Collection of GPT-image-generator 2.0 issues, bugs, and work-around tips (check first post)

When those “blocky / pattern” images are released on the internet, new GPT’s can use those for theire trainingdata.

Then it will be trained on blocky images with repetitive patterns, resulting in models that will generate images like that by default.

Same happened with Image Generation 1.0 which had a very yellow-ish tint in the beginning for almost all images, since it was mainly trained on Instagram-images with those yellow-movie-like filters.

So, it can result in an “Droste effect” where the main image is a copy of a previous image, which is a copy by itself from a previous-previous image, etc…

Anyhow, I created a set of own scripted Photoshop filters, tried the default moiré-filter (Moiré pattern - Wikipedia) (Photoshop), adjusted the prompts to “remove the repetitive pattern” (which made it even worse, since you are altering an already bad image), etc…

Nothing did work.