FLUX 3
Create 20-sec videos in 1080p with the dialogue and the sound design already inside it, just out of one generation. Say what you want to see and the agent works out how to ask FLUX 3 for it.
What's new in FLUX 3
Twenty seconds, one pass
FLUX 3 runs to 20 seconds in a single generation, at 720p or 1080p and 24fps. That's long enough to hold a whole beat — and long enough that one generation can change scene and change camera angle without you cutting anything together.
The sound comes with it
Dialogue, effects, and room tone are generated alongside the frames rather than laid over them afterwards. Everything lands in sync because none of it was ever separate. Shots that don't want sound can have it switched off.
Dialogue that actually lands
Multiple languages, multiple accents, and mouth movement that follows the words. A two-hander can be generated instead of cast, shot, and dubbed.
Motion you set, not motion you hope for
Pin a start frame, an end frame, or as many as 10 keyframes in order, and FLUX 3 works out the movement between them. Prompting for a specific action becomes placing it.
A cheap first look
Draft mode returns a rough version of the shot fast and for very little. The full render happens after you've watched the idea work, so a bad idea costs you a draft rather than a render.
How FLUX 3 compares to FLUX.2
Capability | FLUX 3 | FLUX.2 |
|---|---|---|
Modalities | Image, video, audio | Image |
Video | Up to 20 seconds, 720p or 1080p at 24fps | Not available |
Audio | Same pass as the picture | Not available |
Motion control | Start frame, end frame, up to 10 keyframes | Not available |
Reference inputs | Stills, clips, audio, keyframes | Up to 10 reference images |
Image generation and editing | Early access, following video | Live now, up to 4MP |
Open weights | FLUX 3 Dev announced | FLUX.2 [dev], 32B, released |
How FLUX 3 will work with invideo agents?
FLUX 3 will run inside the same agentic workflow every model does on invideo: you direct in plain language, and the agent does the building. Here is what that will look like.
The prompt is the agent's job
FLUX 3 takes long, dense prompts and pays off on them — scene, action, camera, sound, all specified. You don't have to build that. Describe the shot the way you'd describe it to a DP, and the agent turns it into something the model can read.
Your cast comes with you
Approved characters and locations sit in your project and get attached to every generation. Because FLUX 3 accepts stills, clips, and keyframes as reference, a face or an outfit or a product carries across shots without you describing it again each time.
How much it does on its own is up to you
Let the agent generate and show you the result. Or have it stop at a draft and wait. Or watch it work step by step. Pick one, change it whenever.
What FLUX 3 is built to do best
Dialogue scenes
Speech, accent, and lipsync arrive with the picture, so a conversation can be generated rather than staged and dubbed. The room it happens in comes with it.
Ads and product films
A 20-second spot can come out of one generation at 1080p, voiceover and sound design included. Closer to a cut than a clip.
Action and physical business
Weight and momentum transfer properly between objects, so a catch, a collision, or a whip pan reads as physical rather than approximate. Keyframes let you place the beats you care about.
Looks you'd otherwise have to build
Stop-motion, illustration, handmade texture — FLUX 3 goes well past conventional cinematic output. Styles that would need a rig and a week can be generated instead.
Helping creatives stay creative
Multiplayer mode
Collaborate in real time with live cursors to show what everyone's working on.
Storyboarding
Turn any script or idea into a shot-by-shot plan, then tweak as needed before generating.
Script writing
Write your script inside invideo, and ask an AI co-writer for help if you'd like.
Timeline editor
Picture Premiere Pro with full AI.
Build your own agents
Create custom agents to fill specific roles like cinematographer, music designer, and more.
From solo creatives to creative enterprises
World-class investors stand behind invideo.
Backed by the firms behind Stripe, Spotify, Flipkart, and ByteDance.
Loading pricing…
FLUX 3 FAQs
What is FLUX 3?
FLUX 3 is Black Forest Labs' multimodal generative model. It generates up to 20 seconds of video per pass at 720p or 1080p and 24fps, with speech, effects, and ambience produced alongside the frames, and it takes text, stills, and up to 10 keyframes as input. Image generation and action prediction sit inside the same model. On invideo, you describe what you want and the agent operates it.
What's the difference between FLUX 3 and FLUX.2?
FLUX.2 generates and edits images. FLUX 3 brings video and audio into the same family — 20 seconds a generation, sound in the same pass, keyframe control over how a shot moves. The trade-off at the moment is timing: FLUX.2's image work is live at up to 4MP with 10 references, while FLUX 3's image model is still in early access. invideo's agent uses whichever one suits the job.
Does FLUX 3 really generate 20 seconds in one go?
Yes. Whole-second lengths, audio included, no shorter clips joined together. When a story needs more than 20 seconds the agent generates further takes and chains them, but every take is native.
Will a character hold across a whole clip?
Yes. Your characters and locations live in your project on invideo and get attached to every generation, and FLUX 3 reads stills, clips, and keyframes as reference. A face stays the same face from first frame to last.
Can I see something before paying for a full render?
Yes. Draft mode gives you a fast, cheap version of the shot, and a separate pass renders an approved draft properly. The agent can work draft-first, so you only spend on renders for takes you've already picked.
What generation modes will FLUX 3 support?
Text-to-video, image-to-video, first-and-last-frame, keyframes up to 10 in order, and extending an existing clip from its final frame. Aspect ratios run 21:9 through 9:21. Audio is on unless you turn it off.
How does FLUX 3 handle audio?
It generates in the same pass as the picture. Dialogue, effects, and ambience come back in sync across multiple languages, with lipsync tracking the words, because none of it was ever a separate layer. Switch it off per shot if you'd rather score it yourself.
Does FLUX 3 do images too?
Yes — image, video, and audio are one model. Black Forest Labs has said FLUX 3 Image follows video into early access. Until then invideo's agent handles image work with FLUX.2, at up to 4MP and as many as 10 reference images.
Do I need to know how to prompt FLUX 3?
No. Say what the shot is in ordinary language. The agent writes the version FLUX 3 responds to, attaches your references and keyframes, and runs it.
When is FLUX 3 coming to invideo?
Soon. It'll sit alongside the other models the agent can reach, and you'll direct it the same way — plain language, your cast and locations already attached.
Does using the agent mean giving up shot-by-shot control?
No. Every shot is still yours to approve or reject. The agent does the operating — drafts, keyframes, renders, chaining takes into a sequence — while you decide what's good. A 20-second generation is a longer take, not a longer leash.