#π₯ο½anime
1 messages Β· Page 50 of 1
Yes
I'll give it a try
anyways, using tiled vae using the settings here should work
don't use medvram
i have 2GB less than you and i do 1024 dimensions already
oh 6
Should I delete it completely?
--medvram --disable-nan-check --autolaunch --no-half --no-half-vae --opt-sub-quad-attention
My args
medvram makes it like 5x slower for nearly no gain
--no-half --xformers --opt-split-attention --autolaunch
is what i personally use
If you have NVIDIA it's a totally different story
no it's not
oh right u have AMD
your gpu is literally better than mine
removee --no-half and try
mori calliope?
the mal prompt π
which one is better?
just some random prompting 
LOL
the second is a little more detailed
detailed? or noisy?
It's really fast, but result is terrible, worse than the source
trees are detailed on 2nd
but looks exactly the same
no
guys whats the best prompt of making the subject looks away and not the viewer i tried 'looking away' and 'facing away' some of them works but most of them still looking at viewer, i tried putting this to negative prompts too, tia
if you looks the first, the colors are a little more spread, in the second, are more sharp
1 is blotchier, 2 is smudgier
shouldnt affect results tho,
and should be slower, maybe show us result
one reason can be faster decoder
are you sure you have the same settings
dunno xD
so which one better?
i like 1 better
are just different,
1 - animesharp
2- ultrasharp
2 seems to have should've had a couple of extra denoise strength to get right (upscaling?)
like an image with antialiasing and not
Yes I think
that's the best comparison i can find
it's not like that really, cause antialiasing is better
disable tiled diffusion tho, unless u want to upscale
When I set upscale factor to 2 I got a black square as a result
I do upscale
some people just don't like antialiasing
he wants to upscale
@wispy canopy I'm trying to learn what language it understands from a storyteller
then I can prompt more
did someone try this, seems working well
like, heavily, (Extra details that even when u zoom in, u'll still see details)
is good to use, just in some occasions
what are those???
else, i find it not worth it to use it
upscaler
a X4 upscale for example
often better to upscale 2x2 twice
yeah I know bruh
disable fast encoder if you haven't
where do i place the animesharp? in model\esrgan?
good
My bad, forgot to set an upscaler, anyway, when I choose one, I get out of memory error
i used to set like this, what value u guys prefered to
i set it on 6 or 7
not what I wanted but still cool
i prefer from 5 to 7
how can i make the subject looks away, i tried any methods it keeps looking at me ffs

I did, yes, but it dies with out of memory error
depends on upscaler,
latent just use between 0.5 - 0.6
0.7 will make a new picture out of it
what upscaler do you usualy use?
for hires fix 0.5-0.7 is good
I use 0.6
u are using img2img right ?
source image resolution
latent
what model is it? 
what model are you using?
v2??????????????
since when?
chilloutmix
what
MultiDiffusion Sampling: 0%| | 0/22 [00:00<?, ?it/s]MultiDiffusion hooked into DPM++ 2M Karras sampler. Tile size: 96x96, Tile batches: 2, Batch size: 1
MultiDiffusion Sampling: 0%| | 0/22 [00:59<?, ?it/s]
[Tiled VAE]: input_size: torch.Size([1, 3, 768, 1152]), tile_size: 512, padding: 32
[Tiled VAE]: split to 2x3 = 6 tiles. Optimal tile size 384x352, original tile size 512x512
[Tiled VAE]: Executing Encoder Task Queue: 100%|βββββββββββββββββββββββββββββββββββββ| 546/546 [00:06<00:00, 78.05it/s]
[Tiled VAE]: Done in 7.180sder Task Queue: 96%|βββββββββββββββββββββββββββββββββββ | 525/546 [00:06<00:00, 252.48it/s]
0%| | 0/11 [00:03<?, ?it/s]
...
RuntimeError: Could not allocate tensor with 603979776 bytes. There is not enough GPU video memory available!
that's Darkgemini, the one cryonicus made
idk its just appeared there
when?
give me the prompt you are using
It's just 384x352 and it dies 
oh I known that model but didn't know that
april 24
bruh
i'm hating the fact that his model does just as good, or even better on story prompting than mine π
Yes.
768x512
should i dm it?
his model is great
k
I told you AMD is a different story, this shit can't do neural things
unlike mine, not an accident
I'm making a model for background scene and so impress with that image 
plss accep thanks
most models already can... you just have to prompt it
didn't notice that wow
are u using 6 upscalers same time ? xD
most of the time, waifus always chased me
No, I said I forgot to choose
story prompting is very strong vs waifus
this is on my hentai model. how many waifus do you see?
if u got a browser running or something it might suck up some VRAM, so be careful about running programs
also remove the --no-half flag
and how you guys make a dark version for model, you embed a noise-offset loRA in that?
i use the dark locon, when needed
how to do that, does the checkpoint merger in auto1111 can do that?

a bought a new 1TB one and just 2 months later...
No, only Chrome with one tab of the UI and Discord
Anyway, this works with --medvram, but I got this
As I said before
Just a black\gray square
it works without tiled diffusion ?
that's why you chagne your browser to run in pure software, no more vram hogging π
Yes SD upscale script
Wym
yeah disabling hardware acceleration is a bless
1, sky-scraper-chan
Exactly the same, only one is sliiiiiiightly darker
hello all
first one looks washed out compared to second one
I mean, It works when I use SD upscale script not tiled VAE or tiled diffusion
Mefistic told me that tiled vae is good and will speed up the process, but I tried it once, and get the same results of black and gray square. He suggested me to change some settings and delete --medvram, and I got out of memory error, I add --medvram again and it worked but with the same result of black and gray square.
I'm telling you guys, AMD is a different story, it doesn't matter if it's better in games, even 1660 with 6GB of vram is better than RX6600XT with 8GB of vram (RTX3060 analog)
4x_NMKD-Siax_200k best
did you see the comparison
this one
tiled vae is slower, but uses less vram
Yeah, and gives an empty result on AMD 
The picture on the right gives me a darker feel, the left side is a little light @brittle jay
Also, try to use tiled vae, with no tiled diffusion enabled
which one you like more?
Both are good, it's just a feeling. So if I use them, I use both pictures
first is 200k
siax
second is YandereNeoXL_200k
I'm trying to upscale, what is the point then? It works just fine without those "tiled" extensions.
indeed
to know if the problem is coming from tiled vae ...
It's a tough question, if there are more examples of images with different subjects I will consider.
sd upscale script is good by itself
feel free to try it yourself
Tried on 3 of my merges...
Can't escape anime girls
this is a good prompt β€οΈ
I'm using Ultimate SD Upscale with Tile controlnet to upscale, but it says "could not find upscaler named SwinIR_4x, using None as a fallback" even though I selected ESRGAN 4x anime
indeed
morn'
Try a different
Swinr is weird
mornin
I did, it seems to not recognize my selection.
Strange unsure why
found a model that recommends this, what is it exactly?
An extension
detects faces, fixes them
if a model recommends ddetailer, then it must suck at generating faces
I run a lot of generations across many models with x/y/z plot. I like seeing how the models handle things differently. Anyone else do this? π
really?
XD
Havent even tried it
heyoo
add-detail at -1 or 0.5, who did it better? (ignore the hands xD)
It works without Tiled Diffusion.
So, as far as I can see, it just does tile calculations for you.
I usually do it by myself, like source (after one upscale) 1152x768 scale factor 1.5 so
768*1.5=1152
Calculating tiles not more than 0.4 MP 640x640 (my GPU vram limit)
1728\3=576 and +64 (tile overlap) = 640
1152\2=576 and +64 (tile overlap) = 640
So one tile will be exactly 640x640, no more than 0.4MP and it's fine
It will be 6 tiles```
Tiled vae does this
``` upscaling will process a total of 9 images tiled as 3x3 per upscale in a total of 9 batches.[02:45<00:00, 1.15s/it]
[Tiled VAE]: input_size: torch.Size([1, 3, 512, 768]), tile_size: 640, padding: 32
[Tiled VAE]: split to 1x2 = 2 tiles. Optimal tile size 352x448, original tile size 640x640```
9 tiles which is worse, longer and worse result.
Why worse result? Because program can see less information on each tile, which leads to worse results, ghosting etc.
she was just ensuring the library was 'quiet'
I'm trying to get story prompts to work but Im missing something and Im not using an LLM
I really don't want to
what vae u guys usually use?
@cyan burrow got somethign you can share, i can take a look if you want
I haven't tried Tiled Diffusion yet. That's the controlnet thing?
It's definitely a quiet library 
nobody alive there to make any noise 
a tiled vae shouldnt give different results tho, or thats what i think
unless I learn what to do or change, it won't matter. So unless that's what you're offering, I appreciate it but no thank you
oof... tough learning curve, i hope you'll get it down β€οΈ
looking forward to it
can you give png info pls
close enough, but I never expected a third arm to come from there
almost sure all my pic should have pnginfo
cock hand
can you test and let me know if there's pnginfo
there is none
that's why I asked

1girl, pink hair, bangs, high ponytail, sparkling yellow eyes, round glasses, library, dark, horror \(theme\), bloody, blood on face, blood on clothes,
evil smile,
<lora:scuffedHorror-000008:0.5>
Negative prompt: (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000, censored, monochrome, greyscale,
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 22, Seed: 2976035269, Size: 512x768, Model hash: 1b43df1916, Denoising strength: 0.53, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Dynamic thresholding enabled: True, Mimic scale: 7, Threshold percentile: 96, Mimic mode: Half Cosine Up, Mimic scale minimum: 4, CFG mode: Half Cosine Up, CFG scale minimum: 4
Template: detailed background, best quality, masterpiece, cinematic lighting, highly detailed, high res, absurdres,
1girl, pink hair, bangs, high ponytail, sparkling yellow eyes, round glasses, library, dark, horror \(theme\), bloody, blood on face, blood on clothes,
evil smile,
<lora:scuffedHorror-000008:0.5>
Negative Template: (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000, censored, monochrome, greyscale,```
THIS IS AWESO... for all the wrong reasons xD
from https://www.pixiv.net/en/artworks/107772764
low and high denoising
It does, because of more tiles. I explained why.
First upscales should be with as less tiles as possible. More information on each tile = better. Because first steps you do with the same prompt, and ~0.4 denoise, it will give more details and better results.
Later you delete everything except "masterpiece, ultra-detailed etc." and at this stare, amount of tiles doesn't really afftect quality, it will be sharper, more colorful and stuff.
scuffedHorror?
from holokuki
lower denoising very nice
did you use same thing for this image?
nope, that was with the darkkukiv1
the dark lora for holokuki
oh wow
boil your chibi day
link?
Both aesthetically pleasing
1 is my favorite
NOOOOO
I'm gonna slowly work on getting lora's done for eyes, hairstyles, weapons, armor, poses, advanced poses, comical expressions (you know the exaggerated anime ones)
refined lineart is pleasing on the eyes β€οΈ
Exact value? And what model?
And what did you do? Inpaint sketch?
One of my fav styles
i made a little experiment, even used the lowest tiled vae encoder size possible
[Original | upscale + tiled vae disabled | upscale + tiled vae enabled with 256 encoder tile size]
fuck, i think my LLM throws out better movie concepts than some movies i've seen cough 50 shades cough
I find 1024x64 best for my needs
@wispy canopy fountain pen art ftw
50 shades is good wdym 
I'd rather watch that worse one on Netflix 364 days
where they do nothing but......for those days, it's horrible
says someone who watches/reads isekai (anime fast food trash equivalent)
"I'm a billnionaire, I've kidnapped you but don't worry. I'm rich and sexy and will you give whatever you want. Good yes?"
"OMG YES"
I liked it 
i've hardly enjoy a movie nowadays... but i loved alita... i was ready to be dissapointed only to be really enchanted by it
these eyes are perfect damn
Alita was nice
i should give mars chronicles a new try...
Alita was great. Saw it in 3d
On my vr lol
They should remake John Carter and actually give it good advertising this time. Stick to it's more mature themes and not turn it into a pseudo-disney adventure
pnginfo?
yes please. that one was pretty good, but too disney-fied
gonna go lie down awhile
detailed background, best quality, masterpiece, cinematic lighting, highly detailed, high res, absurdres, (chibi:1.3),
1girl, pink hair, bangs, high ponytail, sparkling yellow eyes, cute and adorable, sitting in boiling water, (in cauldron:1.2),
Negative prompt: (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000, censored, monochrome, greyscale,
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 22, Seed: 3630733727, Size: 512x768, Model hash: 1b43df1916, Denoising strength: 0.53, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Dynamic thresholding enabled: True, Mimic scale: 7, Threshold percentile: 96, Mimic mode: Half Cosine Up, Mimic scale minimum: 4, CFG mode: Half Cosine Up, CFG scale minimum: 4
Template: detailed background, best quality, masterpiece, cinematic lighting, highly detailed, high res, absurdres, (chibi:1.3),
1girl, pink hair, bangs, high ponytail, sparkling yellow eyes, cute and adorable, sitting in boiling water, (in cauldron:1.2),
Negative Template: (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000, censored, monochrome, greyscale,```
hey, the rev animated model recommends bad-quality embedding but its the only embedding that isnt working for me
ive tried all the recommended and all work except the bad-quality one
what is template and negative template?
i think you can just ignore that, it's probably from some extension
not even you know? lol
most important steps in in the world. insert "but why" gif
yup, i have no idea, i was thinking it's from latent couple or tilted diffusion, but both of those isn't enabled

it's not from tiled diffusion for sure
no idea then
but tbf, that's from image browser extension, so might be just image browser being stupid
that's a really good gen for something without a negative prompt
https://civitai.com/models/56519/negativehand-negative-embedding
put this on webui/embeddings
negative_hand Negative Embedding Problem with Negative Embedding's Currently, there are more and more negative embedding, while many are also very ...
and add negative_hand-neg, (worst quality, bad quality, low quality, normal quality:1.5) to the negative prompt
that's like the bare minimum
Where is that
stable-diffusion-webui\embeddings
this embedding improves hands by 5x
Okay, nothing changed. What this supposed to do?
did you put the file on the folder?
Yep
make this your negative prompt
That's all?
Ah, so adding this negative_hand-neg will... Will do something. That's the way of usage.
that is an textual inversion embedding
it was trained to improve hands
it is trained on a ton of bad hands so when you use it in negative it makes hands better
Do you suggest to use KHFB, AuroraNegative as well? I see you added it.
they good yes
they are made for aurora
but work well on other models as well
the thing is, those are made specifically for the model i'm using
though the models should share as much with aurora
do they?
depends how similar the models are
though it should work on most theoretically speaking
I use AbyssOrangeMix3, is there any basic prompt like "masterpiece, ultra-detailed...."
(detailed background, best quality, masterpiece, ultra-detailed:1.5) is what i used
before i got into aurora, which doesn't like this stuff
Do () work without :x.x ?
yes, equivalent to :1.1
quick question, can you run gpt4-x-alpaca-13b-native-4bit-128g on 8gb of ram?
quick question, can you run gpt4-x-alpaca-13b-native-4bit-128g on 8gb of ram?
I think
how, its 13b, are there any settings?
haha
u can select chat mode
its better than writing a whole book to ask a question
Ask ur AI if u should switch to chat mode
chat or cai_chat?
cai chat for pictures
and the name of the bot etc
chat for normal, simple chat
havent used the webui in a month or so, maybe they changed stuff
wee hires x3
Time taken: 6m 25.54s
Torch active/reserved: 1830/2880 MiB, Sys VRAM: 6578/8176 MiB (80.45%)
not bad too now that tiled vae is actually using cpu
so many upscale methods i haven't tried, gotta get on it
Haven't tried ultimate upscale and I feel like that's been around forever now.
We got this new tile upscale thing I need to look into. I watched a couple videos but they were all like 20 minutes long and full of filler... need something that distills down just the information without all the editorializing.
when you find the better ones, share them with us
well I use low denoise hires/img2img a lot, image merging, SD upscale, and they all work quite well

you gotta be pranking me!
MPT-7B-StoryWriter is revolutionary new LLM model that BEATS GPT-4 with an INSANE 65K+ token limit! These new MPT-7B model sets were created from scratch by MosaicML in only 9,5 days and 200k$ for models that can rival the LLAMA 7B model! Impressive! So in this video, let's have a look at these new game-changing LLM, how to install them on your ...
reisen udongein (no lora)
Hi
funny how 6 months ago we had shit, imagine what we have in 6 months from now π
I'm trying to upscale many images but it doesn't do them in order, does anyone know why?
how do you upscale many images?
adulthood in japan is not 21yo?
hmmmm 1000/100mbs internet connection 60 euro, 500/100, 45 euro
eface, that's dark souls?
or 1000/1000 80 euro
100mbps is enough for streaming
500 is also enough
1000/1000 is overkill for most user things
yeah, they are enough, but with so much upload, you can do pretty neat things
maybe if someone wants me to host a server or two for generations
they pay my electricity, I pay my ISP bill, I get to use them too
it's uh... somethign my llm dreamed up
i just asked for an underground figure
i just had an idea, making drawings with stable diffusion and than trying to make action figures out of them
well, it didn't actually follow my prompt at all, but me being me, don't care, got nice image β€οΈ
wtf not gay
you got a gay result?
when u get things u dont want to but it's pretty and you leave it be π
what's the unwanted thing, except k-size breasts?
sitting on the water for example
Exactly that, didnt want humans at all xD
no_humans tag works well
overtuned model. embrace alfahentai, the hentai model that's ... not interesting to do hentai with lol
'hentai model'
hentai indeed
just as hentai as OG nai was
i have so many models developed for very shameful reasons that i use to make artworks
forms are just it, forms
all the loli ones
all the porn ones
although, i suppose the heavy lifting is being done by gpt4-x-vicuna
new anime movie

damn
new gundam
there are gore once out there that put in shame saw the film
woow cool
but they still show very good curvy and fluid forms
that's a good example
i love how it shows the horrors of the war
but it need more black smoke and it's perfect
@steady grail sorry, only got the time of a singl esong to get the llm to the work and then submit it to SD lol π
(those images invoke some pretty strong feelings in me though)
Anyone know?
oh in extras tab
death stranding feeling
my god
i want so much to do something, but i'm eliminating obsolete models from my collection
the hd is exploding of models
so charming, looks like limbus
eface, which promp do you use for these images?
i love these lights
the last one: An image of a person sitting alone in a forest clearing, surrounded by tall trees. They appear lost in thought, with a solemn expression on their face. Behind them, there is a glimpse of a fire, hinting at possible survival or struggle. The colors are muted, with shades of green and brown dominating the landscape. There is a feeling of melancholy and longing in the air, as if the person is searching for something that seems just out of reach.
it's 112 tokens, so the prompt reset is a thing(as always)
so, it's pretty random
but, my bot seems to do quite fine -> it made sure the prompt reset didn't happen on a token that might fuck it up
not quite random at all
π
eheheeh
man it is so fun that I have the ability to rewatch shows as if it's for the first time
feeding my imagebot lyrics -> the output style is tuned to the stuff i know my model likes
it's called Alzheimer
you know them?
you could ask them to quiet down if you do
hehe
try with full moon of sonata arctica
so I downloaded bunch of new models, so far this one is pretty good https://civitai.com/models/14734/store-bought-gyoza-mix
Look what I got
no I don't
Not sure I can even walk the distance to where they are
are you in an apartment building or a house?
apartment
idk then
they're in the stairwell next to this one and up somewhere vertical
you know cryonicus' predicament... not much to lol about π¦
sorry
eface, where i can find this bot?
it's on my own computer
if you can load it, using the oobabooga web ui with gpt4-x-vicuna
you need at least 10gb vram to be able to make it work
too much for me π¦
If I can find a good 3090 second hand, I can sell my 3070 ti for like 400 euro I guess
used pc components are not worth it
mine is a used 3090
i bought is SPECIFICALLY for running LLM's
sometimes they are
a used 3090 is sure as hell worth it
just connect your a100 to internet
I wont get 24gb vram in a good card cheaper anywhere else
"your"
you must be lucky, pc component are set to die in less than 10 years
a bit. i can relate to the lol, but given the stuff cryonicus has been through in last couple of months, i'm pretty sure i'd not speak out the lol, even if i had it my mind
10 years long time
depend how much new is the component
almost got the other unit back up and running, that means some A100 will probably come available, so i can use those again π
lol
and the price
is it confirmed?
I doubt PC components will die, the motherboard and cpu might
gpus get worse over time depending on how much you use them
that's why cooling is VERY improtant
yeah, pretty much, is an estimation based on abuse gaming pc
but GPUs exist far too many versatile uses to kill them
I have a PC which I bought in 2008, still works (only PSU died, I replaced it)
keyword: estimation
I think he meant like....die, as in off the market
yeah, naturally
not literally
wait thats what you meant?
if my gpu gets activated, it will activate ALL the cooling in my pc -> having the case fans push and pull in cold and hot air, it keeps the gpu cooler
....
tell me more?
how does this work?
this mf probably saw a video from LTT or some shit and doesn't know what the fuck he's talking about
from a chemistry point of view please
having a full tower case helps a lot, people that buy med cases its a very bad idea for cooling
my mom took over my old PC, been running since 2010
how current works? the electricity move in the contacts, and this movement just ruin them
just like in a battery
you're not reattaching them, and if you handled it properly, you did not cause ESD (which is erosion indeed)
Haha I just want to know how many Chinese in here
How cars work for decades? You know, they have chips inside as well. And they work even in worse conditions. Yes, not the best example, but still.
cars are much less complicate
ESD is the only in-component erosion that should be happening inside your pc
they need much less precision
precision =/= prone to mechanical defect
component are much bigger and so much more resistant
this remains
for example, with the right insulation, electronics survive much longer
controlled obsolescence is a thing
Anyway, I guess my 15 year old PC must be not knowing about this, still running. I got lucky.
not when they are already afflicted wiht ESD
my last pc survived 6 years for example
my pc is like the ship of theseus
and i use them very much in a very exhausting way
you downloaded a virus or some shit?
no way in hell lmfao
i love to do things
that's def not the cause
natural language best prompting language!
i'm pretty meticulous against virus
there's not a chance in 1000000 that i get one free to do his s*it
there's not a chance of 1 in 10000 that you send a comprehensible english message
it may have been an overvoltage on my old pc that definitelly killed it
sorry, i'm not native eng, just deal with it

most of us aren't, don't worry π
poor power conditions are a surefire killer!
if you live in an area that voltage spikes or brownouts, it's smart to invenst in a surge protector even power regulator
yeah, now i have a power bank under my pc
HEY.... REGULAR ANIME!!!! π
that's very smart!
it's like holy water for me, after the last incident
currently installing vladmandic
hands π¦ otherwise such a good picture
automatic1111 fork that is allegedly faster
can you suggest a good cd anime model? something very flexible?
aurora
alfahentai lol
welll, at least, that's what i'm making all my images on
how come its the porn ones hmmmm
no idea, but that's what i found
hey, what settings did you use or softprompt/character or whatever
I always recommend my Dark Gemini 3, people love it
gpu likes hentai data, so it rewards you

it's not very good for hentai though
well, mine was a happy drunk accident, and the further tuned to get back more flexibility
in the llm?
in the webui
"as an avid image tester with weird prompts, i can also recommend darkgemini 3"
which webui, i'm using 2 xD
the sd prompt, or the llm character?
ogabooga, how did you do it so that it generates a prompt?
i created this character π best used with gpt4-x-vicuna, the best 13b model there is at this moment. if you can only run 7b models, i recommend wizard-7b
on 200 g of model, i only deleted one, and now i have 3 more models to test XD
what an endless battle
ai hoarder
where do you import that?
In oobabooga is a folder called characters
you can copy the yaml file there
then select the character
i need more ram and vram π¦
hello .. how i can generate new image ?
omg, only god knows how much i love to play with ais
is that a good output?
perhaps a bit long, which llm model are you using?
I really don't get it.
Why all my pictures so blurry.
Supel did 1.4 upscale
I tried too.
Attached 2 images, his (1st) and my (2nd) for comparison.
Look how sharp his image is, look how much details added, look at the face. And mine? Blurry af, and looks meh.
AM I DOING IT WRONG???
I put image to tagger and got a propt, added some details and negative prompt
detailed background, best quality, masterpiece, ultra-detailed:1.5, illustration), 1girl, solo, animal ears, long hair, blue eyes, blonde hair, smile, open mouth, red jacket, blush, looking at viewer, heart, dress, ahoge, long sleeves, animal ear fluff, jacket, bangs, cat ears, white dress, cowboy shot, open clothes, :d, arm up, bow, skirt, open jacket
Negative prompt: (negative_hand-neg:1.1), (worst quality, bad quality, low quality, normal quality:2.0), (long eyelashes, eyelashes:1.5), (border:1.5), (grayscale:1.2), (skimpy:1.5), (nsfw:1.5), text, watermark, signature
My settings:
Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1084413999, Size: 460x650, Model hash: 5493a0ec49, Model: abyssorangemix3AOM3_aom3a1b, Denoising strength: 0.45, Clip skip: 2, SD upscale overlap: 32, SD upscale upscaler: 4x-UltraSharp
wizard 7b
yeah, then it seems about right
you might have to regenerate a couple of times
the good 13b models shit on the 7b models in terms of interpretational knowledge
but require a vast amount of vram more
i really can't wait for the moment we can run 30b models on consumer hardware
i only have 8
you can try offloading with layers, so not everthing is put in vram, but it will make generations MUCH slower
like how long?
the next evolutionary chain, a half human with 6 fingers lol
you drop from 15it/sec to 2-3it/sec quickly
or even worse, 1it/sec
1it being one return token

Disturbed - The sound of silence
return to matrix
is that also good?
guys, can you try with full moon of sonata arctica?
not quite, i'll load up wizard 7b and see if it also fails this prompt
first gen with vladmandic lmfao
the prompt was pitch black sky vision?
i fails? it seemed kinda good
wow amazing waifu, you cant even see her
i'm not sure if it woudl work well as an SD prompt though
wizard 7b can do it pretty well, needs a little bit of help for context purposes. i'd cut of the final line of the response, but the rest is pretty useable
i didn't even use the same seed and the second image is identical. weird.
looks like a vae error or smth
stop using the token "(pitch black:3)" lol
how i can generate new image >?
you'd have to install some software on your computer to make it generate the images
is it free ?
from which site i can install it ?
you can even make your dreams come true
Youre mt favorite person
Sonata arctica is my number 1 band on spotify. Been with them since replica
yes
sure, lemme give it a whirl
fuck i cant get vlad to work
from which site ? can you send me a link or website name plz
they saved me during the school years
i was totally mind broken, they truly saved my heart
for more information i'd suggest you try look up in #1072229020520947753 and #1072220168534642768
"I have a right" had that effect on me
When that came out
yeah...β€οΈ
The band is just phenomenal
ok, last model
can someone help me get vlad to work
i get pitch black images
--no-half and --no-half-vae dont work
none of these
@steady grail Full Moon - Sonata Artica
4/4 good results β€οΈ
oh god
can you either help me with vlad or do haunted - laura les
@gritty patrol, do you see the images while they are being generated?
they're.. pitch black
it just captured the loneliness of the song
during generation i mean
hmm
so perfect
:/ got no idea

i....must....use, it!
second option then :D
wish i knew what was going on π¦ and no error output on your console either?
that captures the song theme so well too
let me check
this is why i'm SO fckn amazed with gpt4-x-vicuna 13b, i just fed it the lyrics and asked for a fitting image
try my song pls
was trying
i'm afraid the first description i got is not something SD can handle, or at least, my model can't
unsee
close enough?
did you use this
about this, no error
oof... you did try downloading an extra vae, just to be sure? (kl-f8-anime2)
wdym
i'm using kl-f8-anime2
blessed 0.9 contrast version
also, what's your response
and f1 broadcast starting in a minute -> gonna watch that one tho
first one is the girl on the mirror, new gens running
i asked if you used those lyrics
yeah, already posted single image
it missed some lyrics that are hard to hear in the music
true death covers his face with a veil, is dressed in white and rides a pale horse with no eyes
not exactly what i imagined, but pretty cool, thanks
i think the first gen is more accurate
first gen had a lot of borked results
the LLM interprets pretty well, but the concepts are hard as it goes for specific body positions (gues what SD sucks at...)
hello guys I want to ask is it legal to sell animation made by ai ?
by using img2img
sorry for bothering you dude what i must do after download the files ?
how i can start
just start with the .. uhh. start-webui.bat?
i'm only trying to install this thing because someone said it's faster and i'm a sucker for speed
this one ?
either vladmandic or A1111 with pytorch2, they're about the same speed
ah, in that case, webui-user.bat
ah, you'd first have to install python first. i really suggest you look up how to do the full install yourself, you're going to be tinkering with all of this a lot
If you're doing img2img based on a real artist's work, then you would need to familiarize yourself with laws regarding fair use or transformative art. It isn't as simple as a yes or no answer.
i give up
I use on an image I have its rights
like my drawings
or my models
your own drawing, should be legal, i suppose? then it just depends on the model
Then absolutely that's fine.
@supple raptor i want to talk with you in sensitive topic
maybe is not good to disccs it here
about ai
how do i know if my sd-webui uses pytorch2
guys, aymen have a relation with an ai!!!!
._. no but some servers hate ai
why are you using vlad
that's so sad
I wanted to ask someone to do for me a model using ai image reference
he said sorry I can't
i was told that it was faster
cuz ai is bad and like this
again
sooo I want to know is really ai is bad ?
its not lol, the code base is the same as a111 just with some patchs
when it opens I think it says
just do --reinstall-torch to be safe
no, it's not, ai are just instrument to refine some very laborious details
the rest are just a play thing
you know people say ai steals from artists
Is Dezgo the best when it comes to anime AI art?
ai is good enough to make original pieces
though without controlnets it is not the best
for commercial use at least (I mean when people pay you to make them art)
90 % of that is not refining is just very common
ai is really good but I don't know if I can use it for a bussines
you can, just learn a bit of blender and photoshop
like I want to sell animations made by stable
then use controlnets
oh animations
yeah no give up
I have blender models
art is inspiration, art is communication, art is expression
or wait like a year or 2
Or are there free online services that are better?
with ai you can do cute things, but nothing very inspiratory in it's true sense
it says here at the bottom
imagine I made a really good animation
don't care about quality
are good in a very superficial way
I mean about the rights
there is no such thing with ai yet
I know real artists is better
people have already taught AI how to be part of groups and they create clusters of AI that then create their own clusters from their needs. They are given a mission from the start and ΓΆara
âa¨¨l¨¨
oh about rights
parameters
you probably won't have much rights to it with current laws (usa)
unless you edit it majorly in some way
I think some had AIs running small companies dealing in selling chinese stuff
cheap chinese products
that's the point, we can mass produce anything, but the truth is that what give energy to the things, are the feelings we instill in what we do!
and even then it would be limited probably
you can't instill that energy with mass production
well the great thing is, since we dont own what AI does
if they make a hive mind and defraud for 100s of millions and hide the monet
money
it's not us
it's like dressing a very big shirt, it covers you, but it's not comfy
you mean I should like take time on it so it would be my work?
and even them someone might try to copy you and claim they are fine because you made it using ai
and then you'd have to go to court
it is like on case by case basis rly
lol
who is against the ai, is just a lazy boomer who want to speculate on tedious mundane work.
I mean internet is international
and most sites operate in usa (I think)
so you would have to go to usa's courts no?
idk I am too young to understand law that deeply
if they sue you, then they do it in your jurisdiction
is it better to work with 3d animation and leave ai ?
if you wanna counter-sue
since it's not acceptable
@gritty patrol Where to put hires.fix 4x_fatal_Anime_500000_G.pth
I'm installing AURORA ONE
Why would someone outside usa care about usa laws?
cause you don't have to ever respond to a civil suit and travel to the US for it
if you put your stuff on a platform that is hosted in usa I think you need to go by their laws
4x-ultrasharp is still better, keep using it
idk man
just get a lawyer fr
it is way too complicated rn
civil law is different from criminal law
and laws are still being made
i honestly prefer to use sketches as a base, for curing it, manual details, another cure and than the fine tuning with hands
for them to use criminal law on you they need to have you extradited or tried in your home court
oh so you do know how it works?
nice
AI do what I tell you to do!
so should I go in my bussines and sell animations using ai ?
I mean if you are the one suing
civil court isn't between a prosecutor and a defendant
if you sue, then you sue where the person you sue has themselves or their company located
where their primary location of taxation is
for now I say don't risk it. get a lawyer. and also ai is too bad currently
nice to know
okay I have a question
if you retouch ai, who can give you dirt?
also, no lawyer will ever help you sue someone who has no money
yeah, theyre not gonna get sued in China
can I make 3d animations with my own work ( animating and like this ) but I used a things for the model from the internet
can I sell this animation for a person?
does that look commercially viable to you?
or I should buy that thing so that I can sell the animation ?
if you have no money, no wealth, barely any property. You are pretty much a stone
not really good quality
you can't bleed a stone
for tiktok sure as hell yes
I mean if I am being honest. nobody probably cares, so you can go ahead and sell some on like fiverr or something
bruh
people are fr stupid
I should also respect that who made the thing for the model I use like a t shirt
?
you say I can ingore that and use that thing made by another person who shared it for free
exact ally, didnt say tiktok people were smart, they just consume stupid content, besides no one cares what you think, its what consumer wants, and it works
but I don't know if he doesn't mind commercial use
wee 2048x2048 hires
you mean you make an image with a model?
TikTok is the biggest "you are replaceable" sphere there is, the amount of volatile and toxic nature that breeds from that is revolting
My graphic card wouldn't take it lmao
Oh rly
mmmhmm
Why does it have 2 earholes tho π₯²
cos you touch yourself at night
as I said currently, you need to show that you made enough effort and changes to the image to get rights over it, and I think even then you would still have none over the base image (maybe if you used controlnets you could argue that in court or something)
The better to hear you with, my dear

also I am only talking about the usa
let's say I have animation
I used my models, but I wanted a hat to that modelm, so I downloaded from a person provided it for free
and then I sell the animation I made for a person
I don't know if the person aceepts that I use his hat in a animation and sell the animation
you know I can't sell the free things
but I'm going to sell the entire animation

I mean making 3d animation from zero
hat?
like this
yes a 3d model by example
are you made this video by using this ai ?
"the gap"
so you are saying you make a video based on a 3d model someone else made?
or just the 3d hat you put on your character in this case
no I make 3d animation but I used a 3d model by a person in that animation
3d animation in blender
ai is not using
when good is never truly enough
let's see how much i can push my pc before it explodes
do he mean don't use for an animation that you are going to sell it ?
where was it posted?
it's an asset in booth
booth?
this is in the description
guys, if i dc, my pc exploded
yes does he mean don't use my thing in a 3d animation that you are going to sell it ?
can you link pls
β οΈ Hi! Please make sure to read the product contents and terms of use carefully before downloading them! Thank you! πβ¨ I get this inspiration from: https://twitter.com/RavityStore/status/1633757466189078529?t=z0lSniFEJ1Nf5zQzPJekXw&s=19
huh
I don't know what dc means but how your pc got exploded ?
do you want to use this shirt in a profitable product?
ask them
not really, man, pc don't really explode lol
some do
I want to use it for a model
okay.
π΅ Why Buy an EliteGamingHQ PC -- https://youtu.be/MvnONrEIJKg
π΅ 10 Reasons why PC gaming is better then console -- https://youtu.be/KJ6-sTwsmyg
π΅ CPU Core VS Threads Explained -- https://youtu.be/EjIjvwLt76Y
π΅ Digital storm PC review -- https://youtu.be/PUn1-vCD5oE
π΅ CyberpowerPC review -- https://youtu.be/4ioYlwV11Uw
π΅ How to install a SSD a...
first google result
Ever drank out of AC water collection tank?
did you??
no, but I considered bottling it and giving it to my neighbor
AC Powered H20-EXTREME
yea, some resistances and capacitors did pop, but that's far from exploding
in the past they could explode, the big once
how do i use tilled diffusion to upscale?
you give up
i pushed it to the max, but the size is still the same
what
tiled vae
ohhh



