FLUX 3 Max: 10-second AI video in 12 seconds, fact-checked
On October 19, 2026, fal Research launched FLUX 3 Max, a post-trained version of Black Forest Labs' FLUX 3 video model, with a striking claim: a 10-second clip in about 12 seconds — roughly 22x faster than Seedance 2.5 at what fal calls the same quality. The live leaderboard tells a more careful story: FLUX 3 Max sits #2 on image-to-video and #4 on text-to-video on Artificial Analysis's boards, statistically tied for first. The 75% launch discount ends November 7.
- 12 seconds per 10-second clip — with caveats — fal's product page times a 10-second, 16:9 clip at 720p/768p in 12.45 seconds, against 126.44 seconds for base FLUX 3 and 277.80 seconds for Seedance 2.5. But fal publishes three different render times — 12.45s, 10.95s, and 'about 11s' — and all are vendor benchmarks with no independent timing yet.
- Tied for #1 in image-to-video, #4 in text-to-video — On Artificial Analysis's live boards (checked October 20, 2026), FLUX 3 Max scores 1,138 Elo in image-to-video — second, with a confidence interval overlapping MiniMax H3 and Seedance 2.5. In text-to-video it ranks #4. Post-training lifted it 50 Elo over base FLUX 3 (1,139 vs 1,089).
- 75% off until November 7 — FLUX 3 Max costs $0.0425 per second of output video at 720p — about $0.43 for a 10-second clip — until November 7, 2026. After that, the list price of $0.17 per second makes the same clip about $1.70.
This article is based on fal's official FLUX 3 Max launch materials and product pages (October 19, 2026), Black Forest Labs' FLUX 3 announcements, and Fello AI's launch coverage, which independently cross-checked every launch claim against the live Artificial Analysis leaderboard as of October 20, 2026. All speed numbers are vendor benchmarks. No hands-on testing was performed for this article.
The boldest claim in AI video this month came from fal Research, the generative media platform, working with Black Forest Labs (BFL): a 10-second video, generated in about 12 seconds, at what fal says is quality comparable to Seedance 2.5 — a model it times at 277.80 seconds for the same clip.
The claim deserves scrutiny, because it is doing a lot of work. If real-time-adjacent video generation holds up, it changes the workflow: twenty takes of a shot in the time one used to take, prompting video the way you prompt images. But fal publishes three different render times for its own model, the "tied for #1" claim holds only for image-to-video, and the launch discount that makes it the cheapest frontier video model ends on November 7.
What FLUX 3 Max actually is
FLUX 3 is BFL's video model — announced in early access on July 23, 2026, and generally available through the BFL API since August 4. It generates clips of 5 to 20 seconds with synchronized audio, from text, a single image, first and last frames, up to 10 keyframes pinned to exact times, or an existing clip to extend or edit. Resolutions run from HD to 4K on the BFL API, and the lab reports lip-synced dialogue in at least 13 languages.
FLUX 3 Max is fal's faster, post-trained version of it — launched October 19, 2026, and sold only through fal's API and platform. fal says the extra training targeted visual fidelity, motion, prompt adherence and image-to-video quality, and that this work lifted image-to-video Elo by 50 points over base FLUX 3. BFL CEO Robin Rombach framed the partnership as proof of strategy: "We built FLUX 3 as a foundation for others to build on."
The speed claim: yes, by fal's stopwatch
fal's product page states its method: an average for a 10-second, 16:9 clip at 720p/768p. That gives 12.45 seconds — about 22x faster than Seedance 2.5's 277.80 seconds, and about 10x faster than base FLUX 3's 126.44 seconds.
But fal publishes three different render times for the same model, and each produces a different multiplier:
- Product page: 12.45s (vs Seedance 2.5: ~22x)
- Launch press release: 10.95s (vs Seedance 2.5: 22x stated, 25.4x implied)
- API model page: "about 11s"
Nobody outside fal has published a competing timing yet. The safe planning number is the slowest one — about 12 seconds — and fal's own API documentation warns that actual generation time varies with request settings and operating conditions. a16z partner Justine Moore, a BFL investor, posted test clips at under 10 seconds, but did not disclose which settings she used.
Is it really #1 in quality? Only in image-to-video
On the live Artificial Analysis image-to-video leaderboard (AA-Video-I2V v1.1, checked October 20, 2026), FLUX 3 Max scores 1,138 Elo — second place, behind MiniMax H3 at 1,140 and tied with Seedance 2.5 at 1,138, with confidence intervals overlapping the leaders. "Tied for #1" is a defensible reading of that board and nothing more: Artificial Analysis gives FLUX 3 Max a possible rank of first to fifth.
On text-to-video (AA-Video-T2V v2.1), FLUX 3 Max ranks #4 at 1,159 — behind Wan 3.0 (1,173), Utopai X (1,166), and Seedance 2.5 (1,159). What is unambiguous: Max clearly beats its own base model on both boards (1,139 vs 1,089 in fal's cited snapshot), so the post-training worked.
Two caveats for buyers. First, Max's speed and quality numbers were both measured at 720p — its default output — with optional upscaling to 1080p. Second, it has a reasoning level setting of low or high, with high as default; higher effort increases generation time, and Artificial Analysis ran its benchmark at high.
Pricing: cheap until it isn't
FLUX 3 Max is billed per second of output video, with no minimums and no subscription. Text-to-video and image-to-video cost the same. Until November 7, 2026, it costs $0.0425 per second at 720p — 75% off the $0.17 list price. A 10-second clip costs about $0.43 today and about $1.70 after the discount ends.
For context, fal's own speed table puts Seedance 2.5 at 277.80 seconds per 10-second clip. If your workflow is iterative — try twenty takes, keep one — the speed difference matters more than the per-second price. fal frames it exactly this way: "the take you iterate on is the take you ship."
Notably, neither FLUX 3 video model is open-weights. BFL says an open FLUX 3 Dev backbone is "coming soon" without a date; today only the image model FLUX 3 Image Klein ships weights. Commercial use of videos generated through fal's API is permitted under fal's terms of service.
What to watch
- Independent speed tests. Every timing number is fal's own. The first third-party benchmarks will confirm or puncture the 22x claim.
- The leaderboard after November. FLUX 3 Max is post-trained on the base model, and fal's head of research says it is "continuing to post-train." Expect iterations.
- Whether competitors respond on speed. If video prompting really does converge toward image-prompting speeds, the workflow moat — draft at 12s, re-render finals at full quality with the same seed — becomes the product.
- Open weights for video. BFL has open-weighted every image generation. If FLUX 3 Dev ships open, the local-AI video landscape changes overnight.
Bottom line: FLUX 3 Max is the fastest way to make near-frontier AI video right now, and the 22x speedup at 720p is real enough to change how you work — iterate on twenty takes instead of waiting on one. But every speed number is fal's own, "tied for #1" is image-to-video only, and the current price is a three-week discount. Impressive, genuinely useful, and best trialed before November 7.
- fal — FLUX 3 Max product page and benchmarks (Oct 19, 2026)
- Black Forest Labs — FLUX 3 foundation model announcement (Jul 2026)
- Fello AI — FLUX 3 Video and FLUX 3 Max: Speed, Benchmarks, Price and How to Try It (Oct 20, 2026)
- Hugging Face — black-forest-labs/flux-3-image-klein (open weights, FLUX Kommunity License)
Prices and benchmark figures are standard published rates as of October 26, 2026 and can change without notice. Some outbound links may be affiliate links — see our disclosure. Found an error? Tell us and we'll correct it in public.