#🍥|anime
1 messages · Page 45 of 1
You mean someone has been uploading my failed LoRAs? 
I mean…
Idk…?

Someday maybe I'll figure out good settings for training a LoRA.
there's a new Sheriff in town boys and girls
I mean have several i could lend but no promises
plays slowsong
Honestly I've been too lazy to finish fixing captions on the last dataset that I gathered like 2 months ago.
what's the word for there being a cross between human and another species?
not hybrid
very specific for animals for example
furries especially
often used in prompts too
anthro
like my OC is a stag/wyvern/shark which took allot of weighting to get right
I don't want a dragon to act human
lol
thank god, i was about to send death threats to your dms, ig that won't be needed
need to redo it now im better with SD
random text 👺
I'm just spamming
that much text and that prominent is probably due to a model or lora that was trained with a lot of similar text which may be difficult to get rid of through prompting.
Another one
what does model compile (experimental) mode, do? In latest vlad
lol
What did u need to fix
I need the poses to be captioned and WD tagger didn't do a very good job at it.
Did u use the swin tagger
I don't remember which one I used. I think there's been at least 1 updated model released since I tagged them. Maybe I'll just redo the captions. I should probably sort them into folders so I can control the training depth of different poses too.
If my captions get fucked up i just delete all the txts and redo every image
Sounds tedious but the tagger does batch tagging ofc
It mostly tagged them correctly, just wasn't as accurate with the poses as I want and the poses are the main thing that I want to train.
nobody knows?
I use the Dataset Tag Editor extension to modify them.
Ive actually never touched that lol
Works really well if you want to prune specific tags especially.
do you know anything about model compile mode @supple raptor ?
its in cuda settings
(on vlads auto)
Not a clue. I've tried Vlad 3 times now and it hasn't made it more than a dozen or so batches before completely breaking yet so I gave up on it. Not worth the headaches IMO.
I feel it's not giving me the same quality images as regular auto did
for some reason
Yeah, I've seen a couple of discussions about that. I haven't gotten enough images from it to have really formed an opinion.
and its giving me the ancient type of conjoined twin stuff
that auto hasn't done in a long time
and auto started updating again
like this
it was probably gonna generate 2 dragons but the edges weren't defined well
anyone has tricks for making better eyes?
yeah use my model
why should i?
Which VAE are you using?
ft mse
https://civitai.com/models/6209/dark-gemini-v3 well just check my example images, if you wanna try it that's great. It's very versatile
if not, well ok
i don't like the semirealistic face, does it do well with pure anime?
still gotta refine it abit but i likes where the model is getting
That's your problem. MSE has great color and it's good for realistic images, but it has a hard time with anime style eyes. Try kl-f8-anime2.
klf8 >
there's a few pure anime in examples
klf8 almost always outputs too bright images
comparison
2º looks better imo
Use VAE-BlessUp script to lower the contrast of kl-f8-anime2. https://github.com/sALTaccount/VAE-BlessUp
brighter is better
like I said, my model can do just about most popular styles
not sure on pastel though
i love pastelwaifu
is a good vae
anime.pt no upscaling cos i aint bout waiting rn
why did nobody talk about this before my gowsh
isn't that just the [redacted] vae?
most vae's are just renamed
no idea tbh
i started last month
so all this about the redacted is a mystery to me
same
kfl18
i thought this was an extension lol
TLDR: A certain paid website's model was leaked which has since become the base of nearly every anime model and is included even in many realistic models. We aren't allowed to directly discuss using the leaked assets, so they are redacted.
ooooh, I think I heard about that
catch 22 considering they know what youll be redacting but
we just said redacted and we all knew. Sort of
thats just my opinion on that lol
nobody new to it would know
how exactly do i use this??
i get this error then the window closes 0.1s after
Yeah. The SD staff just didn't want the leak being directly advertised here but (I think) they realized they couldn't do anything about people using it and wanting to discuss how their images were made. By using [redacted] it made talking about it less like an advertisement to people who didn't already know so we started doing that and stopped getting the attention of the mods..
yeah no i totally get and agree with that cos sucks for them
and advertising it would kinda be a salt in the wound thing imo
pip install diffusers or some shit im guessing
Did you create a venv and run pip install -r requirements.txt?
like dude
what is a venv
virtual environment yeah but what
and I even had the wrong resolution on
Virtual environment. It allows you to keep your python stuff separate for different projects so they don't conflict with each other.
and how do i create one?
iirc it has something to do with anaconda
cmd command probably
In CMD run python -m venv /path/to/new/virtual/environment. I always place my venv in the root folder of the project so it's easy to access.
Gemini3?
oops LOL
assuming you place the venv in the root of the project, you can then run .\venv\Scripts\activate from the root directory to activate the venv, then run the pip install command I sent earlier to install the requirements. Whenever you want to run the script, you will need to activate the venv first.
you can just delete that folder and redo it in the correct location.
yeah but kinda funny tho
I think most of us who have created a venv have done something like that at some point 😅
pretty sure this is not supposed to fucking happen
fuck me
please dont tell me it actually did that
did you activate venv first?
i made a venv folder in the root of the vaeblessup folder
It shouldn't have effected your auto1111 install anyways since that is all in it's own venv
do you have a version of your model with light instead of Dark?
ok but how do i do it correctly
wow.
pretty sure this is it
it does light stuff too, I just kept going with dark because its progenitors are dark gemini 1, 2 and 2.1
also this is not correct im guessing
one sec. I'm going to install it so I can show screenshots.
okay :), i'll download it 5Gb pretty heavy
hey i need some advice.. how is this created like is this disco diffusion or stable diffusion with some model??
there's pruned and fp16 version
on the side menu
on civit?
Probs by mixing weights
says 4 files
mb?
quality is the same ?
MEGABYTE


XD
I don't know if fp16 and fp32 have the same quality, most people say it does...pruned does however, but you can't use pruned to train
but yeah pruned 32 is 5gb so yeah
I always use fp32 pruned for models I have no intention of merging with or anything. I think fp16 give pretty much the same quality
i've done it
now it doesn't give an error before closing
BUT WHY DOES IT STILL CLOSE 😭
hang on i think i got it
nvm
why is it asking for a json config file
it says MODEL PATH
damnit
Try this. I'm getting a torch cuda error. python bless_vae.py --model_path .\VAE\kl-f8-anime2.ckpt --model_type compvis --output_path .\VAE\output\kl-f8-anime2-bless --output_type compvis --contrast 0.9 --contrast_operation mul --patch_encoder
manually installed torch2 with auto1111 2 week ago to see auto1111 is updated for it now 
Same here. I'll jump over to the touhou server and try to get help.
The team who makes Waifu Diffusion as well as the guy who made that script are there.
It might be a good idea for me to get bless_vae too
@gritty patrol With your venv activated, run these first two commands in order. The 2nd one may take a little while to finish. If you get an error complaining about "No module named 'pytorch_lightning' when you try to run the bless_vae.py script afterwards then you just need to run the 3rd command
pip uninstall torch torchvision torchaudio
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install pytorch_lightning
What's the last bit of the error say?
try upgrading pip first with python.exe -m pip install --upgrade pip then run that second command again.
still gives the same error
I would probably just delete the venv and build it again. I'm not sure what the issue is.
In the meantime. I can just upload the two lower contrast versions that I made so that you can download them.
please do
a bit lower saturation also would be nice btw
oh wait theres no option for that sorry
do you have comparisons?
what was it called again? and where to find it? the vae thing that fixes the contrast and saturation problems that can arise
dragon t-rex
tysm
fuck my model needs some quadruped injection
WD 1.4 isn't that bad :o
It doesn't fix problems per say, it just allows you to adjust the contrast and/or brightness that the VAE applies.
https://github.com/sALTaccount/VAE-BlessUp
bless09 is def the best vae for anime
that minor adjustment was something we needed for ages
Those are just the WD VAE. I'm using Aurora for the model.
Agreed. And you don't have to run --no-half-vae with it which means you get a little more free vram and slightly faster generations compared to the [redacted] vae.
i see free vram
almost perfect fingers 
his name is henrique
no hand improvement stuff, just pure aurora
just sad that the middle finger being morphed with the collar
didn't even put anything like bad hands, incorrect fingers
i love how adding badhandv4 just hides the hand
bad-hands-5 really does make them into bad hands 
Yeah but I mean like, normal WD VAE isn't that bad :o
Have you tried Nerf's new negative hand embedding? It should change the overall image less than bad-hands-5, while generally still making hands look better from what I've seen.
Just looked it up. That's just kl-f8-anime2 that's been renamed. That is to say, no reason to change unless you want lower contrast, in which case you can grab one of the versions that I just uploaded.
does it take more than 5 minutes to make them?
It took maybe 30 seconds for me. Not sure how much a slower GPU would effect that, but I can't imagine it should take 5 minutes with anything that can reasonably run SD.
i was just making sure that it wouldn't take too long if i asked you to make 0.91 - 0.95 contrast versions
🥺
No, it's super fast. Honestly 30 seconds is probably a pretty high overestimation.
yay!!
free games from epic
https://store.epicgames.com/en-US/p/against-all-odds-409105
https://store.epicgames.com/en-US/p/horizon-chase-turbo
It'll definitely take longer to run the grid to compare them than it will to make them if you use Hires fix for your images.
Yay!!! More games I'll forget I own. 
same buddy, all my games in Epic were claimed for free 
smol sus, this is like the third day in a row that you are posting pink haired girls shopping.

I just really like how it looks

basket in a basket
teen anya needs to eat
does anyone have the link for all those negprompt embeds? and which one does what
You've posted like 40 different versions of this char tho
Simp for my grown up waifu, yes



well, you guys can probably guess who im making for my oc 
it includes bad_hands, bad_art etc
this is my masterpiece tho
beautiful
literally 4HEAD
Looks like that villain from the green lantern movie :o
I need to get that vae bless installed, so it stops burning out my generates
or I could try another vae first and see if that's why
Could also be CFG :o
I don't even remember if i have watched that movie
I'm having like a brain hemorrhage today
it's 7
That looks like either too high of CFG, or too high of weight on something. Definitely doesn't look like a VAE problem to me.
Its a mid movie
concerning.

Yeah, I agree :o
alien ass looking eyes
how much weight is too much?
I stoopid
(Ancient Dragon:1.225), (western style dragon:1.1), dragon emperor horns, symmetrical wings, quadruped, (walks on four legs:1.15), barbed tail, (solo:1.175), (dark fantasy art style:1.14), (2.5d:1.15), (digital illustration:1.15), (digital art:1.15), [hyperrrealism::photorealism:0.65], [pseudorealism::0.70], (masterpiece,:1.2), (best quality:1.2), unreal engine, low saturation, balanced highlights, high contrast, (walks on four legs:1.15),
that too much weight?
Doesn't seem like thats too much
dwayne jhonson or whatever
well... time to prompt alien ass looking eyes
thought so, then not cfg, not weights
what else can it be?
Hmmm
stop cuddling immediately
Could be maybe model
poorly trained model, TI, or LoRA, or a LoRA with too much weight.
well my model isn't poorly trained or merged, since it doesn't always happen. I don't know what triggers it though
stop spending money pink ass influencer
switched from waifu vae to the other one
nevermind the wings being fucked
the colors are stable
stable diffusion 
She looks bouncy, like, as in bouncy to use as a football
even the saturated ones
Since it was merged with MBW it seems very possible that problems like that could emerge only with certain combinations of tokens.
she looks like she just start learnt how to walk
super adorable
That too ye xD

possibly, if I know what does it I can merge it with itself and discard the weight
ah this ones not too bad
CRYONICUS IS ALIVE
same prompt as last though, only vae changed
You're unkillable man
turns out when prompting alien ass looking eyes there is one particular word in there that the AI pays a lot more attention to than the others 🍑
why are you surprised, your nick says I never die
yes I might be

I am believing

I'm so fucking relieved though, so relieved. It's like I've been held underwater for 6 months
Pog
and I can finally breathe

@supple raptor if I only get those washed out colors, the oversaturation, burned edges, color banding etc when kl-anime 2 is used
then what?
That last one you posted is also very burned
Yeah, it's just not quite as obvious due to muted contrast.
I zoom in and see it on wings where the black meets the sky
How can I generate these type of art using this bot?
We aren't using a bot. Pretty much everyone here uses the AI on their own PC. Do you have a decent GPU?
💗
Yeap i have.. but how can i generate this type of art
Yes i have... But how can i generate this
First step is installing this. https://github.com/AUTOMATIC1111/stable-diffusion-webui
Okay its free to use
Yep, only thing you have to pay is your own electricity bill. Once you have that installed, download this and put it in the .\models\Stable-diffusion folder https://civitai.com/models/40199/aurora
and ur gpu ofc
Ohh thats great... Can i use this on mobile...
https://github.com/EmpireMediaScience/A1111-Web-UI-Installer this is an easy way to install it
How can I use this on mobile
The better option is to run it on your PC and access it from mobile. Do you just want to be able to use it while connected to your home WiFi on your phone, or do you need to be able to access it when not connected to your WiFi?
Yes
so weird but kinda neat ig
I want to use it offline also
Something about the style and shading makes me think of early 00s CGI advertisements. Like it's got that bright saturation and roundness going on if that makes sense
Kinda similar vibe but higher quality
Check the documentation for both the --share and --listen commandline_args. Share will allow you to use it from anywhere as long as both your PC and your phone have access to the internet. Listen will allow you to access it from your phone only when both your PC and phone are connected to the same network.
Okay bro thanks let me check
when you realize you're generating 1024x2048 at batch size 2 and forgot to enable tiled vae 
lucky for me, i would have oom by then 
i once upscaled a big ass image (around 1024x576 i'd say) and i did it with batch count 10
wasted so much time
i could've been a billionaire by now
clearly
big image you say? 12288x4096
is BEEG
It's supposed to be.
why the 16x upscale then lmfao
Hello there :)
does something mentally insane
hmm yes society does not accept us
I paid for 12GB of VRAM and I wanted to use it. Sadly, it took less than 9. Maybe I should upscale it another 2x 
go big nubby, x32 upscale
It was only an 8x upscale to begin with (2x 3 times)
use it on controlnet, i've heard it's good
is gon be fun
water
there's gatekeeping here? 
should i put wet on the prompt
don't ask why she's dripping water while underwater
put wet shirt, but not face
doesnt really matter if i put wet or wet shirt, it understands it's wet
Leo, a cat spirit
what should i do next?
negative prompt sweat?
What are you guys up to right now? :D
negative -- water droplets, dripping, sweating, crying, tears, water on face
that should help
that ^
gen some wallapapers 🙂
you can also add weights to prioritize them so like (negative: water_droplets:3.0) to give it higher priority
running some summoners war in second screen
nice nice
sounds amazing
Time to see if I can do an 8x upscale in one shot with tiled defusion.
i love stable defusion
That's much nicer, the rest you can do in a photo editor
why does it looks so blurry
I have a similar problem, discord has a tendency to "lowresify" my images
i'm not upscaling every single one of these
i use 640/448 res
for small gens
I default to 1024x1024 for all my gens, it's nice and simple
less blurry on discord
680/240 upscale x3 
Just a quick question... Does anyone know how to convert an existing .pt file (embedding) into a safetensor file?
@faint marsh ?
omg its Michael Bay
Okay, well will go to bed now. So good bye guys and see you in 3 months :/ My flight is going in 13 hours
Sleep well and have a safe flight later
sleepy tag is not behaving correctly after 1.1 weight
at 1.5 it's really weird
heavily distorted
awww
i think this is as good as it gets
The 8x upscale in one shot was a mistake. It worked, but doesn't look good. I'd try uploading it but discord won't allow me to. So here's a 4x upscaled version instead.
since we were doin dragons earlier, i made my dracthyr pre much
love how almost perfect the right hand is but super fucked up left hand
SirVeggie's models usually do well with these types of backgrounds.
which extentions are u guys using
hey i remember u from lost ark discord lmao
idk ur nick seemed familiar
same
i remmeber u had a pfp of a blonde girl with curly hair
Ya is me lol
yea
anyways, im new to stable difusion, which extension do you guys recommend?
for anime pics
Uhhh, thats more model based
Dw took me abit to get everything more or less understood, only a month or so in
Browse civitai see which you like
Theres allot of anime models
without my dracthyr body loaded in controlnet looks weird lol
ok so i downloaded a model i liked, how do i implement it
yeah true
hovering over something sometimes gives you a little popup explanation,
samplers are just a method that affects the outcome of the img simply put
allot of ppl stick with euler a or dpm ++ karras
all a matter of personal choice in the end
I accidentally left my standard waifu prompt on there but it didn't mess anything up lmao
use minkys prompt as a example
is a nice prompt
ignore the lora tho since you dont use that yet
Without the waifu prompts its just
Positive: no people, landscape, pond, mountains in background, flower petals, cherry blossoms, rule of thirds, sunset, highly detailed background, reflections on the water, soft lighting
Negative: people
what are styles
When you type a prompt, you can save it as a style, and then select it from the drop-down so you don't have to type the same stuff over and over
nice
Ye it's super helpful
why do pictures look way better while theyre being generated rather than the final result?
its weird
atm im getting rly bad pictures, i guess i gotta touch settings
Yeah play around with your prompts, maybe add a vae too if your final images are washed out/desaturated
whats vae 🥲
I don't know how to describe it lol
but they tend to make the colors more saturated
And there's some you can use to get specific styles
its not only the colours
even with 150 sampling steps
the faces are distorted
etc
Oooh, try bumping the sampling steps down
I always stay between 30-50
If it's too high, the image kinda gets overcooked and weird
depending on the sampler you are using, there's no merit in using super high sampling steps
Yeah it's easy to think higher = better but for AI that isn't always the case lol
some samplers converges the generation after certain steps, meaning it will end generation. other samplers continues to add details to the generation, but doesnt mean it's good
adding more "details" when it's not needed will probably create artefacts
https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt one of the very popular vae for anime generation
in which folder does it go

Iss okey

depends. But your newly downloaded vae is almost definitely not applied without you changing your settings
go to Settings in your SD
then search for VAE
better way include adding VAE selection to your quicksettings menu
I still need to set up my quick settings omg 
sd_model_checkpoint, sd_vae, CLIP_stop_at_last_layers, show_progress_every_n_steps put this in your Quicksettings list
i mean, that's my current quicksettings list, you can mimic mine
go to user interface
then do this
yea ui is now different
Oh definitely much better now
Face is still a bit decomposed but yea
best way to learn about prompting, is to learn from examples
wait
what's your gpu?
3060ti
is it not at the top of your ui?
with 12gb vram?
ty
8 sadly
still should be enough to use hires fix
The models are definitely decent now, problem are the faces
and my latest generation has 3 legs lol
XD
you might wanna turn that on, put hires step to 10, denoising strength to about 0.53, and upscale by to like 1.5 or 2 depending on your gpu
did you add any negative prompt?
no...
turn what on
you need to use negative prompt as well, try to mimic what the author is using for the model, and change from there
hires fix
that's kindly like upscaling
oh yea found it
try with this settings
generation might take a lot longer tho
what model are you using?
go find an example generation from the author of that model, copy the negative prompt
you almost always wanna use negative prompt
appreciate a LOT
yes
finally i made a better pfp
you can think of embedding as like a DLC for games
lol
all of them or just 1
3d, cartoon, anime, sketches, (worst quality, bad quality, child, cropped:1.4) ((monochrome)), ((grayscale)), (bad-hands-5:1.0), (badhandv4:1.0), (easynegative:0.8), (bad-artist-anime:0.8), (bad-artist:0.8), (bad_prompt:0.8), (bad-picture-chill-75v:0.8), (bad_prompt_version2:0.8), (bad_quality:0.8) this is one of the negative prompt inside author generation

do i copy that on negative prompt?
you can do that, and see if your generation gets better or not. But do make sure you have those embeddings
like easynegative
bad-artist
etc
anything ping me
I'll be on mobile for awhile
sure, but how do i select the easy negative embedding
like i got it on the files etc
Go to extra network tab like here
And make sure you are on negative prompt
And press the embedding you want to add
It should auto type something to your negative prompt
Make sure it’s on the negative prompt side
Because there’s embedding that you add on the normal prompt side
Yup yup, that’s correct
still getting multiple limbs lol
wait last one is kinda fine
although the mouth is kinda bad...
mouths are always bad
maybe cause i typed smile on prompt lol lemme try and remove it
What’s your prompt?
look at her hands
its damn scary lol
Cute black haired anime girl that is a vampire sitting on a throne, crossing her legs and has a nice detailed background that is dark themed, 5 fingers on each hand
Good generation tho
Well, firstly, try to use word, not full sentences
So example how I would change your prompt will be
yes my previous prompt had only words, but i lost it so i had to type it again
lemme try
If you ever forget an old image's prompt, you can drag the image onto PNG info and it'll display everything for you
And then you can click a button to send the prompt and settings to text to image or image to image
btw do my settings save if i reset pc?
i still got a problem with hands and mouth
lemme add highly detailed mouth or something lol
detailed background, masterpiece, best quality, cinematic lighting, 1girl, black hair, vampire, sitting, throne, crossed legs, dark
Anime girl, Vampire, Cute, Sexy, Big breasts, Black hair, Highly detailed mouth, Has 5 fingers on each hand, Sitting on a throne, Legs crossed, Highly detailed dark themed background I went with this
You can even add ‘highly detailed’
prolly cuz big breasts
Since you want to prompt for anime stuff, use ‘1girl’ might be better than just anime girl
Try my prompt
If your hands deformed, there’s probably a need to tweak your negative prompt
What is your cfg now?
it was default, tried a generation with 30 and its the scariest shit ive ever seen
Sometimes it might be the “fault” of the model
i think it was 7
Some model just can’t do hands well
the cfg
Actually most model still can’t do hands well
Find a known anime model that’s kinda good at hands
i actually got one of the most populars from the website lizzie told me about
Try adding, ‘highly detailed, intricate’
I would suggest using an easy to use model for you to learn first
cursed
Lmao
city
is horrendous
It’s kinda easier to get “semi good fingers” tho it’s kinda the same style
Personal I have never use that rev model
So can’t really help you with that
if i restart my pc
do i get the same settings
as before?
or do i need to set up everything again
Some times the hand rng gods will give you something you can at least fix with inpaint or photoshop
forest
theres an option to use the previous settings im pretty sure, the blue arrow
river
that's for mixing 2 or more models
that's probably not as bad as
jesus
for some reason i cant switch model
look at your cmd
do i need to reload ui
is it loading it?
fully loaded
your model is probably loaded?
what does that mean
Not rly i use it daily lol
if it's a newly downloaded model, press the refresh button
yeye i selected it there
and it charged
maybe its just similar
ill try using another one

hell
Heaven next?? o.o
damn this thing really makes use of my gpu
while generating the usage pretty much fills
XD
life of a 8gb vram 
cosmos
the generation times are too short tho
im happy you posted this
because it was annoying me today that i was thinking to myself "I wish i could fucking merge a bunch in a sequence"
and this is exactly it
but better than i thought originally
whats the restore faces thing
i have never use that, so not sure
just a little annoyed doesnt have the triple sum but ill prolly work around it or use supermerger when i need a specific weight
triple sum? add difference?
okay that's probably a new updated one

dont expect perfect hands on eevrything
I would let her copy my homework, no questions asked
sad
i would not, to make her cry



seed is a randomly generated string of numbers, the will deduce the noise used for generation. Using the same settings, same model and same seed, you will be able to recreate the same generation
okay
This should be s good starting point too
@echo timber
Yw
THIS IS LIKE THE 3RD TIME WHO KEEPS GIVING HER GUNS

LOL
can't have shooting stars

comet storm
maybe
falling stars
also how does /word/ word, i see that weird syntaxing
it's kinda like a way of saying i want something in this but don't actually generate this. example saitama \(one-punch man\), means that you want it to generate saitama in the context of opm saitama
ah fuck thats great
are you not using tagcomplete?
it's an extension that can help you auto complete your prompting
yea
finally got the pink hair gradient to work like that
but yeah merge board is really neat
this vae washes it out abit but i dig the style
man if i'd upscale these they be nice but too lazy
Tried a simple weighted sum mix of neverendingdream and cyberrealistic.
Looks funky, good hands tho :o
@rose tapir Is crypto spam against rules? 
Probs is
Doubt crypto scam people go around manually sending messages
Sure is, thanks!
Alrighty, thanks for the fast response and removing em! 
No problem, and thank you for the alert!
Yeee, reacted with warning but saw you were active recently so decide to @ xD
O, AI can do her too? 
Most of my time is spent in pycharm, so I don't really notice them instantly if mods are busy!
Ohhh, thats fair then! Good to know to ping at times like these when not many people are on :o
Thats pretty neat 
For sure! Definitely report first though, just in case I'm actually not here
Yeee, will do 
hi you wanna buy my NFT
1/0
only 0 in the world but u get the only copy
infinite profit
https://youtu.be/iHfJRON3b-w Reminded me of this XD
Cursed
What a good dog
given the last image, I think we can all guess what it is that you just walked on
Furries these days
XD
I mean, I was trying to get a scene with no one in it, but somehow a mutant dog appeared xD

☆twitter - https://twitter.com/sensenzawa
☆ instrumental used belongs to - https://www.youtube.com/watch?v=_qzo8fRwWHc
Good ol senzawa
beter but a bit too realism still bleh
picnic anyone?
rias \o/
I have an addiction
how do remove the pink light shafts :I
Probably adjust your prompt
Also I'd remove those if I were you, not sure if they are safe enough
oh
what is your prompt? btw these are a bit nsfw
Yeah
maybe your prompt has like light ray or something
That would do it
made it worse xD
Looks like my tabs when I wake up too

I haven't checked civit in a few days, so now I have to catch up
Sakura Haruno from Naruto (full body shot), ((best quality)), masterpiece, anime, ((Extremely detailed))(textured skin),(model) ((downtown background)) (detailed background)(cyberpunk) (detailed clothing) (tattoos) ((night-time)) ((cyberpunk setting)), iridescent accents, neon colours, vibrant, intricate, depth of field, sharp focus, ethereal, rim light, sub surface scattering, global illumination (perfect body)
maybe try removing rim light, global illumination. One of the problems could be that it is baked into the model you're using and its overtrained on the concept
ok cheers mate 🙂
Morning Hoomans
1
I guess you aimed for anime style
hmmm grabber alternatives that have similar functionality and a gui, are there any?
aye, glad to hear that
I tweaked it a lot to do certain things
be able to combat (actively, like impact or kill, dismember)
but it's high difficulty to prompt those, from the start anyway
but gearpunk, dieselpunk and cyberpunk is cool
i'm testing simple prompts for now, i'll go further later
having fun with my model?
yup yup
the best thing imo about it is how versatile it is, you can generate almost any style on it except maybe realism
simple portrait
I've only trained it on 2 realism models, both were nude posing. Why? Because they are obsessed with correct anatomy and it helped lower the amount of times I got 3 legs, 3 arms etc
cool portrait
grand ancient dragon, (western style dragon:1.1), two large horns coming together under the chin, two large powerful horns at top of head, symmetrical wings, reptile tail, (solo:1.175), (dark fantasy art style:1.15), (hyperrealistic:1.175), (photorealistic:1.075), (pseudorealistic:1.025),(masterpiece,:1.2), (best quality:1.2), 8k unreal engine 5 here


















