FLUX 3 Max

Who made it

FLUX 3 Max comes from two companies. Black Forest Labs, known for the FLUX image models, built the FLUX 3 foundation model. fal, the generative media platform, post-trained it through its research team, including reinforcement learning, and optimized it for fast inference. fal released FLUX 3 Max on October 19, 2026.

The naming can mislead. FLUX 3 Max is a video model, not an image model, and it is a distinct model from base FLUX 3. Base FLUX 3 has draft, keyframe, and extend endpoints; FLUX 3 Max launched with two: text-to-video and image-to-video.

Video length versus generation time

Two different numbers come up when people talk about AI video speed, and they are easy to confuse.

Video duration is how long the finished clip plays: 6 seconds, 10 seconds.

Generation time is how long you wait for the model to make it.

Most video models take far longer to generate a clip than the clip lasts. Waiting several minutes for 10 seconds of footage is normal. FLUX 3 Max is notable because fal reports an average of about 12.45 seconds to generate a 10-second, 16:9 clip at 720p. That is roughly real time: you wait about as long as the video plays.

On fal’s own comparison, base FLUX 3 took about 126 seconds for the same clip and Seedance 2.5 about 278 seconds. Black Forest Labs’ launch post rounds this to “22x faster than Seedance 2.5 at the same quality.” These are timings from the companies selling the model, under their conditions, so treat them as indicative. Queue times on a busy platform can easily add more than the generation itself.

Why it matters in practice: fast generation changes how you work. When each attempt takes 12 seconds instead of four minutes, you can try twenty variations of a shot in the time one used to take, and pick the best. Our guide on building AI video one shot at a time assumes exactly that kind of iteration.

What it does

FLUX 3 Max turns a text prompt into a video, or animates a starting image. fal’s benchmarks and pricing are built around 10-second clips at 720p. Other platforms list different options: Krea lists 720p and 1080p output, clips up to 20 seconds, and optional audio. fal’s own launch page does not mention audio for FLUX 3 Max, so check the specific endpoint you are using before assuming sound, longer clips, or higher resolution are available there.

There is no documented support for reference images beyond the starting frame, keyframe control, or video editing in FLUX 3 Max at launch. Those belong to other endpoints or other models.

Quality claims

fal reports that FLUX 3 Max is tied for first place in Artificial Analysis’s image-to-video arena, scoring 1,139 Elo alongside Seedance 2.5, with MiniMax H3 just above at 1,141 and overlapping confidence intervals. That arena ranks models by blind human preference votes and was labeled preliminary at the time. A tie at the top of a preliminary preference ranking means the model is competitive, not that it is clearly better. For text-to-video specifically, fal’s launch materials do not cite a comparable ranking.

Pricing

fal bills per second of output video, with no subscription and no minimum. At 720p the list price is $0.17 per second ($1.70 for a 10-second clip). Until November 7, 2026, a 75% discount brings it to $0.0425 per second, about $0.43 per 10-second clip. Text-to-video and image-to-video cost the same. fal does not list pricing for other resolutions on its launch page.

The fast generation time has a cost angle too. Because iteration is cheap in time, it is tempting to generate many more takes than you need. Budget per finished clip, not per generation.

Access

FLUX 3 Max is available on fal through its playground and through its JavaScript and Python SDKs. It is also offered on other creative platforms, including Krea. fal handles enterprise integrations through its sales team.

Limitations

• Speed and quality claims come from the vendors, and the arena ranking was preliminary.
• Features differ by platform. Audio, 1080p, and longer clips are listed on some platforms but not on fal’s launch page.
• No keyframes, reference images, or editing documented for FLUX 3 Max at launch.
• Launch pricing is temporary. The 75% discount ends November 7, 2026.
• The name causes confusion with Black Forest Labs’ image models.

Alternatives

• Seedance, ByteDance’s video model family with native audio in newer versions
• Runway, a full video creation platform with editing tools
• Veo, Google’s video model
• Kling AI, a widely used video generator
• LTX-2.5, an open-weight video model you can run yourself
• Krea AI, a creative platform that offers FLUX 3 Max among other models

For the basics, see our AI video generator explainer.

Compare: FLUX 3 Max vs Seedance.