Amazon upgrades its AI video model Nova Reel and features the ability to generate video for up to 2 minutes.
Nova Reel, announced in December 2024, was Amazon’s first foray into the generated video space. It competes with Openai, Google and other models.
The latest Nova Reel, Nova Reel 1.1, can generate “multi-shot” videos with “consistent style” throughout the shot, AWS developer advocate Elizabeth Fuentes explained in a blog post. Users can provide prompts up to 4,000 characters long to generate a video of up to 2 minutes, consisting of 6-second shots.
Nova Reel 1.1 also introduces a new mode called “Multishot Manual.” In this mode, the model can refer to the image and prompts to give more control over the configuration of the video shot. According to Fuentes, considering the 1280 x 720 resolution image and the 512-Maximum-Character prompt, the Multishot Manual can generate videos with up to 20 shots.
Nova Reel is only available through Amazon’s AI Dev Suite and Bedrock, and customers must request access. Like most generation AI systems, there is doubt as to whether reels were developed using ethically sound methods.
Video Generation Models are trained with a huge number of video examples to generate new clips to “learn” the patterns of these videos. Some companies train copyrighted videos models without permission from their owners or creators, and expose the users of the models to IP lawsuits when these models “reverse” copyrighted stills.
Amazon has not revealed the source of Reel’s training data and does not provide a clear way for creators to opt out of video to their model’s dataset. However, the company says it will protect AWS customers accused of violating copyright in media generated by the model, in line with its compensation policy.
Updated at 4:57pm Pacific: Previous versions of this article imply that developers must request special access from Amazon to use Nova Reel. That’s wrong. Developers need to request access, but AWS automatically approves these requests. I regret the error.
Source link