If you've ever tried an AI system that generates images from text, you were probably deeply disappointed by your first results. It's not the AI, it's your prompt. As we explained in a previous article ,the "prompt" is the instructions the user gives to the AI system to generate content.
Let's take an example. Let's imagine that, for some reason, we want an image of an evil lawyer1. We can use a basic prompt like "Evil lawyer shouting in a trial", obtaining results ranging from unconvincing to ridiculous2:
(This is how I imagine myself when the profession has completely consumed me, around July 2024).
This difference in the quality of the result has led to the existence of various websites that aim to facilitate the creation of high-level prompts. In addition to jobs described as "Prompt-Enginner" (which in reality require technical knowledge that goes far beyond merely knowing how to make good prompts), we can highlight Ordinary People Prompt, a prompt-sharing forum, Prompt Hero, a repository of AI-generated images accompanied by their Prompts, Prompt Stacks, a platform for learning how to create prompts or Prompt Base, a marketplace where users buy and sell prompts. 5
Faced with this new reality, the question arises: Can prompts be protected by copyright?
To properly answer the question we must bear in mind that a prompt must follow certain rules of structure 6 and vocabulary 7 and that it has a maximum length, which will depend on the AI system that must apply it.
We will begin by reviewing the concept of "work", without losing sight of the principle contained in Article 2 of the 1996 WIPO Copyright Treaty, which states that "Copyright protection shall extend to expressions but not to ideas, procedures, methods of operation or mathematical concepts as such".
Despite being referenced in Article 2.1 of the Berne Convention and Articles 2, 3 and 4 of the InfoSoc Directive, a "work" is not defined in any of these legal texts, just as we will not find a definition of originality in European legislation beyond those applicable to computer programs, photographs and databases8.
The starting point for answering the question, obviously, is the CJEU's Infopaq I [33-38], famous for establishing that a work is"original" in the sense that it is its author’s own intellectual creation9. The decisions of the last decade have further refined this concept to reach the following list of requirements, which we will examine from the least to the most problematic for the protection of the prompt.
In the CJEU's Decision FAPL, [97], among others, it is already implicitly established that exhaustive lists of protectable works are incompatible with European law. Therefore, the fact that the hypothetical work is a prompt does not constitute an obstacle to its protection as such.
The requirement that the object of originality be "formally expressed", established in the CJEU's Decision Levola Hengelo [40]. In the words of the court we find an implicit reference to the principle of non-protection of mere ideas,10 which we will analyze last.
The requirement of free and creative choices is established, among others, in the CJEU's Decision Bezpečnostní softwarová asociace, [49] which states: "where the expression of those components is dictated by their technical function, the criterion of originality is not met, since the different methods of implementing an idea are so limited that the idea and the expression become indissociable.”
Therefore, the question here is whether the rules governing the creation of prompts leave sufficient leeway for the author to make free and creative choices that allow him to reflect his personality.
We could understand that the answer is negative by quoting the CJEU's Decision Funke Medien, [24] which states that: "If military status reports, such as those at issue in the main proceedings, constitute purely informative documents, the content of which is essentially determined by the information which they contain, so that such information and the expression of those reports become indissociable and that those reports are thus entirely characterised by their technical function, precluding all originality, it should be considered, as the Advocate General stated in point 19 of his Opinion, that, in drafting those reports, it was impossible for the author to express his or her creativity in an original manner and to achieve a result which is that author’s own intellectual creation."
That is, to conclude that the ultimate purpose of the prompt constrains its author to follow a certain structure and vocabulary in order to communicate properly with the machine and that this would prevent the author from expressing his personality in the prompt.
On the other hand, the question can be answered in the affirmative in the light of the CJEU's Decision Brompton [26], which clarifies: "It follows that an object that meets the requirement of originality can benefit from copyright protection even if its realization has been determined by technical considerations, provided that such determination has not prevented the author from reflecting his personality in that object by expressing free and creative choices.”
Although the author of a prompt must follow a certain set of "rules" in order for the prompt to be adequately usable by the AI system, these rules relate mainly to the structure of the prompt (which will obviously not be protected), and are not particularly limiting on the vocabulary used.
Taking into account that there are very similar technical restrictions for the development of software (which must be written following a certain programming language) and that the creative freedom of the programmer has not been considered to be restricted to such an extent that the resulting code cannot be protected as a work, we are inclined to understand that this requirement should be considered surmountable.
On the contrary, the answer is not so clear when we analyze the requirement that a work constitutes an intellectual creation of its author in the light of the idea-expression dichotomy.
In the aforementioned CJEU's Decision Bezpečnostní softwarová asociace, [50], this concept is clarified in the sense that the author must be able to "express his creativity in an original manner".In the same sense, the CJEU's Decision PAINER [88] or Cofemel [30].
Given the lax standard that these resolutions set for a work to be original, and bearing in mind that in Infopaq I [38] it is stated that 11 words might be protected by copyright since, as such, they share the originality of the whole work., it might seem that this requirement should be surmountable.
However, we cannot lose sight of the fact that for an author to be able to "express his creativity" he must necessarily go beyond the exposition of the mere idea (however original the idea may be). And a prompt, however specific it may be, is nothing more than a set of instructions that an AI system will use to materialize the idea.
In favor of protecting the prompt as a work, it can be argued that a prompt sufficiently detailed to restrict the possible outputs to variations of the same image with imperceptible differences would in itself be an expression of a given idea. That is, the completeness of the prompt would already contain the totality of free and creative decisions, restricting the activity of the AI model to an execution devoid of originality.
Following this argument, the original and creative expression of the idea would be realized in the set of instructions, being the embodiment of it an activity devoid of creativity, since all its possible results would be almost identical variations of the same drawing. We would speak, in this case, of an "exhaustive" prompt.
Although theoretically possible, given the technical limitations of prompts at the present time, it does not seem that "exhaustive prompts" still exist, so that in our humble opinion, the principle of non-protection of ideas forces us to conclude that it would not be in accordance with the law to protect a prompt as such, since any of them admits multiple divergent expressions for the same AI model.
Another argument in favor of considering the prompt as an original work could consist in arguing that, as a set of instructions for an expression, the prompt is likely to reflect the author's personality, which would be transferred to the materialization.
This would be an analogous reasoning to the one applied in the Barceló, Cala Holmes, and Druet, cases in which the authorship of those who gave instructions for the execution of various works but who did not necessarily participate materially in said execution was recognized 11.
The problem with applying this reasoning to the protection of the prompt (not to the resulting graphic work) is that in those cases the "author" who gave instructions had control over the final result. However, the author of a prompt cannot predict the result of the prompt, so it does not seem that there can be originality in a prompt that is essentially the idea before it is expressed, however detailed said prompt may be.
This indetermination in the expression of the prompt becomes even more evident when comparing it with other works susceptible of being materialized or applied. In our legal system, the authors of entertainment formats and architectural models hold a monopoly right not only over the mere reproduction or distribution of their work, but also over its application or materialization, which falls within the powers of exploitation of the work in article 17 of the Spanish Copyright Law 12.
However, in principle there can only be one way to materialize an architectural model (without prejudice to the possibility of creating derivative works). This is not the same with a prompt, not only because, as we have seen, the same model can make infinite variations of an image due to the margin of expression given to it, but also because the same prompt can produce incredibly different results depending on the AI system through which it is materialized.
Although self-evident, up to now, every graphic expression of an idea has contained only and exclusively that same expression. And by protecting non-exhaustive prompts we would risk protecting a set of expressions of the same idea, specifically, all those that fit within the indeterminacy margin of the prompt. Or in other words, all admissible variations of the non-exhaustive prompts of the prompt. This goes against the principle of non-protection of mere ideas and is therefore inadmissible13.
Continuing with the example at the beginning of the article, we reproduce an image generated using the same prompt through another AI system:
(Unlike the first one, this one can hardly be an image of me in 2024, among other powerful reasons because I don't think one year is a sufficient time frame for me to grow a beard).
Comparing this image with the previous one, we notice the enormous margin of creative freedom that the AI model has, and that the apparently detailed prompt lacks. The images differ in all the elements of expression: the lawyer's features, his clothing, the shape and color of the horns, his expression, the color palette of the image, the stroke, etc. In addition, this second image gives much more weight to the term "Santiago Caruso" and much less to "ultra realistic illustration”, for example. The output of the prompt is unpredictable, even for its creator.
And, given this, we must conclude that any prompt is far from being the expression and instead comes fatally close to being the idea prior to its expression. This conclusion seems to us consistent with Infopaq I, which in our humble opinion cannot be interpreted in the sense of considering automatically original any set of eleven consecutive words. It should be recalled that in that case the CJEU understood that these extracts of eleven words could be protected by copyright since, as such, they share the originality of the whole work. and that "In the main proceedings, moreover, it is common ground that newspaper articles, as such, are literary works covered by Directive 2001/29." However, as we have seen, the originality of the prompt is indeed disputed.
In conclusion:
1. A prompt is essentially a set of instructions for an AI model to express an idea. Therefore, since it is not exhaustive, it does not constitute one of the many possible expressions of the idea. By admitting an infinite number of concrete expressions, whose execution is inevitably reserved for the AI model, it is evident that the prompt is prior to the expression, and is revealed to us as the idea to be realized with a given materialization.
2. Consequently, since originality lies in the expression of an idea, being inadmissible the protection of a mere idea, however creative it may be, unless we are dealing with an exhaustive prompt (nonexistent at present), it cannot be the expression of the unique intellectual creation of its author, it cannot reflect his personality and consequently it cannot be protected by copyright.
Footnotes
- Forgive me for being redundant.
- Generated using the Stable Diffusion model, available at: https://beta.dreamstudio.ai/generate. Note that the eight images generated show a "lawyer" (a term that does not determine gender) as a white male, approximately 35 years old, with a suit and tie and brown hair combed to the side.
- Such as those available at https://stable-diffusion-art.com/prompt-guide/ and https://www.xataka.com/robotica-e-ia/guia-practica-para-escribir-mejores-prompts-midjourney-desbloquear-su-verdadero-potencial
- Prompt: Half demon half lawyer in trial screaming angrily at the camera, red skin and horns and with a three-piece suit with a purple tie, detailed clothing, ultra realistic illustration, Santiago Caruso, artstation, sharp focus, dramatic, low angle shot. Negative Prompt: un-detailed skin, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, ugly eyes, (out of frame:1.3), worst quality, low quality, jpeg artifacts, cgi, sketch, cartoon, drawing, (out of frame:1.1) Making prompts is an art in and of itself ,wink-elbow-wink.
- Paying for a prompt may not be as absurd as it may seem, considering how cheap they are and the time we can save if we find a reference that is close to what we are trying to generate, such as these typically American mascot team logos: https://promptbase.com/prompt/sports-logo
- One of the recommended orders to elaborate a prompt is the following:1.Type of image 2. Description of the main object. 3) Environment and other details 4) Style details 5) Technical parameters (proportion, stylization level, model, etc.).
- Estas serían las reglas que dictan cómo determinadas palabras, estilos o artistas serán interpretados por cada sistema de IA o si serán siquiera comprendidos. For example, before using the name of a certain artist as a style variant, we will have to determine whether the AI model in question knows it and will therefore take it into account. We should also follow some basic recommendations, such as the use of precise adjectives (scary instead of "very scary"), preference for the use of positive rather than negative adjectives, and a long list of etceteras.
- For computer programs, see Article 1(3) of Council Directive 91/250/EECand, now, Article 1(3) of Directive 2009/24/EC. For photographs, see Article 6 of Council Directive 93/98/EEC and, now, Article 6 of Directive 2006/116/EC. For databases, see Article 3(1) of Directiva 96/9/CE. As summarized in the article Copyright at the CJEU: Back to the start (of copyright protection) Eleonora Rosati, Developments and Directions in Intellectual Property Law. 20 Years of The IPKat (Oxford University Press: 2023), which is an excellent and relatively brief summary of the jurisprudential developments that have been permeating the concept of "work" in community law.
- The discussion of whether a "prompt" should be considered software (insofar as, strictly speaking, it serves to give instructions to the machine, as if it were a high-level programming language is as interesting as it is sterile, since the aforementioned Article 1(3) of Directive 2009/24/EC requires that the code be "the author's own intellectual creation". All roads lead to Infopaq
- In the previously cited article Copyright at the CJEU: Back to the start (of copyright protection), is stated that: "The definition thus provided, by requiring that the subject matter be “expressed”, contains an implicit reference to the idea/expression dichotomy. In imposing the conditions of precision and objectivity, it also clearly borrows from trade mark law (this is evident if one reads the AG Opinion in that case53) and its representation requirement. All this further implies that there should be no element of subjectivity in the process of identifying the protected subject-matter. .
- Although without reasoning its decision, the U.S. Copyright Office has said in its document Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence “While some prompts may be sufficiently creative to be protected by copyright, that does not mean that material generated from a copyrightable prompt is itself copyrightable."
- Among others, it is worth mentioning STS 588/2014 of October 22 (ECLI:ES:TS:2014:4623), SAP Madrid 67/2016, February 19, 2016 (ECLI: ES:APM:2016:2280), paragraphs 28-30, SAP BCN 147/2006 (ECLI:ES:APB:2006:3390) FJ Eleventh. The greatest exponent of the foregoing is SAP Madrid 269/2016, of July 8 (ECLI:ES:APM:2016:12864) when it states: “En definitiva, dado que "In short, given that the comparison is made between a television format and an audiovisual work, what must be determined is whether the latter involves, in substance, an application or execution of the plaintiffs' television format,(...)",(...)".
- In this sense, the USCO's decision on the registration of the AI-generated comic "Zarya of the dawn" argumenta lo siguiente para denegar la autoría de la Sra. Kashtanova sobre las imágenes del cómic en cuestión.:"The fact that Midjourney’s specific output cannot be predicted by users makes Midjourney different for copyright purposes than other tools used by artists. (...) Because Midjourney starts with randomly generated noise that evolves into a final image, there is no guarantee that a particular prompt will generate any particular visual output. Instead, prompts function closer to suggestions than orders, similar to the situation of a client who hires an artist to create an image with general directions as to its contents."
Leave a Reply