Capturing the authentic aura of the original character is challenging, as the datasets differ across checkpoints.
However, I am committed to diversifying the personalities and appearances of male characters, and I plan to release updates featuring various characters from time to time.
I have extensively annotated personality traits and facial expressions, so, theoretically, the range of expressions should be richer.
Feel free to leave a comment if you have any suggestions.
"Grumpy face"—if no facial expression prompt is specified, the default is an unhappy expression.
Description
Comments (4)
Oh, this is good...!
Altho personally, I think illustrous checkpoint is great enough to capture each character personality traits with enough samples...
Also, are you using caption rather than tags for the datasets?
I use tags, but I prioritize a fixed order: the character's name comes first, followed by the top 3–5 defining traits or exclusive tags. However, I’m still experimenting with this and am not yet sure what the downsides might be.
@Echo80 Huh, but when I look the metadata, it looks like the using caption rather than the tags... I wonder if you know the differences. Using tags also enables them to use same kind of clothes like in datasets. You can see my models for reference~
@RaidenFan86You are right; I often use natural language for annotation. Due to limited time and energy, I haven't been able to verify every single entry, which has indeed affected the consistency of the clothing feature definitions.