Does it support both text-to-video and image-to-video?

Yes. Both modes share the same post-trained model, so fidelity carries across both at 720p, and switching modes does not require changing integrations.

What resolution does FLUX 3 Max generate at?

The page references 720p/768p output; generation-time figures are averages for a 10-second, 16:9 video at 720p/768p. The priced plan listed is for 720p.

Who built FLUX 3 Max?

fal Research post-trained the model in partnership with Black Forest Labs, working on post-training (for visual fidelity, motion coherence, prompt adherence and image-to-video quality), inference optimization (cutting a 10-second video from 126.44 to 12.45 seconds), and systems engineering on fal's own inference engine. fal Research is continuing to post-train the model.

FLUX 3 Max — Frontier video generation at near real-time speed

FLUX 3 Max is a text-to-video and image-to-video generation model post-trained by fal Research in partnership with Black Forest Labs, starting from Black Forest Labs' FLUX 3. It is positioned as a frontier-quality video model with near-real-time speed: a 10-second 720p clip generates in about 12.45 seconds. It scores 1,139 Elo on Artificial Analysis's preliminary image-to-video arena (AA-Video-I2V v1.1), tied for #1, and is roughly 22x faster than Seedance 2.5 at comparable quality. It is served on fal's API alongside H3 Max, another fal-post-trained model, with the two interchangeable via a single model ID.

Key Features

Tied for #1 image-to-video quality

Scores 1,139 Elo in Artificial Analysis's preliminary image-to-video arena (blind human preference votes), ahead of Gemini Omni Flash 1.1 (1,134), Wan 3.0 (1,127), and FLUX 3 (1,089), and tied with Seedance 2.5.

Near-real-time generation speed

Generates a 10-second, 16:9 720p video in about 12.45 seconds — roughly 22x faster than Seedance 2.5 (277.80 seconds) and about 10x faster than the base FLUX 3 model (126.44 seconds).

Text-to-video and image-to-video in one model

Both modes share the same post-trained model, so the same fidelity carries across both at 720p, and you can switch between them without changing integrations.

Motion and detail fidelity

Fast-moving subjects keep their shape across frames (horses hold stride, wheels keep spokes), and faces, skin, and textures hold up in extreme close-ups where other models tend to slip.

Prompt adherence for complex scenes

Follows full scene descriptions including scale, lighting, camera movement, and framing, demonstrated with detailed single-pass, no-edit example prompts for shots like a Pony Express handoff and a robotaxi at a frontier saloon.

Post-trained by fal Research with Black Forest Labs

Built through three layers: post-training for visual fidelity, motion coherence, prompt adherence and I2V quality; inference optimization to cut latency; and systems engineering on fal's own inference engine for reliable, scalable production serving.

Interoperable with H3 Max

H3 Max and FLUX 3 Max are both post-trained by fal Research and served on the same API, so you can move a shot between them by changing a single model ID.

Playground and agent access

Prompts can be run in the FLUX 3 Max playground on fal, or via fal Agent and ChatGPT integrations shown in the example gallery.

How It Works

Describe your video

Enter a text prompt in the FLUX 3 Max playground, or provide an image for image-to-video generation. Both modes use the same post-trained model.

Generate near real-time

The model generates a 10-second 720p clip in roughly 12.45 seconds, fast enough to iterate through many takes of a shot while another model would return one.

Ship the take you iterated on

Because quality and speed come together, the same model can be used for drafts and finals, and shots can be moved to H3 Max (or back) by swapping a single model ID on the same API.

Pros & Cons

Pros

State-of-the-art image-to-video quality (1,139 Elo, tied for #1 on Artificial Analysis's preliminary arena)
Exceptional speed: ~12.45 seconds for a 10-second clip, about 22x faster than Seedance 2.5 at the same Elo score
Single model handles both text-to-video and image-to-video with consistent fidelity
Promotional price of $0.0425/second of output video (75% off) with no minimums and no subscription
Easy to switch to or from H3 Max via a single model ID change on the same API
Built on fal's own tuned inference engine, so speedups hold in production serving

Who It's For

Best for

Developers and creative teams who need high-quality video generation with fast iteration cycles
Teams already on fal's API who want to swap between top video models without new integrations
Productions that need faithful motion and close-up face detail from generative video
Users generating many short clips (10-second-scale shots) where per-second billing without subscriptions is preferable

Not ideal for

Workflows requiring video output above 720p/768p resolution
Users who need a self-hosted or open-source video model

Use Cases

Generating short video clips (e.g., 6-10 second shots) from text prompts for creative and production work

Animating still images into video with image-to-video

Rapid iteration on shot ideas — trying many takes of a prompt in the time slower models return one

Detail-critical shots like close-ups of faces and fast motion (galloping horses, moving vehicles)

Fantastical or unfilmable concepts rendered photorealistically (e.g., a rocket booster landing on a ranch)

Production-grade video served through fal's scalable inference API

Pricing

Billed per second of output video, with no minimums and no subscription. Text-to-video and image-to-video cost the same. FLUX 3 Max 720p is $0.0425 per second (75% off the $0.17 list rate); this promotional rate ends November 7.

FLUX 3 Max 720p

$0.0425 per second of output video

Promotional rate, 75% off the $0.17 list price. No minimums, no subscription. Same price for text-to-video and image-to-video. Discount ends November 7.

FAQ

What is FLUX 3 Max?

FLUX 3 Max is a video generation model post-trained by fal Research in partnership with Black Forest Labs, starting from Black Forest Labs' FLUX 3. It supports both text-to-video and image-to-video and is served on fal's API.

How good is FLUX 3 Max compared to other video models?

It scores 1,139 Elo on Artificial Analysis's preliminary image-to-video arena (AA-Video-I2V v1.1), tied for #1. Its confidence interval overlaps the leaders, with MiniMax H3 listed first on that board at 1,141. It also leads Gemini Omni Flash 1.1 (1,134), Wan 3.0 (1,127), and FLUX 3 (1,089).

How fast is FLUX 3 Max?

It generates a 10-second, 16:9 video at 720p in an average of 12.45 seconds. That is about 22x faster than Seedance 2.5 (277.80 seconds at the same 1,139 Elo) and about 10x faster than the base FLUX 3 model (126.44 seconds).

How much does FLUX 3 Max cost?