Amazon Nova Reel

Amazon Nova Reel appeared on 3 December 2024 at re:Invent as the video counterpart to Nova Canvas, and it grew in a way the rest of the family did not. Version 1.0 could produce a single six-second shot. Version 1.1, released the following year, kept the same six-second unit but let it be chained: up to two minutes of video, built in six-second increments at 1280x720 and 24 frames per second. Amazon calls the longer mode multi-shot, and it comes in two forms — an automated one, where a single prompt of up to 4,000 characters is broken into shots by the model, and a manual one, where the author writes each shot separately with 512 characters apiece. Single-shot text-to-video is capped at 512 characters. Prompts are English only, and a still image can be supplied as the opening key frame. The two versions also differed in reach, which matters for anyone reading a benchmark. Reel 1.0 ran in Northern Virginia, Ireland and Tokyo; Reel 1.1, the only one that can exceed six seconds, ran solely in Northern Virginia. The model is asynchronous by design — generation is started through the async invoke API rather than a blocking call — and supports neither fine-tuning nor provisioned throughput. On provenance Amazon took the more open route here than with Canvas: output from Reel 1.1 carries Content Credentials, the C2PA metadata standard, and can be checked with public verification tools rather than an Amazon-only detector. Both versions were moved to Legacy on 30 March 2026 and are switched off on 30 September 2026. As with Canvas, no Nova 2 video model has been announced to take over.

▸Videos

No videos yet.

▸Reviews