Who Made This? The Copyright Office, an AI Painting App, and a Lawsuit That Could Reshape the Rules for Human-AI Collaboration

I get asked this alot, now. “Do I have a copyright in what I created with AI?” It depends.

You take a photograph. You open an app. You upload the photo, select "Van Gogh," dial the style-transfer intensity to your taste, and click generate. The app produces a painting-style image that looks nothing like your original photograph. You created something. Or did you not?

That is the question at the center of Suryast U.S. Enterprises, LLC v. Perlmutter, filed May 8, 2026 in the Central District of California (Case No. 2:26-cv-04999). The plaintiff, a California LLC that received an assignment of the underlying copyright claim, is challenging a final agency action by the U.S. Copyright Office refusing to register an AI-assisted image called "Suryast," a Hindi word for sunset. The case is a direct challenge to what the complaint calls the Copyright Office's Human Authorship Requirement, and it raises questions that every creative professional using AI tools should be paying attention to.

The Creative Process

In 2020, Ankit Sahni, an Indian lawyer and photographer, took a photograph of a sunset over a building using his phone. He then fed that photograph into an application called the RAGHAV Artificial Intelligence Painting App, selected Vincent van Gogh's The Starry Night as the style reference image, and set a numerical value controlling how heavily the style would be applied. RAGHAV generated the output. Sahni did not modify it afterward. The resulting image, which he named "Suryast," blends the structural elements of his original photograph with the swirling brushwork and blue-gold palette characteristic of Van Gogh's most famous painting.

Sahni filed for copyright registration in late 2021, originally listing both himself and RAGHAV as co-authors. That decision, made in the spirit of full disclosure, turned out to be the first of several strategic complications.

The Copyright Office Said No. Three Times.

The Office's first refusal, issued in June 2022, concluded that the work "lacks the human authorship necessary to support a copyright claim." Because Sahni had disclosed that RAGHAV "created and rendered" the image, the Office found that any human creative contribution could not be distinguished from the machine's output.

Sahni and his attorneys at Day Pitney LLP filed a request for reconsideration, dropping RAGHAV as a co-author and reframing the argument: Sahni selected the content image, his own photograph; selected the style image, The Starry Night; and selected the degree of style transfer. He exercised judgment. He made choices. RAGHAV, they argued, was an assistive tool, no different in kind from a camera's autofocus or Photoshop's blend modes.

The Office's second refusal, issued in April 2023, reclassified Suryast as a derivative work. That reframing was significant. A derivative work must contain original new authorship beyond the preexisting work it transforms. The Office determined that the new expressive elements in Suryast were not contributed by Sahni. They were generated by RAGHAV.

Day Pitney pushed back again, this time before the Copyright Office Review Board. Their second request for reconsideration made three arguments: Sahni provided traditional elements of authorship; the work was not a derivative because the final image was not substantially similar to the original photograph; and the Office was applying an impossibly demanding standard that would render virtually all AI-assisted art unregistrable.

The Board issued its final decision in December 2023. It was thorough, carefully reasoned, and rejected every argument.

The Board zeroed in on a fact Sahni's own technical description had established: RAGHAV does not function like a Photoshop filter. It does not simply overlay one image on top of another. It uses a neural network, trained on 14 million images, to generate a new image based on features learned from the base photograph and the style reference. The Board concluded that it was RAGHAV, not Sahni, that decided where the building would sit in the composition, how the colors would be applied, and how the swirling forms of The Starry Night would be interpolated against the sunset photograph. Sahni's choices constituted, at most, an unprotectable idea for a work. Copyright does not protect ideas. It protects expression.

As for the strength-of-style-transfer variable, the Board held that selecting a single numerical value is the kind of de minimis authorship that Feist Publications v. Rural Telephone Service Co. has long excluded from copyright protection.

The APA Challenge

Having exhausted the administrative process, Suryast LLC filed suit under the Administrative Procedure Act. The complaint asks the court to set aside the Final Refusal as arbitrary, capricious, an abuse of discretion, and contrary to law, and to order the Copyright Office to register the work with Sahni as the author.

The complaint's central argument is broader than the facts of this particular case. Plaintiff argues that the Copyright Office's AI guidance, which describes a case-by-case analysis of whether a work is "basically one of human authorship, with the computer merely being an assisting instrument," operates in practice as a per se rule against registration whenever AI tools are involved. The Copyright Act, the complaint argues, contains no requirement of exclusive human authorship. Section 102(a) protects works fixed "with the aid of a machine or device." Section 201(b) recognizes that the "employer or other person for whom the work was prepared" is considered the author for work-made-for-hire purposes. Neither provision, the complaint contends, requires that every element of creative expression originate in an unassisted human mind.

Whether the court will find that argument persuasive is a separate question. The Office's position rests on substantial precedent, including Trade-Mark Cases, Burrow-Giles, and the more recent Thaler v. Perlmutter decision from the D.D.C. affirming the Office's refusal to register a work created entirely without human involvement. But Sahni's case is factually different from Thaler. The image in Thaler was produced autonomously by an AI system with no human selection of content or style inputs at all. Sahni, by contrast, provided two photographs and a numerical parameter.

Where AI Authorship Claims Stand Right Now

It helps to think about the pending and recent AI copyright cases as occupying different positions along a spectrum of human creative involvement.

At one end is Thaler. Stephen Thaler's "Creativity Machine" produced an image with no meaningful human input beyond the decision to run the machine. The D.D.C. affirmed the Office's refusal in 2023. The Supreme Court declined to take the case in March 2026.

At the other end is the ongoing dispute over Jason Allen's Théâtre D'opéra Spatial, which placed first at the Colorado State Fair in 2022. Allen used Midjourney to generate images based on detailed text prompts, then further edited the results in Photoshop and Gigapixel AI before printing and submitting the final piece. His federal suit in Colorado challenges a refusal on facts that involve substantially more iterative human direction of the AI output than Sahni's workflow.

Sahni's case sits in the middle. He provided a human-authored content image, a stylistic reference, and a degree-of-transfer parameter. He did not prompt from scratch, and he did not modify the output. The question is whether that level of directed selection is closer to Allen's iterative creative control or to Thaler's fully automated production. The Board concluded it was closer to Thaler. Plaintiff argues it is not close at all.

Why This Matters Beyond Fine Art

The practical stakes extend well beyond fine art litigation.

Graphic designers routinely use AI-assisted tools that apply style transfers, generate background elements, and automate color grading. Photographers use AI-powered editing software that makes expressive decisions about tone, detail, and composition without step-by-step human instruction. Architects use generative design software that produces structural forms based on parameter inputs. The Copyright Office's current framework, if applied consistently, suggests that the human-authored outputs of all of these workflows might be unregistrable to the extent AI contributed any expressive element.

That has implications for licensing, infringement enforcement, and the basic calculus of whether creating AI-assisted work is commercially worthwhile.

The Office has, in fact, registered some works containing AI-generated material where the AI contributions were disclaimed and the remaining human authorship was deemed sufficient. Graphic novelist Kris Kashtanova received a registration for Zarya of the Dawn, though the Office later limited the registration to exclude Midjourney-generated images and retain only the human-authored text and arrangement. A photographer received registration for AI-enhanced images where substantial post-processing by a human was documented. The line the Office is drawing, imperfectly and case by case, appears to be this: documented human creative decisions that survive independent of the AI output are protectable; AI-generated expression, even when directed by a human, is not.

The Suryast complaint challenges whether that line has any basis in the statute.

What to Watch For

The Central District of California will review the agency record under APA standards, which generally means the court will ask whether the Office's decision was supported by substantial evidence and whether its legal interpretation of the Copyright Act is permissible. Chevron deference is off the table after Loper Bright Enterprises v. Raimondo, which means the court will exercise independent judgment on questions of statutory interpretation rather than deferring to the Office's reading of the law.

That is a meaningful shift. The Office's Human Authorship Requirement rests heavily on its own Compendium of Practices and on judicial precedent that predates the current wave of AI image generation by decades. A court conducting independent statutory analysis might reach different conclusions about what "author" means in the Copyright Act than the Office's own registration examiners have reached.

Watch also for amicus participation. Cases like this tend to attract filings from AI companies, creative-industry trade associations, and academic copyright scholars, all of whom have institutional interests in how the authorship question is resolved.

Finally, Congress has held hearings on AI and copyright for three years without producing legislation. Suryast, Thaler, and the Allen litigation are likely to provide additional pressure for a statutory resolution that courts cannot manufacture on their own. Until that happens, anyone creating work with AI tools should assume that the copyright protection they think they have may not exist, and plan accordingly.

Jonathan Phillips is an attorney at Phillips & Bathke, P.C., who represnets clients in copyright, trademark, and intellectual property matters. Contact him HERE.