fal launches FLUX 3 Max, claiming 22× faster AI video generation than Seedance 2.5

Popular enterprise AI multimedia creation platform fal.ai today launched FLUX 3 Max, a video model developed with German AI model lab Black Forest Labs that aims to make high-quality AI footage practical to generate and revise in seconds.

The headline claim: a 10-second clip takes 12.45 seconds to generate, compared with 277.80 seconds for Seedance 2.5 in fal’s published comparison.

That works out to approximately 22 times faster, with both models receiving the same quality score in the company’s cited benchmark snapshot.

For enterprise creative teams, the potential significance is the pace of experimentation. A designer could test a camera move, inspect the result and revise the scene while the idea is still fresh. That could make generated video more useful during collaborative work, where a long wait interrupts the discussion.

The launch also comes with a substantial introductory discount. But the performance claims require context, particularly the assertion that FLUX 3 Max shares first place in image-to-video quality.

A leading score, with an important qualification

The Artificial Analysis image-to-video leaderboard, checked October 19, lists FLUX 3 Max at 1,138 Elo, matching Dreamina Seedance 2.5 and two points behind MiniMax H3.

Those intervals overlap, making a decisive quality advantage difficult to establish from the displayed results. Artificial Analysis places FLUX 3 Max second by its point estimate, while assigning it a possible rank range of first through fifth.

The evaluation measures preferences between videos created from source images, with audio included. It does not establish equivalent leadership in generating footage from text alone, or prove that the model handles every production requirement equally well.

fal’s launch materials use a preliminary snapshot with slightly different scores and describe shared first place through overlapping confidence intervals. The defensible takeaway is that FLUX 3 Max competes with the leaders on this evaluation.

Speed changes the workflow—but the settings matter

fal describes its timing comparison as an average for 10-second, 16:9 clips at 720p or 768p, depending on the model. Its 12.45-second result approaches the duration of the generated footage, although it remains slower than real time.

These are vendor-reported timings. The model documentation cautions that settings and operating conditions affect completion time. A benchmark result should not be treated as a promise that every customer request finishes within the same window.

That distinction matters for software developers considering video generation inside an application. A quick demonstration can establish what is possible; production testing needs to establish how reliably the service delivers that speed under the intended workload.

The text-to-video documentation also provides low and high reasoning settings, with the higher setting increasing generation time. Teams therefore have another variable to test alongside output quality. fal.ai

What fal adds to Black Forest Labs’ model

FLUX 3 Max builds on Black Forest Labs’ FLUX 3 Video. fal describes its additional training as targeting closer adherence to instructions and better-looking output, alongside work to shorten generation time. fal.ai

That division of labor is part of the business story. A company serving models through an API can potentially differentiate its offering through further training and execution improvements, in addition to providing access.

In the launch thread, Black Forest Labs highlights speech, convincing physical behavior, sound effects and unconventional scenes as strengths. Those remain the developer’s assessments; the overall preference ranking does not independently verify each capability.

Developers can access public endpoints for text-to-video and image-to-video. The former starts from a written description; the latter uses an initial image and can accept an ending image.

The documented output range is five to 20 seconds. Videos generate at 720p, with optional upscaling to 1080p. Audio generation can be enabled or disabled, and supported formats include landscape, portrait and square compositions.

A separate keyframes endpoint accepts up to 10 images positioned along the sequence. For a product demonstration or storyboard, that could give teams more control over the visual stages a clip passes through. fal.ai

Introductory pricing starts at $0.43 for 10 seconds

fal’s published rates include a 75% promotional discount ending November 7.

The 10-second prices are calculated from the listed per-second charges. Regular rates are $0.17 per second for 720p and $0.29 for 1080p.

fal says API-generated content may be used commercially, subject to its usage terms. That permission should not be confused with a license to download, modify or operate the model independently; the cited launch materials do not establish those rights.

For buyers, the useful calculation is cost per accepted clip. At the introductory 720p rate, generating 10 versions of a 10-second shot costs $4.25; at regular pricing, the same exercise costs $17. Those calculations exclude staff time and subsequent editing.

Faster generation could reduce waiting, but it could also encourage more attempts. A team that produces substantially more variations without improving its acceptance rate may spend more even as each creative decision happens sooner.

The strongest case for FLUX 3 Max is therefore a workflow test: can a team reach an acceptable result faster, at a predictable cost, while preserving the details its brief requires? The launch gives enterprises a specific proposition to evaluate. A top-tier preference score and a short generation time are useful starting points; the business value depends on how often they translate into footage worth keeping.

VentureBeat’s Agentic Infrastructure Stack Summit, October 28 in San Francisco, is about doing more with less, from agents down to the chip. Attendance is by application.