What is FLUX 3 Max?
FLUX 3 Max is a video model that fal Research post-trained from Black Forest Labs’ FLUX 3 and launched on October 19, 2026. It turns text, an image, or a first and last frame into a 5 to 20 second video at 720p or 1080p with sound, and fal says it makes a 10-second clip in about 11 seconds.
The short version
- Made by: fal Research, with Black Forest Labs, on top of FLUX 3. Launched October 19, 2026, on fal.
- Makes: video from text, from a first frame, or from a first and a last frame; 5 to 20 seconds; 720p or 1080p; eight formats from 21:9 to 9:16; sound on or off.
- Costs on fal: $0.0425 a second at 720p and $0.0725 at 1080p until November 7, 2026 (75% off), then $0.17 and $0.29.
- Ranks: 1,139 Elo in Artificial Analysis’s preliminary image-to-video arena, which fal reports as tied for first with Seedance 2.5.
What it is
fal describes FLUX 3 Max as a post-trained variant of FLUX 3 Video, tuned for stronger prompt adherence and better visual quality and built to answer fast. You give it a prompt, and optionally the first frame or the first and last frames, and it returns an MP4 with sound. It is served through fal’s API; fal lists it under a commercial license.
FLUX 3 Max vs FLUX 3
FLUX 3 is Black Forest Labs’ multimodal model from July 23, 2026. FLUX 3 Max is fal’s faster, post-trained version of its video side. In fal’s launch numbers a 10-second video takes 10.95 seconds on Max against 126.44 on FLUX 3, and the arena score rose from 1,089 to 1,139. Those are fal’s own, preliminary figures.
How to use it
Developers call it on fal and pay per second of video. Without code or an API key, you can use it here after signing in: try 100 credits for $4.99 once (two 5-second 720p videos), then a plan from $10.39 a month billed yearly or a credit pack from $9.99. Make a video.
Videos made with FLUX 3 Max
These are unedited FLUX 3 Max clips we made on fal on October 20, 2026, at 720p with sound on. The prompts are on the prompts page.
Rainy morning vlog
Format 16:9, 10 seconds.
A 10-second ultra-photorealistic live-action smartphone vlog. A European brunette in her mid-20s on a rainy morning. 0-4s: she looks out a rain-streaked window holding a fresh mug of coffee and says to the camera: "Good morning... apparently, it's raining." 4-10s: she walks down a wet city street under a black umbrella, puddles and reflections, traffic passing, turns the phone to herself and says: "First coffee. Then lecture." Natural handheld movement, subtle shake, autofocus changes, realistic rain and wet clothing, natural expressions. Natural diegetic audio only: rain on the umbrella, footsteps in puddles, passing cars, her voice. No music, no narration, no subtitles, no logos, no watermark. Every person is an original fictional adult.
Adapted from a prompt by @nawalsehar
Anime katsudon ASMR
Format 16:9, 10 seconds.
A 10-second fast-paced cinematic Japanese anime cooking video of katsudon. High-quality original anime film style, warm summer light in a Japanese kitchen, ultra-detailed food textures, visible steam, macro close-ups, smooth camera moves, quick match cuts on motion and shape. 0-2s: a breaded pork cutlet is lowered into hot golden oil, intense bubbling. 2-4s: the fried cutlet is sliced on a wooden board, the crust cracks, steam escapes. 4-6s: the slices are placed over onions simmering in amber dashi. 6-8s: beaten golden egg is poured over and sets, glossy and trembling. 8-10s: the finished katsudon in a blue-and-white bowl over rice, topped with mitsuba, slow push-in as steam rises. Audio: bright 1980s Japanese city-pop instrumental with synced ASMR cooking sounds: frying sizzle, knife crunch, bubbling dashi, egg pouring, a soft ceramic clink. No text, captions, logos or UI.
Adapted from a prompt by @Goodmanprotocol
Seoul night camcorder vlog
Format 9:16, 10 seconds.
A 10-second ultra-realistic personal home video of an ordinary summer evening, filmed by a friend on an early-2000s consumer DV camcorder: strong handheld shake, imperfect framing, autofocus hunting, exposure shifts, faded colors, mild noise. A young woman in her early 20s, minimal makeup, long black hair in a messy side ponytail, pastel-blue short top, loose cream pajama pants, black sneakers, thin silver necklace. A quiet older Seoul residential lane at dusk: narrow concrete street, potted plants, bicycles, utility poles and overhead wires, a tiny bakery with warm light. 0-3s: she steps out of the bakery holding a warm pastry and takes her first bite, then smiles because it tastes good. 3-6s: she notices her friend filming and playfully holds the pastry toward the camera. 6-10s: a neighborhood cat follows her for a few steps; she crouches and gently pets it, laughing. Sound: camcorder mic hiss, cicadas, distant traffic, her soft laugh. Keep her face, hair and clothes the same throughout. Every person is an original fictional adult. No text, no logos, no watermark.
Adapted from a prompt by @AIwithkhan
AI videos made with FLUX 3 Max on Flux3Max SI
Want to make one yourself? Try FLUX 3 Max here.
Questions
Is FLUX 3 Max free?
No. On fal it is paid per second, and there is no free video here either: a first try costs $4.99 for 100 credits (two 5-second 720p videos), then plans from $10.39 a month billed yearly.
Who made FLUX 3 Max?
fal’s research team post-trained it with Black Forest Labs, the company behind FLUX. Flux3Max SI is an independent site and is not made by either of them.
How fast is FLUX 3 Max?
fal reports about 11 seconds for a 10-second 720p video. In our own runs, counting the queue, 5-second videos took 17 to 26 seconds and 10-second ones 23 to 31.
Does FLUX 3 Max work from a photo?
Yes. Image to video starts from your picture, and first + last frame ends on a second one. Our test photo of a woman and one of a golden retriever both kept their look through the clip.
Flux3Max SI is an independent service. It is not affiliated with, endorsed by or sponsored by fal or Black Forest Labs. FLUX is a trademark of Black Forest Labs.