On Oct. 19, 2026, fal introduced FLUX 3 Max, describing it as a video model post-trained by fal Research with Black Forest Labs. fal's product documentation calls it a variant of FLUX 3 Video.
fal documented text-to-video, image-to-video and keyframes-to-video endpoints. The first starts with a written prompt. Image-to-video requires a starting image and prompt and allows an optional ending image. Keyframes-to-video accepts up to 10 images positioned along a clip. The pages specify clips lasting five to 20 seconds, with optional generated audio.
Fast in fal's comparison, with a qualified quality tie
fal's published comparison puts average generation time at 12.45 seconds for FLUX 3 Max against 277.80 seconds for Seedance 2.5, about 22 times faster. It covers 10-second, 16:9 videos at 720p or 768p. fal's documentation cautions that completion time varies with settings and operating conditions.
fal also calls FLUX 3 Max tied for first in a preliminary Artificial Analysis image-to-video evaluation. Its table lists MiniMax H3 at 1,141 Elo and FLUX 3 Max at 1,139; fal says their confidence intervals overlap. VentureBeat's Carl Franzen reported a slightly different Oct. 19 snapshot, with MiniMax H3 at 1,140 and FLUX 3 Max at 1,138, second by point estimate and with a possible rank of first through fifth. The ranking concerns image-to-video, not text-to-video.
Black Forest Labs describes voices, physical realism, sound effects and unusual prompts as strengths of the model. Those are the developer's assessments, rather than results established by the image-to-video ranking.
