Amazon has enhanced its artificial intelligence video model, Nova Reel, to produce videos of up to two minutes in duration.
Introduced in December 2024, Nova Reel marked Amazon’s initial venture into the generative video space, where it competes with models developed by OpenAI, Google, and other players in a rapidly expanding market.
The updated Nova Reel, known as Nova Reel 1.1, is capable of generating “multi-shot” videos with a consistent style throughout, as explained by AWS Developer Advocate Elizabeth Fuentes in a blog post. Users can input a prompt of up to 4,000 characters to generate a video of up to two minutes, comprising six-second shots.
Nova Reel 1.1 also introduces a new feature called “Multishot Manual.” This mode enables the model to reference an image alongside a prompt, providing users with greater control over the composition of a video shot. According to Fuentes, given a 1280×720-resolution image and a 512-character prompt, Multishot Manual can generate videos containing up to 20 shots.
Access to Nova Reel is currently limited to AWS platforms and services, including Bedrock, Amazon’s AI development suite, and customers must request special access. As with most generative AI systems, concerns arise regarding the ethics of Reel’s development methodology.
Video-generating models are trained on vast amounts of video data to identify patterns and generate new clips. However, some companies train models using copyrighted videos without permission, which can lead to intellectual property issues when these models reproduce copyrighted content, exposing users to potential lawsuits.
Amazon has not disclosed the source of Reel’s training data, nor has it provided a clear opt-out mechanism for creators whose videos may be included in the model’s dataset. Nevertheless, the company has stated that it will protect AWS customers from copyright infringement claims related to media generated by its models, in accordance with its indemnification policy.
Source Link
