Amazon has enhanced its AI video model, Nova Reel, enabling it to generate videos up to two minutes long.
First announced in December 2024, Nova Reel marked Amazon’s initial entry into the generative video space. It competes with models from OpenAI, Google, and others in an increasingly competitive market.
The updated Nova Reel, version 1.1, can create “multi-shot” videos that maintain a “consistent style” across shots, explained AWS Developer Advocate Elizabeth Fuentes in a blog post. Users can input a prompt of up to 4,000 characters to generate a two-minute video made up of six-second segments.
Nova Reel 1.1 also introduces a new mode called “Multishot Manual.” This mode allows the model to reference an image along with a prompt to give users more control over the composition of each video shot. According to Fuentes, when provided with a 1280×720-resolution image and a prompt up to 512 characters long, Multishot Manual can produce videos with as many as 20 shots.
Nova Reel is available exclusively through AWS platforms and services, including Bedrock, Amazon’s AI development suite, and requires customers to request special access. As with most generative AI systems, concerns remain about whether Reel was developed using ethically sound practices.
Video-generation models are trained on massive amounts of video examples to learn patterns that allow them to create new clips. Some companies use copyrighted videos without obtaining permission from the rights holders, and when these models “regurgitate” copyrighted material, users may face intellectual property lawsuits.
Amazon has not disclosed the sources of Reel’s training data, nor has it offered a clear way for creators whose content may have been used to opt out. However, the company has stated that it will defend AWS customers accused of copyright violations involving media generated by its models, in line with its indemnification policy.
