US Copyright Office: Some AI-Generated Images Can Be Protected – With Conditions

The United States Copyright Office has released a much-anticipated report clarifying how US copyright law applies to artwork created with the help of artificial intelligence. The takeaway: if an AI-generated image includes enough original creative input from a human, it can be eligible for copyright – at least partially​.

However, works that are entirely the product of a machine, with merely a text prompt from a user, are still not considered human-authored and thus not copyrightable by that user. Shira Perlmutter, the Register of Copyrights, summed it up by saying that when a person’s creativity is expressed using an AI tool, that expression “continues to enjoy protection” under the law​. This means that simply using AI as one might use a camera or a paintbrush does not automatically strip the work of copyright – the human contribution is what matters. The Office’s position is that each case will be evaluated individually to determine how much of the final work was shaped by human decisions and originality​.

This guidance was published as part of a broader inquiry the Copyright Office is conducting into AI’s impact on intellectual property. It received tens of thousands of public comments and consulted with various stakeholders, from tech companies to artists’ groups​. The issue has grown pressing as AI tools like DALL-E, Midjourney, and Stable Diffusion allow anyone to create detailed images by just typing in descriptions. Where the line falls between human and machine creativity has been a gray area, and this report aims to draw that line more clearly.

The 'Human Authorship' Requirement

US copyright law has long held that only works of human authorship are eligible for protection. In recent cases, this principle was tested: one high-profile example involved a graphic novel that included artwork generated by Midjourney. The Copyright Office initially granted, then partly revoked, the registration for the images because they were AI-generated without enough human alteration. According to the new guidelines, had the artist significantly edited or guided those AI images to imbue personal creativity (beyond just entering a prompt), those images or aspects of them might have qualified for protection​.

The report specifically states that typing a text prompt and getting an image is not enough on its own to count as human-authored creativity​. On the other hand, if an artist uses an AI-generated output as a starting point and then modifies it, or if they curate and pick specific AI outputs and integrate them into a larger human-created work, those elements can be “expressive elements” attributable to the human author​. Essentially, the Office is willing to recognize the human element in AI-assisted art, but it insists on some perceptible level of human steering or modification.

Real-World Impact and Ongoing Cases

This guidance has immediate relevance to artists and creators experimenting with AI. For instance, an artist named Jason M. Allen famously won a digital art contest with an image he generated using Midjourney, and he has been trying to register that piece with the Copyright Office. They initially refused, saying the image was 100% AI-produced. Allen is appealing, arguing that his selection of prompts (he reportedly used over 600 prompts to arrive at the final piece) is a form of authorship​. Allen’s attorney expressed hope that the Copyright Office’s new nuanced stance might support his case​. If Allen can show he had a meaningful creative vision and control – choosing one image out of hundreds, tweaking it, maybe editing it afterward – the Office might grant a modified copyright that covers the human-authored elements.

It’s worth noting the Copyright Office is not changing the law (only Congress could do that); rather, it’s interpreting existing principles for new technology. Some legal experts say the report doesn’t revolutionize anything but provides “more detailed and nuanced analysis” of how current law applies​. The Office is also cautious, emphasizing that decisions will be case-by-case, and acknowledging that technology will keep evolving, which might necessitate revisiting these issues.

Another part of the report series (forthcoming) will deal with the training of AI models on copyrighted material​. That’s a whole other can of worms: many artists are concerned (and some have sued) over tech companies using millions of online images to train AI without permission. The Copyright Office plans to address whether such training data use might infringe copyrights or if it falls under exceptions like fair use.

Calm Reception with Some 'Wait and See'

So far, the creative community’s reaction to this guidance is cautiously positive. It acknowledges that creators using AI in their workflow aren’t automatically forfeiting rights to their creations. A photographer who uses an AI tool to enhance an image, or an illustrator who uses AI to generate backgrounds that they then paint over, can feel more secure that their efforts are protected. But it also draws a boundary to prevent someone from copyrighting an image that they had no real hand in making beyond a one-line prompt.

There is also the unresolved question: what about AI-generated works that no one claims copyright to – could they be freely used by anyone? By confirming that fully AI-generated works lack human authorship, the implication is those images are in the public domain, open for anyone to use. That has its own interesting implications and might encourage some folks to intentionally distribute content generated by AI as free stock or resources.

In summary, the U.S. Copyright Office is signaling that AI is just another tool – like a camera or a graphics software – and it’s the human creativity wielding the tool that copyright law cares about. This nuanced approach tries to protect human creators without granting monopolies on machine-made art. As artists continue to push the boundaries of AI, these guidelines will surely be tested, and over time more precedents in courts will further define the landscape.