Last update 13-05-2025
Hello fellow prompters, AI engineers and other future people. I tried these tags on a lot more models and updated what works and doesn't. This journal doesn't get a lot of attention, but I hope it helps those who have trouble with this. No credit or buzz needed, keep that to prompt with. :)
I focus mostly on negative prompts for "eyecontact". To reinforces eyecontact without the use of the word "eyecontact" or synonyms like POV, POV-eyecontact, looking-at-viewer, looking-back, etc.etc. More on that later.
negative prompt tha works for all models: Blinking, winking, staring-stares-stare, gazing-gazes-gaze, distracted-distraction, sleeping-unconcious-drunk, eyecontact.
(Pony & Illustrious & XL to some degree understand "winking" however with pony&Illustrious "looking-at-viewer" is all that is necessary in the positive prompt in my experiences, if this doesn't work your base model is "corrupted").
1.5/2.0/2.1/SDXL/3.0/3.5/Flux pos: looking at photographer.
1.5/2.0/2.1/SDXL/3.0/3.5/Flux neg: Unaware, shuteye, candid, distracted, spycamera, cctv, facing away.
Hit or Miss: ASMR eyes, POV solo eyes. these have a 50/50 chance to get the eyes to look at the viewer (you).
With SDXL I noticed adding: "solo" to positive greatly increases succesrates. Also most SDXL models seem to need more guidance and descriptive prompting.
And the opposite is also true, adding team, squad, group, party, mass, crowd to negative helps maintain more control.
Don't use eyecontact in scenes with more than 1 character. Because it often forces them to look at each other.
Hope this helps, and would love to hear your feedback! Happy prompting!