Black Forest Labs announces FLUX 3 for images, video, audio and robotics

Black Forest Labs introduced FLUX 3 on July 23, 2026. Its Self-Flow architecture combines image, video and audio generation with action prediction, producing up to 20 seconds of video with native sound. FLUX-mimic is designed to cut robot manipulation training data collection from 30 hours to 30 minutes and was being tested by Audi. Video and Action launched in early access, with FLUX 3 Dev open weights planned for the second half of 2026.

Visit original source