LEAVE A TIP IF YOU LIKE IT, I AM OUT OF BUZZ!!

What it does:

- Detailed and mostly accurate emotions/expressions

- Separate emotions and expressions for more than one character

- Helps with camera angles regarding viewer perspective height and character angles

- Weights between 0.5 - 1.0 depending on severity of expression

- Also works in portrait and landscape, I am just lazy at changing my sample dimensions

This lora is the third iteration and evolution of my previous expression lora. This has a completely new dataset that utilized exaggerated expressions to allow it to translate better to Krea. It also utilized a separate set of instructions regarding the direction and angle of camera along with the direction of the character's head and body.

Krea is great at so many things aside from emotions and expressions. This still isn't 100% perfect, but my hope was to get it to a place where we can use more detailed expressions for Krea.

How to use:

The training data consisted of a trigger "detailed expression and emotion" along with two sentences for each subject in the photo. The first sentence was a description of the character's physical traits (hair, skin tone, ethnicity, gender) as well as their position and orientation (on the left, looking forward, looking at the other character). For two character's there would be a total of 4 sentences regarding just the characters. After that I would personally do background, location, etc. for a new image.

Struggles/Known Issues:

- Clenched Teeth (it is not super good at this), you might get better result leaving it out regarding "grins"

- Moles (I think due to the variety of skin tones and people used it loves to add a mole), you can add "unblemished skin" to combat it

- Crying/Tears get weird and sus if you have a more NSFW checkpoint

-forehead can get a bit wrinkly on upset or unhappy emotions