Version changes: Higher quality training and addition of trigger words.
Recommended Settings:
Cfg-Scale: 0.4-1.0 ~ Personally I use 0.8
Scheduler: Euler A
Size: 832 x 1216
Optional but recommended: High-Res fix, upscale, and perhaps face-fix
Add 'white pupils' and 'lips' depending on your face preferences. I also advice for either 'adult', 'mature', in your positive prompts, or 'child, kid, loli' in your negative ones, to avoid child-like characters.
Description
Higher quality training and addition of trigger words.
Recommended Settings:
Cfg-Scale: 0.4-1.0 ~ Personally I use 0.8
Scheduler: Euler A
Size: 832 x 1216
Optional but recommended: High-Res fix, upscale, and perhaps face-fix
Add 'white pupils' and 'lips' depending on your face preferences. I also advice for either 'adult', 'mature', in your positive prompts, or 'child, kid, loli' in your negative ones, to avoid child-like characters.
FAQ
Comments (9)
Do you have to use quality words from pony to achieve the effect?
I mean is that I tried the quality word embeddings, but they don't seem to achieve the desired stylistic effect.
I'm pretty sure those mostly affect the checkpoint model you're using and has not much to do with the Lora. At least I didn't use any 'quality prompts' in my training.
@Plantra I tried using [score_9, score_8_up, score_7_up, masterpiece, best quality, very aesthetic, (dynamic:1)] to align more closely with the style.
@JoyDopamine I mean it doesn't hurt, but not sure if it's really necessary... let me know if you notice a difference though :)
@Plantra I uploaded two pictures. Although it's not very obvious, the one with the pony quality word resembles your style more (eyes, tone, etc.).
@JoyDopamine Mhh true, but also could be because it affects the checkpoint and I'm using the same prompt, therefor it looks more like mine... hard to tell
@Plantra Of course, I'm sure Lora is definitely fine, but since the style is more cartoonish, the term "pony" might be more appropriate.
zit version ?