FLUX.3 Video

byBlack Forest Labs

Long takes that hold their subjects, weight and sound together, from a prompt, a start frame, two keyframes or a clip you want to carry on

How FLUX.3 Video works

Brief it like a cinematographer, sketch the take cheaply in Draft, then spend the budget once on the render that matters.

Image node holding the opening frame: a rain-swept glass pavilion at blue hour with lightning over the sea

Brief the shot

Write the camera move, the subject's action and the sounds you want, or drop in a start image, an end image or a clip to extend.

Scene and camera direction text node with a slow push-in prompt through rain, mist and ambient sound, wired onward

Draft it cheap

Set 5 to 20 seconds and your aspect ratio, then switch Draft on to lock camera, timing and audio at 720p.

FLUX.3 model with compact input, controls, and generated output

Render the final take

Turn Draft off, choose 1080p and run the take with picture and sound generated together.

What FLUX.3 Video is good at

Built for shots that need to last: steady camera, faces and signage that hold together, and sound that lands on the beat.

Picture and sound in one pass

Footsteps, room tone and lip-synced speech are born with the frames, in clips of 5 to 20 seconds. Put dialogue in quotes and add an Audio line, or the sound comes out generic.

Two-frame transformations

The first-last-frame variant bridges a start and end image for day-to-night shifts and pose changes. Keep both frames in the same setting and lighting, and prompt the bridge itself.

Extend a clip you like

Extend continues an approved shot from its final beat. Restate lighting, location and speaker to keep the seam quiet, and preview the next beat in the draft tier first, since extend is the priciest mode.

Draft before you commit

Draft fixes the render at 720p for roughly a quarter of the full-tier cost. Settle camera move, timing and audio there, then re-run at 1080p, or step up to the Max tier for a final text or image shot.

Made with FLUX.3 Video

Five takes across fashion, music, product, architecture and reportage, each carrying its own camera move and soundscape.

Fashion editorial in high-contrast tones

A cinematic beat with weather and score together

Product finishes under a moving lens

Interior light that shifts across a long take

Game action frame with bold color pairing

What people build with FLUX.3 Video

For directors, brand teams, fashion labels and social creators who need a shot that plays out instead of a looping flourish.

Film previs and shot tests

01

Directors and cinematographers block timestamped takes of up to 20 seconds, with dialogue and ambience, to test camera move and tone before a shoot.

Fashion lookbook motion

02

Labels and agencies animate a lookbook still from its first frame, so cloth weight and swish carry the garment. Use the Max tier when the key visual is final.

Product and hardware commercials

03

Brands bring an approved product still to life with its design locked in, or transition one finish into another between a start and end frame.

Vertical social moments

04

Content teams set 9:16, draft cheap versions of a hook, then re-run the keeper at 1080p with sound already matched to the action.

Album art in motion

05

Musicians and designers bring a sleeve or poster to life, bridging two key images with a prompted transition and a sound bed that fits.

FLUX.3 Video specs

Plans and credits
Versions
FLUX.3, FLUX.3 Max
Inputs
Prompt (text), Start Image (image), End Image (image), Video
Output
video
Duration
5–20 seconds
Resolution
720p, 1080p
Aspect ratios
21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16
Audio
Generates audio
Cost
827–29,206 credits per run

FLUX.3 Video questions

Draft is a much cheaper, fixed-720p render for iterating, while standard runs at 720p or 1080p for finals. Draft tiers cost roughly a quarter of the full tier. Draft is softer in fine detail, so use it to settle camera move, timing and audio, not skin texture or small text.

FLUX.3 Max reasons about your prompt before it generates, for higher quality, and adds a Reasoning Level and a fixed seed. It covers text-to-video, a start frame and a start-and-end frame pair. Base FLUX.3 adds the cheaper Draft mode and video extend. Settle the idea on base FLUX.3, then render the final on Max where quality matters most.

Use image-to-video to animate one approved still with its look locked in; describe only motion, camera and sound. Use first-last-frame when you know both the opening and closing frames, such as day to night. Mismatched frames produce morph-like results, so keep subject and lighting related.

Extend continues an existing clip from its final beat, following your prompt, with picture and audio carried on. It is the priciest tier, so test the seam with the draft variant first. Restate lighting, location and speaker, because seams can show and errors compound across chained extensions.

Yes. Audio is generated in the same pass as the picture and is on by default, covering ambience, effects and lip-synced speech. Describe the sound in your prompt, and put dialogue in quotes, or it can come out flat and generic. Turn Generate Audio off if you will score the clip elsewhere.

It struggles with fast fight or sports choreography, silent plates for manual sound design, flat vector motion graphics and frame-exact edits of existing footage. It is made by Black Forest Labs, which announced FLUX.3 Video in July 2026 and made it generally available in August 2026.

Try FLUX.3 Video on Fuser

Long takes that hold their subjects, weight and sound together, from a prompt, a start frame, two keyframes or a clip you want to carry on