This model will produce a picture that is a closeup POV blowjob.
It works best with descriptive phrases with a strength of .75 to 1.25. Depending on the prompt, reducing to .75 will produce a more accurate photo. It was trained at various frontal and diagonal angles, and while it does best at straight on, it seems to handle some angles. It also will produce decent pictures without the main trigger word as long as one or more of the trigger phrases are used.
The less detailed the description the worse the picture. It does have some strangeness such as with the football cheerleader photo, when I use the word football jersey it tended to create a football penis or penis in a football about 1/3 the time.
Description
This model will produce a picture that is a closeup POV blowjob.
It was trained from 50 photos captioned by Joycaption-pre-alpha-1
Trained trigger word is: frontview_bj
Trigger phrases: blowjob, penis, looking at viewer, looking at the man, looking directly at the viewer
It works best with descriptive phrases with a strength of .75 to 1.25. Depending on the prompt, reducing to .75 will produce a more accurate photo. It was trained at various frontal and diagonal angles, and while it does best at straight on, it seems to handle some angles. It also will produce decent pictures without the main trigger word as long as one or more of the trigger phrases are used.
The less detailed the description the worse the picture. It does have some strangeness such as with the football cheerleader photo, when I use the word football jersey it tended to create a football penis or penis in a football about 1/3 the time.