Rights, Rules, and Disclosure Obligations for AI-Generated Video

You made an AI-generated video. You likely do not own the copyright in it, you may not be allowed to sell it, and you almost certainly have to label it when you upload it. The legal landscape is a patchwork of agency policy, platform terms, and legislation that applies differently depending on where you are, what model you used, and what you plan to do with the clip. This page routes you by your situation to the guide that names the exact legal instrument, jurisdiction, and effective date. Where the law is unsettled, the child page presents both sides.

Type a prompt and the model generates a clip. The US Copyright Office will not register it. Its policy statements and registration decisions since 2019 have held that copyright vests only in works created by a human being. Thaler v. Perlmutter, the federal court case that tested this principle against the Copyright Office's refusal to register an AI-generated work, affirmed that human authorship is a constitutional requirement. The UK IPO's position on computer-generated works without a human author is distinct. It has a provision in the Copyright, Designs and Patents Act 1988 that could apply, but it is under active consultation and uncertain.

What Counts as Human Authorship

Guides in this section cover the specific arguments: whether selecting a source image and a prompt is enough creative input, whether arranging multiple clips into a sequence creates a registrable compilation, and whether the degree of control in the latest models changes the analysis. Each guide states the jurisdiction, the decision or statute it relies on, and the date from which it applies.

The question "Can I sell this video?" is answered by the model's terms of service or open-source licence, not by your local copyright office. Runway Gen-4, Kling 2.1, Sora, Pika 2.2, Luma Dream Machine, Hailuo I2V-01, and Seedance 1.0 all permit commercial use with a paid subscription. Veo 3 permits it via a Google Cloud agreement. Wan 2.1 and LTX-Video 0.9.5 ship under Apache 2.0, which allows commercial use. Stable Video Diffusion 1.1 ships under the Stability AI Non-Commercial Research Community Licence. It does not permit commercial use.

The Open-Weights Trap

A model that releases weights but not training data or training code is open-weights, not open-source. The licence is the sole document that governs commercial use. Build a product around an open-weights model with a non-commercial licence, and you are infringing before you make your first sale. The guides in this section name the licence version and the date of the terms for each model. Where a model's terms of service contain a commercial restriction clause (the typical buried restriction in unlimited-generation tiers), the guide quotes the relevant language.

Every commercial model's terms of service says "you own your outputs, subject to these terms." That ownership is a contractual grant, not a copyright grant. It means the platform will not sue you for using the output. It does not mean the US Copyright Office will register it. The practical consequence: you can use the clip in a YouTube video or an advertisement, but you cannot sue someone who copies it, because you hold no exclusive rights.

The guides in this section distinguish between output ownership as a contractual right and copyright ownership as a statutory right. They also cover the situation where the source image is itself protected: using a frame from a film, a photograph from a stock site, or a piece of fan art, where the model generates a derivative work that may infringe.

Below is the commercial-use licence status for every model in the fact sheet, using the terms in force as of 2026-09-26. The date matters: licences change with point releases. Each child page in this section carries the date of the terms it documents.

Image-to-video generation takes a single still and produces a short clip. The source image has its own rights chain. If you did not create it, you need permission from the rights holder, a licence, a fair use argument, or a public-domain status. The model's terms of service do not grant you rights to the input. They only govern what you may do with the output.

Animating People Requires Consent

Before you animate a photograph of a real person, secure their consent. The model's terms do not cover this. A right of publicity claim can follow you regardless of the platform's rules. The guides in this section cover the specific scenarios: animating a screenshot from a video game (which may violate the game's terms of service), animating a friend's photograph (which raises a right of publicity concern), and animating a frame from a film (which is a derivative work that likely infringes). Each guide states the jurisdiction's law on derivative works and the consent obligations that apply.

Model Training Data Rights

Your source image may have been used in the model's training data without the photographer's consent. Several class-action lawsuits in the US and the EU AI Act's training-data transparency provisions address this. The child page on this question names the specific cases and the dates of the relevant filings.

YouTube, TikTok, Facebook, Instagram, and X each have a synthetic or altered material disclosure policy. The threshold for triggering the label differs. YouTube's Creator Studio requires an "altered or synthetic content" label when the material is realistic and a reasonable person could mistake it for real. TikTok's synthetic media policy requires a label for any AI-generated material that is realistic, regardless of intent. Facebook and Instagram use a combination of metadata detection and self-declaration.

The guides in this section name the specific UI location where the label is set, the date the policy came into effect, and the enforcement mechanism (demonetisation, reduced distribution, removal). They also cover the mismatch between model-level watermarking and platform-level detection: a model's metadata watermark may satisfy its own compliance but the platform may use a different detection system that does not read that format.

Disclosure Threshold and the EU AI Act

The EU AI Act imposes transparency obligations on deployers of AI-generated video. If your video is published in the EU, you must disclose that it is AI-generated or manipulated. The disclosure must be clear and prominent. The threshold is not "is it realistic?" but "is it generated or manipulated by an AI system?" The child page on this question states the relevant articles, the application dates, and the penalties for non-compliance.

If your goal is to clear a generated clip for broadcast or advertising, start with the guide on commercial-use terms. That page names the exact licence clause for each model and the jurisdiction's rules on derivative works.

If your goal is to register a copyright in a generated clip, start with the guide on the human authorship requirement. That page states the US Copyright Office's policy, the Thaler v. Perlmutter holding, and the UK IPO's alternative position.

If your goal is to comply with platform disclosure rules before uploading, start with the guide on platform synthetic material policies. That page names the specific UI element and the threshold for each platform.

If your goal is to check whether you have the rights to the source image you plan to animate, start with the guide on source image rights. That page covers fair use, derivative works, right of publicity, and the consent obligation by jurisdiction.