MODEL PROFILE · Black Forest Labs · served on fal

FLUX 3 FAST Edit Video

A profile of FLUX 3 FAST video editing on fal: short MP4 inputs, the published 720p rate, preservation claims and the audio details still unverified.

Source reviewed: 2026-10-02 · Video editing · documented API endpoint

Provider documentation summary. No hands-on results or independent ranking are claimed.

Documented facts

The following fields come from the official provider source, reviewed on the date above.

Exact identity: fal labels blackforestlabs/flux-3/edit-video as FLUX.3 Edit Video [FAST]. This profile covers that route, not every FLUX 3 video mode.
Inputs: Required fields are prompt and video_url. The API schema specifies an MP4 under 50 MB and under 15 seconds.
Controls: The published input schema lists safety_tolerance from 0 to 4, default 2. It does not expose a resolution, duration, mask or reference-image input.
Output: The schema returns a video file and the seed used. It does not document a user-supplied seed input.
Claimed behavior: fal describes re-rendering the source while preserving motion, timing and framing. We have not measured preservation or edit accuracy.
Access and dates: Public API documentation is accessible as of October 2, 2026 UTC. This is a documentation review date; the exact launch date and account-specific access were not verified.

Published rate and what it does not establish

The fal listing quotes $0.03 per second of generated video at 720p. A 10-second output would therefore cost $0.30, assuming exactly ten billed seconds and no additional charge or rounding. Three such attempts would total $0.90 before deciding which to accept. The listing’s approximate five-minute generation estimate is not our latency measurement or a service guarantee. Its broad upload-widget format list should not override the narrower MP4 requirement in the API schema. fal endpoint listing and pricing.

The practical decision

The practical question is whether an existing shot can survive the requested change. Before evaluating a replacement, write down what must stay fixed: camera movement, object position, the timing of an action, or a face visible near the edited area. Then separate those requirements from the desired change. A convincing new object is insufficient if the shot no longer cuts correctly into the surrounding sequence. Treat the original clip as the reference for continuity, and assess the edit against it rather than judging the output alone.

What to record in your own comparison

Compare against another video editor using the same source clip and intended change. Match the delivered resolution where possible and disclose any preprocessing, cropping or frame-rate conversion. A text-to-video result from a different scene cannot establish editing fidelity.
Review the entire clip at normal speed, then examine corresponding frames around movement, occlusion and the beginning and end of the edit. Record background changes, unstable edges and altered motion separately from whether the requested subject appeared.
Listen to source and result separately when audio matters. Record whether a track exists, whether speech or ambience changes, and whether synchronization survives. Do not assume an unchanged soundtrack because the visual edit looks local.
Keep the route, input file, instruction, request date, returned seed and billed duration with each saved result. Count rejected attempts in the cost per accepted shot; keep generation time separate from human review time.

Limitations and unknowns

No model calls, generated examples or independent scores were produced for this profile. Audio preservation or generation, exact output frame rate, maximum input pixel dimensions and a fixed checkpoint version remain unverified. We have not verified downloadable weights, a self-hosting license or hardware requirements for this FAST variant. The name FAST does not establish a measured speed advantage. The public listing marks commercial use, but that label is not a complete licensing analysis for an input asset or intended distribution.

Compare your saved results

Use the DualView editor to inspect files you already have. The accepted-output calculator uses your own rates and counts. Neither link runs a model or purchases generation.

Related reporting

Related model profiles

All model profiles and dated changes · Subscribe via RSS