The following is a list of ten popular AI image generators and a detailed explanation of the relevant copyright regulations for their use.
Most information on this can be found in the terms and conditions/terms of use of the programs, although these are always quite extensive and not always easy to find. They also do not guarantee reliable legal certainty with regard to any copyright infringements. Instead, the risk of any legal infringements is regularly passed on to the user.
As far as possible, the origin of the training data, so-called data sets, is also discussed. The only service that actually discloses this is Stable Diffusion. For most image generators, however, the origin of the data is unclear. Scraping, i.e. the reading and saving of information from the Internet, can be assumed. There is no filtering, and in particular no consideration is given to possible copyright infringements.
Dall-E
Dall-E was developed by OpenAI – the same company that is responsible for the (also) sensational ChatGPT program. Version Dall-E 2 is now available, which achieves up to four times better resolution than Dall-E (1).
The name of the application is a combination of the artist Salvador Dalí and the movie robot Wall-E.
Dall-E states in its terms of use that the program may not be used in a way that infringes the rights of others. Whether and how this is done is not explained.
The generated image may also not be used in a way that creates the impression that the image is man-made. This gives the impression that OpenAI is trying to enforce a kind of machine copyright. This is not possible under German law.
Furthermore, the terms of use stipulate that all rights to the user’s input commands (“input”) remain with the user. The rights to the generator product (“output”) are also transferred to the user, provided that the output was generated in accordance with the terms of use.
It is unclear where OpenAI obtains the photos it uses to train the AI.
Neuroflash
Neuroflash is a German AI content tool that focuses on the creation of texts, but also offers a tool for generating AI art, which is currently still quite simple. It should be noted that Neuroflash is a partner company of OpenAI and therefore – based on the English term “fair use” – transfers responsibility for legal conformity to the user in the terms and conditions under the point “Fair use regulation”. Reference is also made to the “limits and restrictions” imposed by OpenAI (see above).
With regard to the images generated via Neuroflash, reference is made to the software of a stable diffusion model from Stability AI LTD and their ownership provisions are linked. However, there is no safeguarding regulation there either. The user merely undertakes not to violate any laws.
Neuroflash thus evades all responsibility.
Regarding the origin of the training data, Neuroflash only writes in the FAQ that most companies train the algorithms using real photos that are already available on the Internet.
Runway
The start-up company Runway mainly offers an image/video editing program, but also deals with “text-to-image”. In contrast to Dall-E, Runway mainly generates realistic situations. For example, if you give Runway For example, if you give Runway the command “a horse wearing boots”, the output is a photo of horse legs next to a person wearing boots. Dall-E, on the other hand, actually generates horse legs in boots.
Nevertheless, Runway does have one innovation: the company recently released an AI model called Gen-1, which also enables the generation of videos.
What is interesting about the startup AI is that the terms of use explicitly state that the use of Runway’s services for the sole purpose of sending copyright warnings or similar is expressly prohibited. Furthermore, the rights of use to the content transmitted by the user to Runway (“User Contributions”) are transferred to Runway and its contractual partners. The user is responsible for ensuring that the content has been used in accordance with the law and that the user is also entitled to transfer rights to it to Runway.
StableDiffusion
Stable Diffusion, launched in August 2022 by the company StabilityAI, is another popular generator of artificial images. The startup Runway (see above) initially played a key role in its development, but there is no longer any collaboration.
By generating images, the user grants open licenses to the software operators and all other users. This is followed by a free exchange of the uploaded and generated images. As a result, copyright infringements occur particularly quickly, extensively and uncontrollably – stable diffusion has been the subject of several negative headlines recently.
The website www.stablediffusionlitigation.com is even promoting a class action lawsuit by international authors.
The problem was brought to a head by the software’s seemingly lax handling of copyright-protected photos. In contrast to other generators, researchers succeeded for the first time at the beginning of the year in extracting the AI training data from an image generated with Stable Diffusion and proving that, for example, portrait photos were used to train the AI without the permission of the respective copyright holder. The major photo portal Getty Images also filed a lawsuit against StabilityAI for copyright infringements by Stable Diffusion. At the end of last year, it was discovered that the program had been significantly trained on data from Getty Images – the AI was obviously trying to imitate Getty’s watermark.
The dataset used by Stable Diffusion for AI training is the “2b English language label subset of LAION 5b”. This is a publicly accessible dataset that was created using a tool for general internet searches.
Mindverse
Mindverse is a German provider that offers a self-attested “all-in-one tool”. It enables research, brainstorming, image creation, text generation and AI analyses.
Mindverse has created its own terms and conditions for the image generator and has listed comprehensive copyright-relevant conditions there. They state:
“All content protected by copyright and/or competition and/or trademark law [relevanten] content is the intellectual property of Relativity GmbH or licensed to us and protected by law.”
The user is then granted a license for use.
MidJourney
MidJourney describes itself as an independent research laboratory that discovers new thought media and imaginative powers of the human species. The operators are a small, self-funded team specializing in design, human infrastructure and AI.
By using the software, the user grants MidJourney non-exclusive but full usage and exploitation rights to the artworks generated by the user. MidJourney also grants itself the right to grant sublicenses.
A distinction is made with regard to the granting of rights to the user: Midjourney offers two service models, free and fee-based. When using the software with a payment plan (“Pro” membership), MidJourney grants the user all possible rights to the generated art in accordance with the current legal situation.
If the software is used commercially by a company with an annual turnover of more than USD 1,000,000, a paid “Pro” membership must be taken out for each employee who is to use the software.
When used without a payment plan, MidJourney only grants the user a simple license to use the generated images. However, the image itself is also displayed to other users and made available to them for viewing or further processing. With the paid membership, the public sharing of the generated images can be prevented.
Finally, MidJourney offers a contact point for “takedown requests”. Here you can request that content that infringes your copyright/trademark rights be removed from the database.
StarryAI
StarryAI is an art and design app. In order to generate an image, one of three algorithms must first be selected, each of which produces different results.
By accepting the terms of use, the user undertakes to license the content generated by him/her to StarryAI in full, but not exclusively. This license is explicitly revoked as soon as the user deletes the content or account.
StarryAI expressly assumes no liability for any legal infringements that may occur through the use of the program.
Nightcafé
This program offers the possibility to generate images with AI or to revise existing images to be fed in by the user in certain styles. A practical feature of Nightcafé is the ability to order the generated image as a wall mural for immediate delivery to your home.
The terms of use state that all rights to the AI art are transferred to the user and that the user can use the images extensively – explicitly also for commercial purposes. However, Nightcafé accepts no liability for any infringements of third party (copyright) rights committed as a result.
Canva
Canva sees itself as a platform for visual communication, but focuses mainly on various design methods. The well-known design tool has now also integrated an AI generator into its services.
Due to the fact that Canva offers a lot of different web content (besides the image generator also learning platforms, t-shirt/greeting card designs, etc.) the terms of use are very extensive. Moreover, they do not refer specifically to the image generator. For all tools, Canva distinguishes between user content (i.e. content uploaded by the user) and licensed content (available in the Canva database). For user content, the user bears the risk with regard to his authorization to use and license the content to Canva, which is mandatory when uploading. There are separate license agreements for database content, which are available separately.
Otherwise, Canva clearly mentions that the platform operators exclude warranties “to the fullest extent permitted by law”.
Craiyon
Craiyon is an AI-based artist application and was formerly known as Dall-E Mini.
The terms of use prohibit the user from violating laws through the specific type of use – including copyright law. If this condition is violated, Craiyon reserves the right to restrict/change/delete/etc. the account.
By using the AI, the user grants Craiyon extensive usage and exploitation rights to all input commands and generated works. This license shall be maintained even if the contractual relationship between the user and the software ends.
Here, too, it is not possible to determine exactly which data is used to train the AI, but the data set is less extensive than that of OpenAI.
Overall legal situation uncertain
The overall picture shows that most software operators themselves do not know for sure how the legal situation is to be assessed in a binding manner and how the terms of use must be structured. The regulations are superficial, and statements regarding authorship are for the most part only made vaguely. More detailed requirements for the copyright law formulations are to be expected in the future from case law. Accordingly, all terms of use contain a passage in which future changes are reserved.
Focus on AI & copyright
In our thematic focus “AI & Copyright“, we deal with the copyright aspects of current developments in the field of artificial intelligence. So far, this series has included introductions to the applications ChatGPT and the use of AI art, an overview of the terms of use and training data of popular AI image generators and the Mandelbrot debate in the context of the legal discussion. We shed light on the concerns and fears of press publishers in the article“Ancillary copyright and AI” and explore the fundamental question of whether AI can be an author in a separate article.