Amazon sued by YouTube creators over training Nova Reel on third-party videos

Three YouTube creators filed a collective lawsuit against Amazon in federal court in Seattle. They claim the company downloaded their videos from YouTube without permission and used this material to train Nova Reel — Amazon's own video generation model available through Amazon Bedrock.

Who filed the lawsuit

The plaintiffs are Ted Entertainment — the company behind the channels h3h3 Productions and H3 Podcast Highlights — as well as golf channel creator MrShortGame Golf Matt Fischer and the Golfholics team. According to the lawsuit, they collectively have more than 2.6 million subscribers, approximately 4 billion views, and over 5,800 original videos on YouTube. The lawsuit was filed on April 3, 2026 in the federal district court for the Western District of Washington. Formally, this is an attempt to turn the case into a class action on behalf of a broader group of content creators. The plaintiffs are seeking compensation, recovery of unlawfully obtained profits, and an injunction that could affect the further distribution of the model.

"If such circumvention of protection remains unpunished, authors will be less likely to publish their work on

YouTube".

How the scraping is described

The key claim centers around section 1201 of the DMCA — a provision about circumventing technical protective measures. The authors are not merely disputing the fact of viewing public videos. Their position is stronger: Amazon allegedly used automated download tools, virtual machines, and IP address rotation to massively extract video files and circumvent YouTube's limitations. The complaint specifically mentions academic datasets HD-VILA-100M and HD-VG-130M. They contain not the videos themselves, but links to YouTube videos. According to the plaintiffs, in order to convert such URL lists into training data for the model, Amazon had to download actual files from the platform — and this step, in their view, violates the law and the platform's terms of service.

• The plaintiffs believe YouTube already contains technical barriers to mass video extraction.
• Amazon, according to the lawsuit, circumvented these barriers using automated infrastructure.
• The dispute is not only about copyright, but also about the method of data acquisition.
• A separate requirement is to prohibit the use of the model if it was trained on the disputed set of videos.

Another important argument of the plaintiffs concerns the irreversibility of training. They argue that once content is incorporated into a model, it cannot simply be "removed backward", so the damage is not limited to the fact of downloading. This makes the case uncomfortable for Amazon: even if the disputed dataset is no longer used, questions about the already trained version of Nova Reel do not disappear.

Why this matters for Amazon

Nova Reel is Amazon's video model, unveiled in December 2024 and available through Bedrock. According to AWS documentation, it can generate videos from text and images, supports clips ranging from 6 seconds to 2 minutes, and version 1.1 added multi-frame mode and better style consistency between scenes.

For Amazon, this is not a side experiment, but part of a major bet on enterprise generative AI. Against this backdrop, the lawsuit strikes not only at reputation. It raises questions about the origin of the data the model may have been trained on.

If the court accepts the plaintiffs' logic, the risk will arise not only for Amazon, but for any company that took academic datasets of YouTube links and then mass downloaded the original videos to train video models. This is also part of a broader series of lawsuits. The same plaintiffs have already filed similar complaints against Nvidia, Meta, ByteDance, Snap, OpenAI, and Apple.

The overall strategy is clear: attack not abstract training on the open internet, but the specific technical mechanism through which companies allegedly circumvented platform protection and turned publicly available videos into commercial AI assets.

What it means

This case could become one of the most significant disputes over video data for AI. If the court agrees that bulk downloading videos from YouTube for model training violates the DMCA even when the content is publicly available, developers will need to reconsider not only datasets, but the entire pipeline for collecting data for generative video.