Flux 3 Is Here: Black Forest Labs' First Video Model Lands on Apiframe

Black Forest Labs spent two generations of FLUX becoming the default name in open image generation. Flux 3 is the first time they've turned that architecture on video, and it arrives with something most video models still treat as a second step: sound, generated in the same pass as the picture. It's live on Apiframe today, behind the same endpoint and API key as the rest of the catalog.

What Is Flux 3?

FLUX 1 and FLUX 2 were image models. Flux 3 is a multimodal foundation model: one set of weights trained jointly on images, video, and audio. It builds on Self-Flow, Black Forest Labs' method for aligning generation and understanding inside a single architecture, scaled up across all three modalities at once. Video prediction accounted for more than 95% of the training compute.

That design has a practical consequence. When audio and video come out of the same pass, they line up: an impact sound lands on the frame where the objects actually meet, and lip movement matches the syllables being spoken. Models that generate a clip and then score it afterwards have to infer that alignment after the fact, and that is usually where the seams show.

The same backbone also drives FLUX-mimic, a robot control model Black Forest Labs built with mimic robotics and put on Audi production lines. It is a useful hint at how much the model had to learn about contact, weight, and cause and effect in order to predict video at all.

Pricing That Scales With the Clip

Like most video models, Flux 3 bills per second of output: 25 credits per second at 720p and 42 at 1080p, with audio included in the rate. Ask for a 5 second clip and you pay for 5 seconds; ask for 20 and you pay for 20.

One thing to know: duration is optional. Leave it out and the model picks a length that fits the content — but since credits are deducted when you submit, before that length is known, an omitted duration is billed at the 20 second maximum. Pass an explicit duration whenever you want exact pricing.

What You Can Actually Do With It

Text to video, with sound

A prompt on its own returns a clip with synchronized native audio: room tone, footsteps, dialogue, score, whatever the scene implies. Audio is on by default. Set generate_audio to false when you're laying your own sound design over the top.

Animate a single image

Pass one image in images and it becomes the first frame. This is the path for putting motion behind art you have already approved, whether that's a product shot, a character render, or a Flux 2 still. You keep the look and generate only the movement.

First and last frame

Pass two images and Flux 3 reads them as start and end keyframes, then generates the transition between them. It is the most controllable mode in the model: you fix where the shot opens and where it lands, and the model solves the middle.

Storyboard from up to ten images

Three or more images act as an ordered storyboard, and the model paces the clip to hit each beat in sequence. A storyboard carries no implied length, so this mode requires you to set duration explicitly. Give the beats enough room to land.

Continue an existing clip

Point start_video at an mp4 of up to 50MB and 15 seconds, and Flux 3 continues from its final frames, carrying the scene, characters, and audio forward instead of cutting to something new. Chain it, feeding each output back in as the next start_video, and you can build multi-shot sequences well past the 20 second single-generation ceiling. Continuation is priced separately from generating from scratch; see the table below.

Draft mode

Set draft to true for a fast pass at roughly a third of the cost. Drafts are always 720p, so resolution stops applying. Use them to find the composition and camera move you want, then re-run the winner at full quality with a single field changed.

Specs at a Glance

Where This Actually Gets Used

• Paid social and ads: a 20 second cut that already has audio is a deliverable, not a silent clip waiting on a sound pass.
• Product motion from existing stills: animate photography that has already been signed off, instead of regenerating the product and risking one that looks different.
• Multi-shot narrative: chain continuations to run a scene past a single generation, with characters and audio carrying across the cuts.
• Storyboard previz: hand over an ordered set of frames and get back a timed animatic that hits each beat.
• High-volume prompt exploration: draft mode makes it cheap to try fifty directions before committing to a full-quality render.

How to Use Flux 3 on Apiframe

Same endpoint as every other video model on Apiframe. If you're already generating video here, this is a one-string change:

curl -X POST https://api.apiframe.ai/v2/videos/generate \
  -H "X-API-Key: afk_your_api_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "a lighthouse keeper climbs a spiral staircase at dawn, lantern swinging, waves booming against the rocks outside",
    "model": "flux-3",
    "flux3Params": {
      "duration": 20,
      "resolution": "1080p",
      "aspect_ratio": "16:9"
    }
  }'

You get a job ID back immediately:

{
  "jobId": "c3d4e5f6-a7b8-9012-cdef-345678901234",
  "status": "QUEUED"
}

Poll /v2/jobs/:id, or set a webhookUrl on the original request and let Apiframe call you when the clip is ready:

{
  "id": "c3d4e5f6-a7b8-9012-cdef-345678901234",
  "status": "COMPLETED",
  "model": "flux-3",
  "result": {
    "videoUrl": "https://cdn2.apiframe.ai/videos/c3d4e5f6-a7b8-9012-cdef-345678901234.mp4"
  }
}

No Black Forest Labs early-access application, no second SDK, no new billing relationship. Flux 3 sits in the same catalog, credit balance, and job API as everything else.

Pricing

Total cost is the rate times the clip length: a 10 second 1080p clip is 420 credits. Omitting the duration bills at the 20 second maximum, so set it explicitly for exact pricing. Continuation costs more because the model has to ingest your source video and stay faithful to it while generating new footage. As always, you only pay for successful generations. The model docs page has the full parameter reference.

Try It Today

Flux 3 is available now on Apiframe, alongside every other major video model, behind one endpoint and one API key. Get an API key and generate your first clip, or read the docs for full parameter details. Questions? Join our Discord.