Yes, I have used AI to generate images.

[Re-published with edits July 2nd]

As the discussion on AI generation evolves, we’re seeing the edges of the dialog sharpen into focus. It’s becoming clear that artists are indeed being disenfranchised and their original works appropriated. This is unequivocally true and also obviously wrong. The problem with the AI morass is that it’s possible to source plenty of open-source imagery to train image generators. We can be reasonably certain that the image generators have scraped everything there is to scrape, which means that tangled up in the work of hard-pressed artists and creators is a mountain of free-to-use content.

I would prefer that Dall-e and ChatGPT and the others start with only free-to-use content and have a specific toggle that enables the theft of intellectual property. That would go some way to settling the discussion over theft, and artists could reasonably expect that an AI wasn’t going to go into business producing images that people would use in lieu of paying them for their talents.

I had originally been somewhat snarky about how that Pandora’s box has been opened, but with the recent improvements to ChatGPT, I’m more comfortable knowing that I can specify that the images produced are not taking away someone’s livelihood.

Yes, the discussion isn’t that simple, but we’ll need to take the improvements where we can.

The images on this blog were all re-generated from scratch using a prompt directing ChatGPT to reference only public domain references or sources with clear Creative Commons licenses or open data sets. If readers have suggestions on how to further refine my sources I’m open to ideas.