fal speeds up FLUX 3 to near real-time

fal Research and Black Forest Labs released FLUX 3 Max, a fine-tuned version of the FLUX 3 video model. According to fal's measurements, it generates a 10-second 720p video with audio in 12.45 seconds, compared to 277.8 seconds for Seedance 2.5, and costs $0.425 per video until November 7, 2026.

Compiled by Sergey KostenchukPublished 2026-10-21Updated 2026-10-21

2026-10-21 Multimodal

🎨 fal speeds up FLUX 3 to near real-time

fal Research and Black Forest Labs have released FLUX 3 Max, a fine-tuned version of the FLUX 3 video model that generates videos with audio from text or image prompts. According to fal's measurements, a 10-second 720p video is ready in 12.45 seconds, compared to 277.8 seconds for Seedance 2.5—approximately 22 times faster.

🌍 The improvement came from post-training, not architecture: Elo on the AA-Video-I2V arena rose from 1089 to approximately 1139, while latency dropped from 126.44 seconds to 12.45 seconds. Production iterations are becoming cheaper, and competition may shift toward speed—though this is currently confirmed only by fal's measurements.

👤 Until November 7, 2026, one second of 720p video costs $0.0425—a 10-second video with audio costs $0.425 on fal.ai, rising to $1.70 after the promotion. It is suitable for quick scene tests, portrait animation, and product videos.

Source 1: https://fal.ai/flux-3-max Source 2: https://fal.ai/models/blackforestlabs/flux-3-max/image-to-video

Sources