Supreme Court Rules Against Warhol Estate in Copyright Dispute Over Use of Photo of Prince for Magazine Cover — Potentially Wide-Reaching Implications for Generative AI for Visual Art

From The National Law Review:

On May 18, 2023, the US Supreme Court affirmed the Second Circuit’s decision that artist Andy Warhol’s silkscreen portrait of Lynn Goldsmith’s photograph of musician Prince, used for a Vanity Fair cover, was not a fair use under US Copyright Law.

In a 7-2 decision, the Court found the “purpose and character” factor of the copyright fair use analysis did not weigh in favor of a finding of fair use where the use of a new work encompassing an original work shares the same purpose as the use of the original work and is commercially licensed.

Although the addition of a new meaning or message is a relevant consideration in assessing the purpose and use of a work for purposes of determining fair use, it is not dispositive. According to the Court, the “purpose and character” fair use factor must consider “the reasons for, and nature of, the copier’s use of an original work,” balanced against the original creator’s exclusive right to make derivative works. If the new work achieves the same or similar purpose to the original work, and the new use is of a commercial nature, the first fair use factor likely weighs against a finding of fair use, absent another justification of copying. Ultimately, the inquiry is objective and questions how the original user has used the original work.

The case comes with much anticipation and could have significant implications on the availability of the “fair use” defense in copyright infringement cases, including potentially with respect to the hotly contested use of generative artificial intelligence to create new works.


Vanity Fair magazine originally commissioned Warhol to create an image of Prince for publication alongside an article of the musician in 1984. Photographer Lynn Goldsmith granted the magazine a “one time” license for it to use one of her Prince photographs as an “artist reference for an illustration.” Warhol created a purple silkscreen portrait using Goldsmith’s photograph. The magazine credited Goldsmith for the “source photograph” and paid her $400.

Unbeknownst to Goldsmith, Warhol created 15 other silkscreen prints and pencil drawings from Goldsmith’s same image of Prince. Goldsmith claimed she did not learn of the Prince series until 2016, when Condé Nast published some of the images in a posthumous tribute to Prince. In particular, in 2016, the Andy Warhol Foundation for the Visual Arts, Inc. (AWF) licensed an “Orange Prince” silkscreen image to the magazine for $10,000 for its cover. Goldsmith did not receive any compensation or credit.

(PG note. Both images are from the Opinion of The Supreme Court.)

In response to Goldsmith’s objections, AWF sued for a declaratory judgment of non-infringement of Goldsmith’s alleged copyright. Goldsmith counterclaimed for infringement. On Goldsmith’s counterclaim, the district court granted summary judgment for AWF, finding all four copyright fair use factors, under 17 U.S.C. §107, weighed in the foundation’s favor. The Court of Appeals for the Second Circuit however reversed and remanded, finding all fair use factors favored Goldsmith.


The only issue before the Supreme Court was whether the “purpose and character of the use” weighed in favor of AWF such that its commercial licensing to Condé Nast would be considered fair use. The Court ultimately agreed with the Second Circuit, finding in favor of Goldsmith, and that the factor did not favor AWF’s fair use defense to copyright infringement.

Goldsmith had licensed her works for years, with images appearing in LifeTime, and Rolling Stone magazines. People magazine had also paid Goldsmith to use one of her Prince images in a special collector’s edition of its publication following Prince’s death. Because Goldsmith’s original photograph and AWF’s work licensed to Condé Nast were both portraits of Prince, “used in magazines to illustrate stories about Prince[,]” the Court found the works had the same purpose and that AWF’s use did not favor a finding of fair use.

A determination of fair use requires consideration of the following four factors set forth in §107 of the Copyright Act:

  1. The purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
  2. The nature of the copyrighted work;
  3. The amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
  4. The effect of the use upon the potential market for or value of the copyrighted work.

The Court explained that the central question in determining the first fair use factor asks whether and to what extent the new use adds something new, with a further purpose or different character. The larger the work supersedes the objects of the original creation and goes beyond that required to qualify as a derivative work, the more likely the purpose and use factor will weigh in favor of a finding of fair use. Works that merely supersede the objects in the original creation — without adding something new, with a different character or further purpose — are unlikely to be found transformative and thus fair.

Included in the Court’s fair use analysis was significant discussion of its 1994 Campbell decision, a case involving 2 Live Crew’s copying of Roy Orbison’s song, “Oh, Pretty Woman,” and from which it created a rap derivative, “Pretty Woman.” There, the Court found the new work clearly transformative in that it was a parody, had a different message and aesthetic, and had the distinct purpose of commenting on the original.

Conversely, Warhol’s image was substantially the same as Goldsmith’s original photograph in that both were used to illustrate a portrait of Prince, AWF received compensation for doing so, and the degree of difference (i.e., the flattening, tracing, and coloring of the photo) was not enough for the first factor to favor AWF, according to the Court.

The Court compared AWF to other would-be users who may attempt to seek the shelter of the fair use defense, including musicians who sample songs, playwrights who adapt novels, and filmmakers who create spinoffs, noting that a finding of fair use in the case might authorize a range of commercial copying for use in a substantially same manner as the original work.

Link to the rest at The National Law Review

PG would have come down on the fair use side of the argument, but nobody ever nominated him to serve on the US Supreme Court.

12 thoughts on “Supreme Court Rules Against Warhol Estate in Copyright Dispute Over Use of Photo of Prince for Magazine Cover — Potentially Wide-Reaching Implications for Generative AI for Visual Art”

  1. Usually, I can kind of figure out who won in a legal argument and why; this time, no can do.

    Copyright – especially now that AI is being fed huge chunks of copyrighted material – is the next BIG THING.

    I won’t be using AI to help me write fiction; but I’d still like to understand why I could and why I couldn’t.

    Music and visual arts – and writing – are all having these questions come up for grabs – it behooves us to understand, because it is going to hit like a load of bricks.

    Can someone kindly explain in Kindergarten-level legal words? Thanks.

    • Try this: is the output of the software a direct substitute for somebody’s *specific* product?
      If the answer is yes then you shouldn’t use it. If the answer is no (and the thing is good enough on its own) then you should be home free under current precedents.

      There seems to be the misguided impression by the “AI” pearl clutchers that any connection, however faint, is a copyright violation. Like the racist “one drop” rule. Absolutist thinking.

      That’s not how US copyright works. It is more of an “it depends” situation.
      There’s the matter of Fair Use and the Four Factors test, all four of which are judgement calls. Along the lines of the OP, intent matters: a derivative intended as a one for one substitute for the source might be infringement for one use but proper for another. Parodies are expressly permited as are annotated editions (for the most part) if meant as a scholarly product and not an excuse to skirt valid copyright.

      Which is where SCOTUS seems to have focused on in the OP: Warhol was given a photo licensed for *one* commercial use to use as a source for a magazine cover. Creating multiple derivatives as part of his artistic process for that comission was likely fair use but using those derivatives commercially went beyond the scope of the license granted to the magazine and extended to Warhol. Same product, different use.

      Should any creator attempt to sue over an “AI” model that analyzed their product along with 100Trillion others they will have to get very specific as to how the scan harmed them. That strikes me as far from doable. But even if they manage that, there is still the suite of fair use tests to navigate to say nothing of *quantifying* the harm. (I’m reminded of the lawsuit by Diesel books vs Apple and the BPHs over being put out of business by the Agency Conspiracy. They proved actual harm but not the extent of the damage or intent.)

      For now the main reason not to use a software creation tool isn’t legal, just that the output isn’t likely to be good enough to use. The tools are barely at the proof of concept stage.

    • Try this:

      Adobe Photoshop now has integrated “AI” that, among other things, lets you use natural language instructions to edit an image. The example shown lets you swap the background of, say, a deer in a field to a grimy NYC alley. Or day to night. Or take out a plane flying overhead.

      Useful, no?

      Or say you have a photo of a young girl playing with a chihuahua and you want to swap in a dire wolf.

      Tons of uses. And this is still a first generation implementation.

      “AI” is at its best as hidden plumbing in the background of an existing productivity app. A feature instead of a product.

      As to copyright concerns Adobe claims that “Firefly is designed to generate images that are safe for commercial use and is trained on Adobe Stock’s hundreds of millions of professional-grade, licensed, high-resolution images, helping ensure Firefly won’t generate content based on other peoples’ or brands’ intellectual property (IP).”

      Let somebody try to get past that.

    • Making a go at kindergarten level legal words…

      In 1994, the Court ruled that parody could be a protected fair use of another’s copyrighted work. In what turned out to be a poor choice, when describing the nature of parody, they included comments about how parody was a “transformative” use. Everybody that wanted to justify their use of someone else’s work now said their use was “transformative” and therefore fair under the 1994 case. Generative AI developers argue that their uses are transformative (and they are generally more transformative than Warhol’s use in this case). A key problem is that pretty much all derivative works transform the original in some fashion. The 1994 case was not supposed to abolish copyright holders’ right to control derivative works. The statute that provides the right to control derivative works uses in its definition of derivative works the phrase “…or any other form in which a work may be recast, transformed, or adapted.”

      This Court used this case to untangle some of the confusion from the 1994 case’s sloppy language. They focused on the first prong of the fair use test, though some have argued they did so in a way that heightens the fourth prong or merges it with the first prong. The first prong concerns the purpose and nature of the infringing work. This Court said both the derivative work and the original work were used to illustrate articles about Prince. Both were licensed for commercial use for this purpose. The purpose and the nature of the works are therefore effectively the same and that works against a finding of fair use for the derivative work.

      The primary implication for AI is that the closer your work is replicating the purpose and nature of the original work, especially in a commercial context, the less likely you are to have a fair use. Did your training set include works by Booboo the Magnificent and people are now paying you for pictures of X in the style of Booboo the Magnificent? You may well be skating on thin ice after this decision. A second implication is that even if work A is a fair use derivation of work prime, work B that is derived from work A may not be a fair use derivative of work prime; chains of fair use are now a bit in question.

      For generative AI, the case raises more risk than there was before for creating work based on unlicensed original works incorporated in the training set.

      • In recent cases the most powerful test appears to be substitution.

        Note the case of the Internet Archive that was arguably transformative of a printed presentation into a digital rendition and not offered for sale, but it substituted for versions that *are* for sale. To the extent of being identical presentations of the content.

        Applied to generative software output, you would need to have a direct link to a specific product by Boo Boo the magnificent to be liable. (Say, a watermark suggesting a link of provenance.) And then you would have to be substituting for said product. And finally, the aggrieved would have to show significant damage. Just evoking a similarity to a copyrighted work would be akin to an homage or citation.

        Copyright isn’t for ideas or processes but for specific implementations of either.

        Style is not specific enough to be copyrightable. And it would have to be specific enough to be patentable under trade dress. So “in the style of” lawsuits are likely to be rare. If anything, most of the exemplars called upon are likely to bd defunct to start with. (“In the style of” Rembrant or Russ Manning, rather than Starving Artist Lee.)

        For all the griping over IP law (usually by those at the extremes) the existing framework of copyright, patent, and trademark seems to be holding up fine. If anything, any useful changes might run along the lines of extending the lifetimes of certain patents (and voiding speculative ones) and reducing the duration of post mortem copyright. But the world won’t end without changes.

        • Style is not specific enough to be copyrightable

          Sounds right for images but how about if the AI is generating music and we call it “feel” rather than “style” (as differences in melody, harmony and rhythm apparently don’t matter)?

        • Minor note: “Trade dress” relates to trademark, not patents; I think you were invoking the design patent. But a design patent can’t apply to music because the statute requires that design patents be entirely nonfunctional… and there’s nothing that’s distinctive enough to be “original” in music to be nonfunctional (where it’s actually been tested), and probably almost nowhere across the visual arts either.

          All of that said, what we have here is the eighteenth century’s law made even then by constipated old white men who had no originality anywhere in them, and more-recent conceptions and evolutions being forced into those 250-year-old (US and Europe!) categories. It is, to say the least, an interesting argument as to whether software should be protected (as source code, as it is) by copyright law, or (as truly original concepts rising above “mere algorithms”) on occasion by utility patents, or (as was attempted by Lotus 1-2-3 and failed miserably) design patents and trademark/trade dress. One of those constipated old white men, musing after a few too many sherries (Over There)/pints of ale (Over Here), might have condescendingly mused that the law is not fit for the purpose of encouraging mere trade amongst the grubbier parts of Society† — and he’d be right, if not necessarily for the reasons he’d declaim from the bench.

          † When copies of the new US Constitution made their way into the Inns of Court, there was pretty universal disdain for the Intellectual Property Clause. None of them believed that “mere tradesmen” could possibly advance the useful arts and sciences merely because they were getting paid to do so.

  2. As usual, the National Law Journal missed the point on an IP matter.

    The real dispute in the Warhol matter was actually about whether one aspect of a claim of “fair use” by a licensee’s successor in interest (the estate) that exceeded the terms of its license could provide a defense when the licensor objected to the excessive use. Put another way: If the licensee agrees to restrictions substantially narrower than the widest boundaries of fair use as it was understood when entering into the license,† is the licensee’s successor in interest allowed to use a newer, nonstatutory (that is, the “new” is “new judicial interpretations of the same statutory language”) concept of “fair use” to essentially rewrite the limitations in the license?

    It’s not as easy to answer as it sounds; I can give a definite tenative hypothetical “maybe?” as the best response. That is, it’s even wishy-washier than the usual lawyerly “it depends.” On the facts here, there appears to have been some dubious lawyering all the way back to the licensing transaction (and of course there was dubious lawyering involved much farther back in drafting § 107, the fair use provision of the statute… and now I’m really gonna scare y’all by saying “VARA,” § 106A which was drafted utterly without regard to § 107 Because Fine Art Is Special, and nobody really knows whether portrait photography is Fine Art).

    tl;dr It’s as clear as mud and doesn’t cover the ground because there’s a prior agreement to restrict use that is not entirely consistent with current conceptions of use that would be allowed absent that license. So this is not a groundbreaking decision on fair use that will change the internet, social media, and the fabric of the spacetime continuum (presuming, that is, that it’s not discontinuous anyway around Warhol).

    † Also seldom noted: There’s a stench of deception involved in how the license was negotiated in the first place. The statute of limitations on that particular part of this issue ran a long, long time ago, however, particularly since the licensor foolishly (admittedly, with some element of 20/20 hindsight) accepted a flat payment.

Comments are closed.