#๐ฅ๏ฝanime
1 messages ยท Page 122 of 1
in my cmd, keep saying no target word is detected in prompt mode
Yeah

or default, i suppose
getting close tho ๐ฎ
this be cool shit ^^
i think i might've figured it out ๐ฎ
"uwaaaaa"
okay, i don't understand where im getting wrong
so basically threshhold is how hard it should take notice of the activation regions

do you get the masking images?
No target word is detected in Prompt mode So what decides target word?
im getting this
it should be the last prompt after BREAK
girl in forest BREAK cute, girl
as far as i can tell
the first prompt has all the words that will be scanned for activation regions
then after BREAK , you specify what you want to do with it, and the last word after a comma, is the activation word to look for
so i'm trying to replace girl with cute, in this case
that's a weird sentence if you remove the context/comma
๐ค
1girl BREAK face BREAK legs
resulting image
BREAK (red leaves:1.5) winter, forest
BREAK yellow demonic smile, moon
BREAK angelic wings, girl
BREAK purple sexy summer, little chibi sticker angelic wings, dress ```
so basically, what happens, i have the prompt girl in a dress in winter, walking through af forest at night, sky moon
that's going to be the initial generation
right, so the important part that i was missing, you're going to have to set the treshholds for EVERY break
example, tell me if im wrong ```(masterpiece, best quality:1.2), high resolution, detailed background, perfect lighting, extremely detailed, realistic, photorealistic,
1girl, white hair, long hair, red eyes, shirt, forest BREAK
(sun:1.3),shirt```
so this prompt, should have only 1 threshold right?
1 treshhold
and 1 keyword
and you should check that the mask is not 100% black or white
which is shirt
but it's not even detecting the word
like straight away, so i just interrupt
you're in prompt-ex?
i sometime use line breaks to keep my prompt nice looking but never tried regional prompt
hmm i just turn on SD after 2 months, now it have this
might wanna wiggle those treshholds a bit...
vae, etc is normal as last time i'm used
It's trying to communicate with you
There should have been 3 regions
the hands is the least of your concerns
every images gen have that color fragments
nope

cfg high?
model issue?
nope
controlnet right?

angelic leaked again to the moon
welp
yes
do i make like a outline of the letters and use canny or lineart
most probably, people are putting those qr generators to hard use
i think that's the QR modules
try with softedge and depth
okk
the qr modules have images clamp to bright and dark regions, that's how they do it
i had a feeling it wasnt controlnet for the really good looking ones but i just realized controlnet would probably do it just fine
i was just thinking about how ai is supposed to be bad at text
oh yea u right thats the qr controlnet thing
imma update my extensions 
oh qr is controlnet
oh ha
there's no qr module for that yet
thats ok 1.5 is the future
https://civitai.com/models/90940/controlnet-qr-pattern-qr-codes its prob this one
oho neat
thx for the treshhold tip @gloomy scroll โค๏ธ
Np
downloading that ill try a fe ways once i get some energy
Well now I got magic circle behind her but beach disappeared
some mask took over, pretty sure

it shows only one mask
and how many BREAKS do you have?
WITH detected words?
i clearled up like 10 gb of storage with that model converter extension last night and only did like 3 models
after installing extension, did you restart a1111?
i already have that extension previously
but i just updated
ah ok
Can you check what's broken with my mess of a prompt
if it works, it should straight away say something like target word detected right?
@lyric zealot what was wrong? ๐
got 2 masks going now
trying to figure out why 3 and 4 not working yet
my Use common prompt was on
and so it should have another ADDCOMM
but it didn't
so it took the BREAK as ADDCOMM?
or someshit
I believe that's what it does
ADDCOMM is just for visualization
oh yeah
BREAK does same thing
ADDCOMM and stuff is all 'labels'
it internally translates to BREAK
but it makes reading it a whole lot easier

because previously i was doing the trio waifu, so was using ADDCOMM
then now for testing, i forgot to turn it off

now it's working

I can't get mine to work 
Now it kinda works but I don't get any masks

jesus
Okey that prompt is broken somehow
ehhh
It's missing ADDCOMM and added first line after every BREAK
i think i see the problem
like example beach, BREAK
but you shouldn't have the ,
i mean its fine for first one
but
sorry was afk for a bit, still investigating vils promt

Well "working"
A. FUCKING. COMMA

Well I don't think I changed anything else
ahhh, then it's okay to have , there
good to know
Yes, I know who was making those
I found them very funny
oh. and prompt ordering is important too
anthing that comes in a later BREAK will overwrite it's region on earlier ones
@gloomy scroll ```anime
ADDCOMM
girl,
magic circle behind her,
standing on beach,
BREAK magic spell, runes, transparent, , circle
BREAK (shorts, short white hoodie, sneakers, modern clothes, small breasts, open hair, long hair, straight hair, white hair, hime cut, blunt bangs, azure eyes, pointy ears, dark, choker, fangs, blush,(full body), standing, facing viewer, happy, :D:1.3), girl
BREAK warm, bright day, sky, clean sky, outdoors, scenery, beach
@wispy canopy U know how supermerger works, right?
(hands:4.6)
nah, usually, those blocks all have a piece of the puzzle -> so the stuff for hands is in all blocks, but mostly concentrated in the middle blocks
early blocks for composition, middle blocks for "filler" , and then the high blocks that determine outcome style
in general

Why is there 3 black and white images
i'll give it a try
those are the masks
region prompting, in prompt mode, makes it's own masks to where to apply the prompts
so i have one for the circle, one for the girl, one for the beach
and black and white images are where it thinks the things where
Does it make them itself
yes


But why
totally working 
Literally my type of image when I try it 
yeah, i moved beach in front
beach in front with a lower threshold was the right choice
BEHOLD! vil's prompt!
This is what I'm trying to replicate with this different method
fkn wicked prompt @gloomy scroll
so basically, that means we can now say what stuff goes where, the bright sunny day applies to the beach, but not to the magic circle
the magic circle has runes, and should be transparent, but this does nto apply to the girl
the girl should have some features like white hair and stuff, but that all should not apply to the beach
being able to split the attention to the right areas is a HUGE boon
In this one is like the magic circle replaced the sun for the light source?
what happened here, was that the beach was the lastest and i gave it very high treshhold because of that. what happens is that it didn't apply a lot of the beach part of the prompt
waifu?
ai generated burger
(burger:3.9)
bee-hoon?
where you from that that's a thing?!
2/10
@nova remnant and when all is said and done, i can determinestically make the images ๐
I'm from Singapore

never knew it's called bee-hoon
i'm not a real noodly guy xD
ummm
pancakes with whipped cream!
there's no pancakes at this time
I have no clue how this works I need to read
wait there is
you in japan or smth?
mcdonald
I'm in SG
nice!
๐ธ๐ฌ
well fuck
uhh...
I'm not sure if i want pancakes tho
i hope you didn't 1:1 copy my prompt ๐
lol
well, i'm here for a bit, so if you need help ๐
Can you give me a very simple prompt and screen shot the settings
For the basic one
Not this attention stuff sun and circles 
So I mess around with it
and still, i send this, because i got the threshholds right now
@nova remnant you can see the masks below pretty wwell now, the first one is "beach", the 2nd one is "circle" and the 3rd one is "girl"
magic circle behind her,
standing on beach,
BREAK warm, bright day, sky, clean sky, outdoors, scenery, beach
BREAK magic spell, runes, transparent, , circle
BREAK (shorts, short white hoodie, sneakers, modern clothes, small breasts, open hair, long hair, straight hair, white hair, hime cut, blunt bangs, azure eyes, pointy ears, dark, choker, fangs, blush,(full body), standing, facing viewer, happy, :D:1.5), girl
those are the LAST words after each break
break * , beach
break * , circle
break * , girl
Yeah this is my circle mask
check your treshhold
you're gonna have to lower it
This is amazing to see. ๐ฎ
so to simple it down
girl on beach BREAK red hair, girl
what would happen is that regional prompt will search for everything matching girl, in the main prompt
then it will apply red hair on that area
but not other areas
BUT I CANNOT USE THIS WITH MY SONG LYRICS 
now my circle likes feet
I like feet too
waifu feet... hmmmm

I just can't get that magic circle to right position
Is this time for the breakthrough
OH SNAP
Twice in a row
he got it!
Third time's the charm...
it just had to bail ๐ฆ
Nice mask 
double songmaster-wildcard with simple left/right mask
Any word to describe upper body with out word body
You think this would work for region?
source?? ๐ค
is it civitai dot info?
no
years and YEARS of uhh... black and white line studies
and which settings should I use for this ?
performed on uhh... uhh... uhh... some site which has all sorts of 6 digit codes... xD
Not really civitai but... same purpose ๐คทโโ๏ธ https://tusiart.com/
regional prompter -> prompt
you have prompt ex selected 
yeah, prompt-ex ๐
you should see which works best, but i think prompt-ex works best

sometimes you get weird shit
ok stuff
and then there's whatever the fuck these are
should this be giving me the black and white images
she trying to glitch out of the map
when done correctly, yes

show ur prompt ๐
should I be using styles ?
@wispy canopy Here's my magic word behind_her_upper_half
no styles at first
okay wait I'll try once more
i mean, i never used styles in the first place -> it's just tokens you know ๐
very nice โค๏ธ
Earlier I use behind_her_upper_body which caused her whole body to shine
uhh... i'm not sure 'the power of heart' is supposed to work like that
ok, this is decent ๐
so, what's next, is tuning the treshhold levels
right now your mask is almost pure white, so it's activating way too much
increase your threshhold level by 0.2 and see if that improves it a bit
hm
its almost not pure white
another minute for +0.2 wait
but still looks like synthwave ps1 error
what's the full prompt?
@wispy canopy

just the simple one
girl on beach
BREAK red hair, girl
oh no, vil figured out the magic!

Art.
BREAK not art
eface are my settings correct ? 
they seem correct, something else might be up
i'll replicate in my instance, it oom'd again anyway
my a1111 went out of memory again
i have to restart it in that case
i think it's memory fragmentation
aight, tested it, and that prompt seems to work
girl on beach BREAK red hair, girl
Whats this base ratio for ?
that's for if you're using a base prompt
when I try prompt ex I get RuntimeError: expected scalar type float but found struct c10::Half

Can you try the other one
that one makes those ps1 errors
nah, i'm on fp16
did you install extentsion and then restart a1111?
somtimes extensions want that
ye its been a while since installed
Gimme a minute
@wispy canopy
with --no-half
it works on prompt-ex but no black and white image
same thing without base prompt select
Whoa. ๐ฎ
I'm getting some weird stuff too
well, my lyrics prompts are always 'like a box of chocolates' , you'll never know what you'll get
I give up getting that magic circle to work is too hard
I can see that...
Still pretty nice results tho
they just take a while to bake xD
dafug does this means? RuntimeError: shape '[1, 60, 33]' is invalid for input of size 2040
It's completely different if you just prompt the power of heart tho
obvioussly, it's a lyrics, this is the one that caused it: You need not to fear us Unless you are a dark heart A vile one who preys on the innocent I promise You can't hide forever even into the darkness For we'll hunt you down like the animals you are And pull you in the very bowels of hell
yeah ik
@wispy canopy what dis?
on the regional prompter?
ah, that's the weighting factor of the base block
if i'm not mistaken, the first block of the model
2 minutes of render time on sdxl brightprotonuke vs this: 2 seconds of render time on sd1.5 base model, same prompt, same seed and euler, because we didn't ahve restart back then
uhhh... it tried
I found a spare mouse...
well, damn, with restart 50 steps it's even decent ๐ฎ
but it's double clicking 
nah, i suppose i'll be sticign wwith brightprotonuke

how often does it happen
"oh just one more generation"
and you end up hours later
"yes"
SHHHH!!!!
one more gen, one more model mix, try out one more lora
is there an anime hands fixer lora for sdxl?
not that i know of, but i don't really use SDXL tho
i haven't seen one yet, then again, i'm not really looking for it, most hands are decent enough
imma go to bed now tho, sleep tight you all โค๏ธ
what are you trying to do?

regional prompting things
ADDBASE (abstract art:1.3)
ADDCOL (sun:1.2)
ADDCOL (clouds:1.2)
(moon:1.2)
elp 
the things that comes before ADDBASE, is added as BASE
ADDBASE (abstract art:1.3)
ADDCOL (sun:1.2)
ADDCOL (clouds:1.2)
(moon:1.2)```
abstract art, ADDCOMM
1girl, black hair, ADDCOL
building```
(steven seagal:4.5)
something like that, masterpiece, best quality is added to all segments, per your ratio, so if your ratio is 0.2, then it means, (masterpiece, best quality:0.2) and then :0.8 for others
for ADDCOMM, it's just added, as it is, so just abstract art added
if I have more columns I add more addcols
and just keep going ?
yup, but you usually can just let the template let you know how many you need
like here
she lost a leg in the abstraction 
waiit

I changed yours to this
masterpiece, best quality ADDBASE
abstract art, ADDCOMM
sun, ADDCOL
clouds, ADDCOL
moon, ADDCOL
i think you probably only want ADDBASE or ADDCOMM
not both
i can't really think of a reason to add both
it's gonna complicate things
ADDBASE will depends on your base ratio
@lyric zealot
yup so basically, you only have ADDBASE, ADDCOL and another ADDCOL
think of the line separating purple and green, that vertical line as the first ADDCOL
the second vertical line, between green and whatever that reddish color is, the second ADDCOL
so, if your prompt is this lalala ADDBASE tututu ADDCOL popopo ADDCOL hahaha
basically, lalala is added to all segments, purple green and reddish
tututu is before the first line, before the first ADDCOL, so only purple
popopo is between first ADDCOL and second ADDCOL, so green
hahaha is after last ADDCOL, so reddish
that's why this is probably the best way to write your prompts for regional prompter
hopefully you understand

I'll try something else
try something like
first segment is girl with red hair
then another is girl with blue hair
or something
or even
try this
there's metadata inside
let me slowly finish this gen

abstract art, masterpiece, best quality, ADDBASE
rock, ADDCOL
girl, portrait, blonde, ADDCOL
garden
o:
i think it works?
i think it will be best if you try with super distinct stuff
like different colors
I'm using ADDCOMM tho
procceeds to download his own image (that has no metadata) and try to png info it
I'm very sleepy
hi very sleepy, im dad
@lyric zealot does it work if I send it to text to image ?
with the image?
ye from png info
not sure if it will fill up the regional prompter or not
1,1,1 divide ratio for this ?
i believe so
The Metadata for this is powerful.
lmao, how is the metadata powerful? 
@lyric zealot 

it's working
you can definitely see hairband on region 1
not in others
yellow hair, twintails
black hair bobcut in middle region
pink ponytail forehead in third

i would say works out well
This was the prompt
(masterpiece, best quality:1.2), high resolution, detailed background, perfect lighting, extremely detailed, realistic, photorealistic, ADDBASE
3girls, yellow hair, twintails, blue eyes, hairband, choker, open mouth, smile, gyaru, ADDCOL
3girls, black hair, bob cut, red eyes, (tired:1.4), expressionless, ADDCOL
3girls, pink hair, high ponytail, forehead, yellow eyes, light smile, nose blush, jirai kei,
It didn't work with something in there I deleted a line 
the ADDCOMM one
I need the use common prompt enabled ?
like, for my model, it's hard to get someone to smile, and another to be sad
yup, if you have ADDCOMM
yup
my model, it doesnt
or at least
i think it's likely because of hires fix
changing all faces to happy
or something
or adetailer




HOW did you do that? ๐ฎ
A Fabulous image there. ๐ฎ
Regional prompter and I copied a prompt! 
o we doing multiple girls again?
@lyric zealot Why does the darker-haired one look like she could be a part-time hair model, and no one would be any 'wiser' about it?
i don't even know what you meant

minimized negative prompt to 62/75 tokens...
bad anatomy, bad proportions, malformed, bad hands, (fused fingers, interlocked fingers:1.2), asymmetric, conjoined, distorted, cloned, (low quality, worst quality:1.5), low contrast, (monochrome, grainy, pixelated:1.1), borders, (bad artist, bad art, bad fan art:1.1), (watermark:1.1)
Those fused finger negatives are just increasing your generation times, they don't do anything
That's a lot of Ru... Mikenekos
hmmm? what?
Fabulously speaking, the three of them work together.
It looks like they are smiling for the camera.
Funny plans on October await them.
like a svaroski pokemon
well, it jsut kinda took me there ๐
๐
having so much funw ith regional prompter tho, being able to dictate what goes where
LOL
i mean, this kind of stuff was just 100% impossible a year ago
it was possible but u had to do the inpaiting by hand now its automated 
kinda automated
for regional prompting you still need to balance the threshholds a bit
still, being able to place elements without them bleeding into other elements is a huge boon
uh... it's not quite doing what i want but oh well
he looks like this ๐คฃ
But it's same prompt from last night
Oh it's just on ADetailer step that gives that no target
Punk tomboy gf
Was images without highres. fix always this bad

Morning people
And night for you on the other side of the world
Big moon butt 

Its 5:56 AM for me 
Same time
Very suspicious

I rolled a new elf character tonight in AD&D and a friend jokingly called him Alphinaud Leveilleur so I decided to roll with it.
Stable is amazing for d&d :p
And elves best race
yep, I made a database of about 10k character portraits for my friends to pick through for PCs and my DM friends to grab for NPCs.
Where you from? 
Meanwhile I'm making eldritch body horror waifus 
Oh
You're from Argentina
Just remembered
I'm close, up a bit if you can guess :p
Fair XD
Brazil I presume? :o
Yup
Here's the rest of my current stable of characters in AD&D. male human magic user, female elf fighter/thief/magic user, and malf half-orc fighter/assassin
which one's half orc?
ngl they both look human to me
last one. AD&D half-orcs can pass for human.
lol I guess
I mean, it's not an opinion, it's written by the creator. 1st edition was a lot different than later editions.
Budget circle 
Ayy fellow morrowind player here
In a game I play there are some areas with museum style decoration
In those you can find human skeletons like that in display, pretty weird like some mutated demon, mad scientist stuff
pretty wicked prompt tho ๐
@nova remnant did you get the regional prompter working a bit, or is still acting up?
I couldn't get the prompt ex and the other one to work
๐ฆ
sadge
ah, the area prompter?
ye the one that divides things 1,1,1
you can a LOT with that tho
aight, imma try to break this regional prompter, how crazy can we make it
what can i replace, what can't i replace
trying for heart eyes, so far no luck
it's trying though
strength at 2, lets go for 3 now...
I got heart eyes to work before
forgot which prompt it was
found it
also that's a mighty sharp chin
don't get me started on the actual failures...
nah, this is what i do best tho -> roll the dice ^^
shhh
I hate how by the time it finished half the mask gonna be gone
you're disturbing the crickets
๐ญ
and don't make create some crazy ass crickets while i'm at it
see? not even finished and 60% already gone
the silence?
the mask gone thing
๐ฆ
did you see how much more mask 1st preview got
which sampler you using, restart?
yes
yeah, restart is excellent for image quality
but for some of those more unique features, it tends to remove them ๐ฆ
here's finished product
I HATE NIPPLE BUTTONS XD
lol
no way to reliably prevent them
I know a way
they're secret camera
do i look like someone who wwill actually take the time to clean up his images? ๐
i appreciate the offer 4gb cat โค๏ธ
you know what
there could be some adetailer model to detect those and smooth it
usually having nipple in negative helps a lot too
he angy
lava beads xD
o:
@nova remnant also are you a poe player
long time ago, i did buy tabs and a couple cosmetics though
dude, you got a squire?!

i haven't been playing too much
I'm poor as hell
you got a fucking squire!
i don't know how to craft ๐ฆ
other than that I'd have nothing lmao
even after so many years
my weapon is like 5x the price of a squire
i never can be bothered finishing builds
they are cheap now
and then just like give up
dude you have more currency on that setup than i made most leagues
my meow avatar is microsoft paint powered by me
damn you hands

Why is that?
it's not ๐
it is for porn
can't disagree there
she escaped?
i think yue's drinking milk tea again idk
kinda working 
its ugly probably because I cant do much more than 512x512 with the regions

the out of vram nightmare comes to reality
have you tried installing more ram? have you tried multidiffusion?
might work in your case
amd says no, it won't work
I tried a linux install for that
ROCm doesn't like my gpu either its too old

time to rob a bank


๐

Which type
looks like you worked out regional prompting ๐
the prompt type


the other one with masks didn't want to work in any way
Mayybe I need to update to 1.6
I'll do the test later
you just started, im sure you'll get the hang of it soon enough

Why are there so many 
I just can't figure out how to separate characters in prompt type RP
I don't think it's possible 

Can you do the vertical regions one?
rate her pet
unless you can use both modes together in regional prompter
?

oh i thought you meant if you can change from vertical to horizontal regions
I was doing prompt mode for regional prompter
so you can't really specify a region
girl in one region and statue in the other sounded simple I'm my mind
yup, but i will not be able to get that girl, since im using prompt mode for her
maybe it's possible, but need testing
how did you switch it tho?
Why is she slowing ass
I swear that happens all the time for me when I gen someone lying down
instead of 1,1do 1;1
ahh
or switch main splitting to Rows
there's a statue now

I... I mean...
the butt is up
statue of nothing
Wait a minute
They are in the same position
What trickery is this
lol openpose controlnet
guys, i didn't turn on my adetailer 
what position?
still looked cute
the eyes is kinda fucked 
torq's images

the eye is hardly a defect
I know whose just what position
zombie be sneaky
?
How did you go about getting the position
Scanning random Google images?

usually search and collect from this site https://www.freepik.com/?log-in=google
Ioohhhgggg
what resolution u using? 512xx760?
I want itttt
512x768
Can you make them do something instead of just stand there kek
always try to have it divisble by 64
Smol
the statue isn't even well done yet 
"dynamic pose" works in your model?
It's just a prompt
I wanna see
smol sus im using 960x640 just for fun, the 3.2 aspect ratio works nicely

















