timeline editing, dubbing, and manual cleanup.
• Direct it like a shot: in FLUX 3 Video, camera angle, mood, and pacing all come from the words you write.
• Sound in the same pass: every clip arrives with synchronized native audio, not a separate dubbing step.
• Fits real deliverables: results are shaped for ads, previz, and social placements, not just test footage.

FLUX 3 Video at a glance#

A quick read on what FLUX 3 Video can do:

• Clip length: up to 20 seconds in a single generation, roughly double the 10-second ceiling common to most video models, with room to chain shots into multi-minute sequences.
• Audio: native, synchronized sound generated together with the picture, and optional when you want silent footage.
• Inputs: text prompts plus image, video, and audio references (up to about ten image references), so it edits and remixes as readily as it generates from scratch.
• Resolution: 720p during early access, with higher resolutions expected as the model matures.
• Aspect ratios: a wide spread from vertical 9:16 to ultrawide 21:9 for social, film, and product formats.
• Styles: candid camcorder, animation, and polished cinematics from one model.
• Status: limited early access, so FLUX 3 Video is coming soon on RunComfy.

FLUX 3 Video Key Capabilities#

Black Forest Labs describes FLUX 3 Video as one multimodal model with several generation modes, and every output carries native audio. The officially confirmed tasks are:

• Text-to-video: generate a fully voiced clip up to 20 seconds from a written prompt alone.
• Image-to-video: animate from a starting frame, or use supplied images as visual references to shape the motion.
• Video-to-video: carry central elements of a source clip, the same character or object, into a new scene or context.
• Video + audio continuation: extend an input video and its soundtrack instead of starting over.
• Keyframe-to-video: define moments and let it fill the controlled transition between them.

Beyond those modes, FLUX 3 Video also delivers:

• Multilingual dialogue with mouth movement that tracks the speech.
• Agentic multi-shot chaining that links clips into sequences running for minutes while characters stay consistent.
• A broad range of styles and aspect ratios, from candid camcorder footage to animation and cinematics.
• Strong typography and animated design rendered directly in the frame.
• Grounded realism: Black Forest Labs notes FLUX 3 Video is already standout at expressive faces and at matching sounds to on-screen physical events.

Native audio and world-grounded motion in FLUX 3 Video#

The reason FLUX 3 Video stands out is that picture and sound are designed together, not stitched on afterward. In a single pass it can produce dialogue, foley, ambience, and music timed to the action: a door slams with its own impact, an alarm pulses in step with a warning light, and a character can speak in the same shot the camera moves through.

That synchronization is a by-product of how the model was trained. Black Forest Labs reports that video prediction accounted for more than 95% of FLUX 3's training compute, which forced the model to learn contact, motion, weight, and cause and effect, in other words how the physical world actually behaves. Audio is then learned as a consequence of that world model, so sounds line up with the on-screen events that cause them and speech tracks lip movement. For creators, that means motion reads as physically plausible and the soundtrack reinforces the edit instead of fighting it.

FLUX 3 Video Prompt & Reference Tips#

These tips help you get predictable output while FLUX 3 Video is previewed:

• Be specific about camera and motion (medium close-up, slow push-in, handheld vs locked-off).
• Use images for what must stay stable (face, costume, logo); use text for what should evolve (action, mood).
• Keep reference videos and audio within 2-15 seconds, and reference audio under 15 MB.
• With audio on, name the dialogue tone or ambient sound you want to hear.
• Fix the seed while iterating so changes come from the prompt, not randomness.
• Simplify the prompt if the result feels noisy or contradictory.

How FLUX 3 Video compares to other models#

Because FLUX 3 Video is still in early access, the clearest evidence comes from Black Forest Labs' own preliminary evaluations, 10-second 720p clips generated with audio. In those side-by-side tests, reviewers preferred FLUX 3 over a broad field: Luma Ray 3.2 in 93% of comparisons, Runway Gen-4.5 in 77%, Grok Imagine Video in up to 69%, Kling v3 Pro in 60%, and both Seedance 2.0 and Gemini Omni Flash in 52%. The numbers are early and expected to move, but they anchor the comparisons below in capability rather than hype.

FLUX 3 Video vs FLUX.2#

• Image-only to motion: every prior FLUX, including FLUX.2, generates stills only; FLUX 3 is the first to fold video and synchronized audio into the same foundation, so FLUX's reference handling, prompt adherence, and typography now carry into clips.
• Shared DNA: the traits that made FLUX images dependable, holding a character, product, or logo steady, are what FLUX 3 Video extends across a shot.
• New territory: motion, temporal stability, and on-beat audio are fresh problems for an image-first lab, which is exactly why the early-access window exists.

FLUX 3 Video vs Seedance 2.0#

• The data: in Black Forest Labs' early tests, FLUX 3 was preferred over Seedance 2.0 in 52% of comparisons, essentially even with a mature, proven multimodal video model.
• Capability: Seedance 2.0 mixes text, image, video, and audio in one request and returns roughly fifteen seconds of multi-shot 480p-4K video with native, synchronized multi-track audio plus edit and extend loops. FLUX 3 Video matches that multimodal ambition and, per Black Forest Labs, is particularly strong at expressive faces, matching sound to physical events, and multilingual dialogue.
• How to choose: if you must ship today, Seedance 2.0 is proven; FLUX 3 Video is the one to watch as its public release approaches.

FLUX 3 Video vs Kling, Runway, and Luma#

• The data: FLUX 3 was preferred over Kling v3 Pro in 60%, Runway Gen-4.5 in 77%, and Luma Ray 3.2 in 93% of Black Forest Labs' early comparisons.
• What drives the gap: those margins track FLUX 3 Video's stated strengths, expressive human faces, audio locked to on-screen physical events, multilingual dialogue, and consistent multi-shot chaining, the details that make a short clip read as finished rather than as a test render.

What to expect from FLUX 3 Video in early access#

FLUX 3 Video is a preview of a model still in limited early access, so it helps to set expectations:

• Strong today: long, coherent clips, dialogue-driven and dramatic scenes, and native synchronized audio all hold up well in early hands-on testing.
• Still maturing: image-to-video reference adherence can be inconsistent in early builds, and very fast, high-energy action does not yet carry the kinetic intensity of the most motion-focused rivals.
• Rolling out: resolution starts around 720p with higher resolutions expected, while pricing and open weights are still to be confirmed by Black Forest Labs.

Treat FLUX 3 Video as a look at where the model is heading: its capabilities and limits will keep shifting as it moves toward general availability.

FLUX 3 Video editions and the wider FLUX 3 AI family#

Black Forest Labs is rolling FLUX 3 AI out as one multimodal foundation with several editions, each released in phases after its own early-access period. Based on the official launch plan:

• FLUX 3 Video: video and audio generation and editing through APIs and private weight access, the experience previewed on this page.
• FLUX 3 Image: image synthesis and editing, also delivered through APIs and private weight access.
• FLUX 3 Action (FLUX-mimic): action prediction for robotics, built with partners such as mimic robotics and already tested on production lines at Audi.
• FLUX 3 Dev: an open-weight multimodal backbone spanning image, video, and audio generation plus action prediction, for teams that want to run and adapt the model themselves.

Black Forest Labs has confirmed open weights are planned, and once the FLUX 3 Dev backbone is available, a community FLUX 3 LoRA workflow, including FLUX 3 dev LoRA fine-tuning for recurring characters, styles, and brand looks, is the natural next step, mirroring the large LoRA ecosystem that grew around earlier FLUX releases.

For most teams, FLUX 3 Video will handle day-to-day clips, while FLUX 3 Dev and FLUX 3 dev LoRA open the door to self-hosted customization.

FLUX 3 Video Official Resources#

• Black Forest Labs

In short, FLUX 3 Video is coming soon on RunComfy: one multimodal foundation that turns prompts and reference media into cinematic clips with synchronized native audio.