Skip to content
Luma AI Help Center home
InboxAsk a human

Flux 3: FAQ

Updated by Martin

1. What is FLUX 3?

FLUX 3 (Black Forest Labs) is a video generation model with synchronized audio. It runs in two modes; Keyframe (text-to-video, or start/end frames) and Reference (continuation from an existing video) and can produce clips up to 20s.

2. What resolution and duration does FLUX 3 support?

Resolutions:

  • HD (faster iteration, lower cost)

  • FHD (the same generation run through BFL's video upsampler for a higher-resolution finish)

Durations (clip lengths):

  • 5s, 10s, 15s, or 20s — plus auto, which lets the model fit the length to the content

  • 20s is Keyframe-mode only; Reference (continuation) mode caps at 15s

3. How does FLUX 3 handle audio?

Audio is generated and enabled by default — disable it for a silent clip. Describe sound in layers (ambient, music, speech). For a spoken line, quote it, describe a visible speaker, and add "no on-screen text, no subtitles" to keep captions out of frame.

4. Can you combine inputs with FLUX 3 in one prompt?

Yes, within a single mode:

Keyframe mode:

  • A Start Frame (the video opens on this image) with an optional End Frame (the video closes on it; requires a Start Frame).

Reference mode:

  • One reference video (mp4, up to 60s / 100MB) to continue from — the generated clip picks up from its final frames. Only the tail matters, so trim long sources.

Note: the two modes are mutually exclusive; a reference video can't be combined with keyframe images.

5. What aspect ratios are available?

  • auto (derives the ratio from the prompt and any reference frames)

  • 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16

6. What does FLUX 3's continuation (Reference) mode do?

Reference mode lets FLUX 3 continue an existing clip. Feed it an mp4 and the new generation picks up from that video's final frames, so you can extend a shot. Only the tail influences the result; trim long inputs first.

7. Are keyframes available with the FLUX 3 model?

Yes, in Keyframe mode:

  • Start Frame: the video opens on this image.

  • End Frame: optional, requires a Start Frame, and the video closes on it.

Modes at a glance

Mode

What it does

Keyframe

Text-to-video, or start/end frames. Supports up to 20s.

Reference (continuation)

Continues from a reference video's final frames (up to 15s).