FLUX 3 is Black Forest Labs' multimodal foundation model, trained jointly on images, video, and audio in a single architecture. FLUX 3 writes video and sound together — up to 20 seconds in one generation — and follows keyframes, reference clips, and multilingual dialogue.
Required · dialogue, ambience and on-screen text can all be described in one prompt
Integer, 5 to 20 seconds
Default hd - draft previews are hd only
Default auto - seven fixed ratios available
On by default - a silent clip is not cheaper
Try one of these prompts
Credit-based — 1 credit = $0.001 USD. Pay only for completed generations.
One 5-second sample at the cheapest tier.
FLUX 3 is Black Forest Labs' multimodal foundation model, and its video capability is the part callable today: reAPI serves it as one async endpoint with three modes — text-to-video, image-to-video from up to 10 reference images, and video-to-video continuation from an existing clip. Clips run 5 to 20 seconds at HD or FHD across seven fixed aspect ratios plus auto, with generated audio on by default and a draft mode for cheap previews before the full render. Billing is per second, at a rate set by mode, resolution and whether draft is on.
Real-world workflows and production use cases you can build and ship with this model.
Text is the starting point, not the whole instruction set. Hand FLUX 3 a first frame to animate, reference images that pin down a character, a source clip whose elements should carry into a new scene, or two keyframes that define where a transition begins and ends. FLUX 3 can also continue an existing video and its audio, so a shot can be extended instead of regenerated from scratch.
Read the FLUX 3 docs
FLUX 3 handles multilingual dialogue and renders readable text inside the frame, so one creative direction can travel between markets. Signage, titles and animated typography are generated as part of the scene rather than composited on afterwards, and the spoken track is produced with the picture — which is where FLUX 3 is already strongest, according to Black Forest Labs' own early evaluations of facial expression and multilingual work.
The same FLUX 3 backbone synthesizes and edits images across styles, aspect ratios and resolutions, then extends those stills into motion. Shape, colour, material and branding stay recognisable between the packshot and the clip, because both come out of one model instead of a still generator bolted to a separate video generator.
FLUX 3 learns from images, video and audio inside a unified architecture built on Self-Flow, Black Forest Labs' approach to aligning multimodal generation and understanding. The modalities constrain each other: the sound has to match the impact, the motion has to obey the mass.
Every FLUX 3 video output carries native audio from the same generation — dialogue, ambience and impacts land on the frame they belong to, instead of being timed back onto silent footage in post.
Most video models make every attempt cost the same, so exploration is expensive. FLUX 3 splits it: draft previews run at roughly a third of the rate, and finalizing re-renders that same draft at full quality using its own prompt, timing and inputs — you pay the full rate once, for the take you actually keep.
Black Forest Labs' own early evaluations put these two closest together of everything it compared — viewers preferred FLUX 3 in 52% of matchups, effectively a coin flip. The split is scope: both are callable on reAPI today, but Seedance 2.0 is a video model, while FLUX 3 is a broader multimodal foundation whose image and action stages are still rolling out.
Comparison reflects publicly documented behavior at the time of writing. The 52% preference figure is from Black Forest Labs' own preliminary evaluation of 10-second 720p text-to-video clips with audio, on a model it describes as still in development; sample size and methodology were not published, and it is not an independent benchmark.
Sign up at reAPI and generate a key. One key, one balance, and the same request shape across every model on the platform.
OpenThe FLUX 3 page tracks what Black Forest Labs has actually published — capabilities, input types and rollout stage — and gets the endpoint, parameters and rates the moment they are confirmed.
OpenVideo models with native audio are already callable on reAPI. Build the integration against one of those now, and switching to FLUX 3 later is a model id and a parameter map.
Opencurl -X POST https://reapi.ai/api/v1/videos/generations \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "flux-3-video",
"prompt": "A rain-soaked cafe terrace at dusk, a tram rolls past",
"duration": 5
}'Rejected synchronously with the failing field named. Check enum values, ranges, and URL fields — nothing is charged.
401 means a missing or invalid Bearer key; 402 means the reserve exceeds your balance. Manage keys and credits in the console.
Reference material and outputs pass automated moderation. Rejected tasks fail with a clear error and a full refund.
A task that reaches a failed state is never charged — the reserve refunds automatically. Keep the task id and retry when ready.
Common questions about this model.
Pricing
Calculate FLUX 3 video costs across Draft, HD, FHD, keyframe, and continuation modes, including the draft-to-final workflow and 5–20s totals.
Comparisons
Choose FLUX 3 or MiniMax H3 by duration, keyframe control, continuation, resolution, local weights, license limits, native audio, and current API cost.
Comparisons
Choose FLUX 3 or Seedance 2.5 by maximum duration, ordered keyframes, references, editing, native audio, Draft workflow, resolution, and API cost.
Explore more models in the same category.
ByteDance
From $0.071 per second
Alibaba Cloud Bailian
From $0.146 per second
Kuaishou
From $0.077 per second
Vidu
From $0.053 per second
Try it in the playground or grab an API key to integrate now.