FLUX 3 API Is Here: Video With Audio, Then FLUX 3 Image

The timeline

• July 23, 2026: FLUX 3 announced as a multimodal foundation model, with an early-access program for video and for FLUX 3 Action, a robotics model built with mimic robotics.
• August 4: POST /v1/flux-3-video launches: up to 20 seconds at Full HD with synchronized audio.
• August 17: video continuation capped at 15 seconds; text-to-video and image-to-video stay at 20.
• September 10: native QHD (2560×1440) and UHD (3840×2176) output, with no separate upscaling step.
• September 23: FLUX 3 Action released as an open-weights 7B world-action model.
• October 1: FLUX 3 Image launches on the same API, with a 50% introductory discount until October 8.

What FLUX 3 Video does

FLUX 3 is BFL's first video model, and it arrives with features that took other labs several generations:

• Up to 20 seconds per generation, more than Veo 3.1's 8 seconds or Kling 3's 10.
• Native audio: dialogue, sound effects and ambience generated with the picture, including multilingual speech with lip-sync.
• Four modes: text-to-video, image-to-video with keyframes, video continuation (extend an existing clip) and draft enhancement (turn a cheap draft into a full render).
• Up to 4K: HD, Full HD, QHD and UHD output from the same generation, so composition stays consistent across sizes.
• Draft mode at a fraction of the price, to iterate on prompts before paying for a full render.

BFL also published a preliminary preference study on 10-second 720p clips, in which raters preferred an early FLUX 3 candidate over Luma Ray 3.2 in 93% of comparisons, Runway Gen-4.5 in 77%, and both Gemini Omni Flash and Seedance 2.0 in 52%. These are the vendor's own numbers on a pre-release model; treat the last two as a tie. Our FLUX 3 comparison looks at where it fits.

And FLUX 3 Image

The image side launched on October 1 through the same API, where the prompt decides whether to generate or edit. Two features stand out for developers:

• JSON layouts: you can describe a composition as a list of elements, each with an ID, a description and bounding-box coordinates on a 0 to 1000 grid. Layouts stay the same at any output size.
• Pixel-locked editing: when you edit inside a bounding box, pixels outside it are kept unchanged, which makes iterative product and ad edits predictable.

FLUX 3 Image accepts up to ten reference images and renders natively up to roughly 4K. Pricing runs from about $0.041 per image at 768 px to $0.607 at 4K, before the launch discount. Details in our FLUX 3 pricing guide.

Open weights

BFL built its following on open FLUX image models, and it intends to keep part of FLUX 3 open. FLUX 3 Action is already available as open weights. An open FLUX 3 Image is expected "within weeks", and FLUX 3 Dev, an open multimodal backbone for video, audio and image, is planned for later in 2026. No license terms or hardware requirements have been published yet, so if self-hosting matters to you, wait for the actual release.

Using FLUX 3 through VideoGenAPI

On VideoGenAPI you get FLUX 3 Max, fal's post-trained variant of FLUX 3 Video, tuned for prompt adherence and visual quality with lower latency. It does text-to-video and image-to-video (with an optional end frame) from 5 to 20 seconds at 720p or 1080p, with native audio. It is billed from your balance at $0.25 per second at 720p and $0.40 at 1080p, so a 10-second 720p clip costs $2.50. The request is the same as for any other model:

curl -X POST "https://videogenapi.com/api/v1/generate" \
  -H "Authorization: Bearer $VIDEOGEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "flux-3-max",
    "prompt": "A hand places a coffee bag on a marble counter, the label reading SINGLE ORIGIN stays sharp, slow push in",
    "duration": 10,
    "resolution": "720p"
  }'

Poll GET /api/v1/status/{id} until the status is completed. Failed generations are refunded to your balance. For 4K, video continuation or FLUX 3 Image, use BFL's API directly. See the generate reference.

Frequently asked questions

Sources