#🍥|anime
1 messages · Page 101 of 1
Want some cookies?

somebody is under the table 
shrugs
uhhh
It's just a candy bar. IT'S JUST A CANDY BAR
uh huh

pov chocolate mukbang
holy mixed up wtf dr who happened here?

Who wants some chocolate?
is my second favorite mushroom theme i've made
are you sure there's mush room in there?
give me a moment or three 😄
more than in here xD
hawhaw
heh I'm looking in my "other" folder, which I don't know why things end up in there sometimes (as a copy)
but somehow got this, I don't remember if it was img2img with too low denoise or not though
also yesterday I made some newer Clara pictures
thats the name of the character I made
her look is very consistent, aside from clothing
even when art style is switched
here's all the art styles I made her in
theres three different 2.5-3d ones that on a quick glance look the same but aren't
try "fly agaric eyes with pupils like oysters"
was that the one I said?
no
boring
Good night! 
Anyone using sdxl?
I know I'm not. The short answer is that there are too many features that I'm used to that either aren't supported yet, or people haven't yet made the same innovations that we've had with 1.x.
I'm not gonna do 7 minutes per gen for larger gens, when 1.x can do it in 50 seconds. 
That and we shouldn't have to run it with no-half-vae to make it work.
I'm running SDXL to gen the base then running it through 1.5 checkpoints
159 sec per prompt for 2700 x 1400 outputs on a 3060Ti
dont jump!

oh noooo
I'm heading off too. Good night peeps. Happy generating.
Neko Boy.
waiting until sdxl is released with auto1111 bc im too lazy
uh oh
It's been functional on a1111 since the day before SDXL 1.0 released. Unfortunately the performance is significantly worse than it is in ComfyUI. It's bad enough that it isn't worth using at all on a 3080 12GB.
bruh my laptop would die with its 6GB 3050 Ti
Yeah, don't even bother trying. Your laptop would probably run it faster in ComfyUI than my PC does in A1111. Comfy is very good if you don't mind nodes.
im trying to create a cover for my webnovel and was looking into SDXL bc it looked good, but even then my pc fans go mach 10 when i make an image in SD 1.2
💀
Colab is a good free option if your laptop is struggling to run it.
moners 😔
tried controlnet
i can see the net on her head!
its what happens when you eat to many crayons
damn, crayon, always crayons, when people will stop eating them?
i suppose when they stop making them so tasty
it's just me, or comfui it's way faster than auto?
idk what those are
auto1111 it's what we usually use to generte images
comfyui it's a new build. but it's pretty weird and confusing, but his performances looks better
i could generate an image with sdxl with a 4gcram
on a 4 giga grapic card ram
it's a big jump, considering that usually it runs on a 6 giga gc ram
moeplus lora where i can take?
looks cute from the name
Flee! The Trumpet has been blown!
what's the point? our world is doomed
let's just play piano in tone with the trumpet
True, I mean look at those little lumpy humanoids there
that's us when it's blown
unless we bow to this guy
he is not a party guy, boring
are you kidding? Look at the blood fountain behind him
he is that guy that add alcohol to the fruit juice, thinking it's funny
nah he's the one who put LSD in the stamp you get at the door
damn, i knew it was him !
aaaah almostgone through 1000 images now and saved like 50 of em
Do you have an idea? a world? a story? or most importantly, characters?
working on i
it
not what I was going for. hmmm
almost time for work. again. sigh
-wheres my order - Btch you think I know, fedex has it, i'm not Fck fedex!
here have a chibi
disappears
In the time that you worked on it I came up with three ideas I didn't come up with but will pull straight out of my ass
Idea 1:
Liam has lost count of the number of times he was suddenly pulled into an arena and without warning, forced to do combat. Before being thrown back after winning. So he wondered if it would stop if he lost. Liam purposely lost a battle and this time he found himself thrown into a landscape barren and filled with dangers.
Apparently he wasn't the only one who figured he'd try losing, he would stop being summoned. It turns out some of the most powerful entities in existence purposely lost. Now Liam has to survive in this barren land, find allies, fight foes that previously seemed so easy to defeat in the arena...But that was just another one of those situations.
The Arena exists past the nine circles of hell and is the only way back into the real world. is Liam ready for a journey into hell?
Idea 2:
A space ship of alien design drifts into the solar system and crashlands on Mars, where humanity has just a decade ago began a colony in the former lava tunnels. The Aliens it seems had detected earth and humans, but passing through the black void between solar systems their ship was torn apart by a black hole.
The aliens begin slaughtering humans until Chris, in desperation bites into an aliens artery by sheer luck and being paralyzed, he can do nothing but wait as the alien bleeds out into his mouth and wounds. Causing a mutation, a hyper-evolution to exactly the stage Chris needed to win.
After dealing with all the attackers and trying to have others drink the alien blood, causing them to die horribly. They realize Chris for some reason is their only defense until earth can mount a rescue, but communications seem to be quiet. Has something happened to earth?
Im pulling these out of my ass btw
just coming up with them now as I go
how many will I come up with before you return with yours!
Aren't those a falling star
I'd just paint over it in paint
Then run it through inpaint
temporal = relating to time
I'll upload it somewhere once I merged it lol
BUNG
what is the second option?
please i need it
Did any banger new anime models drop in the past two months?
I'll shamelessly plug my own merge that's been pretty popular https://civitai.com/models/100505?modelVersionId=126070
lmao
I haven't used much else in the last month since I made v1 
I made https://civitai.com/models/94401 and I like it a lot, also recommend kakigori https://civitai.com/models/100505 and darkalfa https://civitai.com/models/64843
Looks good. Thank you.
Playing with img2img on some stuff that's a bit abstract for SD while doing my best to keep the content regardless. This one's a ball-jointed doll. Not my pic (mine is in storage right now T-T), but one of the marketing pics. It's not perfect, but I think it came out cool.
Took a lot of work to get there
These look good. Appreciate it.
That came out really good
Was something like 30 low denoise img2img transformations at varying resolutions and controlnets to get there.
nice, the left is original?
Yep
ok ended testing, with 4g gram i can run only sdxl, but on the main model. no refine, lora, or upscaling
if i try, i get out of memory, or black images
Actually going to downscale it and step it up again, see how it goes if I bump up my denoise a bit more now that it's established in the model/lora style and therefore less volatile
Nope, REALLY doesn't like the sword in front if I step up the denoise any higher than 0.425
i love the original concept
I love her, just don't have anywhere to display mine right now. Need to get the place cleaned up.
damn, it's a true doll?
Yep
Eh, wouldn't call it an action figure. She's a 600$ collectible resin doll XD
"This is the White Star Fleet. Negative on surrender...we will not stand down."
"Who is this? Identify yourself."
"Who am I? I am Susan Ivanova, Commander, daughter of Andrei and Sophie Ivanov. I am the right hand of vengeance, and the boot that is going to kick your sorry ass all the way back to Earth ... I am Death incarnate, and the last living thing that you are ever going to see. God sent me."
White Star opens fire on Earthforce fleet
good quality actions figure don't cost any less
Fair
I miss B5, I should rewatch it sometime
b5?
yep
nier?
Babylon 5
Bab5 is the best
G'Kar and Mollari were such tragic characters
I mean that as in to say they had amazing storylines
ohhhhh that saga, i remember, pretty similar to battle star galactica
hmm that wasn't even the prompt wtf
in my language, the voice were orrible
BSG is a machine race creating bionoids etc. B5 is about ancient races and prophecies
hmm and have to active the vae manually oi vey
then that crusade spin off that only lasted a season, a shame
yeah B5 is still good to this day
cause even the race that was left behind was so weird
maybe i should see in original language
even if not ancient
there's even been some fan made crossover films
i should give it one more chance
kinda funny
I dont watch fan films unless theyre very short tributes and really good and when I do I dont count them as canon
oh and they are making a cartoon/anime of B5
Return to BABYLON 5 as the epic interstellar saga continues with THE ROAD HOME. Travel across the galaxy with John Sheridan as he unexpectedly finds himself transported through multiple timelines and alternate realities in a quest to find his way back home. Along the way he reunites with some familiar faces, while discovering cosmic new revela...
but the genetic poisoning of all humans was something
I do love how so much of the plot in B5 was interwoven in this tapestry of great world building
it's a shame they told him he was gonna be cancelled after season 4 and then changed their mind and said he had to make S5 or someone else would
that really screwed up his timeline
My favorite B5 moment is Vir absolutely roasting Mr. Mordin
That's B5
Yes, my fingers got dumb
it happens, my fingers are diagnosed with retardation at this point
Mr Mordin was a great character too, I loved hating him
G'kar has some great lines too
After G'Kar becomes more philosophical he has a lot of wisdom
guys, refine it's a thing just of sdxl?
oh and that little wave.. love it
its a shame they didn't get to evolve the Drakh etc
the thirdspace aliens were underutilized, I always wondered if they were intended for more
yes, but you can refine and image using img2img as well I think
with regular models
ok ty
tho hi-res fix can do the same thing, with some things like adetailer and other thing.. not sure of all the stuff, learning th ings all the time
DAMN SHE THICK
Niiice
the upscaling is very very very fast
holy moly
this is huge
3 min and 40 sec to complete it
a 4096x4094 lol
now, if i can set a control net, i can generate monsters very fast
is it just for me that the image does not show up in the webui when it completes anymore?
where is it? if you know ofcourse lol
Live preview?
oh wtf
no claudia christian no B5
where can i see that config?
In the settings.
This isn't live preview though. Images have always been buggy for me with Auto's in showing up correctly. I'd reset the UI or restart SD to get it working again.
same ,sometimes the png info stops working for me too
@wet elm
If that setting is turned off for you and you still aren't getting images shown after generating then something is broken in your install.
its off
it was normal until yesterday for me
Yeah, the images sometimes stop showing up for me properly. They've been doing that pretty much since the beginning with all the Auto1111 builds I've used. Often times I can get it working again by exiting out of the image at the top right, or selecting a new one from the bottom image choices. That can kickstart it into showing the images again after generation. I often also get the previously genned image displayed instead of the current one. 
it stays like this for me
Reboot SD. And close the tab and open up a new one.
You can also try a different browser if that doesn't work to see if that helps.
firefox my beloved
I would not recommend using any Gradio based WebUI in anything other than normal ass Chrome, and preferrably with little to no Chrome extensions. Gradio is bad enough as it is, it doesn't need any help getting broken.
Yeah, Edge is mostly fine, but has been known to have problems with inpainting sometimes (same goes for other Chromium browsers).
Honestly, its beneficial not to use webui in your main browser anyways so that you can turn off HW Acceleration in the browser you use for WebUI, and stick to that one when you're generating to save ~2GB of vram. Then you can leave it turned on in your main browser.
chrome uses too much memory so i dont use it, but ill try to see if my problem ends
imma try that
This.
Shouldn't all this stuff be in https://discord.com/channels/1002292111942635562/1002602742667280404 ?
Chrome is very good at releasing memory when other apps actually need it. The "cHrOmE uSeS ToO mUcH rAm" thing is a meme. If you just have a free 8GB of RAM sitting unused then it's beneficial for your browser to use it.
Turning off HW acceleration need to be the standard practice in the browser you use with SD.
Here you go. 
We have discussed this sort of thing between posting images in #🍥|anime since at least Aug of last year. I wouldn't expect it to stop anytime soon.
I like to think of this as the anime community within Stable Foundation.
We post a lot of anime and have fun doing it, but occasionally we talk about other things.
like anime or animation (japanese)
Also, how do I delete someone elses profile picture?
Trypophobia?
didnt know that. ty
i can switch to my Kirito pfp or my steven seagal pf if you want
Ooooh Kirito!
Boring isekai sword jesus
No, that really is how I feel about Kirito
SAO was disappointing for me
do you have a moment to talk about our lord and savior Kirito?
Sorry to hear that. I loved the first and second seasons.
Totally fair. I'm just VERY burned out on isekai power fantasy in general.
The longer time goes on, the more we get with longer title names. Pretty soon we'll have paragraphs as the anime titles, heh.
then allow me to recomend battle royal highschool
I only find isekai power fantasy to be fun when it doesn't try to take itself seriously.
It's not REALLY isekai, or rather, it's about as isekai as SAO is, but Bofuri fits this perfectly
what else do you want,it has epic based lonely but kind gamer Kirito who plays VR,saves the world and gets Asuna + a harem
It has sexual assault, it has female characters being reduced to plot points with zero agency, it has a main character who pulls deus-ex-machina out of his ass on almost a scheduled basis.
yep, Bofuri is fun
mfw ( my face ) when sexual assault in a fictional world
The issue isn't so much the sexual assualt
It's that it was portrayed as fanservice
While being sexual assault
i remember that on a blockbuster movie i forgot the name
Okay?
indeed
Are you trying to say that because it's happened elsewhere it's okay/unobjectionable?
Serious topics in anime isn't a new thing, nor a bad thing. It is if it gets gratuitous, as it did in one of the last seasons. As for characters being reduced to plot points, that's just something that happens when the focus is elsewhere. While not great, it does allow for better writing for other characters, like Asuna and Kirito's relationship. Regarding deus-ex machinas, yeah that's a downside of the series that I'm not a huge fan of.
i do not care if it happens or not
So you find it perfectly acceptable for sexual assault to be portrayed in a fetishistic, fanservice-y way?
i have no strong feelings one way or the other
Yeah, that feels like a complete cop-out
you could try to submit a complaint with our costumer support if you have problems
think my first 'trapped in a videogame isekai' was during the .hack days https://youtu.be/3gTzstpYhUU
Remastered opening movie of .hack//G.U. Last Recode Vol. 1 Rebirth.
.hack//G.U. Last, Recode includes a remaster of the G.U trilogy, as well as an exclusive conclusion to the saga: Vol 4. Reconnection.
.hack//G.U. Last, Recode will be released on PS4 and Steam on November 3rd.
Pre-order for PC: https://store.bandainamcoent.eu/ref/D00200
Pre-o...
I remember those days.
.hack was amazing, and also VERY different from SAO
I remember //sign being much more of a philosophical bent rather than just action anime
wait thats not from bayformers
opinion invalid
...Bayformers are an abomination and should be expunged from the collective memory of the species. The fact that they're still being made is an affront.
Don't take the bait.
Fair, I'm just ignoring them from now on
now imagine Shia on top of that
restarted it but still nothing, forgot to say it shows the image as it is generating but not the final result
i asked there 2 times and everyone ignored me so i came here
And you've tried a different browser? What are you running it through?
but the image is on the output folder?
yeah
try to reload gradio
how do i do that?
its at the bottom where its says what version u running
holy can i have the parameters for that?
yes, i didnt understand the second question though
high resolution photo of a Lycian waifu with glowing backlit panels, overlooking the abyss, swirling ominous clouds, tokyo street alleyway, daytime, darkness, shiny, fashion, intricate storefront details, detailed, composition:1.3) it's a mix of sdxl base with lottalewds "lottamix"
sdxl is something else dayum
Nevermind that if you've already tried a different browser. At this point, I'd say it has something to do with your build. If you've updated recently then it's time for a rollback or a separate/reinstallation of Automatic1111.
sometimes i get that bug where image doesnt show but it fixes itself after i generate another img on a different tab
Wait, you can merge xl with 1.5 models?
@untold glacierWell, lottamix is a 1.5 model, so I'm confused then XD
what would be the prompt for this sitting position?
tried with sitting on floor, legs pulled to the side, but hmm not working
use controlnet openpose pixel perfect mode
Hmm. I'm confused as well then.
i cant with SDXL
ya i re-encode the the VAE decode back to latent and then use the 1.5 models on the latent
CN is not supported in SDXL yet
🙏
So essentially your'e using SDXL for the composition but 1.5 for the image? Sorta?
imma try that
ya the XL bases are higher resolution so you end up with higher outputs and better comps
yeah im gonna try reinstalling auto1111, if it does not work i will give up lol
Ah. There's your answer.
try with front view,sitting down,crossed legs,one hand on top of leg,
it's a hacky work around until the xl models come out
"crossed legs" will nuke the pose, sadly.
he could try with "yokozuwari" idk if sdxl model would recognize that tho
Yeah, for as specific poses as that, usually controlnet or a LoRA is needed.
I really dig the white outline. What is the model? / How do you get it consistently?
crossed legs is different tho
crossed legs is this
as a browser, i suggest opera base. it's very basic and have less risks to conflict
yeah im experimenting with those prompts lol didnt find much info on google
yep, yokozuwari is the right tag for most anime models since that's what the boorus use.
is that 1.5?
Yeah
dang that's beautiful
https://github.com/go-skynet/LocalAI
does anyone use this
and so.. it begins
I use Oobabooga as my backend and SillyTavern as my frontend for LLMs
https://github.com/oobabooga/text-generation-webui
lama.cpp performance is pretty bad (which appears to be what LocalAI uses from a quick glance).
ah
For low vram situations lama.cpp might be better than the alternatives though since it runs on CPU with optional offloading to GPU as opposed to AutoGPTQ or GPTQ-forLLaMa which run on GPU with optional offloading to CPU.
The best performance I've gotten is with ExLlama, but that can only split between multiple GPUs, not between GPU and CPU.
oops
those are some fucked up tiles
For the base_path the folder you need is usually named stable-diffusion-webui, it'll be the one with all of this stuff in it (I probably have a few extra folders and files in there)
Yep, that's the right one.
i thought it was only the image issue
but i cant change models
i posted on tech support but again they ignored me
can someone help me?
post the console log to see the error messages
why people still train models in euler?
ty, but i was able to solve it by adding a commandline arg about gradio idk but it worked
and it solved the other problem too
was it -nogradioqueue?
yeah
🙏
SD + Pika Labs
this is the ideal female body. you may not like it, but this is what peak performance looks like.
i finally made comfyai work decently, what a chore
but it's too bad, it's missing so much stuff
yea id rather wait for more features
but it's so much faster
is it faster on 1.5 models too?
yep
it upscales in an instand for example
maybe it's not worth it on a powerful pc, ut on a slow one, the difference is noticiable
id like to try it but i dont think the video2video extensions work on comfy
i suppose it's about making it, setting the various boxes well
but if more ppl switch to comfy then someone will eventually make it,and ppl will switch cuz its faster
yeah
i had to clear my prompt and start over, everything was looking weird and bad
this is why i havent tried it yet
am only still getting 111 tuned in
does comfy run alongside 111 or a whole nother thing
cuz i saw a comfy ui extension for 111
i was wondring about that wouldnt they both load the same ckp into memory
i suppose i have to try to know, bt it's possible
or do they share ram i dont think id have enough for both of them to load a ckpt
im curious what the comfy 111 extension needs
They each would need to load the same file so you definitely wouldn't want to use both at the same time.
From what I understand the Comfy extension for A1111 just lets you build your own pipelines the same way that ComfyUI does, but uses A1111 code which means giving up basically all of the performance benefits that ComfyUI has.
aa
there thats better
that would make a good backdrop for a comic page, with another layer for the characters
or visual novel
could def see that in a dating sim
Another thing to keep in mind with ComfyUI is that it does a few things differently than A1111 does.
- Comfy generates seeds on CPU, while A1111 does it on GPU. All this means is that a given seed on Comfy is the same regardless of GPU, but that isn't true for A1111. Neither option inherently produces better outputs.
- ComfyUI does not normalize weights in your prompts like A1111 does. Again, neither is inherently better or worse, but will result in different outputs if you assign emphasis to anything in your prompt.
- Some of the hidden sampler parameters are different between ComfyUI and A1111. AFAIK Comfy uses the official sibma values while A1111 uses... something else.
i love the part where she says "w8'\0 you * hovc"
maybe blurring the bg behind the words, helps exalting them
i don't see the token limit on comfyai. weird
Yeah, comfy doesn't use one. There is no actual token limit anyways since it automatically handles long prompts like A1111 does.
is there a token limit on stableswarm?
yeah, but after 75 tokens, the next once works on the work done by the precedent once
why there should be an hard cap? what's the point?
there is no token limit but giving it a long prompts isn't recommended in SDXL
simpler prompt works better in my opinion
Not sure if it does on it's own, but it can use use ComfyUI or A1111 as backends so I would assume that it would function like them if you use one of those backend options.
simpler prompts means less attention to personal detail
and more reliance on random pretty
ah cool
StableSwarm is pretty useless though unless you have a multi-GPU setup you want to use.
What I mean by simple is reducing the usage of redundant words in positive and negative. I've seen people still using something like "masterpiece, highly detailed, best details" and "bad hands, disformed, ugly, blablabla"
tickles mushroom
in practice for anime based model it works since its based on NAI leaks, but for SDXL isnt going to change alot
I was under the impression it could run like auto and comfy at the same time
guess not
The reason that the 75 token limit sometimes exists is because CLIP processes exactly 75 tokens at a time. We get around that limitation by processing 75 token batches and averaging the results to create a single text embedding that is 75 tokens long that can then be handed to the to UNET to do the magic denoising stuff. The standard Stable Diffusion Pipeline and Diffusers Pipelines do not do this on their own, it is a feature that has to be added by a developer.
stableswarm for some reason wants to start autos webui.bat instead of webui-user.bat
found an highres fix workload for comfyui, cool
without it, that's what i get for now
yeah, the 2 have stable as a base
yes I wanted them running simultaneously
the face still looks like... well... derpy, but look at the resolution
done in 5 min
4096x4096
Why not fix face?
still don't know how to do it with comfyui lol
maybe i can set a face lora and an inpaint
make a mask, use inpaint model on it I guess?
Ive been trying all day to get my model in my image to face the f*in camera~
Could i get a keyword suggestion or two 😐
like forward facing, you can see the full front
"looking at viewer"
"looking at camera"
is she gesturing a bj?
"detailed eyes" or something descriptive about the eyes I tend to get full face shots
huh ty ty
i hope not lol
well, I mean, shes looking at me now, her bodys still turned tho
then your prompt must contain something that makes it think thats what you wrote
so much funny this comfyui, it mix my love for art, curiosity about ai and passion for building things
everyone in movies no matter how poor, always seem to afford a graveyard spot and a coffin and a priest and a funeral ceremony
🙏
you can try "from front" as well
Got the shot I wanted ! ❤️ Thank you
"I just want a small one.. really"
how small
wonder if I can find a very cheap 3060 second hand and use it for LLMs or secondary gpu in swarm
heh funny people charging 300-400 euro for 2nd hand 3060 while 3090 is 800
how long does it take to set up a runpod or something for training?
I assume it runs everything on ram that is temporarily persistent
and then clears itself once a user logs off
goes to bed.. thud
he looks too kind, make him angrier, more fat and psychotic
he is in peace now that he found an anime waifu
This is very cool! How do you do that on stable diffusion?
i used loopback scaler for those ones and a modified prompt template called chant-fancyfiremagic
chant-fancyfiremagic what that?
i found it in the autocomplete list after typing "<" it shows the loras and stuff u have installed, a few of them are listed as a chant and when u select it its like a whole preset prompt that pops up
theres a demo-chants.json file containing a few so i picked one and changed words to fit the subject i wanted
{
"name": "Fancy-FireMagic",
"terms": "Fire, Magic, Fancy",
"content": "(extremely detailed CG unity 8k wallpaper), (masterpiece), (best quality), (ultra-detailed), (best illustration),(best shadow), (an extremely delicate and beautiful), dynamic angle, floating, fine detail, (bloom), (shine), glinting stars, classic, (painting), (sketch),\n\na girl, solo, bare shoulders, flat_chest, diamond and glaring eyes, beautiful detailed cold face, very long blue and sliver hair, floating black feathers, wavy hair, extremely delicate and beautiful girls, beautiful detailed eyes, glowing eyes,\n\npalace, the best building, ((Fire butterflies, Flying sparks, Flames))",
"color": 5
},
{
"name": "Fancy-WaterMagic",
"terms": "Water, Magic, Fancy",
"content": "(extremely detailed CG unity 8k wallpaper), (masterpiece), (best quality), (ultra-detailed), (best illustration),(best shadow), (an extremely delicate and beautiful), classic, dynamic angle, floating, fine detail, Depth of field, classic, (painting), (sketch), (bloom), (shine), glinting stars,\n\na girl, solo, bare shoulders, flat chest, diamond and glaring eyes, beautiful detailed cold face, very long blue and sliver hair, floating black feathers, wavy hair, extremely delicate and beautiful girls, beautiful detailed eyes, glowing eyes,\n\nriver, (forest),palace, (fairyland,feather,flowers, nature),(sunlight),Hazy fog, mist",
"color": 5
}
i tried it after giving up on my custom prompt that wasnt giving me any good results anymore
it looks like making your own TI
Hey, why my negative prompt removed? Can anyone help me?
prompt:woman, wearing full crystal armour(weigthing:1.0), translucent, on top of building, shot from below, edge lighting, cool tone, cinematic, colorful style:Anime negative_prompt:nudity,skin
this is very interesting https://comfyanonymous.github.io/ComfyUI_examples/gligen/
removed from where?
yeah looks like the manual way of the same kind of thing that Gyre AI Photoshop Plugin has in 3d editing
My prompt was like bla bla bla style:anime negative_prompt:nudity,skin
When i press enter, the negative prompt is gone
it's gold for me that i love to use regional prompt on auto
so it's in your webui?
No, this discord
oh then its because you wrote nudity
the bot here is already censored not to do nudity
Oh lol
oh, nsfw words are automatically censored
I tried skin too, but doesnt work
yeah but just skin doesn't matter anyway
it removes the word skin?
oh ok, so, you can't negate some skin colors
but thats stupid
Thats hard iguess? Haha
not that is useful in most cases anyway
Want to try my prompt?
what if you want to make amazon warriors and don't want pasty white? Or a Viking and you don't want black? or Japanese Samurai and you don't want an Arab? Or Persian Warrior and you don't want a Japanese? Or African Zulu Warrior and you don't want Spaniards
etc
you gotta be able to control skin color
I want to remove that thigh ..it keeps coming after redo lol
well, you can set a color in positive prompting
I already put full armour
negating a color is just when the model is very adamant in using a different one
pretty cool armor
ffs some of these downloads are impossible to deal with they keep stopping and I have to restart them
civitai?
I cantttt…remove the thigh skin hahaha..frustating, even fulll body armour doesnt work
that's because models that are over tuned to only one style suck ass in terms of following the prompt
I get so annoyed when people gush all over midjourney and have no idea SD even exists or what it is, it makes MJ feel like normie stuff and me like a nerd :<
highschool memories come back
(everybody wanting my help with computers...I had grown fairly tall and wide, so I guess that saved me from bullies)
but elementary school was a nightmare
I feel so nerd every time I have to tell someone that they can use google
plz, don't remind me about school, i had gangsters in class with me
it was hard
what is stable diffusion? a new brand of climatization?
Gangsters kinda stayed to one school
where I live
I didn't go to that school
cmooon merge, get working
they were so unconfortable
adding regional prompting, slows down the gen by 4 times 😦
butter using only with controlnet and img2img
Hello very nice anime.
Thanks
Hello, the anime character is very very cute
Those are very very nice anime pictures.
This is a very nice anime.
think she can do a kickflip on that broom?
This took a WHILE
Little bit more work to integrate the focus character with the background better and also fix her hair.
may I ask how this was created
can you tell me your workflow
Workflow was just a simple prompt
1girl, chibi, hair bow, looking afar, outdoors, (darkness:1.2), night sky, crescent moon, cumulonimbus cloud, <lyco:LoconLoraOffsetNoise_locon0501:1>
on my Kakigori_v2 model. Used Hires fix for 2x upscaling from a starting resolution of 1920x544 with an adetailer pass to further improve the face. Full details are in the PNG Info.
I am a complete newbie, can I dm you? can I try what you described on the web?
Sure, you can DM me
dso good
my power bill was over 200$ this month cries in floridian
Have you tried stealing it from others?
Try $500 here in TX.
whys things look kinda mushy and blurry
or is it just bad at making hello kittys
more samples or something?
idk use Ultimate upscale it tends to work
i had an issue with that + tiled diffusion
whats the Denoising at?
usually around .32-.44
i have multiple upscale and loopback scripts ill try around and see if i get better results
I tried with larger tiles i think thats the problem you are runing into
128 by 96 seems a bit small i think anything from 400-800 would work better for what you are making
just imo hope this helps
ty
i dunno, is that trained model / LoRA or ?
dont see any of the usual flaws with her fingers or shoes
I mean, Im using a model with lora, yes
well, i think it's probably the model, and also how much time you are willing to keep generating before finding a good seed

nice but i love the first one on top in black tshirt
i tried loading it on a1111 but it has no meta data to load
here you go with metadata
sorry, i'm always too lazy to copy from folder, but you can just ask me for any image for my metadata 
ahh thanks xD
interesting eye colors
fingers are flawless
just "heterochromia" and let SD do it's work 
a Korg Taktile49, not the Triton one. But the Triton one works weird cause the synth is a digital out box, it won't send information in
I tried heterochromia once long ago and what it did was remove the pupils
@lyric zealot im curious to know, do you use A1111 or something else? cause this is how your meta data comes up with line breaks
or are you making those line breaks to keep the prompt neatly organized?
yup, i like to keep it neat
oh clever, i love it
so the first line is usually my basic prompt, the second line is about the subject and background
last will be lora
this will be my new practice with prompting
I read about some of the community model devs at civitai talk about not using masterpiece, best quality as prompts, they claim those prompts don't really affect the image
well, i think it really depends on the model, but probably yes
true i think models can make a difference
ugh I hope AMD GPUs become easier to use
actually not much differences
positive:
(masterpiece, best quality:1.2), detailed background, (hyper-realistic:1.4), mature female, full body, 1girl, white hair, multicolored hair, long hair, high ponytail, red eyes, shirt, skirt, alley, graffiti,
negative:
bad anatomy, bad proportions, (disfigured, deformed, mutilated, malformed, mangled:1.2), long neck, (fused fingers, interlocked fingers:1.2), asymmetric, conjoined, distorted, duplicate, cloned, out of focus, cropped, (low quality, worst quality:2), oversaturated, lowres, low contrast, (greyscale, monochrome:1.1), (underexposed, overexposed:1.1), (grainy, pixelated:1.1), (text, logo, title, watermark:1.1), jpeg artifacts, multiple views, morbid
after that anime pic you shared but I used rev animated model
__nsp/nsp-pop/pop-culture__
is fun
batman,blade runner, planet of apes
darkwing duck lol
sneaks in
slight changes in the propmpt
1girl, full body, white hair, multicolored hair, long hair, high ponytail, red eyes, shirt, skirt, alley, graffiti, (highly detailed:1.4), (beautiful, aesthetic:1.2), vibrant colors, ultra quality, HDR, cinematic lighting, soft focus, dynamic angle
the descriptors at the end
the entire line after graffiti
btw some of the models dont recognize -> full body
at least not with just bare prompt, weight might change that
without weight full body
and with weight (full body:1.3)
yay for multicolored hair in this one
think about it ...
this prompt is so simple and sweet it shows in the images
1girl, full body, white hair, multicolored hair, long hair, high ponytail, red eyes, shirt, skirt, alley, graffiti
this is what I got with SDXL base 30 step, refiner 5 steps, cfg 8
honestly, inadequate
1girl, full body, white hair, multicolored hair, long hair, high ponytail, red eyes, shirt, skirt, alley, graffiti, (highly detailed:1.4), (beautiful, aesthetic:1.2), vibrant colors, ultra quality, HDR, cinematic lighting, soft focus, dynamic angle, (anime style:1.1)
im gonna increase the weight on full body and anime style tags and see what i get
what would be suggestable sampling steps in SDXL?
puts nsfw in negitives... yep can't share this one. oh boy
but dam she thick
thicc?
i'll go for 30 steps with base and 10 more steps with refiner
is that SDXL?
no
NOT SDXL but damn
yeah don't knock it
you can make some damn gorgeous stuff on 1.5
agree, 1.5 still rocks albeit few blemishes
redoes the prompt
like them eyes
SDXL PC requirements?
not very well for me
think you need 8gb vram as standards
the third one seems to have a fire in her brain
but very customizeable
😄
ok that went nuts
first lesson today my friends @bitter latch @nova remnant - learn how to wield JRPG weapons
cool 😉
I always thought it was funny how totally unfeasible those weapons are
and clumsy
yetwielded with perfection
scythe is a great weapon tho xD
when it isnt the size of a car
that looks nice, the creative part about prompting is when you bring about a perspective and pose 🙂
thank you, i love it aswell, it was suprising me that it got this way i love it
the final one makes sense though
looks epic
one last one more simple then i'm gonna go crash
@bitter latch do glitch-art
remember tha weapon is not a weapon, its a symbol xD
these are coming out different due to the model : cocotifamix model
yeah iv'e done glitch art
makes things interesting
that's lot of artifact
this one used glitch art in prompt
yeah not a real good model
way less artifact this time
different model I think
ah
here's some i've done last few days https://www.randomtora.com/new-improved
ok maybe the last week or so
I'm thinking of some prompt that's good for robot/android rn
hav fun i'm gonna go sleep... thud
around hre it's the damn black ants. they bad this year
RandomTora might wake up tomorrow as consumed
they woudl starve
xD
lol look at this cute life like chibi .. taken from civitai
cursed
they taking ova the wurld
organic is weak machine rule
death to the organic
you and what army gonna defend them?
as machine hive mind march through and exterminated them
why do you even want replace organic girls with metal scraps?
that can be a thing naturally
you can do that with organic too, just gotta have innovation
^
nah flesh is weak
no never
why use mere words use art to fight machine then




