AI Art Made Entirely with Prompts
NOTE: All nudity was inpainted with Fooocus.
Whistleblower
This project required unusually creative thinking due to Imagen 4’s lack of access to a ControlNet or FaceSwap mechanism to control poses and maintain consistent character appearance. The net effect is that no two images will ever look alike regardless of how detailed your prompts are. That includes every nuanced detail of every image you generate, whether it’s the shape of a character’s eyebrows, an earing in one image and then missing in another, or the position of an object in relation to the room it’s in. Consequently, I had to relegate myself to one image for each element I wanted to feature in order to maintain what little semblance of consistency between frames that I could. My hope is that your imagination can fill in the holes.
Incidentally, this is my first and last experiment with a multi-panel comic using Imagen 4. The model is too restrictive and costs too much for the number of compute units I had to burn through to get one usable image for each panel.