#🍥|anime
1 messages · Page 58 of 1
Generally that chromatic aberration like effect is a sign that either the model is overtrained, or your CFG is set too high.
Stole ur prmopt. My model forgot what wariza is. Will this make up for it? (you DID put large breasts in the prompt...)
Update
ohhh, right, the cfg, it's a life i never set it over 9
I use txt2img, and cfg is 7
which model?
Hey
And to counter those massive mammaries, i shall use... massive ... uhh... this dude is just massive.
not the best way to show it since I'm using Hires fix which obviously makes more changes after the upscale, but the first is fatal_Anime and second is AnimeSharp
I always use 4x-ultrasharp
Likely a problem with t he model then. It can also happen when the emphasis is set too high on a word in the prompt.
This one is also a good choice. Ultimately none of them make a huge difference for Hires fix though since the image is fairly significantly changed by the img2img pass that happens after upscaling.
I fixed the problem
was it too high of emphasis on a word?
But still, usually more coherent than pure latent upscale which tends to go wonky
Yeah, I mean any of the ESRGAN upscalers will give pretty similar results. As long as they aren't highly specialized ones anyways.
CHISELED!
oh like that. yeah, you're pretty much right about that. in that same vein, you could use lancoz, which usually runs cheaper, although i do feel it does give lesser results
Latent is randomly either really good or really bad with very little middle ground and no good way to predict which it will be.
if it doesn't screw up, it's the best option, not even worht considering others
it screws up. a lot. often
I made resolatuin 1024x1024
Yep, definitely bad more often than it is good. I assume it probably performs a bit more consistently at lower upscaling multipliers.
yeah, i feel it's sweet spot is around 1.4
any higher and you're just asking for weird artifacts
i think there was a script that did multiple smaller latent upscale steps
so it could keep image coherence a lot better
Makes sense.
guys what's the best method to make faceless character?\
I like this style
It's hard to remove always the face
TBH I think you'd need to find or train a LoRA for it. The closest thing that most models have had any significant amount of training on is the faceless male which is basically only used for sexual images and is basically just a black silhouette of a person.
Edit: to be clear, I'm not recommending you try faceless male. I don't think it is what you want at all.
or to do fan art of the slenderman
how to train my own model?
with a very good gc
ugh, now comes the really hard part... which 20 images to do i add to my model page 😮
You don't have to draw your own images. You just have to come up with a collection of images in the style you want to train.
isn't a stale ?
some images are freeware
using others' images to train a model isn't illegal, it's just a matter of if you feel it's right or wrong to do it
The ethics are up to you to decide. If you have a problem with using an artist's artwork to train an AI model then may want to consider how all of the models that you use were trained in the first place though.
I use models from someone made it
not I who made the model
but yes I really like AI but I feel maybe the artist doesn't aceept , so that makes me unallowed.
usually, models are made with images posted on public galleries, anime, manga and films
Which were trained on artist's artwork without their permission. Personally, I don't have a problem with that.
what if the artist doesn't accept ?
in art, plagiarism, improvement and personalization, are a thing
Then that's an ethical decision that you have to make for yourself on whether or not you are willing to continue using AI with that knowledge.
I don't know I think I need ai for my work
Then it sounds like you've made the decision that you feel ok with using another artists work for training an AI model.
You probably need to restart SD to make them appear.
refresh the gui
Also, you need to move them from the RealESRGAN folder to ESRGAN.
you can do it in setting, or extensions
i'm looking the video of silent hill 1, i totally forgot how much scary that game was!
@untold glacier @hearty heart @scarlet kelp @sage kiln @modest anvil headsup. Darkalfa model release 😄 https://civitai.com/models/64843?modelVersionId=69479
The continuation of a weird model! Having used alfahentai for lot random art, it was only being held back by it's own confusion at times. This mode...
oh
@fallen sundial make sure to do this before restarting.
your model?
Great will test it later 👍
Can't wait to test... 
i NEED COFFEE
yeah, there's some, but they're... sketchy at best?
so, you will use what to improve your prompt
some kaleens using different models
yo those eyes are sick
hello how to get controlnet models ?
Improved workflow test. Added the autodetect-autoface inpaint so now the faces and eyes are fixed by itself. Input image from blender render:
Output for more detail and so it looks 2D:
Problem is that it keeps changing the skin tone too light. Also, I don't know how to make images get line-art (like on the buildings for example)
ah 840 dollar 3090 (second hand), if I sell my 3070 ti for 400-500 that wouldn't be too much to pay for an insane upgrade
I had that idea and tried to sell my old PC but it only went for 150 xD I ended up getting a PC with 3060 for 600 total
well I've researched the market and currently people are paying 400-500 for 3070 ti and 3090 goes for 800-1000
somebody will buy it
and a 3090 enables me to play much higher settings on games and still get high fps and I can do more AI shit 😛 If I want to. Though I need a big think on this
Yep, 3090 is a good deal compared to the 40 series at least
I refuse to buy the 40 series, the power draw is too high
my 750W will be enough for a 3090 but barely for a 4070
that's ridiculous
40 series is also physically large I heard. I'll have to check if I can upgrade to a 3090 but there isn't much space in there
40 series isn't upgrade friendly
there's enough room in mine
https://www.bequiet.com/en/case/1855 this is my computer case
mesh flow on front, top and bottom
yes but i m using it just like the effect in above, there's a lot of gas around the edges
you need to set it well nd than edit the rest by hand
~Ara Ara
oh, sakura
Its Rin......
hello what's the model?
my own train
so you don't offer this model for downloading or using ?
shame on me!
How'd you get double exposure like that? Wheneve i try comes out weird
editing?
i think it's a lora he's using
Know which? Thats cool af and hard without one i imagine
was in the png info: blue background,(masterpiece:1.10), (best quality:1.10), 1girls,(hatsune_miku),(dynamic pose:1.2), (arms up:0),(full body) (double exposure:1.25), (simple background:1.2), dutch angle, messy hair, (low contrast:1.10), (flat color:0.6), starry, <lora:doubleExposureLocon_v10:1:BG>, <lora:hair_with_scenery:.7>,sky, galaxy,,cloud,cumulus cloud,<lora:scenery-000016:1>,scenery,(by blindArt),(by Ayari Warabi),from behind
probably not very effective
I used two double exposure lora,you could DM for the pic info if you want.
Png info works thx 🙂
absolute thiccness lol
this is a very specific... eh... look 😛
oh my. i think she's into slavery...
Sad
Can someone GIVE ME A PROMPT about there expressions!? I can't think of it
ahegao?
heart-shaped pupils ?
You have to be the dude be posting tiktok slide shows about this
@stone onyx
nooo not me, but I think I got decently close
lora:Helltaker:0.6 using expmix_line v2
Thank youu
Anybody at all
heart-shaped pupils
as wpatzz said
The expression
ahegao
^
Ahegao?
Ahegao is the sus expression?
sus expression?

This is an ahegao
I think I would get another warning if I explained Ahegao
not sure, ahegao being the most used one, you'd best google it... i'm not a google bot 😛
you could try ecstasy?
let's just say its the face that made Belle Delphine famous
ugh.
same character with different prompt yeah
Ima try it tmrow
but hey, she probably made more money doing that than some of us will ever make...
She's a really smart business woman and troll 😛
she worked her way towards the big thing and milked so much money out of people
I think she made a million a month
holy shit
THE TONGUE OMG
can't hate the hustle
Lewd expression, mouth wide open, tongue out
If I was a young woman at age 18-19 I would've done the same thing. For a few years
then hopefully become a millionaire and disappear from the internet
So pretty ❤️
What are you tryna say
Nice pfp
Uh thanks?
that he isn't a young woman aged 18-19 XD
Kinda weird you need to ping me about it considering i never seen you but thx lol
@stone onyx blud try short haired, tomboy, messy hair like mine
yeah
Not accurate prompt of mine
@wispy canopy if you want really gorgeous eyes, try Cobalt Blue or Lapis Blue
I just say that as a habit
Try out teal colored
nah that doesn't do it for me
Can u give a good eye color
Cerulean is better
When i try to get heart pupils i get heart signs
heart pupils is not really strong in models, you'd best use a lora for that
Does it have a lora innit?
you'll have to search.
i don't know, i don't use loras tbh
But whyy
Try out thicklines it'll help if u want to do 2D stuff
Quite exquisite
Not mine btw
@wispy canopy liking your model so far
got a prompt chief? ^^
futobot, cyborg, mechanical wings, masterpiece, best quality, ultra-detailed, illustration,fullshot, sitting, 1girl, 1machinegirl, dress, (city skyscraper :1.6), <lora:Futuristicbot4:0.6>, <lora:add_detail:1.0> was the base prompt. But I did a lot of inpainting and editing
Basic negative prompt
the model shouldn't need masterpiece, should do fine without it 🙂
Ah ok
I noticed that I still have to edit out the artifact in the top right
i've included the negatives i use on the model page, the model gets a lot sharper images using those. did img2img with my negatives
I should retreat from SD and AI stuff for awhile, I really need to write on my novel and not just fuck around with images that I have no use for
🙂 addiction!
It's an unfortunate trait of mine that's never been good for me
How much gb yall even have
vram? ram? disk space?
Well the reason my gb is so low because of genshin impact 
Storage
storage? 10tb?
got a beast pc
most of it full
Isn't 10 tb 10k gb?
got a few storage XD
Aint gigabytes like storage?
Soo your telling me 900+gb storage is not close to a gigabyte
Like one of your storage
Blud how tf can a 7.27tb be in your pc
I have terabytes, which is counted from 1k gb = 1 tb
I really like how your buildings and wings turned out. What denoising level was that for img2img?
0.4 loopback x4
it's easy to get 7.27tb into your PC...Just buy 4tb or higher harddrives
or a lot of em
How much is a 7.27 tb ?
7270 gb
No money
Like buying it
1 terabyte is 1k soo i can download 50 games in steam without worrying about my storage being full
I think they cost like 100 dollars now, if you buy a docking bay you can also just switch em out and use usb 3 or usb c to attach it
Do u call your storages beshikkbar?
it's Dutch -> beschikbaar = available
Oh
How much is a usb these days?
Spotted the native english speaker 
I lost the school usb
Im a Filipino 
I wonder why the hell taker lora has perfect or sometimes perfect habds
Isn't Filipino 50% english 50% your own language?
the model I'm using is pretty good with hands
- I'm using hand loras
Send the lora hand BOY
I mean sir

Idk our words are very deep and they just end up meaning apple in English
that's tagalog, there's also another language
These are loras
I meant textual inversions

there's a hand lora but I haven't gotten to using it yet
My storage is crying rn 
Despite being a LoCon, which is real nice
But its for the GENERATE
I just edited the color abit
But the tongue god dammit
MAKE IT LEWD LIKE THIS

Pretty sure there is a LoRA for that
For ahegao and heart eyes
open mouth, tongue out doesn't work?
Does ahegao mean the expression they're making?
Yes
Try it
I just type
Mouth wideopen, tongue out
YESS FINALLY
It is literally the face
Thank you
I'm pretty sure someone told you earlier tho
I guess try open mouth instead of that one
It sounds native

Honestly my type of girl when i die
That japanese symbol is kinda accurate
I mean, it is
I imagine Ai probably look like this in 30s
Are these lors ? can I use them?
they're textual inversions, you put them in your embeddings folder
How do i steal the prompt?
@zinc flax blud how did u make the hands.
Yk what i wouldn't be able to get it or have it
@zinc flax whats the model?
Lora strength to
Good news i found a good ahegao lora
Ayyy, nice :o
What is your model i need a new that's good with generating stuff like this
That's a lot of pings lol
No lora, just model
What's the prompt did you add outlines , thick lines??
I love her
Can you tell me abit about the model?
Interesting
There are previews on the model card
Definitely need a concept lora for that I guess
Ima delete that before i get muted
Hands are literally rng sadly
Sad can you tell me what's the prompt though like small bits of it
Now all
Not"
Try to get gabimaru
What kind of girl is that? Forgor
What about the thick lines?
1girl, open mouth, tongue out, white background, pink hair, yellow eyes, spoken question mark, half-closed eyes
Lucky seed I suppose
No clue
It's from the model itself
I'm not gonna attempt making male characters
I suck at that

TYSM BTW IMA DOWNLOAD THE MODEL TMROW
Have fun
gnight my man

what are textual inversions?
files that activate with certain keywords in ur prompt and do magic
easynegative is one of them if you've heard of it
you mean it will make my image ai better?
well, you can put it that way yes
Do you guys know what's going on? I downloaded Counterfeit to try it, but I only get this:
Which is really bad compared to the other examples I've seen.
Am I missing something?
are you using hiresfix?
No, only using 512x768.
Try entering the example prompts of the model. (Positive and negative, and set the CFG correctly)
It's exactly what I've done, resolution wasn't included though.
Don't think hiresfix alone solves it
Everything default settings?
What should it look like?
This, lol. I'm far away from it.
It should help a ton though that's for sure
I use these settings personally
Yeah but I thought something else was wrong since the base looks too bad
But yeah @arctic pewter try hiresfix
It's not even close to the image above but yeah, sometimes it does look mega terrible without hiresfix
also one thing with hiresfix
I DID try to replicate the showcase images with no luck
idk what the issue is
Already looks a lot better though
In terms of colors, yes. I think I'll delete this model anyway, can't use it properly.
welp
I meant to run wildcards on highres fix for about 4 hours while I got half a night's sleep to test my model
accidentally did what my doctor keeps telling me to do and slept all night
now I have about a gb of images and it's super hot in here 😒
Yeah idk what's up with counterfeit, I didn't like it either
helltaker lora?
Dont u hate it when u accidentally do the right the thing
I think I'm gonna tweak it to be a little less 3d but so far 😍
it's the worst feeling in the world
Anybody awake who uses dynamic prompting?
Hello Anime community, I'm in a hurry, does anyone know when a prompt is saved as a style, where are they saved?
you mean this?
yes i need the path where it go ...
That'll be your styles.csv in your main SD folder.
You can just edit it with any text edit app. Like notepad or notepad++
Just be careful to keep the formatting.
😦
I had to do a fresh installation from SD and move all the models manually but the styles do not appear in the new
Did you copy the styles.csv file over to the new folder?
i not sure --i move some styles files but i can t not remeber the paths .. i go to check all again

I love having deterministic xformers so much.
@untold glacier
ok thanks I found them where you indicated, I think that before I moved some backup or some sub folder that was inherited from another SD installation, again tyvm.
No problem. Glad to help. 
thanks
guys I did genarate gif but where did it go ?
I can't find it here.
She spilled the Ketchup 
uh, mind sending png info for that?
i mean you can always save and download it right
What Mal said, but for your question.
yeah, that's a pretty common problem
Search in the other folders inside outputs
Probably in extras
Oh wait
I think it's in img2img grids
Just to be clear, discord formats those information
Not recently.
Try it. The info is there.
What do you mean by that? Like prompt, model, cfg?
Try updating discord perhaps?
WAIT YOU CAN DIRECTLY GET THE PROMPT OF A GENERATED IMAGE?
no, directly from the client
I didn't expect my ojou-sama pose lora to suddenly become a betterhand lora.
That'll get you the info then.
I don't think that matters as I just download it directly from discord half the time and it works
I found it
Call me crazy, but I believe it does matter. I've not found info before when doing it that way.
Nice
Idk, worked for me both ways
Maybe I am the crazy one
Maybe we're both crazy
Crazy kazoo if I do say so myself
for me none lol
Try updating discord
Only fixes anything that have her touch her own face though lol
already on latest
Then it may be your version of Auto1111, I'd think.
i dunno, but for me or my country, they still get stripped, i found an addon that is specifically made for that, now i can read the metadata
What UI are you using?
Just use automatic1111
i tried both, online and client, i needed this addon to get the problem solved anyways:
https://github.com/ashen-sensored/sd_webui_stealth_pnginfo
How did you download automatic1111?
github
Do git pull
Lol @ the stealth pnginfo readme.
stealth info was nice though since copying the image worked instead of having to drag the file directly
does anyone know how much pruning a checkpoint affect its outputs?
Not much
^
what was that extension plugin for mwb called that shows preview of your settings?
Hey, anyone willing to try out a new model and give feedback, it's here
https://civitai.com/models/61493/vizionize
I uploaded my model with both pruned and non pruned. Pruned makes a bit more errors imho.
It talks quite in depth about her backstory, it's worth a read

a little blood but no violence or gore
demon transformation can be painful 😛
wings breaking through the skin
Promte: https://cdn.discordapp.com/attachments/1082173390430949456/1092473963285860432/IMG_8178.png , show the man in the picture in a realistic photography and in a suit in front of a white hospital building.
@tame fern
Currently, there is no public bot on the server that generates images. There is an experimental bot available for early server members & Stable Society. You can obtain the Stable Society role by winning our weekly events #⭐|pow-info #1087493421209485393! However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝|tech-support for any problems you encounter while installing it locally!
A succubus taking human form and having fun dancing and drinking blood and not a care in the world for washing it off, eventually she's drank so much she begins to transform
who this
just need to color correct the first one and there's a smile, if I use after effects interpolation to create new frames based on current and front, and then some duplicates. I could get a full second smile 😛
maybe add a few more, add some light changes in there in photoshop, so it's like the train is moving, add a shakey camera (subtle) so it's like obvious it's a train
nah go for it the gens are nice
I love that LORA
she definitely is about to steal from you
or sell you something
or end your life
last one is the most probable one
ruthless 
guys is there any good loras I can use ?
based on the artist モ誰. Probably my best quality LoRA thus far.
This is best used as an inpaint tool, since it's a little overtrained and it can affect the overall image quality if used unconditionally (nothing ...
what is this lora extactly ?
style of an artist
I don't think I can get useful of it
I liked it
https://civitai.com/models/55153 this one i used recently its pretty cool. i accidently send the wrong lora there sry
SMUGO HAS GENEROUSLY STARTED HOSTING THIS LORA: https://smugo.ai/ <3 An ATTEMPT at a LoFidelity (not VAPORWAVE) Aesthetic / Pastel style Lora. T...
is there other useful loras ?
oh okay it's okay
any other loras ?
If there is lora for correct hands or poses it would be nice
btw talking about poses and hands. can controlnet be used for hands and fingers? does anyone know
I've seen it done but I have never done it myself
if you have seen it done i might have to try it then
I tend to just bang my head against the wall re-doing the gen 100 times until the hands are right ;)
yes
i do this too but it takes very long and its just a massive time waste and then maybe the hands are right but something else is wrong and i just get frustrated with SD..
so control sounds attractive now. i mean the stuff looks very good CS1o posted
it depends, you can get crap output with controlnet too xD
you need a good source image
I feel I get lower quality with control net
using control net with inpaint to fix details sounds better though
I can't use it sadly
why is that?
@wispy canopy about to download your mix
testing my model against different ages and ethnicities and the columbian woman is just a close-up on her ass 👀
be sure to use the negative prompt / model settings i posted too, it really helps bring out the quality
got this dark skinned beauty 😄
that looks really nice
being able to do dark skin well is a hard requirement for me now
got the prompt from aiska, it has dark_skin:1.8
mrrreow?
oh wait i said fox girl
gonna recreate with cat now
these eyes ❤️
woah nice style
one of my favorite prompts ever
here is the cat:
That's a nice cat.
thx, indeed it is ^^
your sleep paralysis demon
An eerie nighttime scene unfolds under a full moonlit sky. At the center of the illustration stand a group of undead creatures known as "The Dancing Dead." They wear tattered clothing reminiscent of olden times and move in unison, swaying their bodies back and forth as if caught in some kind of trance. Their faces showcase empty expressions devoid of emotion. In the background, ghostly apparitions emerge from between the shadows, watching the spectacle unfold. The overall atmosphere remains dark and ominous, punctuated by the silvery light cast by the moon. you know you want to prompt this!
those who prompted this song, know which song this is!
my sd is kinda borked atm lol
oof... well, it's ... dancing dead!
they're borked by definition!
slightly more terrifying tbh
it's glorious!
imagine those things jumping after you

in dancing motion pouncing on you
that thing will never die, will it? 😄
Noy♥️
noy ❤️
these dudes are certainly vibin'
i mean, they are totally vibin man, what's wrong with evil undead just having a good time
spoopy
this looks like the beginning of some tragedy about to unfold
got the inspiration for that prompt because of this song: https://www.youtube.com/watch?v=vX4y3rXwy4M

i got something different...
Their lord and savior
THIS DUDE VIBES XD
undead rave
seal girl from AMB that i tried training for fun
oh ignore that reply
shes emote material
i dont know about you....but this rave looks sus
does anyone know which anything model is the real deal cuz theres so many and idk which one is which anymore 
anything is pretty outdated by this point...
just a normal rave at 6 in the morning
ik but im getting bored with all the mixed models and everything 
well, i got bored of Anything XD
damn
ok
Anything(except for some things)
is it a mix containing aom3
nope -> i removed that specifically
respond to natural langauge prompts better
AND it doesn't need masterpiece 😄

this is on darkalfa
damn
oh nice will try out as well
if antyhing, it has gorgeous lines ❤️
trying out the new controlnet "reference_only" preprocessor. This uses no controlnet model and my prompt was just 1girl, purple hair, closed eyes, open mouth, smile, mid shot, full body, outdoors, bright sunlight with your image as the controlnet input
damn 😮
those controlnets are getting very spicy
Yeah, this is pretty cool. it does a really good job of capturing the idea of the input image without just copying it.
I could see it being really useful for expanding small datasets to use for training.
This new controlnet 1.1 tech?
Yep, here's a short video covering it. https://www.youtube.com/watch?v=ddcySSbjkNQ
The new Reference Only ControlNet Method is very Powerful. It can create similar images from just a single input image. You don't need to train a Model or a Lora. You don't even need a complex prompt. While other ControlNet Methods rely of composition or image Details, this method give you artistic freedom.
Links from the Video
Refer...
Thx sir
do I actually NEED it NOW?
Depends
is this new newest version?
Damn, I need to try this later.
can't wait to use
omg
on what artist is this based on?
I wanted this style for so long now
oh wait I know this model
I talked about it when it came out
omg it is trained on the artist I thought I saw in this image
shion
this model is too good for it's time fr
maybe even good now
idk
need to test
man I tried making a shion lora before and failed
well it was my first lora ever
but still
I go sleb, GN
O, good nite kazooni :O
Swimming with the fishes
My type of girl
hi everyone, I downloaded this image from someone of you guys a long time ago and still have no idea how to achieve this style. Do you know any model or loRA that will help?

Sir they asked for help, not replications XD
masterpiece, best quality, high res, detailed background, intricate, detailed eyes, cinematic, highly detailed, absurdres, (realistic:0.6), small breasts, 1girl, round glasses, cat ears, speech bubble, piercing, lora:tafIsAllYouNeed_taf2022B4:0.8@molten shadow
1girl, animal ears, solo, long hair, yellow-framed eyewear, glasses, cat ears, looking at viewer, round eyewear, black hair, parted lips, blue eyes, shirt, off shoulder, upper body, bangs, earrings, bare shoulders, black shirt, jewelry, animal ear fluff, teeth, speech bubble, simple background, off-shoulder shirt, eyelashes, grey background, lips
wallpaper, extremely detailed, intricate, cinematic lighting, Realistic, highres
metadata use png info to see
😄 is for help him
nah, of course I know the metadata, but the model is secret guys
it look like the Aoi Art style but I'm not sure
How do u get such pretty hands 🥺
Anybody know which model is best for anime example these kinds
that a Kpop idol or sth, not sure
luck
Some just use rng or depth library
I feel I got closer to the original style not using the meta data
Does everybody show there prompt?
@molten shadow sometime the lora bleed the style of artist or model..
.> I don't show my prompt because I am lazy
Png info
ye, I will give it a try, tks for advices 
that is what I mean. PNG info my images and you will get none
Did u use something to make that hsppend?
So you basically dont wanna share your prompt?
No I'm lazy and copy paste is the fastest
So copy pasting wont be showed?
🤓 looking mf
yes the clipboard removes meta data
Computer clipboard or sd clipboard?
you mean copy/pasting from the UI?
I don't think that's right
I mean....
pretty sure its a setting that turns off params
it is what I am doing
their may be a setting that turns it off
I don't know what it is and I never set it
do you all just... save the image and then drag it over or have the output folder open to share the images with meta data?
I copy/paste from the UI
Sometimes
people have 100% taken my prompts before too ❤️
Can u steal this prompt?
yeah
Check.
I did.
Do u y'all just save the negative prompt? Or is it just me
is this unchecked?
ngl I've tried to take your prompts before but I assumed you just wanted to keep them private
not that it's a bad thing
lol well now that I know I might bug you for prompts sometimes
you post really nice stuff in here most the time
because I'm not opening this folder
feel free to ask me for prompt info and I'll get it I'm never intentionally hiding things
I just >.> 🦥
I should state that this only applies to recent images because if you ask me about an image I posted a month ago...

lol ofc
Same
You guys never use subfolders?? 
I run my gens through cafe aesthetic and then the good ones go into keep folder
then I delete everything in my gen folders
Maybe so you wouldn't have the problem of explorer trying to load 27 GB worth of images, lol.
It is only a problem if I acknowledge its existence
Ah, the head in sand approach. 
I mean i dont really run into any loading problems, i just really dislike clutter
Oh I run into loading problems. Windows explorer isn't the best file management application, and it shows quite prominently when dealing with large quantities of image files.
I only start having problems when the file count gets to 5 figures
which admittedly is only about 2 weeks of gens when I'm active
but it's probably better to just use the image browser extension to navigate your gens anyway, since you can filter exif data
Ehhhh not always
Thats not always up to par
Like rn, for me when i use it, i cant delete images one by one without it exiting the picture im on and making me find it again and reclick it
I can only remove like a few pictures and then it just goes back to page view without me saying so
The Infinite image browsing extension is fantastic, and far better than the others but is still far slower than explorer when it comes to file management.
I love testing these models by pushing strange ideas via prompt. It can yield such creative results.
seconded
esp for linux
Cadia Stands!
So I tested two different models thinking that they were specific models--and found out at the end that they were different models than I thought they were after double checking things.
In short, I did a double-blind testing all by myself... 
So is that a lightsaber or a tanto?
Tanto-saber. The Eastern Jedi sect got some neat things

If family guy can have light saber butter knives, why not tanto saber
Pretty neat

How do yall get good eyes
Pov your training your prompt:
basically you have to find a nice balance in the model, settings, and get lucky to start getting good aspects like eyes, hands, etc. also, being able to difuse larger images can produce better images overall. I would also upscale and play with settings there too to refine things
Ill ss that ty for your time
Wait damn it i forgot what weight does again
And none of that (((((prompt:2))))) stuff
So does ((prompt)) do nothing?
how do you make the art style anime-like? when I try to type something in "styles" it just clears it immediately
Lora or something train it
(Blue:0.6, red:0.8)
By () youre linking the words within together and with the numbers youre giving numerical value of importance
Basically this was two days ago prompt
why are there so many more text input fields than on my screen 👀
So example i say ((big opai:1.2)) it gets important?
Exactly
did you learn all those prompt tricks through trial and error?


what's the SD VAE field at the top center of your page?
there are a lot of good vaes these days so shop around
hm my page doesn't have that for some reason. did set something up wrong?
like the entire text field doesnt exist
so i want to look around for a VAE i like and also LoRA
Civitai is a platform for Stable Diffusion AI Art models. Browse a collection of thousands of models from a growing number of creators. Join an engaged community in reviewing models and sharing images with prompts to get you started.
thank you, ill take a look at these!
You really like these sort of things don't
you
is .safetensors the right extension for lora files?
What are you tryna do though?@tepid geyser
uhhhh like
okay to be frank right now i have no specific higher goal, i just want to make an image that looks like it's from an anime and doesn't look like shit
everything i've tried so far looks like it came out of a horror story
Umm
Okay download thicker lines
It'll help you with 2D stuff
Add 2d in the first prompt
After you download thicker lines
Go to models to lora
And add outline,thick lines
How did you get good eyes blud
that style is from a old embedding i did
I have to like include the filename of the lora model in my prompt right
i dont think it's recognizing the VAE i added to the folder
it has .ckpt as the file extension is that right
ok im just gonna try it
see what happens
@glacial bane there are two methods to adding emphasis to a part of your prompt. You can either do (word) or (word:1.1) format. In those When no number is specified, each set of parenthesis multiply the emphasis by a factor of 1.1, so (word) = (word:1.1), ((word)) = (word:1.21) and so on. Unless you're exceptional at mental math, you shouldn't combine the two formats because they will still multiply off of each other and get out of hand quickly. It's best to simply use the (word:1.1) format because it allows you to be more precise with no math required to figure out how much added emphasis it is receiving. In general, you can safely add up to 1.3 emphasis to most parts of a prompt without major problems. Going higher than that will sometimes cause images to start degrading in quality. You can also specify negative emphasis in order to still prompt towards a thing but with less emphasis towards it. The two formats are [word] = (word:0.9). Note that all words inherently have an emphasis of 1.0.
what's the difference between ckpt and safetensors?
I have no idea how i understood this but damn ty for writing all that
Ultimately nothing that you will notice, other than them being different python filetypes. Behind the scenes, .safetensors are considerably safer to use than .ckpt. Pickle imports are a method used in python that allows importing additional code that is needed, the problem is that they are written in binary and are not readable by humans, meaning that malicious code can be imported and executed by them without the user's knowledge. Safetensors do not allow any pickle imports, while .ckpt uses them. WebUI has a pickle detector that is intended to detect unsafe pickle imports in .ckpt files, however it could be possible for unsafe pickles to slip through the cracks in this type of system. TLDR: use .safetensors if given the option between the two.
This might be easier to understand: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#attentionemphasis
@supple raptor quick question are you okay with using your time typing large chats to help the membere?
As long as I'm not busy with something else, yeah. I enjoy helping people out.
Good to hear we are really grateful for your help
thank you! I put a .ckpt file in my VAE folder, it's the klF8Anime2 one, but the webui doesnt seem to recognize it as it doesnt give me the option to enable it/use it?
i could be looking at it incorrectly
Where are you trying to select the VAE from?
are those parentheses or brackets?
actually im not sure where im supposed to look
(( parentheses))
I'd recommend going to Settings > User Interface and add ,SD_VAE to the end of the box labeled "Quicksettings list". Apply that setting and then restart the WebUI. This will give you an additional dropdown at the top of your WebUI that is called SD VAE and will allow you to use that dropdown to change your VAE at any time, just like you do when you select a model.
thank you!! i'll do that
The sort of glassy look that her eyes have is unexpected (in a good way) for that artstyle. It looks nice.
because im so new i dont know if my results are bad just because my prompts are bad or because i set it up badly
i just generated the most cursed image ive ever seen
also I did this but no dropdown menu appeared for some reason
yes i reloaded it and also restarted it
You may need to close the CMD window completely. Sometimes that change doesn't take effect when you just use the restart button. While you're modifying your quick settings list you should also add ,CLIP_stop_at_last_layers, so the full list will be sd_model_checkpoint,sd_vae,CLIP_stop_at_last_layers. This will add a slider to select your clip skip. Nearly all anime models will perform better at Clip Skip 2 than they will at Clip skip 1 (which is the default value.
thanks for the tips!
let me try to restart the cmd window
oh the clip skip slider showed up bot not the VAE dropdown
strange
ahhh
it was a case sensitive thing
i did all caps like here
but when i switched to all lower here it worked!
thank you!!
oh lol. WebUI code is made of spaghetti so that doesn't surprise me now that I think about it XD
when you say "models" like all anime models will perform better, what exactly are you referring to?
so i can understand better and also make sure im doing things correctly
models are any of the files that you select in the "Stable Diffusion checkpoint" dropdown menu. The big .ckpt or safetensors files that are ~2GB+
ah
where can i find different models?
and where do you put them folders wise?
i only have the v1-5-pruned=emaonly one
https://civitai.com/ and https://huggingface.co/ are the best places to find models. CivitAI is much more user friendly. Those files go in .\models\Stable-diffusion













