Finely tuned by fal Research using FLUX 3, the model ranks joint first in image-to-video generation whilst being 11.5 times faster than its predecessor.
On October 19, 2026, the inference platform fal launched FLUX 3 Max, a video model resulting from a post-training conducted by its research team, in partnership with Black Forest Labs, on the FLUX 3 foundation model. The work focused on visual fidelity, consistency of movements and compliance with instructions, but above all on speed: according to fal, a 10-second clip in 720p comes out in 11 to 12.5 seconds, compared to more than two minutes for FLUX 3, a gain of a factor of 11.5.
Tied at the top of the standings
On Artificial Analysis' image-to-video ranking, FLUX 3 Max achieves an Elo score of about 1,139, up about fifty points from FLUX 3, and is neck and neck with MiniMax H3. The model accepts text, an image or keyframes as input.
It is available via the fal API, billed per second without a subscription: $0.17 per second of 720p video at the regular rate, reduced to $0.0425 until November 7, 2026. The content produced can be used commercially.
Read the full article on the original source
Go to the source →