Rights in the Source Image You Animate: Permissions Required Before Generation

The most common mistake is assuming that animating a photograph you found online is fine because you are not copying the whole image, you are just making it move. That assumption is wrong. An image-to-video model takes your source image and produces a new clip built on top of it. Under copyright law in the US, UK, and EU, that generated video is a derivative work of the source image. Creating a derivative work without permission from the copyright holder is infringement unless an exception like fair use applies, and fair use is a defence you raise in court, not a right you have at the keyboard. You need permission in advance. What follows covers the permissions you require before you hit generate, what happens when you animate a photograph you did not take, and where the consent obligations for animating a real person's face sit in the workflow, not in a footnote.

If you did not press the shutter, you do not own the copyright in the source image. The person who took the photograph, or their employer if it was a work for hire, holds the exclusive right to prepare derivative works. Running a still through Runway Gen-3 Alpha, Kling 1.6, Luma Dream Machine, or any other image-to-video model creates a derivative work. You need a licence from the copyright holder that explicitly permits modification, not just display or sharing. A standard social-media terms-of-service grant does not cover this.

Fair use is sometimes raised as a justification for generative use of found images, but it is a narrow defence. US courts consider four factors: purpose of the use, nature of the copyrighted work, amount used, and effect on the potential market. Animating a news photograph for commentary may weigh differently than animating a fine-art portrait for a commercial. In the UK and EU, fair dealing provisions are narrower still and do not include a general transformative-use exception. The safest path is to obtain a licence to modify and create derivative works before you generate.

Unsplash and Pexels: The Free-Image Trap

Stock image platforms like Unsplash and Pexels have become a primary source for source images, but their licence terms have shifted on AI generation. Unsplash historically offered a blanket CC0 dedication to the public domain, no rights reserved. That changed. Unsplash now prohibits use of its images for generative AI model training or derivative generation. The current Unsplash terms require you to check the specific licence on the image; CC0 images from before the change may still be usable, but the platform's terms of service now restrict AI generation on newer uploads. Downloading an Unsplash image and animating it without verifying the licence risks breach of contract even if the copyright has been waived.

Pexels operates under a similar model. Its licence permits free use and modification for most commercial and non-commercial projects, but the Pexels licence explicitly states that you may not use images to create a competing image service or to train AI models. Whether animating a single image into a video clip counts as 'training' a model is contested; the safer reading is that the licence to modify does not extend to AI generation. A CC0 dedication on a Pexels image is only valid if the image carries a true CC0 dedication that the platform has not retroactively restricted, which is legally uncertain.

Paid Stock: What Your Licence Actually Covers

For commercial exploitation, you need a licence that explicitly permits derivative works via AI generation. Standard royalty-free licences from Getty, Shutterstock, and Adobe Stock do not automatically grant this right. You must read the End User Licence Agreement for the specific image and look for clauses on 'AI generation', 'automated processing', or 'derivative works via machine learning'. If the licence is silent, the default rule applies: permission has not been granted.

Copyright in the image is separate from the right of publicity of the person depicted. A stock image may come with a model release, a signed consent form allowing commercial use of the person's likeness, but that release was signed for the specific purpose stated in the licence. Animating that person's face with an AI video model and causing them to move, smile, or perform actions they did not consent to may exceed the scope of the original release. The consent obligation does not disappear because the image was licensed for commercial use. If the release mentions only static display and not animated or generated content, you need updated consent.

This applies even if the person is deceased. The right of publicity survives death in many US states, typically for 50 to 100 years after death. Animating a deceased person's photograph for a commercial video, such as a historical figure in a brand advertisement, requires consent from their estate. If you animate a photograph you did not take of a person you do not know, you are operating without either copyright clearance or consent. That combination is the most common source of legal exposure in this space.

Every major image-to-video platform requires you to warrant that you hold the necessary rights to the input image. Runway's terms require you to warrant rights and state that you own the output but that Runway may use it for service improvement. Kling 1.6's terms require the same warranty and grant Kuaishou a licence to use your output for service improvement. Luma Dream Machine, Pika 2.0, Hailuo Minimax, Seedance, and OpenAI Sora all include a similar warranty clause. Google Veo 2's terms require you to hold necessary rights and state that Google claims no ownership of the output. These clauses shift liability to you: if the copyright holder of the source image sues, the platform points to your warranty.

The input rights requirement is not a suggestion. Platforms enforce it through content moderation filters that reject images containing recognisable branded products, copyrighted characters, or identifiable public figures without a release. If your source image is a photograph you took yourself of a subject you control, you satisfy the requirement. If it is a photograph you did not take, the warranty is a promise you cannot keep.

Why a Static-Image Release Is Not Enough

Animating a real person's face requires two separate permissions: copyright clearance on the image and a model release that covers the generated use. A model release for a static photograph does not automatically extend to animation. If you animate someone's face and the generated clip shows them performing an action, expressing an emotion, or appearing in a context they did not approve, you risk a claim for misappropriation of likeness or false endorsement. The consent obligation is not a minor detail; it belongs in the body of this page because it is as important as the copyright question.

Moral Rights and Jurisdiction

Moral rights add another layer. In the EU and UK, the photographer and the subject have moral rights that include the right to integrity, the right to object to distortion or mutilation of the work. Animating a photograph in a way that changes the subject's expression or context could be a distortion that violates moral rights, and those rights cannot be waived in all jurisdictions. Jurisdiction matters: what is permissible under US fair use may still violate moral rights in France or Germany.

If you plan to sell the generated video or use it in client work, the commercial exploitation right in the source image is the critical permission. A licence that permits personal or editorial use often excludes commercial use. Stock images on free tiers of Kling, Pika, Luma, and Runway are licensed for non-commercial use only; the paid tiers grant commercial use. Even on paid tiers, the underlying source image must itself be cleared for commercial use. Animating a personal photo of a friend and selling the clip requires their consent, not just your copyright ownership of the image.

Attribution requirements may also apply. Some CC0 licences require attribution even though no permission is needed. If you animate a CC0 image that requires credit, the generated video must include that credit. Failing to attribute does not create copyright liability, but it breaches the licence terms and the platform may remove the content.

Transformative use is a concept within fair use that asks whether the new work adds something new with a different purpose or character. Animating a still image is often described as transformative because it adds motion and time. But transformative use is a factor in the fair use analysis, not a standalone right. A court will weigh it against the other factors, including market harm. If the animated clip is used in a way that competes with the original photograph's market, for example selling the clip as stock video, the transformative argument weakens. You cannot rely on transformative use as permission before you generate. You need permission from the copyright holder.

The single thing that most often goes wrong here is the assumption that an image is free to use because it is available. A search engine result, a social media post, or a free stock site does not grant the right to create derivative works. If you animate a photograph you did not take, you are creating a derivative work without a licence. The platform's terms of service require you to warrant you hold the rights. If you cannot make that warranty truthfully, the generated video is an infringement. The only safe source images are ones you photographed yourself, ones with a licence that explicitly permits AI generation, or ones dedicated to the public domain under a licence that has not been retroactively restricted. Any other source carries legal risk.