Sign In

how to prompt in Stable diffusion. hair\poses\emotions\gestures\camera\effects\themes

Basic


First things first, try find a good quality checkpoint.. try avoid checkpoints merged with lora because creativity and compliance with the request are significantly reduced. for example, can produce very few varieties of hairstyles or faces. can ignore many tags on clothing or background and so on ad infinitum. checkpoint merged with lora is the same as you use that loras just for an extra model. They can be distinguished by complex verification tags, for example 225 tokens. in most cases a bad model will ignore 90% of the request here few examples with three different checkpoints with 221 token request at same seed. compare to tags - first work great, and assemble all prompts entered other ones seems like wont know what i want from them. first is mine checkpoint yutamix13. second yutamix11 merged with lora and last one 777 private model merged with loras also. as you can see difference is huge.

MANY TAGS GOOD OR BAD? - THIS IS ONLY DEPENDS ON CHECKPOINTS. AND LORA USED INSIDE. GOOD CHECKPOINT WITHOUT HIGH STRENGTH LORA OR WITHOUT LORA AT ALL CAN PRODUCE AWESOME DETAILED ART WITH ANY HUGE AMOUNT OF PROMPTS. SO YES MANY PROMPTS GOOD AND CAN PRODUCE UNIQUE ART. BUT FOR OUTDATED CHECKPOINTS OR WORST CHECKPOINTS MANY PROMPTS INPUT MAY BE DIFFICULT OR IMPOSSIBLE AT ALL.


TAG STRENGTH HOW IT WORK - FOR EXAMPLE WE MAKED FINAL PROMPT AND SOMETHING WE WANT DOES NOT APPEAR ON ART, NOT A PROBLEM. JUST EDIT "(YOUR TAG:1.2)" MAXIMUM STRENGTH MUST NOT EXCEED :1.9 OR YOU GET REALLY WEIRD PICTURES

now lets talk about hair prompting...

Multicolor hair in stable diffusion


except just common hair color you may want not just pink for example? maybe you want pink and blue?

this is not easy task. type (pink hair:1.1), (blue hair). remember about some hierarchy. main color must be higher than secondary and also you may require more strength to hair color. like (pink hair:1.5), (blue hair:1.4). and remember also SD accepting all colors prompted. not in exact that place you prompted., this is random. off course worst quality checkpoint cant handle such tags as multicolored hair, also in that cases you may require use hight strength prompt (solo:1.5) if you use high strength hair color, because it just generate two girls with that different colors. so choose good checkpoint for these tasks, like mine

hair color intensity...

example you dont want just pink hair, but you want light pink, just write to prompts as it is, and dark pink for opposite "dark blue hair" for example. also there are "Pale" prompt example - pale pink hair works like light pink hair

hairstyles prompts\tags in stable diffusion

Straight hair

ponytail

side ponytail

twintails \ alternate (Pigtails)

low twintails

braid

hairbun

half updo hair

messy hair \ bedhead

ahoge

\

bang over one eye \ hair over one eye

blunt bangs

shaggy cut

drill hair

floating hair

bob cut

chignon

wavy hair

two colored eyes - heterochromia in stable diffusion

this one simple enough but may wont work on some checkpoints just write prompt like this example -

red blue heterochromia

Emotions prompts\tags in stable diffusion

I won’t show completely banal emotions. I'll show only the interesting ones

worried

grin

naughty smile

smirk

scowl

sulky

whimpering

tearful

bashful

embarrassed

nose blush

fang

Poses prompts\tags in stable diffusion

some interesting poses what base model can generate without loras

hug own legs

Bent over

open arms

arm over hair

arm on chest

hiding arms

crossed arms

(arms behind back:0.9) - not very stable pose but work in most cases correctly

dynamic pose - adds some effects on backround

reaching towards viewer

reaching towards viewer hug

looking away

all fours

squat

claw pose

paw pose

hands up, elbows on table, face palm

arms on hip

Gestures prompts\tags in stable diffusion

finger pointing

facepalm

put index finger on mouth

thumbs up

peace fingers

Camera settings in stable diffusion \ view settings


close up

from above

from below

from side

full body

from behind

camera effects in stable diffusion


bokeh

fisheye

grayscale \ as a negative works opposite, less gray color

short theme setup \ theme prompt \themes in stable diffusion

Let's say you don't want to bother writing prompts for each thing separately. and here prompt "theme" comes to the rescue, how it works just write in prompt what you want in one word. watch examples..

wedding theme

halloween theme

christmas theme


japanese theme

Ending

this should be enough to know some sd features... if you have something to add here, feel free for share in comment. hope this guide helps unexperienced users get better use of stable diffusion

34

Comments