Flux 3 video — FLUX 3 Video GeneratorFlux 3 video
NewMulti-shot sequences are live — several cuts from one prompt

Generate FLUX 3 Video from a prompt, an image or two keyframes

flux3 runs the Black Forest Labs video model end to end — text to video, image to video, reference-driven shots, keyframes and multi-shot sequences, with the soundtrack generated in the same pass. Write a prompt below and your clip comes back in minutes.

Try a prompt
See what it generatesCredit pricing

No install, no waitlist · Sound generated with the picture · Pay with credits, not a subscription

Model
FLUX 3 Video
Built by
Black Forest Labs
Audio
Generated in-model
Open weights
FLUX 3 Dev pending

Independent platform

flux3-video.net is an independent generation platform. We are not affiliated with, endorsed by or sponsored by Black Forest Labs, and "FLUX" is a trademark of its owner used here descriptively.

Generation modes

Five ways to get a clip out of FLUX 3

Each mode takes different inputs and solves a different problem. Pick the one that matches what you already have — a sentence, a photo, a character sheet, or two frames.

01

Start with nothing but a sentence

FLUX 3 text to video reads camera language — focal length, dolly direction, time of day — so a prompt that specifies the shot gets the shot, instead of the same slow push-in every model defaults to.

  • Long prompts hold up; detail keeps paying off well past 200 words
  • Same prompt and seed returns the same shot, so you can iterate one word at a time
Generate a text to video clip
Text-to-video result

prompt · Handheld 35mm, following a courier through a rain-soaked night market, neon reflected in puddles, camera drifts left as she turns the corner.

Mode
text-to-video
Ratio
16:9
02

Animate a still you already own

FLUX 3 image to video keeps the source frame's identity, lighting and grain instead of re-rendering your subject into someone adjacent — which is the whole game for a product shot, a headshot or a piece of artwork.

  • Upload the frame, describe only the movement — the image carries the look
  • Faces and logos survive the full clip rather than melting after the first second
Upload an image and animate it
Image-to-video result

prompt · Source: studio portrait, 85mm. Motion: subject exhales, turns toward camera, hair settles.

Mode
image-to-video
Input
1 still
03

Keep the same character across clips

FLUX 3 reference-based video generation takes images of a person, product or location and carries them into new shots, so a series looks like it was shot on one day rather than assembled from strangers.

  • Character, wardrobe and location references can go into the same generation
  • Consistency holds across separate runs, not just inside one clip
Generate from reference images
Reference sheet + result

prompt · References: [character sheet], [jacket]. The same character steps into a freight elevator, fluorescent light overhead, doors close on the shot.

Mode
reference
Inputs
2 images
04

Pin the first and last frame

Give FLUX 3 keyframe video a start image and an end image and it solves the motion between them — the way to get a clip that cuts cleanly into footage you already have, or loops without a visible seam.

  • Both frames come back exactly as supplied; only the middle is generated
  • Use the same image at both ends for a perfect loop
Generate between two keyframes
Start frame → end frame

prompt · Start: empty diner booth at dawn. End: same booth, coffee poured, sun higher. Slow drift, no camera move.

Mode
keyframe
Inputs
start + end
05

Get a whole scene, not one shot

FLUX 3 multi-shot video generation returns several cuts that share a subject and a location, so a 15-second story beat lands in one generation instead of four clips and an afternoon in an editor.

  • Describe the cuts in the prompt and the model cuts where you said
  • Subject and lighting carry across shots without stitching anything by hand
Generate a multi-shot sequence
Multi-shot sequence

prompt · Shot 1: wide, a lighthouse in fog. Shot 2: close on hands winding a mechanism. Shot 3: from the lamp room, down at the sea.

Mode
multi-shot
Shots
3
Sound

Your clip arrives with its soundtrack already on it

Sound is generated with the frames, not added afterwards — which is why a door the model decided to close still gets its latch click on the right frame.

Ambience, foley and dialogue in one pass

FLUX 3 video native audio means no separate sound-design step for a first cut. Prompt the mix the way you'd brief an editor — "rain on a tin roof, no music" — and that's what comes back.

Extend a clip without a seam

FLUX 3 video and audio extension continues a finished generation past its end point, carrying picture and room tone forward together. Chain extensions when one clip needs to run longer than a single generation allows.

Audio example — play with sound on

One generation, no sound design added afterwards.

prompt · A workshop at night: a woodplane scrapes, shavings drop, distant traffic, a radio in the next room. No music.

Audio
in-model
Post
none
How it works

Three steps from prompt to finished clip

No install, no node graph, no waitlist. The generator at the top of this page is the whole workflow.

  1. 01

    Describe the shot

    Write the prompt, or drop in an image, reference set or keyframes depending on the mode you want. Prompt presets are one click if you'd rather start from something that already works.

  2. 02

    Set ratio, length and sound

    Pick 16:9, 9:16 or square, choose the clip length and resolution, and decide whether audio is generated with the picture. The credit cost updates before you commit.

  3. 03

    Generate and download

    The clip renders in the background while you keep working. Download it, extend it, or send it back through image-to-video for another pass — everything stays in your creations.

Signed-out visitors can build a prompt and see the cost; generating needs an account.

Prompt library

Prompts that work, and what they returned

Real generations published with the exact prompt and settings, so you can copy one, change a word, and see what moves. Nothing here is graded, re-cut or sound-designed after the fact.

Prompting notes

  1. 01Name the camera: focal length, movement and speed change the shot more than adjectives do
  2. 02Describe sound explicitly if it matters — silence is also a valid instruction
  3. 03Change one thing per run; a seed plus a prompt is reproducible
Physics example
Mode
text-to-video

Weight and contact

Prompt

Cloth, dust and impact — the prompt shape that gets physics right on the first try.

prompt · A heavy canvas tarp is pulled off a motorcycle in one motion, dust catches the light, the tarp folds as it hits concrete.

Consistency example
Mode
reference

Same character, five scenes

Prompt

One reference set reused across generations to build a series instead of one-offs.

prompt · References: [character sheet]. Five scenes, same subject and wardrobe, varied lighting.

Text rendering example
Mode
text-to-video

Legible text in frame

Prompt

Signage and packaging that survives the render — what FLUX has always been good at, now moving.

prompt · Slow push toward a hand-painted window reading "CLOSED FOR THE SEASON", street reflections sliding across the glass.

Audio sync example
Audio
in-model

Sound locked to the action

Prompt

Four strikes, four sparks, four clangs — how to prompt audio that lands on the frame.

prompt · A blacksmith strikes hot metal four times, sparks on each strike, quenching hiss at the end.

Media slots are wired and waiting — drop exported clips into /public/showcase and fill in the paths in src/i18n/pages/index/en.json.

Compared

How FLUX 3 Video holds up against Veo 3, Kling AI, Runway and Sora 2

Same prompts, five models. The differences that actually change your workflow, not the ones that change a benchmark score.

CapabilityFLUX 3 VideoVeo 3Kling AIRunway Gen-4.5Sora 2
Reference-based consistencyPartialPartial
Start + end keyframes
Multi-shot from one promptPartialPartial
Native audioPartial
Extend clip with audioPartialPartial
Open weights announcedDev expected

FLUX 3 Video vs Veo 3

Both put sound on the picture, so it comes down to control. Veo 3 is the better single-shot cinematographer; FLUX 3 gives you references and keyframes, which is what you need when the clip has to match something that already exists.

Hear a native audio generation

FLUX 3 Video vs Kling AI

Kling has been the dependable workhorse for image-to-video and start/end frames. FLUX 3 matches that control and pulls ahead on audio and multi-shot sequences, which is where most editing time actually goes.

See the flux 3 image to video mode

FLUX 3 Video vs Runway Gen-4.5

Runway Gen-4.5 lives inside an editing suite and half its value is that context. FLUX 3 is a model you drive from a prompt box or an API, which is faster if your timeline lives somewhere else already.

Read about the flux 3 video api

FLUX 3 Video vs Sora 2

Sora 2 sets the bar for long narrative coherence. FLUX 3's answer is per-shot control plus multi-shot in a single generation — and, if the Dev branch lands like earlier FLUX releases, weights you can run yourself.

See multi-shot generation
Access, API & pricing

Generating today, and where the model is headed

What you can do right now, and what Black Forest Labs has and hasn't announced.

Available now

How to get FLUX 3 Video access

Create an account here and generate — no invite code, no waitlist, no local install. That's the fastest route for anyone who wants output rather than a seat in a queue.

Invite only

FLUX 3 Video early access

Black Forest Labs' own early access programme is invite-gated and carries lower usage limits. Applying there and generating here aren't mutually exclusive — most people do both.

Per clip

FLUX 3 Video pricing

Credits, not a subscription. Cost scales with clip length, resolution and whether audio is generated, and the exact price is shown before you press generate — no surprise metering.

Available

FLUX 3 Video API

Drive the same generation modes from your own pipeline: submit a job, poll it, retrieve the asset. Useful when video generation is one step inside a larger product rather than the whole task.

Not announced

FLUX 3 Dev release date

Nothing official yet. Earlier FLUX generations shipped a Dev variant after the hosted model settled, so weeks to months is the pattern to expect — precedent, not a promise.

Expected

Will FLUX 3 Dev be open source?

Every previous Dev release shipped downloadable weights under a non-commercial licence — open weights rather than OSI open source. Video models are also far heavier, so plan for serious VRAM if you intend to self-host.

FAQ

Questions before you generate

Write a prompt in the generator at the top of this page, or upload an image, reference set or keyframes depending on the mode. Choose aspect ratio, length and whether sound is generated, then press generate. The clip renders in the background and lands in your creations, where you can download it, extend it or send it back through image-to-video for another pass. No install and no waitlist — an account is all that's required.

Generation is priced in credits rather than a monthly subscription, and the cost of a clip scales with its length, its resolution and whether audio is generated alongside the picture. The exact credit price appears before you press generate, so nothing is metered behind your back, and unused credits don't expire at the end of a billing month the way subscription quota does. Current rates are on the pricing page.

Start from whatever you already have. A sentence and nothing else means text to video. An existing still you want to keep exactly as-is means image to video. A character or product that has to appear in several clips means reference-based generation. A shot that has to begin and end on specific frames — a loop, or a hand-off into footage you've already cut — means keyframes. When a single beat needs several cuts, multi-shot generation returns them from one prompt.

Yes. Ambience, foley and dialogue are generated in the same pass as the frames rather than layered on afterwards, which is why sound events line up with actions the model itself invented instead of only with the ones you described. You can direct the mix in the prompt — asking for rain and no music gets you rain and no music. Clips can also be extended with picture and sound carried forward together, so a longer cut doesn't announce itself with a seam.

There is an API for driving the same generation modes from your own pipeline: submit a job, poll for status, retrieve the asset. On usage rights, clips you generate on a paid plan are yours to use in client work, ads and social content — check the terms of service for the specific wording, and note that this is separate from the licence question around FLUX 3 Dev weights, which are expected to ship non-commercial like earlier FLUX Dev releases.

No FLUX 3 Dev release date has been announced by Black Forest Labs. Precedent from earlier FLUX generations points to a Dev variant arriving weeks to months after the hosted model and shipping downloadable weights under a non-commercial licence — open weights rather than open source in the OSI sense. Video models are much heavier than image models, so self-hosting will demand far more VRAM than running FLUX image weights ever did.

More

Elsewhere on the site

Text to video generatorThe dedicated page for prompt-only generation, with more prompt structures.Full studioLonger sessions, generation history and every mode in one workspace.Credit pricingWhat a finished clip costs once length, resolution and audio are counted.FLUX 3 Dev release trackerOur running log of announcements and what changed in each build.

Generate your first FLUX 3 clip

One prompt, sound included, back in minutes. Credits only get used when you press generate.