This originally comes from a bounty (https://civitai.com/bounties/486/anime-chibi-discord-react-emoji-lora). Please thank the author of the bounty as well if you liked the LoRA!
A LoRA producing anime-style chibi with thick lineart, and focused on creating emoji-like images in the style of Discord/Twitch/Vtuber emotes.
The checkpoint will have a significant impact on this LoRA. Most tested checkpoints produced good results, but some might produce "less cartoon-esque" results. Choose the one closest to the style you want to emulate. The recommended one is "AuroraONE", which suggestively produced the best results. I've had results almost as good, or even better, with "AbyssHellVer3".
You do not need to go high with the resolution to produce good results. Stay 512x512 for best results.
The LoRA should be compatible with most characters LoRA. Some emotions will depend on:
how the character LoRA was trained
the checkpoint you are using
Note: Use the tag close-up
or portrait
if it keeps generating the whole body and you don't want it.
Your prompt MUST include the tag chibi emote
, then you can add optional triggers on top of that.
Use weights ranging from 0.7-1.
angry
crying
(optional use with streaming tears
)
disgust (underfit)
drooling
(highly dependent on the checkpoint)
eating
+ the food
embarassed
(recommended with flying sweatdrops
/ blush
/ blush stickers
)
laughing
(recommended with +++
, but the latter triggers is highly dependent on the checkpoint)
one eye closed
(use instead of wink)
panicking
(highly dependent on the checkpoint)
pout
shaded face
(depending on the checkpoint, you might want to set the weight to 0.7-1)
sleeping
smug
(random to get it right)
sparkle
, sparkling eyes
(use instead of excited)
thinking
(recommended with hand on own chin
)
thumbs up
(some checkpoints might have trouble getting the thumbs correct)
Copy the .safetensors file to stable-diffusion-webui/models/lora
Add <lora:chibi_emote_v1:1>
to your prompt (or use a plugin)