Amazon has unveiled an updated AI model, Nova Reel 1.1, that can generate videos up to two minutes long based on text prompts from users. Its previous version, Nova Reel, was announced in December 2024, marking the company’s first foray into the generative AI video-making market.
Image source: Amazon
Nova Reel 1.1 can generate “multi-frame” videos with a “consistent style” between frames, AWS developer advocate Elizabeth Fuentes wrote in a blog post. Users can provide a text prompt of up to 4,000 characters to generate a two-minute video made up of six-second frames.
The updated model also has a new mode called Multishot Manual. In this mode, the model can use an image along with a hint to better match the video frame composition request. Given an image with a resolution of 1280 x 720 pixels and a hint with a maximum of 512 characters, the model can generate videos containing up to 20 frames, Fuentes said.
Nova Reel models are only available on AWS platforms and services, including Bedrock, the company’s AI development toolkit, and customers must request access.
Video-generating models are trained on huge numbers of video samples to “learn” patterns for creating new clips. Some companies train models on copyrighted videos without permission from the owners or creators. As a result, these models can “spit out” copyrighted footage, exposing users of the models to the risk of being sued for intellectual property infringement.
Amazon did not disclose the source of the data used to train Nova Reel, but promised that it would defend AWS customers if they were accused of copyright infringement under its indemnification policy.