[Forenote: the title of this post references a volume of Spike Milligan’s memoirs, and its claim should be taken with the same degree of seriousness]

Back in September 2022 I posted some initial thoughts here about AI Art, trying to be as objective as I could.

Things have changed since then: the AI Art “industry” has ramped up its hyperbole (and energy consumption) significantly. And my opinion on AI — specifically on how AI is being developed and used under the demands of late-stage capitalism, and on the low quality of its output — has only become more negative.

Written AI spam now has its own name — slop — and AI-generated images are the Office ’97 Clip Art of today - instantly recognisable, and rapidly becoming a telltale sign of a lack of quality of whatever they’re adorning (and, by turn, of any product they’re promoting).

As you can probably guess, I’m not in favour of my artwork being used to train AI image generators without my permission, so here’s what I’ve been doing/investigating:

Opting out on third-party services I use (where possible)

I recently completed the “Object to Your Information Being Used for AI at Meta” form accessible via a notification from Meta on my Instagram account, and received an email from Instagram saying We’ve reviewed your request and will honor your objection.

Chances of it stopping them using my images without my permission: possibly this will work in the short-term unless/until Meta decides to rewrite its Terms and Conditions to force people to have their data used. At that point it’s a decision between closing my Instagram account (which does draw in a few sales of my artwork) and giving in to Meta’s use of my work.

Telling AI scrapers that they can’t use my website’s text/images (done)

My robots.txt file tells multiple known bots/scrapers that they are forbidden from accessing my site, and additionally I’ve also set up an ai.txt file, based on a proposal that’s been going round.

The trouble with the robots.txt approach is that it’s whack-a-mole - you have to individually exclude scrapers when you discover them.

Chances of it stopping them using my images without my permission: Very low - I suspect the AI wantrepreneurs are too busy dreaming of becoming the next billionaire Bezos or Zuckerberg to worry about whether they’re following long-standing conventions such as robots.txt (or indeed laws).

Protecting and poisoning my image files (investigating)

The Glaze Project has released two interesting pieces of software that you can use to post-process your images before uploading them.

The first, Glaze, disrupts style mimicry by AI systems. The second, Nightshade, works in a similar fashion to Glaze, but instead of just disrupting AI’s ability to mimic the style of an image, it “poisons” the AI learning model.

I’d use both of these in a heartbeat, but unfortunately they’re only available for Windows and MacOS, and my entire set-up is running on Linux.

There is an online web version called WebGlaze, but it’s invite-only and the process is cumbersome (it emails you the “glazed” images).

Personally I think that the ideal solution for the Glaze Project is to see if it can get Glaze and Nightshade incorporated into a library such as ImageMagick (or their own separate library with a PHP extension, Java Interface etc) so that people could configure adding Glaze and/or Nightshade protection as an automated part of the server-side image upload process, in the same way that WordPress resizes your images on upload to a blogpost.

I suspect that the library idea of mine is not going to happen soon, if ever, so it would be great if Linux desktop versions could be developed and released soon.

Chances of it stopping them using my images without my permission: it won’t stop AI companies scraping and using my artwork against my explicit wishes and without my consent, but it might make them wish they hadn’t. Poisoning the well is looking like the most viable option.