BFL launches FLUX 3 Max with 75% off for two weeks

• FLUX 3 Max is fal Research's post-trained version of Black Forest Labs' video model, with a 75% launch discount announced in the launch thread.
• Ten seconds of video can render in about twelve seconds, according to gizakdag's test.
• A 20-second, 720p clip cost $0.85 and took 51 seconds in ozansihay's Turkish village test.

A Rocket League parody sends a thruster-equipped Roadster over a football stadium. A Turkish village sketch asks for autofocus mistakes and clumsy zooms. The keyframes API accepts ten images pinned to specific moments on a 24 fps timeline.

fal Research built FLUX 3 Max by post-training Black Forest Labs' FLUX 3, according to fal's launch page. The headline comparison combines two measurements:

• Generation speed: 12.45 seconds for FLUX 3 Max versus 277.80 seconds for Seedance 2.5, roughly 22 times faster, for ten-second, 16:9 clips at 720p/768p.
• Image-to-video preference: both models scored 1,139 Elo in Artificial Analysis's preliminary arena results.

“Tied for #1” refers to overlapping confidence intervals. MiniMax H3 was listed first at 1,141 Elo, rather than sharing the exact same score.

A roughly twelve-second turnaround also appears in gizakdag's hands-on report.

There is a separate timing detail in the API documentation: the returned timings.inference field measures denoising only. It excludes prompt expansion, decoding and upload.

PJaccetturo said his Rocket League-inspired parody took two hours to make. He explicitly labels it parody and proposes a movie featuring zenrll and SquishyMuffinz, with the famously wholesome Squishy as the villain.

PJaccetturo's Rocket League-inspired flying-Roadster parody

His follow-up asks who wants a Rocket League movie featuring the game's top streamers.

ozansihay reported 51 seconds and $0.85 for his 20-second, 720p village sketch. The post is marked as promotional.

His prompt's best choice is the old consumer-handycam look, complete with autofocus hunting. It divides the clip into five explicit four-second scenes:

1. 0-4 seconds: an older man holding tea outside the village coffeehouse.
2. 4-8 seconds: a woman kneading dough beside a tandır oven.
3. 8-12 seconds: a shepherd standing with sheep on a dusty slope.
4. 12-16 seconds: a grandmother sitting in her doorway.
5. 16-20 seconds: a young man leaning against a red tractor.

Each character gets scripted Turkish dialogue with a rural Anatolian accent. The sound request includes camcorder microphone handling noise and village ambience; subtitles and on-screen text are explicitly excluded.

BFL's launch thread links three generation routes. Their controls cover different amounts of visual direction:

• Text-to-video: a written prompt supplies the scene in the text-to-video playground.
• Image-to-video: a starting image establishes the opening composition, while a prompt directs motion in the image-to-video playground.
• Keyframes-to-video: up to ten images receive unique frame_index positions, according to the API schema. At 24 fps, frame 120 lands five seconds into the clip.

The schema also exposes:

• Duration: 5-20 seconds, with a five-second default.
• Resolution: 720p by default, with a 1080p option.
• Audio: generated by default, with a switch to disable it.
• Reasoning: low or high effort; high is the default and increases generation time.
• Seed: optional, with the used seed returned in the result.

venturetwins said every clip in this montage was text-to-video and generated in under ten seconds.

fal's prompting guide publishes 50 single-pass examples with their exact prompts. Its five-part structure is:

1. A camera move.
2. The main action, written as ordered beats.
3. Secondary motion in the environment.
4. A lighting or time change.
5. A closing “Sound” line.

BFL advertises “75% off for the next 2 weeks”, while fal's endpoint pricing note specifies October 15 through November 7.

| Output resolution | Promotional price per second | Price after promotion |
| --- | --- | --- |
| 720p | $0.0425 | $0.17 |
| 1080p | $0.0725 | $0.29 |

Text-to-video and image-to-video cost the same. Billing is per second of output, with no minimums or subscription, according to fal's launch page.