AI-Generated Art and Copyright Infringement

Scott-Martin Kosofsky

The New Yorker recently posted an article about the class-action lawsuit filed in November by attorneys Matthew Butterick and Joseph Saveri against the AI image generators Midjourney, Stable Diffusion, and DreamUp. The legal matter at hand has nothing to do with typography, per se, but some of you will recognize the name of Matthew Butterick, who began his career as a typographer and type designer, and was part of Font Bureau in its early days.

Here is the link: https://www.newyorker.com/culture/infinite-scroll/is-ai-art-stealing-from-artists

I am interested to know if any of you believe that there is a potential link here to AI-generated type design, if not from a legal standpoint (that's a lost cause in U.S. law), then from a practical one.

Thomas Phinney

Scott-Martin Kosofsky said:

if not from a legal standpoint (that's a lost cause in U.S. law), then from a practical one.

I disagree.

The US copyright office accepted registrations for fonts for about 25 years (~ 1992–2018) , before recently beginning to reject most of them, asking for a form of authorship that makes no sense at a technical level. It seems very much unresolved.

Scott-Martin Kosofsky

Tom, while I am sympathetic with your feeling about this, I’m afraid that you are engaging in wishful thinking. When the U.S. Copyright Office published its revised Compendium of U.S. Copyright Practices, in 2021, it went to some length to specifically exclude typefaces (see §313.3[D]), though it did suggest a right to register some part of the work as a computer program, as it had earlier. However, as you point out, the revised requirement to separate the original authorship portion of a font from the preexisting code of the design program instructions is very near impossible (see §723), at least at this time.

Perhaps more to the point is to ask whether it would be possible to defend in court what would be, at best, a very limited copyright registration—or  whether those registrations filed before 2019 were ever defensible. I don’t know of any case law in which those registrations were defended successfully, or ever defended at all. Or have I missed something?

U.S. Copyright law is, and always has been, centered on the idea of authorship. The growth of regulations around that central notion has always been focussed on establishing de minimis standards of what authorship means. Unfortunately, what looks like a big thing to those of us professionally involved in type design is often subliminal to those outside the field. The Supreme Court has always been averse to ruling in cases in which only groups of experts are able to discern one thing from another. I highly recommend reading all of Chapter 300 of the Compendium to get a better idea of the standards for what is copyrightable in the U.S. (I am aware that many of these things are viewed differently in other countries.)

I’m not an attorney or a legal scholar, but I’m very much involved in multiple aspects of publishing, so I’ve made it my business to have a good understanding of copyright law, both to avoid legal problems and to protect myself from legal bullying.

Christopher Slye

As long as I’ve been in this field, it’s been clear that U.S. copyright was about the font program, not the typeface’s visual identity. (That was/is covered by a design patent.) So it’s not at all surprising to see this strongly asserted in §313.3 there. While I understand the current disagreement about authorship, my reading of the above sources (§723) still suggests (to me) that copyright could apply to a font program. There are various references to source code, and a lot of font developers here know they can put a lot of unique authorship into a font’s source code (e.g. layout features).

I’m not saying I’m sure about what is or is not viable with today’s copyright guidance, or what would prevail in court, but I’m with Thomas in thinking that it seems “very much unresolved.”

Thomas Phinney

Scott-Martin Kosofsky said:

I don’t know of any case law in which those registrations were defended successfully, or ever defended at all. Or have I missed something?

Yes, the elephant in the room: Adobe v SSI (1998)
https://en.wikipedia.org/wiki/Adobe_Systems,_Inc._v._Southern_Software,_Inc.
This decision was a very big part of the basis of people in the font business thinking that their fonts were protectable by copyright.

Thomas Phinney

Additionally, typefaces and fonts are not the same thing. The general belief in the font industry in 1998–2018 was that typefaces (the designs) were not protected by copyright, but that the code of a font was, as software.

Dave Crossland

A v SSI was a lower court tho, so that may turn out to have been wishful thinking.

Thomas Phinney

Oh, absolutely. The Adobe v SSI decision was a lower court and not a precedent. But as (AFAIK) no other US court ever specifically ruled on the digital font copyright issue at any level, it has been all we had.

John Hudson

It was a lower court, yes, but has the benefit of Judge White’s investigation of the technology and subsequent summary judgement being remarkably thorough. He didn’t rely on inapplicable analogues, and actually took the time to learn how the fonts were made and recognised a significant level of authorship in the creation of the digital outlines within the very general concept of software as defined by USCO. So while that doesn’t establish a legal precedent in the way a case decided in trial would have—SSI settled after receiving the summary judgement—, it does provide a model of how such a case can be argued and evidence that if the characteristics of works in a particular technology are properly examined, instead of ignored in favour of lazy analogues, a proper appreciation of their authorial signficance is possible even for a non-expert.

JoyceKetterer

My understanding is that design patents are very tricky to maintain, and only last 15 years.  But I've considered it several times.  Maybe AI is a good reason to do it now.  I'm not sure. 

That said, I don't think it's true that copyright of font software is settled either way.  If this Zazzle/Laatz case goes to court I think she has a good chance of winning. 

Double that said, I've long believed that at the point where you're explaining fonts to a judge you've already lost (regardless of the purpose of the explanation). Metaphorically lost, of course, not literally. Or perhaps spiritually lost.  

John Hudson

I’ve not yet registered US design patents—or gone through the equivalent Canadian industrial design registration—for our library fonts, but some of our clients have registered design patents for typefaces we have created for them. Notably, Microsoft obtained design patents for a lot of typefaces, which I suppose is relatively easy for them to do because they have a large legal staff to look after such things (also, I suspect managers get brownie points for increasing the quantity of Microsoft’s patent assets).

Am I right in remembering that patent protection is a bit like trademark, in that one is obliged to defend it (unlike copyright, which persists whether actively defended or not)?

Stuart Sandler

A font design application such as Glyphs or FontLab need only add a small toggle-able UI window that shows the font as live XML code which can be edited in realtime and/or allows actual XML files to be opened, edited and saved via the application to conform to the US copyright office requirement of hand coded readable source code.

John Hudson

or allows actual XML files to be opened, edited and saved via the application

The FontLab JSON language .vfj source format is probably the closest to what you describe.

Thomas Phinney

FontLab 6+ has EXACTLY what Stuart describes. Just go to Windows > Panels > Source

- the Source code window displays the JSON code of the current glyph
- changes made in the drawing interface are instantly reflected in the Source code view
- changes made in the Source code view are displayed in the drawing interface whenever you hit the “apply” button at the top of the window

Additionally you can decide whether to save files in the FontLab binary format or the functionally equivalent VFJ JSON format (which uses the same code mentioned above), import/export fonts as VFJ, etc. You can also do things like copy/paste the Source (VFJ) text for a glyph into a text message or email text, and then somebody at the other end can drop it into a glyph. It is rather cool.

John Hudson

Very cool.

We’ve also used VFJ as an inter-tool source format, in preference to UFO, when converting FontLab data to VOLT, and wrote some scripts to operate directly on VFJ sources rather than to run internally in FontLab. I’d like to do more of the latter, and build out a complete VFJ toolkit.

https://github.com/TiroTypeworks/TiroTools/tree/master/VFJtools

Dave Crossland

https://llmlitigation.com/case-updates.html hasn't been updated yet but it looks like the court rules partially on the first Butterick suit, this week:

https://www.theregister.com/AMP/2024/02/14/tremblay_vs_openai_update/

https://regmedia.co.uk/2024/02/13/tremblay_vs_openai.pdf