FLUX 3 Max

FLUX 3 Max is a video generation model launched by fal Research in October 2026, built in collaboration with Black Forest Labs (BFL). It is a post-trained version of BFL's FLUX 3 foundation model: fal took the FLUX 3 weights and applied its own post-training and inference optimizations to push both quality and speed.

It is available on fal for text-to-video, image-to-video and keyframes-to-video generation. It follows the same recipe as fal's earlier H3 Max, which was post-trained on MiniMax H3 open weights.

Key points

• Who made it: fal Research (the research arm of the generative-media platform fal.ai), together with Black Forest Labs, the lab behind the FLUX model family.
• What it is: not a new base model but a post-trained FLUX 3. The base model's research comes from BFL; fal adds post-training plus inference optimization.
• Speed: in fal's own benchmark it generates a 10-second video in 12.45 seconds, versus 277.80 seconds for Seedance 2.5, which fal describes as 26.5× faster with 96.2% lower latency.
• Quality: in Artificial Analysis's image-to-video evaluation, fal reports a 47-point Elo improvement over the original FLUX 3, placing it alongside the highest-ranked video models.
• Ongoing: fal says it will keep post-training FLUX 3 Max, so quality and speed may keep changing.

How it relates to other names

• FLUX 3: BFL's foundation model that FLUX 3 Max is derived from.
• FLUX 3 Image Klein: a separate, smaller FLUX 3 image model from BFL.
• H3 Max: fal Research's first post-trained frontier video model, built on MiniMax H3.

All benchmark figures above are fal's own claims from its launch post.

How to use it

1. Read the launch post: Introducing FLUX 3 Max by fal.
2. Sign in at fal.ai and open the FLUX 3 Max model page from the post (text-to-video, image-to-video or keyframes-to-video).
3. Try it in the browser playground first, then call it through the fal API or SDK (Python / JavaScript) using your fal API key.
4. For image-to-video, upload a still image plus a text prompt; for keyframes-to-video, provide start and end frames.

Good fits

• Fast iteration on ad clips, social video and storyboards where latency matters.
• High-throughput pipelines that need many videos quickly.

Tips

• Check fal's pricing page before batch jobs, since video generation is billed per output.
• Compare against Seedance 2.5 or other models on your own prompts; vendor benchmarks don't always match your use case.