Getting the right facial expression for the situation is not always easy.
⠀⠀
I was just about to create a new wildcard for myself and, as always, first tested exactly how each prompt would affect me. I knew beforehand that I had never really been successful with certain facial expressions, so I had to document that. Maybe you're interested too...
⠀
Some people create wildcards with supposedly billions of prompts, which is absolute nonsense. Just for the very important reason that 90% of these “prompts” don't work at all. But hey, big numbers always look better than a few that are good!
⠀
So I have tested prompts like >amazed< and also with the increases like very amazed and (((amazed))) or in the new spelling (amazed:1.5). I immediately noticed that amazed doesn't actually work at all. So it's bad for all those who want to create those exaggerated YouTube faces for clickbait videos 😖
⠀
What doesn't work with a prompt, whether SD 1.5 or Flux, can often only be realized with a LoRA....
⠀
⠀
Now to the result
My favorites
⠀
My rarely used
⠀
The expressions that rarely or never work
⠀
New Update Face Expressions
⠀
⠀
For my test I used a solid Pony model and this is the prompt for it:
⠀
Positive:
score_9, score_8_up, score_7_up,
a realistic photo of a woman, >amazed<
⠀
Negative:
score_3_up, score_4_up, score_5_up,
⠀
Metadata:
Checkpoint: realDream_sdxlPony11, Sampler: Euler, Steps: 40, CFG: 10, Res: 1024x1024
⠀
⠀
⠀
The amplification as in the example (amazed:1.5) should only be used with caution as such a high amplification can overwrite other prompts, but an attenuation such as (amazed:0.7) can be used without problems. I am only very slightly amazed :)
⠀
I still had a column with >very< in the table, but i deleted it because, for example, happy and very happy hardly differ at all. For all expressions. And because you can't write properly and show something well here anyway due to the upright smartphone format in the Articles, the table only has 3 columns instead of 5 like mine.
Are you missing any expression that works well? Let me know! otherwise I would be happy about a like.
⠀
⠀
⠀
⠀
⠀