April 8, 2011 - Last December's debut of theAmazon AI Video Model Nova Reel Nova Reel 1.1 has been upgraded to produce videos of up to two minutes in length.
Nova Reel 1.1 is capable of generating "multi-camera" videos with a "consistent style," according to a blog post by Amazon Web Services (AWS) developer advocate Elizabeth Fontes.Users can provide cues up to 4,000 characters long to generate videos up to two minutes long consisting of six seconds of footage.
1AI notes that Nova Reel 1.1 also introduces a new mode called Multishot Manual. In this mode, theThe model can be combined with an image and cue words to generate video footageThis gives the user more control over the composition of the video shots. Fontes says the Multi-Camera Manual mode is capable of generating videos with up to 20 shots when supplied with an image with a resolution of 1280×720 and a cue with up to 512 characters.
Currently, Nova Reel is only available through AWS platforms and services, including Amazon's AI development suite, Bedrock, and customers need to request access.
However, as with most generative AI systems, there are questions about whether Nova Reel was developed using ethical methods. Video generation models are typically trained on a large number of video examples to "learn" the patterns in those videos and generate new video clips. Some companies use copyrighted videos to train their models without the permission of the copyright owner or creator, exposing the user of the model to intellectual property lawsuits when the model "replicates" a copyrighted still image.
While Amazon has yet to reveal the source of Reel's training data or provide a clear opt-out for creators who may have contributed videos to the model's dataset, the company has said it will provide protection under its indemnification policy for any AWS customer accused of copyright infringement through the use of media generated by its models.