Black Forest Labs launches FLUX 3 Video, claims SOTA in text-to-video and ties top model in image-to-video
AI video model releases
Black Forest Labs launches FLUX 3 Video, claims SOTA in text-to-video and ties top model in image-to-video
Black Forest Labs made FLUX 3 Video generally available via its API and select partners, generating clips up to 20 seconds long in HD, with Full HD available through upscaling and native audio generated alongside video.
Internal ELO evaluations show FLUX 3 leading all-vs-all text-to-video comparisons with a score of 1135, outperforming existing SOTA models by a solid margin.
In image-to-video, FLUX 3 ties Seedance 2.0 and beats all other tested SOTA models, based on human rater preference.
The model supports Video Continuation (extending up to 4 seconds of existing video and audio), multi-shot generation with coherent scene/camera switches, and dialogue lip-syncing across languages including English, Chinese, Spanish, French, German, Japanese, Hindi, and Punjabi.
Black Forest Labs used third-party partner Cinder to evaluate the model for misuse risks including non-consensual intimate imagery (NCII) and CSAM before release, and previewed upcoming FLUX 3 Image and open-weight FLUX 3 Dev variants.
cho.sh