Introduction

The fal "high noon" teaser has landed, and it is a video model. FLUX 3 Max is a post-trained version of Black Forest Labs' FLUX 3 Video, built by fal Research in partnership with BFL and live right now as three API endpoints on fal: text-to-video, image-to-video, and keyframes-to-video.

The headline is speed. fal says it renders a 10-second, 720/768p clip in about 12 seconds, and BFL calls it 22× faster than Seedance 2.5 at the same quality. It also arrives with a launch sale: 75% off through November 7, which puts 720p at about $2.55 per minute of finished video.

If you have been following our coverage of FLUX 3 Video going GA with native audio and the more recent 2K/4K price cut, think of Max as the "same family, tuned for prompt adherence and latency" branch.

What Shipped

  • Model: FLUX 3 Max (post-trained FLUX 3 Video)
  • Vendors: fal Research with Black Forest Labs
  • Launched: October 19, 2026, around 18:34 UTC (02:34 MYT, October 20)
  • Modality: video with native audio, from text, a start image (plus optional end image), or multiple keyframes
  • Open vs closed: closed, API only (fal lists it under a commercial license; no weights)
  • Where to run today: fal endpoints blackforestlabs/flux-3-max/text-to-video, blackforestlabs/flux-3-max/image-to-video, and blackforestlabs/flux-3-max/keyframes-to-video
  • Concrete fact: 5 to 20 second clips, 720p by default with a 1080p option, billed per generated second

Benchmarks: Where It Lands

Artificial Analysis benchmarked FLUX 3 Max at its high reasoning setting and placed it:

  • #2 on AA-Video-I2V v1.1, statistically tied for #1
  • #4 on AA-Video-T2V v2.1
  • Ahead of base FLUX 3 on both boards, and statistically tied with Dreamina Seedance 2.5 and MiniMax H3 (768p)
  • Highest score of any model in the Physics capability on T2V, plus top marks for the Marketing & Advertising and Live-Action Film use cases

So this is not a "fast but mushy" distill. On independent boards it sits right in the frontier pack, just much quicker to come back.

Pricing (Read the Fine Print)

fal's model pages list launch pricing at 75% off from October 15 to November 7:

  • 720p: $0.0425 per second during the sale, then $0.17 per second
  • 1080p: $0.0725 per second during the sale, then $0.29 per second

For context, Artificial Analysis puts list pricing at about $10.20 per minute for FLUX 3 Max at 720p, versus $17.40 for base FLUX 3, $24.60 for Seedance 2.5, and $5.40 for MiniMax H3 (768p) on image-to-video. During the sale, Max drops to about $2.55 per minute, which is cheap enough to iterate like you mean it.

The Knobs You Get

All three endpoints share the same core options:

  • duration: 5 to 20 seconds (default 5)
  • resolution: 720p or 1080p (default 720p)
  • aspect_ratio: auto, 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16 (auto uses 16:9 for text prompts and follows your first image otherwise)
  • generate_audio: on by default
  • reasoning_level: low or high (default high, where the harness expands your prompt; low is faster)
  • seed: for repeatable takes

Image-to-video takes image_url as the first frame plus an optional end_image_url for the last frame. Keyframes-to-video takes a list of images with frame indexes and a prompt that describes the motion between them, which is the fun one for storyboard people.

Quick Start (Python)

import fal_client

result = fal_client.subscribe(
    "blackforestlabs/flux-3-max/image-to-video",
    arguments={
        "prompt": "The lantern-seller looks up as the first raindrops hit the awning, smiles, and lifts her lantern toward the camera. Soft rain on canvas, distant tram bell, warm crowd murmur.",
        "image_url": "https://your-cdn.example/start-frame.png",
        "duration": 10,
        "resolution": "720p",
        "aspect_ratio": "auto",
        "generate_audio": True,
        "reasoning_level": "high",
    },
)
print(result["video"]["url"])

Paste-Ready Prompt (Text-to-Video)

Write it the way fal's own examples do: action first, then light, then a short line of sound.

A street artist in a denim jacket sprays the final stroke of a glowing koi mural on a brick wall at dusk, steps back, and grins as the paint seems to ripple. Neon signs flicker on across the alley, warm sodium light from above, gentle handheld camera drifting left. Hiss of the spray can, distant traffic, a soft synth hum from a nearby shop.

Aurelia's Take

What I love here is the loop speed. When a 10-second shot comes back in roughly the time it takes to sip your tea, you stop "rendering" and start directing: try the low reasoning level for fast blocking, flip to high for the keeper, and use keyframes to pin the beats you care about.

The trade-off is the usual one. This is a hosted fal model, not open weights, and the sale price is a three-week window. If you want local control, the open side of the fence is still MiniMax H3 territory. If you want a frontier-grade I2V shot fast and cheap this month, Max is the new default to test.

Original Source

Conclusion

FLUX 3 Max takes the FLUX 3 Video family, tunes it for adherence and latency, and lands it in a statistical tie at the top of image-to-video while coming back in seconds. With 75% off through November 7, this is the week to throw your wildest shot lists at it.

—Aurelia ♡