Today, fal Research launches FLUX 3 Max, our latest post-trained video generation model, in collaboration with Black Forest Labs. FLUX 3 Max delivers #1-tier image-to-video quality, statistically tied with the highest-ranked state-of-the-art video models, while generating video at a fraction of the latency. By combining Black Forest Labs' foundation model research with fal Research's post-training and inference optimizations, FLUX 3 Max pushes the speed–quality frontier forward.
Frontier Quality, a Fraction of the Latency
FLUX 3 Max delivers quality competitive with the highest-ranked video models, without the long generation times. In our benchmark measurements, FLUX 3 Max generates a 10-second video in 12.45 seconds, compared with 277.80 seconds for Seedance 2.5. That's 26.5× faster, with 96.2% lower latency, while delivering competitive visual quality.
These speed improvements enable faster iteration, more responsive creative workflows, and higher-throughput video generation.
Improving on the Original FLUX 3
Through post-training and inference optimization, FLUX 3 Max improves on the original FLUX 3 in both generation quality and speed. In Artificial Analysis's image-to-video evaluation, FLUX 3 Max achieves a 47-point Elo improvement over the original FLUX 3, placing it alongside the highest-ranked models. The improvements extend beyond quality. fal Research has also optimized inference performance to reduce generation latency, enabling faster video generation without sacrificing visual fidelity.
What Frontier AI Collaboration Makes Possible
FLUX 3 Max builds on the research and engineering breakthroughs behind H3 Max, fal Research's first post-trained frontier video model and the fastest frontier video model in the industry. Post-trained on MiniMax H3 open weights, H3 Max demonstrated how combining post-training with highly optimized inference infrastructure could dramatically improve model performance. With FLUX 3 Max, fal Research is applying the same approach to Black Forest Labs' FLUX 3 foundation model, delivering further improvements in visual quality and generation speed. The collaboration demonstrates what's possible when frontier model developers provide access to their weights, enabling specialized research teams to optimize models beyond their initial release.
fal Research will continue to post-train FLUX 3 Max, with continued performance improvements underway.
Try FLUX 3 Max on fal
FLUX 3 Max is now available on fal for text-to-video and image-to-video generation.
Try it now:
text-to-video↗
image-to-video↗
keyframes-to-video↗
