Black Forest Labs announced FLUX 3 Max, a version of its FLUX 3 video model that fal adapted and optimized. The companies describe it as a collaboration, with fal listing the model for text- and image-to-video generation.
Black Forest Labs says FLUX 3 Max can generate a 10-second video in 12 seconds. It also claims the model is 22 times faster than Dreamina Seedance 2.5 at the same quality. Those are company-reported performance figures, not results from Digg testing.
What the independent leaderboard shows
The Artificial Analysis AA-Video-I2V v1.1 leaderboard gives FLUX 3 Max an Elo score of 1,138, with a 95% interval of 1,123 to 1,153, based on 5,054 samples. That places it second by point estimate, two points behind MiniMax H3, but its statistical rank range runs from first to fifth.
Seedance 2.5 also has a point estimate of 1,138, with a wider confidence interval. The benchmark ranks image-to-video models using Elo scores derived from pairwise human-preference votes by validated private evaluators. This supports a statistical tie near the top of that quality leaderboard, while the 12-second generation time and 22-times-faster comparison remain claims from Black Forest Labs.
The broader FLUX 3 platform on fal supports clips generated from text, still images, first and last frames, keyframes or an existing video, with native audio and durations of up to 20 seconds. FLUX 3 Max is the post-trained variant built with fal Research.
Comments