#🍥|anime
1 messages · Page 126 of 1
I've been figuring that out a bit this week, not with qr codes though
what do you all do with the images you generate that you prefer to keep?
for my stash
stack them on local drive or upload ?
i sometimes upload them to X/pixiv but not all of them
ahh
usually i upload all of the ones i like to keep on tumblr
then free up my ssd
i'm gonna be honest i hate seeing AI stuff on pixiv
it got saturated pretty fast
a lot of the time the person posts each image as its own post instead of just having them all in the same post and it floods tags
aibooru exists
yea there too but with all the oversaturation of imgs dont even feel like posting them anymore
i enjoy messing around with AI but i still think art websites like pixiv should be for handmade art only
no way for them to really check that since i can just draw on top of it and call it handmade
i think merging human creativity with Ai to leverage on finer ambition is what we have to look for
ive seen some real artists using AI generated backgrounds behind a drawn subject
ya i think they use the qrcode controlnet model but with an image instead of a qr code
or with a black and white image or greyscale idk if colour images work
this generative Ai basically puts unlimited variation of beauty at the hands of its users
at the bottom of the guide it mentions it can be used for any image b/w not just qr code
the guide is overcomplicated though in a bit ill try to compile a screenshot of some settings that worked for me
i have an idea to cut out an ai generated character and put into a 3d environment in blender or something
apparently people are using this
https://huggingface.co/spaces/AP123/IllusionDiffusion
not local
which runs a QR code controlnet
ah ok
are you sketching them by hand or using sketch on controlnet?
It's just a sketch LoRA.
oh, you had me thinking if you were drawing them yourself
Nah, I'm ok at fixing little things in images, but I'm terrible at drawing from scratch.
me too, i can't draw well either
u could make it look like a real legit sketch if u photoshop a real pic of a pencil there and some real notebook details
I used to draw a little, but quit because I never finished them.
i totally lack the ability to draw human figure
especially the face
but i think i can sketch female legs fairly ok
the AI can add the pencil and the notebook details now
for every one of those I get, I get like 5 of these
Part of what I like about the LoRA is that it reduces the amount of notebook edges and pencils that you often get when trying for sketch style on most models.
It's just a merge. I put the models and LoRAs I used in the description. Info for v1 is in the main description and the extras I used in v2 are in the about this version section.
ahh ok
i have favorited your page, so any new update will show up in my civitai notification
quality of the image tone, color, contrast is pretty awesome
I might work on the new version a bit more in the morning. Next version will hopefully be less biased towards loli without loosing the overall style.
i would say there is this obvious aesthetics to it, hope you dont mess that up
there is a distinct level of innocence and cuteness that blend in nicely with the overall colorful image output
you could however work on the different one you just mentioned but perhaps name that different to set that aside from this version
so people can download both as they need
Yeah, that's the goal. I've had a couple versions that looked pretty nice, but the style got too close to the other generic anime models that are already out there so they got scrapped.
ok so there are lot of authors on civitai with multiple different mods for different purpose
you can still work on this one w/o actually upgrading it to something different
and have different mod for the goal you have now
I'm not just going to upload something just to do it. If it doesn't differentiate itself from other models in a meaningful way or at least do something better than them then there's no reason to have them.
ok thats nice, but i would like you to keep this mod too, its something unique from the various mods i have encountered
It's not going anywhere. At the very least v1 and v2 will still be available even if v3 isn't to yours or someone else's taste.
what happens if you use a vae on a model with a baked vae
does it override the baked vae, not use the selected vae, or does the selected vae overlap the baked vae?
Selected VAE will override it.
can you gen a couple of images with the following prompt using the negative you normally use ?
1girl at a beach, walking on the boardwalk
sure
where is the "ControlNet conditioning scale" slider?
that sounds like something that would be used to train a model
?
weight is like the main setting basically controls the overall stregnth
probably wat ur looking for if the efefect is too weak
this is nice
it won't let me set it any higher than 2
hehehe
oh nice
this is with it set to 2
thats bit high tho
yeah but it isn't really applying the controlnet image as much as i'd like
does controlnet work with SDXL models?
ive usually had to crank it for the qr code model aswell
you could also try with
amazing
one thing thats usuallyw orked nice is set the weight super high like 2 but make the steps less than full
which model and prompts did you use?
like if ur at 30 steps set the controlnet ending stept to like 25
this is identitydisorder model
masterpiece,high resolution,highly detailed,best quality,4k, sun beams,white clouds,volumetric lighting,brightly lit,sunny day, 1girl at a beach,walking on the boardwalk,petite,small breasts, comfy,saturated,abstract,Cutester art Negative prompt: badquality,lowres,normal quality,bad anatomy,misshapen,missing limbs,deformed,bad hands,worst quality,low quality,large breasts,FastNegativeEmbedding,UnrealisticDream,cropped
if you do can you render couple more wit that prompt setting up there?
or just one image is fine
i downloaded 3 on accident before i knew what i was doing
ohh i use this btw....
you can grab it later if you want
ok i think thats the one on the left
ok ill try that
i used hirez fix too on that
yeah hires fix is fine
what im trying to get at is how much the prompts impact the output considering how much you can trim the prompts
and have a quality output
so far your renders look pretty awesome
what if you used something like ....
EasyNegativeV2, negative_hand-neg, (low quality, worst quality:1.5), (low contrast, oversaturated), (monochrome, grainy, pixelated), borders, (watermark:1.2)
kay
im gonna render a couple on dark sushi as well
your positive prompts are vital too, i think you should keep them as it is
hang on
im trying that one again with the alternative negative prompt
hmmm how about this using just the following...
Positive : 1girl at a beach, walking on the boardwalk
Negative: (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, oversaturated), (monochrome, grainy, pixelated), borders, (watermark:1.2)
oh shit thats cool
preferably not using any other prompt if you can
its easier to compare when fixed seed tho
want to see how we compare
came out good... group shot
1181541726
thanks
gonna try that condensed one now
just the last 2
katie try a new seed with the basic prompt i just shared
for some weird reason my character is showing the rear view
on a pedistal?
i'm gonna test to see if controlnet will work on SDXL
"Potter< harry Potter.. here t o save the day"
CN for SD 1.5 and XL are different
lol
you'd have to download the newer CN releaes to make them work on SDXL
im getting the same rear view too lol
yes very diferent
given this
if the model recognizes cutester art it does a big effect
these are the SDXL controlnet mods https://huggingface.co/lllyasviel/sd_control_collection/tree/main
your other images you started out with looks much better
i love that prompt lol
i keep the top and bottom the same and just change out the middle 2 parts mostly
well, interesting outcome, t2i only, no controlnet
which negative prompt did you use or did it have TI?
1girl at a beach,walking on the boardwalk
Negative prompt: (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, oversaturated),(monochrome, grainy, pixelated),borders,(watermark:1.2)
just that
ahh ok yes
not sure why it shows rear view
edited it slightly ...
1girl at the beach on a boardwalk
rendering now
even facing the viewer has the character looking over her shoulder
those look so good tho
yep just used and rendered a couple now
1girl at the beach on a boardwalk, front view,
and natural theyre like a vacation photo
try emphasis and cfg+
thanks katie, yours look better, stunning imo
i think the model was probably trained on qwuality tags like masterpeice and best quality etc
could be
😄
what if you add these into positive prompt
(highly detailed:1.2), aesthetic, (vibrant colors:1.1), cinematic, hyper realistic
been a while since i did one of these ^^
looks like the prompt matrix i was using earlier for the first time
that time already? 😮
yeah it's like 5am. sleept for a little bit , need more sleep , can barely seen screen, stupi d eye issues
last one . haha
👋 sleep well @bitter latch ❤️
gooonight
need something in place of the moonlit beauty
nvm, already got something stupid in my head ^^
is that combinational generation
ive been watching beekeeping videos
Inspection, remove queen, and then ultimately decide to euthanize the colony. Shows the process undertaken, including the euthanization.
The beekeeper is EAS Master Beekeeper Kevin Inglin
it's prompt matrix
i have 5 prompt elements, which each get turned on and off for a single generation
makes it fun to see what a tag has for effect on an image
@gusty palm added beekeeper. 2nd row, last image XD
invasion of the sunflowers
there in every picture lol
i liket he middle area the space suit looking beeker suit
Guys, please tell me which upscaler is best to use for realism or photo style?
4x-ultrasharp? possibly
i`ll try this is...
kek
lovely bees uhh... how the f do you pronounce your name?
just some randoom idea from yesterday
intereupted render
looking like a watercolor sketch
Guys, who knows how to tell the aDetailer to recognize only faces, for example, those in the foreground?
it looks like theres a place to type in adetailer prompt and thats where u tell it what to look for
next to where u select the mod, im not super sure
llooks like it just requires simple prompt like 'face'
for example, in an image with a depth effect I have two persons, I want the detailer to pay attention only to the person in the foreground
it might do that automatically just by looking for the most prominent face
damn, all faces are the most noticeable to him
if there is a head then there should definitely be a face on it
depending on how big the face in the foreground is theres like a min/max thing for
Mask min/max ratio Only use masks whose area is between those ratios for the area of the entire image.
like if the face in the background is very small and the one in the front is much bigger its probably within a certain ratio
this is ?
yeah
thats will be a bit finnicky , to figure out how much of an image the face takes up
and then set the minimum to just below that , so it ignores anything smaller
i might have found the cause for rear view images using dark sushi...
possibly cause of canvas width x height
sound strange but thats what im noticing
canvas that are vertically bigger are all showing the character front facing
but this is all true when using the process img-to-img. not for text to img?
trying random qr code t hingies
1girl in joggers at the beach on a boardwalk,
that looks so good
i havent had a good render in hours since i got distracted earlier
thanks
yea thats difficult to do, but in my case i think the model was trained to do that
@wispy canopy you on 1.5?
nah, sdxl based
i see
wait you don't use comfy right?
thta's correct
just a1111
two things happened about the rear view and getting them to look front view, first i needed to increase height of the canvas and had to use out of focus in the negative prompt
and i think its model specific with dark sushi
im using two different versions of dark sushi, using same settings, prompts, and seed for these two images... left one is Dark sushi mix 2.25D and the right one Dark sushi 2.5D v4.0
the left one (dark sushi mix 2.25D) looks more cartoony than the right
Guys, tell me what tokens will generate a tan with white breaks from a swimsuit?
reverse tan, bathing suit tan, or bikini tan
thanks
tan lines too possibly
i got the oobabooga thing installed and stuff and got a model but either i got the wrong model or it doesnt work anything like diffusion
oh wait its actualyl responding 😮
Omg the lighting texture style everything looks so well together you're definitely getting better than ever
HOW!?
But I think the best is the night art with the yellow golden highlights
i just did the default install and loaded the smallest model i could find
thanks, yes that prompted for evening
what the time of generation?
Output generated in 155.16 seconds (0.37 tokens/s, 58 tokens, context 67, seed 118432895)
mostly because of to less vram, i couldnt even run oobagooba on my 1080 8gb
how you run SD with use CPU?
hes testing an llm webui not SD
but yea its also possible to run SD on cpu. its super slow
i dont this is the text thingy
2023-09-27 07:55:02 INFO:Using ctransformers model_type: llama for E:\text-generation-webui-1.6.1\models\mlewd\MLewd-Chat-v2-13B.q5_k_m.gguf
2023-09-27 07:55:02 INFO:Replaced attention with xformers_attention
2023-09-27 07:55:02 INFO:Loaded the model in 4.24 seconds.
it says its using xformers let me try again
can you ask it about the prospects of a resource based economy?
;]
for sure
im trying to see if i need to install cuda in the conda env or something
i give up on trying to get .gguf or whatever , i guess the normal models are broken into multiple 9gb .bin files im gonna try that,. that seems to be the standard im gonna hope that means its better compatability
guys, tell me why СFG has a value of 30 if after 10 the image is “burned out”?
some models can probably go that high, or used to
u can edit a ui config in the sdui dir to change some of the value ranges and stuff and defaults
or in what cases is it possible to use this parameter with a value greater than 10?
like almost all my pic used cfg of 22
prompt not cooperating :3
i regurarly ran like 15 ish cfg for along time im still not sure i want to change to 8
if u modify your ui config make a backup first though in case u put a space in the wrong area
Doesn't it make sense to use parameter values greater than 8?
yeah but 8 has been giving me nice results sometimes too
i dont really understand what it means for 'lower values mean more creative and higher values mean more prompt dominance'
What about the lower values ?
basically, low cfg, it doesnt follow prompt as much, that's why it's more "creative"
but higher value, follow prompt more closely, as such, there's not much room for "creativity"
a xyz plot may be in order
what is this?
it lets u compare settings in a coherent manner
I would not say. Today my results with a value of 2 turned out to be almost accurate to the prompt
that's why the question arose
well, that's what cfg actually means tho, so yeah
every time i do a high cfg, my contrast is super high
might help to add oversaturated to neg
oh contrast
i know what u mean though its like a comic book
hmmm, im going to try that
i usually stay anywhere from 2 to 5, rarely do 5, often do 2 or 4
Isn't it possible to get high contrast using a prompt?
probably, not really a fan of high contrast in most situations
2-5 CFG?
ye
anywhere from 30 to 50
hmmm
lately, it's been always 50/4
50/4? 12 for HiresFix?
steps 50,
cfg scale 2.5
and to what extent does the prompt match
take a guess 1girl, bbq in the backyard, sun is setting
my hires through comfy upscales it to 4x, downscales it back down to original, runs through same step/cfg, then upscales again
steps 50
cfg scale 22
i think higher cfg benefits if your pompt is longer/more specific
if you have a short simple prompt, high cfg may not be as useful
just a theory
lately my posprompts hava been nearly 100 tokens i think i need higher cfg but i havent tested
but in higher CFG and long prompt my result completely broken
i could be wrong , theres also those cfg mimic extensions which im not really sure how to use
1girl in a bikini at a beach resort. She has long flowing hair, big sparkling eyes, and a beautiful figure, walking along the beach as the sun shines brightly and the waves crash against the shore
i dont' suppose that's a complex prompt tho
CFG 22 steps 25
high sample steps looks so good
girl that aint no plate, that's a grill
(30/2)
can always count on cranking up the samples
high cfg also means making SD follow prompt more closely
im loading my ui to test with yall
i wonder how i can improve fingers by high cfg
high cfg might make negative ti more stronger
not sure
yeah but watch this
her fingers are quite well drawn
i used cfg 22
wait 10 cfg looks pretty good
SD fingers - is a Magic thing
50/10
fingers are fused
i have a problem where the fire is behind the item and not in front
like i can do house fire and the fire will be behind the house
going to do a comparison with cfg 22 and 7.5
cfg 22
im always happy when hands come out nice
50/7.5
im hungry
surprisingly calicomix is a nice mod that usually generates good hand
Is there any dependence CGF value on the sampler?
50/11
It seems to me that fingers and hands depend on the model
50/2 & 50/10
50/10 (right) looks much vibrant
probably should have stated a setting but w/e lol
50/10 - not look good
you doing hires fix too?
nope, i rarely do that (i also use comfy and not A1)
and SDXL
i shouldnt have uncluded clip skip
cfg scale 11 failed to follow prompt
i dont even know what to make of this result
clip skip 2 looks bad in all of those
where change Clip?
trying 50/15, and clip 1
i forgor, but in settings > user interface >quicksettings list< you can add it and itll be at the top of ur UI when u restart
ye last laster stop clip
50/15 .. clip skip 1
this is really cool setting method! i am dont know about it
token merging ratio is another mysterious one
katie.. here is 50/15 with clip skp 2
tell me more about
and pad conds unconds
50/20, 30/20, 50/2, 30/2
i really need to setup xy plot in comfy
1-2 not look good really
https://huggingface.co/docs/diffusers/optimization/tome
this 'explains' token merging but
i dont really understand it
me too
1girl at an amusement park, sitting on a bench, wearing t-shirt and mini skirt, with a cone ice cream in her hand,
chat gpt told me it will speed up generation and thats what its intended for but it does seem to change results
She is strong.
Also how come part of her hair is blue?
1girl at an amusement park, wearing a t-shirt and a mini skirt, buying a cone ice cream from a stall
this one is really nice
thanks! 
that's AI doing its RNG thingy
oh really 
think i saw one of a subway or something earlier on instagram or something i forget
nice, what cfg scale did you use?
4
ok the pic is missing the stall
i think cfg 15 comes close to following prompt
her shirt looks so 3d :D
it's just off camera 
can you like
do really @lyric zealot resolution and then upscale twice or more using high cfg so it follows the prompt and looks good ?
smol
hello

this is a great picture
great rng not gonna lie 
definitely save it
i found the threshold of cfg scale where sd follows prompt specifications more precisely ... CFG scale 20
how can i do something like this? with controlnet lineart? or canny?
mj niji lol 
actually wonder if you can do it without using this
i only need these?
think so yeah
think that peta tweet, they probably literally prompted mcdonald's two all beef patties on a sesame seed bun... line

aniverse has some conflict with extended prompts and messes up hands
hallo

ive found using preprocessor is unnecessary for the qrcode things
@gusty palm i dunno if it was answered before but i saw you doing llm?
gguf = cpu based model
although invert or gaussian blur can somewhat modify your results
you can use the offload to gpu
i couldnt seem to find info about what the different file types mean or how to use them
im currently downloading the .bin that are like part00001
oof
lets dm 🙂
ok
i never heard of gguf
it's the successor to the other cpu model
gguf has all the needed settings packed inside
took me soke time eh
they even stated as such that the llama.ccp team won't even bother supporting ggml as gguf simply beats it in every category
c plus plus?
they use c++ now?
llama.cpp is a cpp thingy for llama
thought everyone moved to C#
what
or you mean just the file name convention
what
that's what im thinking, never looked into it though
"Inference of LLaMA model in pure C/C++"
ahh ok, i was thinking people use C# now
man we need pnginfo viewer bot
I suggested that like 4 months ago
like in projectai
same

they were making their own and was told it will be out soon. But it's still not out

They said that like couple months ago
yup yup yup
there's the webversion https://supagruen.github.io/StableDiffusion-CheatSheet/
Talking about this kind of inspectors
https://github.com/sALTaccount/PromptInspectorBot
idk how discord bots work, couldn't someone set this up and invite it here? 
Yes
But SD wanted to make their own bot and didn't want to use Salt's

help, how 😭
you can't invite a bot into server which you don't have "admin" power to
unless you self-bot which is against tos
TOS
I was just meme-ing
you really like this 1girl
If I wasn't lazy I would fix her leg and this would be good image
white hair, elf, smol
Elf 
I rly like this so here's repost

looks clean
spaiship
the others were too much booba

was testing add detail lora its nice, that one is at 0.25
🌻
prompt!!! 
it's mostly drawn by hand and ran through a model
I swear i've seen that armor before :d
i have a lot of these
Iz cat with 4gb vram send help
cool image but has 3 hands
opens krita
🔎
You are gonna be disappointed in the prompt, metadata included
Now iz cat girl with 4gb vram send help
might have to bump that weight up
I agree with the AI's choice of topping order
@lyric zealot how about something like this
like start from 2 and work your way down
i know i did it right if you can read it
SDXL btw

i am using realistic vision
maybe is that?
i would not eat that burger
x2
needs green cat eyes
nah
try end control at 1, starting control 0, weight at 2
all control type
My OC got red eyes 
same
needs smug expression
add the " :3 " face
might need a lora for that 
hmmm, i know i got it working on 1.5 before with my suggestions, maybe its your main gen
i found that if you're making a 512x512 image, the cn image should probably be 512x512
no, ":3" works as a tag

flowered ✅
I forgot small breasts
the image with the text is 512x512, maybe is because i am using version: v1.6.0 of a1111? i have seen that that versions have many errors
that is pretty cool I have never seen someone hiding behind a shield like that o:
Giant weeburger.
kinda? maybe?
can you gib prompt 
i am not entirely sure sry
hamburgers again?!
thats all correct sometimes it just doesnt seem to work randomly for me, but to ensure its working at all, i recommend turning weight up, and start step to 0 and end step to 1
they're mostly drawn by hand but prompt was
Negative prompt: (worst quality:1.4), (low quality:1.2), bad anatomy, bad composition, bad hands, more than 2 hands, low effort, mosaic censoring, censored, censor, easynegative, horns, animal ears, skirt, cape```
Can easily be defeated with a laser pointer, 0/10 waifu
GREASEBURGER
maybe set to resize only until its shown to work
I'll try that holding mace and shield like that
theres another model that ive had better luck with besides qrcode monster as well may be worth trying
cheddar is terrible
meat is murder though
i know
it won't work 
AI can't hold
Don't kill my hopes like that
https://civitai.com/models/90940/controlnet-qr-pattern-qr-codes
https://civitai.com/models/90472/controlnet-qr-code
theres actually 2, make sure u get the sd1.5 one
Controlnet model for use in qr codes Conditioning only 25% of the pixels closest to black and the 25% closest to white. Move the model and the .yam...
"bruh-are-you-even-trying-burger"
I'm a cat I swear
try preprocessor set to invert
ive found better results with white text on black background
ok thanks i will try
imma try this with the sdxl one, see how well it does
although both work its all a slot machine
Amazing picture there.
show if u doo
also if i find a better cn model ill share it, i know these are trained for qr codes so i imagine a cn model could be made specifically for optical illusions
this burger is called "chef's confusion"
nobody reacted to my gnome
Gnomes aren't real
should be called "the end" cause you die

instead of a quarter (1/4) pounder, have a 4 pounder burger!
this one will end you too
people
@lyric zealot maybe you
Do a thing for me to see if it works in any way shape or form
with inpaint
right is 1 hires pass but i like the white on black text better
guess it depends on what kind of image you're making
what do you mean?

i'll give you this in exchange
mature Yue!
Hmmmmm can you take a black picture and throw it in inpaint, you mask everything and generate with one of your prompts
my gpu cant inpaint, its dry
what's the purpose of this?
I've seen people doing that I think looked cool

accidentally on purpose
https://twitter.com/peta/status/1706435937331192168?s=46
is working but why i am not getting something like this
not a lesbian yet still works as a thirst trap
some people like tomboys/ girls with mjwcle
muscle
rare find 
this nice
nooo his prompts are fake with manual work
?
wdym
I'll get my gens to hold weapons perfectly you'll see
oh
this :p
im so confused
sad
her titties will escape
like this?
mew
They did for about 70% of the generations
Delete system 32, might fix it
1shot sword
let me try that
same effect, not sure why
10 seconds in paint
interesting
i have no gods!
problem might be, it's a lot more vram intensive to do this at same res
Dveon
and this is me sober?!
This actually looks very good
well, th eprompt is now very versatile, so i should be able to swap in most animals
do a starfish
this could definetly be used to create sprites for a mobile game out of thin fucking air
GUARD-DOG!
@nova remnant that starfish is gonna be hard
I said it as a challenge
transparent effect works at base but doesn't work further up
this one hella cool tho
i can already see the microtransactions
Starfish girl level 1 5 stars
uhh... i think he leveled up
me irl
paint dot net, specifically
looks like am extracted outline of an existing image tbh also there are way too many artifacts
@nova remnant PEAK. PERFORMANCE.
bruh
👍
"I can no see, am legally blind"
YESS
That is the boss class
uber viable starfish
!!
but they are holding it so well
idk why but I just thought that if someone were to propose to you it would be "Vil, will you marry me?" and this is because vil sounds like will
imagine a guy named will aswell
will, will you marry me
damn
anyeays
i should probably go to slep
Cuté
But where is her body? I only see head, hands and hair
Big pile of hair
Wait
I think I found bigfoot
Her body is under her thick hair
Literally bigfoot
Oh shit wait
Literally fits the definition perfectly
Hairy ✓
Human-like ✓
Mythical creature ✓
Wow this is so insanely good
Just remove the girl
Would be much better
Plato was applauded for his definition of man as a featherless biped, so Diogenes the Cynic “plucked the feathers from a redacted (male chicken), brought it to Plato’s school, and said, ‘Here is Plato’s man.’ ”
This server literally blocked me from sending my message
Because it used the word for male chicken
guys tell me what VAE good for Realistic models?
no I only play games
What
Yes
I am philosophy student
Idk reference
clearly
thanks
Np
Well I really gotta slep, I have like less than 4 hours
And tomorrow (today) I fly back home
So yeah
good night
Gn
nighty night 
Maybe watch The Princess Bride during the flight.
I downloaded the entire no game no life light novel
And some chapters of blame
(manga)
Gonna read those
Maybe after I get home
Anyways gn

pewpew
you should try the word infrared with this style, wonder what it looks like
tried to make a borg. not there.. sigh
closer
the new borg?
I would so watch that anime

sdxl ?
i used sdxl for base then feed it into 1.5 model
Time to make more models

@wispy canopy
pls end me
what did u do?
I was looking at used gpu prices someone selling a 2080 super hall of fame something something for giga cheap
I can't buy 
And some others 
what's ur budget?

(money, in my bank:9.5)
💸

aight, with these, imma go to bed
So how this lady keep on getting more fabulous and expressive?

That is a Succubus.
doesnt affect much 
i see 
(Infrared:4.4)





