How do AI Video Generators manage copyright issues?

AI Video Generators

Artificial intelligence is transforming the way we produce content. However, the question of how copyright applies to this new type of creative work has yet to be determined. Until it is, it is likely that generative AI companies will face a lot of lawsuits from creators who feel their works have been stolen.

A common complaint is that AI Video Generator generate images in the style of other artists without their permission. This is because to create these images, the AI model has to be “trained” (a process referred to as “machine learning”) on large quantities of existing art. In the case of an image-generating AI like Stability AI, this is achieved by showing it hundreds of millions of existing images and text descriptions, often sourced from the internet, which are then used to create an algorithm for producing new artworks in a particular style. The artists who have filed a class action against these companies claim that they were wrongfully shown their work to train the AI software, and that the subsequent images produced in their style compete with their own — all of which violates copyright law.

Similar issues have arisen over the production of music with generative AI tools. Although the resulting works are unlikely to be protected by copyright as they do not contain any elements of human creativity, rightsholders may be able to bring actions for copying in relation to the training data on which the AI tool has been trained. In particular, if the statutory exception for “text and data mining” (“TDM”) (s. 29A CDPA) was used in a jurisdiction that does not prohibit TDM for commercial purposes, and if the AI tool was trained on a collection of music that was not authorised by the rightsholders, then there is a risk that the entity responsible for the TDM could be held liable for copyright infringement.

How do AI Video Generators manage copyright issues?

Firstly, many AI Video Generator platforms implement robust content filtering mechanisms during the training process. These filters identify and exclude copyrighted material from the training dataset, ensuring that the model does not learn to replicate protected content. Additionally, metadata analysis tools may be utilized to cross-reference video content with copyright databases, flagging potentially infringing material for review or exclusion.

Another approach involves incorporating copyright detection algorithms directly into the AI Video Editor itself. These algorithms analyze generated content in real-time, comparing it against copyrighted works to identify potential violations before the output is finalized. If copyrighted elements are detected, the generator may offer suggestions for modification or replacement to mitigate infringement risk.

Whether TDM can be a defence to copyright claims will depend on how the courts see the relationship between the creator of the AI tool and the user of it. In the case of a prompt-based AI image generator, it is possible that the user inputting the text prompts could be seen as the author of the resulting work, because they are engaging in an act of creativity in creating the prompts themselves. If the courts do view the users of these AI tools as authors, it would mean that any future legal battles over AI-generated works could be fought on much more solid ground than in the US. In addition, it will also be easier for the industry to develop licences to enable the creation of AI-generated content.

Leave a comment

Your email address will not be published. Required fields are marked *