Amazon has advanced its AI video technology, Nova Reel, empowering the model to create videos lasting up to two minutes.
Initially unveiled in December 2024, Nova Reel marked Amazon’s entry into the burgeoning generative video space, competing with significant players like OpenAI and Google.
The newly upgraded version, Nova Reel 1.1, can produce “multi-shot” videos with uniform stylistic integrity across scenes, as detailed by AWS developer advocate Elizabeth Fuentes in a recent blog post. Users are allowed to input prompts as lengthy as 4,000 characters to craft videos comprised of six-second segments.
A noteworthy addition to Nova Reel 1.1 is the “Multishot Manual” mode, designed to enhance user control over video composition. This mode enables the model to incorporate both prompts and reference images, offering a tailored approach to shot creation. When using a 1280 x 720-resolution image alongside a 512-character prompt, Multishot Manual can generate videos with up to 20 distinct shots.
Currently, Nova Reel is accessible solely through AWS platforms, such as Bedrock, Amazon’s AI development suite, and is subject to a request-based access policy.
However, as with many generative AI systems, Nova Reel’s development raises ethical considerations, particularly concerning its training data.
Video-generating AI models rely on extensive collections of video examples to learn from patterns and produce new content. Issues arise if these models use copyrighted materials without prior authorization, posing a risk of intellectual property disputes for end users.
Although Amazon has not disclosed the origins of Nova Reel’s training data or provided an opt-out mechanism for creators, the company assures its customers of protection against copyright infringement allegations through its indemnification policy.
