Fal launches FLUX 3 Max with faster video generation

Fal launched FLUX 3 Max on October 19 in collaboration with Black Forest Labs. The video model builds on FLUX 3 with additional training and changes to how it runs.

Fal said its tests produced a clip lasting 10 seconds in 10.95 seconds. It also cited preliminary human preference testing that gave the model a 1,139 Elo score from 4,868 votes, with uncertainty ranges overlapping those of the leading models.

The company said the updated model gained 50 Elo points over the original FLUX 3 while running inference 11.5 times faster. The results reflect work on an existing foundation model through reinforcement learning, further training and system optimization.

Developers can access the model through fal using text, images or keyframes to guide video generation. Fal is offering a 75% discount during the first three weeks and said further improvements remain in development.