OpenAI recently unveiled an image generation service integrated with ChatGPT, capable of producing realistic visuals and videos in iconic styles such as those of Studio Ghibli, Disney or Pixar, or even the famous Simpsons. This new tool has aroused massive interest, but also concern. What legal lessons can we draw from the global buzz generated by this new feature linked to Chat GPT?
OpenAI recently enhanced its image generation service by integrating its DALL-E 3 model directly into the ChatGPT interface. In concrete terms, since autumn 2023 for subscribers (Plus, Pro, Enterprise) - and early 2025 for the general public - it has been possible to describe a scene to ChatGPT and see the assistant generate several corresponding images, all in the usual conversation thread. The user can then refine or modify the result while chatting, without leaving ChatGPT. OpenAI points out that this conversational integration makes it much easier to create and, above all, revise images: "Describe your vision, and ChatGPT will bring it to life by providing a selection of visuals for you to refine and iterate upon" . In other words, you can ask for corrections or variations in natural language, and ChatGPT will automatically reformulate the detailed prompts for DALL-E 3.
On 25 March 2025, OpenAI reported that it had significant improvements thanks to refined training: the GPT-4 system that drives ChatGPT "vision" (sometimes called GPT-4o) trained on high-quality text descriptions associated with images, enabling a finer match with user requests. However, the way in which the AI has acquired this know-how has raised legitimate questions from rights holders in the world of animation: it seems clear that DALL-E 3 has been trained on thousands of images taken from Ghibli and Disney films, without authorisation or licence.
One of the outstanding features of DALL-E 3 is its ability to imitate recognisable artistic styles on demand, which immediately sparked off creative uses by the public. Just a few hours after the opening of image generation in ChatGPT, a veritable viral trend emerged around the style of Studio Ghibli films. Social networks were flooded with images "à la Miyazaki", reproducing the dreamlike aesthetic of classics such as My Neighbour Totoro and The Voyage of Chihiro. Users have started to "ghiblify" their own family photos or selfies, in other words to represent themselves and their loved ones as characters from Japanese animated films. The result is often stunning, with a disturbing fidelity to the studio's style, to the point where some images could pass for original illustrations by Hayao Miyazaki. Sam Altman (OpenAI's CEO) himself has accentuated this Ghibli mode: he has changed his profile photo on X (Twitter) to a Ghibli-style version of his portrait, and has encouraged his followers to do the same.
Sam Altman illustrated the popularity of this new feature by saying that the company's GPUs were "melting" under the demand, underlining the pressure exerted by millions of enthusiastic users. Open AI even had to stop its free image creation feature because of the flood of requests.
It's not just the Ghibli anime style that's in demand. DALL-E 3 generally excels at imitating the graphic worlds of many famous studios and artists.
The most frequently requested examples include :
The arrival of these new features has met with mixed reactions. Among the general public and the media, the enthusiasm was undeniable: many users praised the multiplied creativity offered by ChatGPT coupled with DALL-E 3. In just a few days, social networks were flooded with original creations, ranging from memes twisted into artistic versions (for example, historical or viral scenes recontextualised as cartoons) to more serious uses such as logo prototypes, product designs or professional illustrations generated in the blink of an eye.
However, this enthusiasm is accompanied by criticism and major legal issues, particularly around intellectual property. The phenomenon of "Ghibli-style" images has agitated the artistic community and rights holders. On the one hand, Miyazaki's admirers found themselves torn between the pleasure of seeing their favourite world revisited, and the unease of knowing that these pastiches had been generated without any intervention from one of the studio's artists. On Studio Ghibli's side, no legal action has (yet) been taken, and the studio has refused to comment officially on the situation. Nevertheless, Hayao Miyazaki's former statements on AI have resurfaced and taken on a particular resonance. Back in 2016, the Japanese director criticised computer-generated animation, calling it "insult to life itself"an .
On the other hand, there are questions to be asked: what training images did he use to reproduce the Ghibli 'touch' with such precision? The issue of copyright soon came to the fore.
In Europe, the use of protected works to train AI is regulated by the Digital Single Market Directive (DAMUN). Article 4 authorises text and data mining for training purposes, including on copyrighted content, unless rights holders exercise their opt-out rights. This mechanism allows creators to oppose the exploitation of their works, for example via technical tags that comply with the text and data mining (TDM) protocol. For OpenAI, this means that training its model on works by Ghibli or Disney could be legal under European law, in the absence of an explicit opt-out.
In Japan1, according to article 30-4 of the copyright law, there is a broad exception to copyright for the learning of generative artificial intelligences, which applies even when the use is commercial. However, according to article 30-4, this broad exception is called into question for generative AI when the learning "would unreasonably prejudice the interests of the copyright owner". Furthermore, according to a document2 from the committee of the Japanese Agency for Cultural Affairs, in principle style is not protected by copyright. However, according to this document, imitating a designer's style that includes a "creative expression" would constitute an infringement of copyright3. Furthermore, this exception does not apply when the AI has trained on infringing content4.
In the United States, AI publishers such as Open AI rely on the doctrine of fair use, codified in 17 U.S.C. § 107, to justify the use of copyrighted works for training large language models (LLMs) and other generative AI systems. Fair use permits limited use without permission for purposes such as criticism, comment, news reporting, teaching, scholarship, or research, evaluated through four factors: (1) the purpose and nature of the use, (2) the nature of the copyrighted work, (3) the amount and significance of the portion used, and (4) the effect on the potential market for the original work
Open AI has already used the principle of fair use in its defence against the New York Times' complaint about the unauthorised use of Times articles in ChatGPT training. OpenAI is also defending a broader interpretation of fair use with the Trump administration, hoping to legally consolidate the integration of protected content into its AI models, while limiting the risk of litigation.
According to some American experts5, Studio Ghibli could also argue that OpenAI's generation of "Ghibli-style" images constitutes trademark infringement or "false endorsement", which falls under the Lanham Act (15 U.S.C. § 1051 et seq.) and more specifically Goodwill, which refers to the reputation, customer loyalty and positive associations that a company has developed with its brand.
According to Rob Rosenberg, a legal expert in artificial intelligence law in the United States:
"Ghibli could argue that by converting users' photos to "Ghibli style", OpenAI is exploiting the Goodwill of Ghibli's trademarks, using Ghibli's identifiable style and creating a risk of consumer confusion as to whether this feature is approved or authorised by Studio Ghibli [...].If OpenAI were marketing its platform as a place where consumers could go to turn their photos into "Ghibli-style" anime, that fact would weigh more heavily in favor of a court finding that OpenAI's conduct violated the Lanham Act".
OpenAI could also be penalised under trademark law and the Lanham Act if it presented Chat GPT as a means of generating Ghibli-style images. In addition, the photos published by Sam Atlman on X could be considered as promoting the possibility of generating images in the style of Studios Ghibli.
French copyright protects original works, in particular the characters in the works (both their image and their personality traits) as well as the sets and designs.
For example, the character Dogmatix (Idéfix in french) was recognised6 as an original creation protected by copyright, and a company that had depicted Dogmatix without authorisation in the Pop Art figurines it was selling was penalised for copyright infringement7 and ordered to pay €100,000 in damages.
In another case, the painter Xavier Marabout produced a series of paintings based on Hergé's work in unprecedented situations in which Tintin is depicted in amorous and sensual settings in the style of paintings by the American artist Hopper. Once again the painter was condemned for copyright infringement8 and the parody exception9 that he claimed was rejected10 because the painter's approach was considered to be commercial on a large scale.
However, graphic styles cannot be protected as such under French intellectual property law. Thus, generating an image in the Ghibli style does not necessarily constitute an infringement, unless specific elements such as characters (Princess Mononoke, Totoro, etc.) or sets are used.
Faced with an uncertain legal situation, OpenAI has adopted a cautious stance. Officially, the company's policy is to refuse requests aimed at copying "the style of a living artist". DALL-E 3 has been configured to filter prompts of the type "in the style of [Name of contemporary artist]". On the other hand, the styles of studios or deceased artists are tolerated, which OpenAI justifies as being "broader styles" that are part of general visual culture. Thus, requesting the generation of an image "in the style of Hayao Miyazaki" should theoretically be blocked (since Miyazaki is a living artist), but requesting content "in the style of a Studio Ghibli film" is accepted - a nuance that has raised legal and ethical questions. Indeed, behind a studio style there are often very real artists (in the case of Ghibli, Hayao Miyazaki himself is the co-founder and artistic driving force). So there's a fine line between what's allowed and what isn't, and some see it as an attempt by OpenAI to minimise the legal risks while still taking advantage of the appeal of these famous styles. Especially as it is possible to get round this filter limit on prompts, as shown by the generation of Ghibli-style images.
If studios or creators feel that OpenAI (and/or its users) are taking advantage of their reputation or investments without authorisation, they may also consider making a claim on the grounds of parasitism (Article 1240 of the French Civil Code).
Economic parasitism makes it possible to condemn the behaviour of an economic operator who substantially usurps the "economic value" of another, identified and11 individualised. It is up to the person claiming to be the victim of acts of parasitism to identify the individualised economic value he is invoking. For example, ChatGPT users could themselves be found liable for such parasitic behaviour if they commercially exploit content generated by artificial intelligence that clearly imitates protected visual styles, thereby capturing individualised economic value derived from the investment and reputation developed by the famous animation studios.
Users of this tool are not exempt from risk. If protected elements - such as a specific character - are reproduced, the Chat GPT user may be prosecuted for copyright infringement and parasitism. The OpenAI terms of use specify that the images generated belong to the users , but they do not offer protection against third-party claims. Users should therefore act with caution, particularly if they are monetising their creations . Users who use ChatGPT and DALL-E 3 to generate images are not immune from legal action. Indeed, if they reproduce, even indirectly, elements protected by copyright - such as specific characters from Studios Ghibli - these users could be prosecuted for counterfeiting or economic parasitism.
This "Ghibli" case will, however, have had one important merit: it will have raised awareness among economic operators of the intellectual property issues surrounding AI in the strongest way since Chat GPT! was launched to the general public.
1 Article 30-4 of the revised 2019 Japan Copyright Act, stating that "It is permissible to exploit a work, in any way and to the extent considered necessary, in any of the following cases, or in any other case in which it is not a person's purpose to personally enjoy or cause another person to enjoy the thoughts or sentiments expressed in that work; provided, however, that this does not apply if the action would unreasonably prejudice the interests of the copyright owner in light of the nature or purpose of the work or the circumstances of its exploitation: (i) if it is done for use in testing to develop or put into practical use technology that is connected with the recording of sounds or visuals of a work or other such exploitation; (ii) if it is done for use in data analysis (meaning the extraction, comparison, classification, or other statistical analysis of the constituent language, sounds, images, or other elemental data from a large number of works or a large volume of other such data; the same applies in Article 47-5, paragraph (1), item (ii)); and (iii) if it is exploited in the course of computer data processing or otherwise exploited in a way that does not involve what is expressed in the work being perceived by the human senses (for works of computer programming, such exploitation excludes".
2 General Understanding on AI and Copyright in Japan – Overview page 9.
3 See the analysis by Elodie MIGLIORE - AI et fouille de textes et de données : retour sur la souplesse de l'exception japonaise à l'ère de l'AI générative - Revue Francophone de la Propriété Intellectuelle ISSN 2490-8347 - N° 18 - Décembre 2024.
4 General Understanding on AI and Copyright in Japan – Overview page 11.
5 According to Rob Rosenberg, founder of Telluride Legal Strategies, which specialises in US AI law.
6 6 February 2025, Tribunal judiciaire de Marseille, RG n° 22/12168
7 Under the terms of Article L. 122-4 of the French Intellectual Property Code
8 Again according to Article L 122-4 of the French Intellectual Property Code
9 According to Article L 122-5, 4° of the French Intellectual Property Code
10 Rennes Court of Appeal - 1st Chamber - 4 June 2024 - no. 21/04257
11 Cass. com. 31 March 2015, no. 14-12.391
