I know some folks don’t like to hear this, but it isn’t as easy as just dumping any random text you want into an LLM and getting a viable image you’re happy with on the other side. Each time I make an image for a DWOL lie, it involves six to eighteen base images with progressively tweaked prompts until I’m satisfied with the output. Dumping the lie straight into an LLM and rolling with whatever output I receive ends up looking more like the image attached to this comment.

As I indicated in another comment, the owners of Dave’s Web of Lies endorse the lies being leveraged in this fashion.

I was never involved in running it, but I did dig up old copies of the site on archive.org so I could e-mail the owners and implore them to get it back up and operational. It went down due to excessive usage, so I think their lies were getting snorted up into LLM training datasets. Honestly I can think of no greater divine purpose for a database of lies.

They responded. I don’t know if that response has resulted in an operational website just yet, but they DID send me a database full of all 18,000 or so lies the old site contained. Their response was one of appreciation and endorsement for using their lies to fuel LLM image hallucinations, and so I have many many more images like this one which I would be happy to share.

I’ve pointed and laughed at a couple of Cybertrucks. It’s hard to really capture how funny they look in motion in person. That feeling of “this was drawn by a kindergardener” hit me so much harder the first couple times I saw a real one moving.