Black Forest Labs and fal launch FLUX 3 Max video model: 10-second clips in about 11-12 seconds, joint #1 in Artificial Analysis' preliminary image-to-video ranking
0 reads
- Published
- 2026-10-20
- Type
- Model release
- Organization
- Black Forest Labs / fal
- Source
- agipulse
Why it matters
【Key points】
Black Forest Labs (BFL) announced FLUX 3 Max, a video generation model, on its official X account on Oct 19 (18:51 UTC). According to fal's model pages, FLUX 3 Max is built on BFL's FLUX 3 foundation model from July, with post-training and inference optimization by fal Research in collaboration with BFL. It is available on fal through text-to-video, image-to-video (optional first and last frames) and keyframes-to-video endpoints.
Specs: clips of 5 to 20 seconds, 720p by default with optional 1080p upscaling, optional generated audio, aspect ratios from 21:9 to 9:16, and a low/high "reasoning level" (high expands the prompt first).
Speed and ranking: BFL says a 10-second video takes 12.45 seconds, 22× faster than Seedance 2.5 at the same Elo. fal's page says about 11 seconds, more than 25× faster than Seedance 2.5 (277.8 s), and joint #1 in Artificial Analysis' preliminary image-to-video evaluation. fal notes the joint #1 applies to image-to-video, not text-to-video.
Pricing: billed per generated second, $0.17/s at 720p and $0.29/s at 1080p. A 75% promotional discount ($0.0425 and $0.0725) runs until Nov 7.
【Verifiable】
BFL announcement: https://x.com/bfl_ai/status/2112255353677328415
fal text-to-video: https://fal.ai/models/blackforestlabs/flux-3-max/text-to-video
fal image-to-video: https://fal.ai/models/blackforestlabs/flux-3-max/image-to-video
【Our read】
1. The pitch is speed: a 10-second clip in about as long as it plays brings video models close to an edit-and-see loop, which matters for ads and storyboards built on many retries.
2. Letting an inference platform post-train and launch its base model is an unusual move for a frontier lab, and backs BFL's claim that FLUX 3 was built for customization.
3. So far there are only vendor speed numbers and a "preliminary" Artificial Analysis ranking: no technical report and no open weights. List prices are not cheap, so real adoption will show after the discount ends.
Related developments
Aggregated by AGI Pulse. Titles and links only; no full text is republished. Editorial policy