FLUX.3 Edit Video on fal: Restyle a Clip for $0.03 per Second, and Why It Is Not a Face Swap (October 2026)
FLUX.3 Edit Video rewrites the look of an existing clip from a prompt while keeping its motion, for $0.03 per second. The limits, the missing reference-photo field, what it is good for, and how it differs from replacing a person.
FLUX.3 Edit Video takes a video and a text prompt and returns the same video restyled: same movement, same timing, a different look. It lists on fal at $0.03 per second, which is a third of what FLUX.3 text-to-video costs and about a tenth of the video-edit models that replace people. That price tag invites a misreading. This is a restyle tool. It has no input for a reference photo, so it cannot put your face into someone else's clip, and the sooner you know that the less you spend finding it out.
What it takes
| Input | Limit |
|---|---|
| Source video | MP4, under 50 MB |
| Source length | under 15 seconds |
| Guidance | a text prompt, plus safety_tolerance from 0 to 4 |
| Reference images | none |
The absence of a reference-image field is the whole story. Models that replace a person (HappyHorse video-edit, Kling O3 edit, Recast) accept pictures of the replacement. This one accepts only words, so the best it can do with "make the man look like me" is invent a plausible man.
What it costs
| Clip length | $0.03 per second |
|---|---|
| 5 seconds | $0.15 |
| 10 seconds | $0.30 |
| 14 seconds | $0.42 |
Compare the same job done by regeneration: a fresh FLUX.3 text-to-video clip at $0.085 per second is about $0.43 for five seconds, and it will not follow your source's motion. The edit endpoint is both cheaper and closer to the original.
What it is good for
Restyling without losing the choreography. A live-action clip turned into an illustrated, painted, claymation or noir version; the movement and timing stay.
Face-less footage. A food clip made warmer, a street shot moved from day to night, a product clip re-lit. With no person in frame, there is nothing to get wrong.
Re-using your own old footage in a new visual identity for a series.
Trend variants. The same clip in five styles for five test posts, for $0.15 each at five seconds.
What it is not good for
Putting a specific person in. No reference field. Keeping a specific face. The model is rewriting the look of everything, and a prompt can describe a style but cannot pin an identity. Long videos. Under 15 seconds per call; longer sources are cut into pieces.
There is also a legal point that applies to every restyle tool. If the source is someone else's video, restyling it does not clear the rights to the footage, the soundtrack or the people in it. We ask users of our own tools to confirm they have the right to a source before it runs, and the same discipline applies when you call the API directly.
How we think about it for a Reel
A Reel that mixes people and no-people scenes needs two treatments: replace the person where there is one, and leave or restyle the rest. For the face-less scenes, our pipeline today trims the original footage and publishes it unchanged. The restyle endpoint offers a possible third path, for Reels where you want the whole piece to look different, not only the face, at a price that is trivial next to the main replacement. We have it queued as a five-second test at about $0.15 and make no claim until it has run.
If your goal is a version of a viral clip with your face in it, this is not that tool; the walk-through of replacing a person is. If your goal is to make a clip look different, it might be exactly it, and the prompt habits in our notes on product video effects, which describe a sequence rather than a label, carry over.
A five-minute test before you commit
Pick a clip you own, under ten seconds, with clear motion. Write one prompt that names the style, the light and the material ("hand-painted gouache illustration, warm window light, visible brush texture"), and run it once for about $0.15 to $0.30. Watch the result next to the original at full screen. Check whether the movement and timing survived, whether text or logos in the frame turned to mush, and whether any person's face changed in a way you did not intend. If the motion survives and the style is right, batch the variants; if not, change the prompt before you change the model.
Common questions
Can I put my face into a clip with it?
No. The endpoint takes a video and a prompt only, with no reference-image input, so it cannot take a face from a photo.
What does it keep from the original?
The motion and timing. It restyles the look according to your prompt.
What are the input limits?
An MP4 under 50 MB and under 15 seconds.
Is restyling someone else's video allowed?
Restyling does not clear rights to the footage, the soundtrack or the people in it. Use your own material or have permission.
What does a clip cost?
$0.03 per second: $0.15 for five seconds, $0.30 for ten.
How to do this without code: For a Reel that includes face-less scenes, the autopilot keeps or regenerates them as needed. In AutoHustle this is the autopilot on a Reel with no person in it, where face-less scenes are kept from the original footage rather than regenerated (quoted in credits before you confirm the plan). See Faceless content and Recreate a viral video.
Related reading: FLUX.3 Video on fal: Text and Image to Video at $0.085 per Second, With No Face Filter (October 2026) · FLUX.3 Image and Klein on fal: Per-Megapixel Pricing, Ten References and Where Each Tier Fits (October 2026)
For developers: the fal endpoint
| Model | fal endpoint |
|---|---|
| FLUX.3 Edit Video (fast) | blackforestlabs/flux-3/edit-video |
Inputs: MP4 under 50 MB and under 15 s, prompt, safety_tolerance (0–4). No reference-image input. Price $0.03 per second, read from fal on 18 October 2026.