Amazon’s AI video model can now produce minutes-long clips

Amazon has a wide range of products. Nova Reel was upgraded to be able to generate videos of up to two minutes. Nova Reel was Amazon’s first attempt at generative video. It was announced in December 2024. It competes against models from OpenAI and Google in a market that is rapidly becoming crowded.

According to AWS developer advocate Elizabeth Fuentes, the latest Nova Reel can generate “multishot” videos that have a “consistent” style across shots. Users can enter a prompt of up to 4,000 characters to create a video that is composed of six-second clips.

Nova Reel also introduces a “Multishot Manual” mode. In this mode, a model can reference an arbitrary image and a prompt in order to give users more control over the composition of a video shot. According to Fuentes, given a 1280 x 720-resolution image and 512-maximum-character prompt, Multishot Manual can generate videos containing up to 20 shots.

Nova Reel can only be accessed through AWS platforms, including Bedrock – Amazon’s AI development suite – and customers have to request access. Reel, like most generative AI systems is generating questions about whether it was developed ethically.

Video generating models are “trained” on a large number of videos in order to “learn”the patterns of these videos to create new clips. Some companies train models using copyrighted videos, without obtaining the permission of the owners or creators. When these models “regurgitate”copyrighted stills are exposed to IP lawsuits.

Amazon does not reveal the source of Reel’s training data nor does it provide a way for creators to opt out. According to its indemnification policy, the company has said it will protect any AWS users accused of violating copies rights with media generated by their models.

Updated at 4:57 p.m. Pacific. An earlier version of this post implied that developers had to request special permission from Amazon to use Nova Reel. This is incorrect. While developers must request access, AWS automatically accepts these requests. We regret the error.

Kyle Wiggers, TechCrunch AI Editor. His writings have appeared in VentureBeat, Digital Trends and a variety of gadget blogs, including Android Police and Android Authority, Droid-Life and XDA-Developers. He lives in Manhattan, with his music therapist partner.

View Bio

www.aiobserver.co

More from this stream

Recomended