#🏞|general-with-images
1 messages · Page 45 of 1
I gen one image now and my card OOMs on me. close down and restart auto and can do one more
where do i find pose image?
by "pose image" he means the image you are using as an example for the pose
okay...
but controlnet pose model would also work
while i am trying this how do i obtain controlnet pose?
yeah i mean that's close but i would like her to have the weapon in the photo be the gun she's using
Yeah
anyone know how to get the imagebrowser in a1111 to show the prompt? this thing's been broken for a while for me
nope, I never upscale
stable diffusion ?
Yes
👍
this is her weapon
This is what I'm confused with, there is no way to "mash up" images that I understand yet
evenin'
did you mask out the weapon and then inverse img2img in that one?
yeah it broke for me too and I replaced it with the fork mentioned there which worked: https://github.com/yfszzx/stable-diffusion-webui-images-browser/issues/90#issuecomment-1514130713
there's also another extension that's called something like "infinite image browser" which embeds a small gallery file browser, but that one is much less alike
@icy dirge thanks so much
Looks great @dense tapir , haven't even tried food with my model yet
Try it
gotta load it up first
what'd you add?
It's getting kinda late and I have standardized testing tomorrow, but I really want to start working on v2 😭
actually it might be an EOC
I am not sure tbh
Although it's bio which is really easy for me, so I am not too worried.
I plan to revise the dataset, add some images, delete some images, and modify the captions to include specific tokens for either digital art or photography.
Said that earlier.
Which is why I said I am not sure what you will do for V2
if you want realism remove any anime/cartoons
No, I want both digital art and photorealism, which the model can already do both, but it's not captioned properly for that.
cinnamon roll!
whats even in this salad 😭
oh might have had "crunchy" still in the prompt from the taco image... woops
Just as I was starting to work on an AI project, another one of my siblings is in quarantine. So I can't access my desktop for another week.
😦
why with a1111 continuous preview does it show blurry or weird quality until the very very end?
that's at 96%
\
Yes, it's a feature to increase speed while generating. If it's rendering your preview at full res every x steps there will be a much larger slowdown. No preview is the fastest, so I don't use it.
I see. I thought it had to do the output anyway for the comparison to the prompt vector stuff
lots of new options there too.. neat.
(in Live Preview settings)
just what is wrong with BLIP
pew pew pew pew
it's just a woman standing in a church
church benches are called pews
over there imagining BLIP going MORE DAKKA when all its trying to do is give an accurate description 😉
hahah blip2 def is better about this. Blip 1 loves to do these repetitions. I rec the opt6.7 model tuned on coco caps if you are trying to captions a set. its default outputs are pretty decent

pew pew pew!
Now you need to get a stromtrooper in the church
Pewfection!
man
now add anime girl running from pew pews by stormtroopers
sigh
it just has to fuck up somewhere
now alien has bit better hands 🙂
That made me laugh 😆
luckily at least they arent flawless and you can avoid deepfakes more or less
and MJ too
didnt play around enough with photorealism for Firefly tho
i did, but not with trying to make actual looking photography
this native man wants to know your location
maybe its a transsexual 
that applies to pretty much any AI image
some more, some less
Negative prompt: watermark, text, plastic doll, render
Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2788102398, Size: 768x1024, Model hash: 063e9ed089, Model: SDArt_v2_768, Denoising strength: 0.45, Hires upscale: 1.5, Hires upscaler: R-ESRGAN 4x+```
not mad I think, just someone VERY passionate about SDXL
I tried to show the beauties of 1.X and 2.X too, to no avail 🙂 to each their own, and I have hopes for XL too, but I can't compare those
imagine the AI art race between Stable Diffusion, Adobe Firefly, Midjourney, Nvidia Picasso, DALL-E (3?) in a year or so
let's not attack people there though
don't start saying people are full of shit or silly
hey guys. if you want to keep on chatting on this animosity level, then you'll love the PMs. No need to go one after the other, the throat is not the goal here
I only see you going after him now, so let's block each other now if you need to go on
or move to PMs
but damn, limits, respect, not accusations and name calling, what the hell
he is expressing an opinion, not name calling anyone. you can have a diverging opinion, don't start attacking like that because you disagree
thats not an insult tbh
to tell the devs that the previous version may look better
but depends on how he says it
an "hahaha" in between for example may make a difference
thats mocking
we got a report system if you see attacks like this that would go against the rules. Use the ⚠️ emote, or Right click=> report to staff
but don't go rule breaking, name calling and accusing like this.
a realistic photo of an old black african man in traditional clothes
Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2927337387, Size: 768x1024, Model hash: 063e9ed089, Model: SDArt_v2_768, Denoising strength: 0.45, Hires upscale: 1.5, Hires upscaler: R-ESRGAN 4x+
Time taken: 23.28sTorch active/reserved: 6182/10136 MiB, Sys VRAM: 12479/24564 MiB (50.8%)
(base 2.1)
#🚀|pow-discussion message
finetuning on all the PoWs
let's get practical once again into moderation then, because it's continuing to escalate.
1/ @paper salmon stop attacking them. if you think they are genuinly making things up, block them and move on, or continue in PM if they wish to also
2/ same for peter, if you wish to continue this dispute, please move it to PMs.
Can we move on to picture sharing in this channel ?
Why not come into a sharing spirit, take the tips you still find useful in the general "flawed" pipeline they are using ? You don't need to be right on a random public channel. You've had multiple encounters with multiple peoples in the last few days that end like this, escalation until the point your messages get reported, or we intervene directly.
lets better share some images 😄
and stop those attacks. I already gave you an ultimatum before a time out earlier today
how about a cross platform 
i mean across SD, MJ, FF and co.
as you wish, can be both too 😄
skin is too smooth
plastic like
I'm trying to alter the prompt for it
prompt was "a portrait photo of a woman with white wool hair, black 20 year old woman"
sorry
yeah, 25-30 range
but the red head is too orange
my family has lots of red heads and this feels unatural
too vivid
i love to edit AI images with Photoshop
"synesthesia" token is strong :p
same prompt overwise
it's been so long I haven't made weed pics
my prompts are limited due to the software behind it compared to Stable Diffusion
a lot yeah. given the blue that you took for the eyes, some over saturation like this is OK.
Still for a more photorealistic, you can even point toward a little more "brass" like color
copper even
more than brass
it has lots of shades for sure, some more clear than others
no, family member
she's gone from a more vivid orangelike younger, to an almost brown at times, but always had those coper colors
That's a charming personality lol
love that one 🙂
a realistic photo of Autumn in a bottle
a realistic photo of Spring in a bottle
a realistic photo of Summer in a bottle
a realistic photo of Winter in a bottle
you know you can post multiple at once, right ?
I'll try some roads to infinity
less fractal maybe
Can your models generate mask? Like a theater mask or this draw that i made time ago, i test 1.5 and 2.1 but all looks like white skin
like their example
this is 2.1 and looks like white skin
a realistic photo of a man with a delicate mask of jewelry, someone
a digital drawing of a man with a carnival mask, someone
no need for SDXL beta to get impressive results 😉
That's made with realistic Vision V2.0
Yeah, it's going to be awesome in a year, or even just a few months.
can't wait for proper hands and fingers 😄
i like the service exist, but dont like, not sure how it describe... some weird smoothnes mostly seen on B&W images
sdxl
the model will be open source eventually
and eyes looking at same direction
I'm afraid you are right
so many good results I've had, only to be garbage because of hands or fingers. sometimes foots
again, fingers
anyway, it's super fun to generate this art.
I'm doing this just for two days. Still figuring out what prompts are good
but that looks damn impressive.
first day only with cpu, damn that was a waste of time 😄
even with 24 threads
what model do you use?
I never said that SDXL is not better, only that it is not needed to get impressive results.
well, that does not look very realistic 😄
I made that yesterday
now that I have set up my GPU for the render, I guess I'll get back to it and try to improve it
and yes, I know it does not look realstic, but that was not my goal anyway
I tried SDXL yesterday with some credits. but I would rather play a while "for free" on my pc. If I'm good at prompts I can still spend credits
lol on those 4
cracks me up
satisfactory's player know what's up
crystal flower 🙂
#🌠 род
Finally a Luna Lovegood that doesn’t look like a eldritch creature.
Gamers Are FORCING Nvidia To LOWER GPU PRICES
it's happening for Nvidia RTX 40 series GPUs like the 4070, 4070 Ti, and even AMD too!
Remember to Subscribe and smash that like button if you enjoy it!
Amazon Affiliate Links: (Only if they're cheap, prices fluctuate)
Nvidia RTX 3060 https://amzn.to/3gBfMTm
Nvidia RTX 3070 https://amzn.to/3GLp...
if an alien civilization is more advanced than us
out there somewhere
did they already maximize the potential of the transformative nature of reality, the property that gives rise to the ability for matter and energy to compute
some species out there must have reached a technological singularity, or we are the first ones to do it ?
The universe is vast and even using hyper drive speeds, or warp speeds, they are so far away that it could take over 1k years just to make the journey one way.
which I suspect we are all stuck with
theories
a theory is just some dude hoping, wishing, something to be true but cannot yet prove it. Until they can prove it I remain based in reality and current scientific facts.
i think we know nothing about the universe.the amount of data and potential patterns to collect are enormous. we have only seen a tiny fraction
you see, all theories are verified by experiments that produce data. the statistical analysis then concludes a certainty to which the data is potentially amount to randomness
as the certainty rises and the probability for something to not be random to shrink into a certain threshold
we call it true and verified and proven
Yeah, and the moon is made of green cheese too.
i agree btw
we are stuck with theories
afterall what we call theories are actual abstract objects inside brains
patterns
subject to the same laws as the theories describe
We also do not know much about the ocean and its immensity, only small vertical immersions in specific areas, we do not know much about its history either, the few species of which we know their existence because they were lucky enough to leave fossils should not be even 5%. the human being is quite ignorant but here we go with the little we have at hand
if we want something absolute we'd have to go outside of reality, and see it as an isolated system where our interaction would change nothing so we could actually measure the state of things. but each time we look, we interact and interaction changes reality
so where do all the species go that fully research and understand the phenomenon of intelligence
or atleast, understand it enough to improve it, in order to then improve it further
forever
without every reaching a final conclusion, but getting infinitely close to it
what do they do with that knowledge
this is weird. looks like a little 8 years old girl with boobs 😮 crazy
awesome!
I think these look pretty good.
It’s the Playground AI model with the cinematic filter.
I think this one came out pretty nice
I think you just haven't tried good models 
👀
okay guys my 3090 is here, probably guy who sold it to me saw this and thought gpu was dying so wanted to get rid of it asap. now here's the deal - board itself is brand new, nothing wrong with it, everything works like a clock but i need advices about cleaning this, it looks like mixture of white and black rusts and whole radiator is covered in it.
Unfortunately yes. Is there a model or a website that can generate people who actually look like people?
blue shiba dog
Hello guys I don't know how to generate image?
Hey there Peter 🙂
We don't have a bot for it, but there are multiple ways
here is the FAQ on this
@dry wolf
Currently, there is no bot on the server that generates images. 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!
there are some free credits but yes
Sad.
you have ways to run it on your computer directly if you want, or buy credits to run it more online
There are also other websites that give some credits and ways to access in #1080946152318443610
realistic vision,
realbiter, realdos, protogen, Guofeng shampoomix if you want something between realistic and art
(technically with good prompt those can do realistic better too)
models that do realistic on 1.5
yea that's a list of models which can do realistic gen, most people use realistic vision if they want pure realism
realbiter ❤️
I'm usually somewhere in between realism and art
that's a strange comparison to evade checking those models yourself
I'm happy for you that you know you are right. no need to check then
I don't see where the user you were responding to talked about 2.1. Still, I don't care, like the models you like. Stop on dissing on others because of their choices.
No, I want a completely free alternative.
I don’t understand why I should pay for AI.
depending on the computer you have (you need 4GB dedicated vram), Stable Diffusion can be free to run, outside from the electricity cost of course
in the #1080946152318443610 , I also push you to check "Stable Horde", that tends to give some free ways to access bots for it
Lastly, lots of people use replicate, or directly (very slowly) Hugging face itself for free
I have an Acer Swift 1 laptop with 4GB of RAM and a Pentium N4200, that’s clearly not enough to run SD.
I can confirm, it won't be enough.
There are some harder way to run it for free for limited time in the cloud, but outside of that, and what I added to my last comment, not purely free way because of processing power needed
Playground AI gives you 1000 images per day, I think.
yep, and implements SD
it's a good place to start, depending on how much you want to make
But why is 2.1 limited to 512*512 though?
on there ? I would say to reduce the VRAM they need
there are 2 versions of 2.1, one is 512 and the other one is 768
The problem is that the site is very slow for me compared to others of that kind.
And it wasn’t as slow a month ago.
you shoudl check Stable horde maybe
2.1 is far from outdated, loui is just fanatic of SDXL today ^^
The waiting times are very long.
not long ago, google colab was a good alternative but they pulled a little the plug, on the UIs at least. You may find tools that don't have UI that run SD for free in there for around 2 hours per day, but it's not going to last.
for free, you won't have fast :/
feels Mars Attacks
🙂
I think this is SD 1.5. It’s not perfect, but OK.
Artbreeder never failed me.
SD 1.5 still does ok. it sure has seen some upgrades, but on its own it's already a magic carpet to fly on
Any sites that can help me do that?
Oo this one is really good
is there a channel here or something that explains how I could install stable infusion locally?
Playground is still pretty good.
Playground still looks fairly believable if you don't look hard enough.
Definitely. But the website feels slow at times.
No, what makes it unique?
Ooh.
Gonna check that out then.
are you going to antagonize the whole server @paper salmon ?
I have given you that advice 3 times now today only. Read the #✍🏼|rules-and-tos , focus on rule 2, and go calm down for a while, this is no respect to have to others, that do have sent you this picture, that do have given you models names you didn't even check
Come back with a better mindset to participate inside a community, not culture of trolling and drama.
plus you have been posting the same pics again and again now.
ooo spooky
nice spooky gens
I hate that memleak auto has
is this SDXL ?
Even gets the 48 gig cards
yep, loui is doing almost exclusively XL from what I get
a digital painting of a horse in a field, rainbowshift trees, Synesthesia
sorry for the noob question but.. where can i find the model ? or is it on a website ?
no problem 🙂
Currently, SDXL isn't done yet. there is only a beta available, and you can use mainly on the official website
Alternatively, using the API of the official website, but yeah, it's if you need to specificaly
A jar inside a jar
a detailed painting of a cat mouse creature underwater, Encapsulated, masterpiece
thanks ! :D
do you know if the vram requirment for SDXL will be higher than 1.5 or 2.0 ?
oh and.. how do you integrate links into text ?
how do you integrate links into text ?
discord has started using markdown syntax, so[title](link)
do you know if the vram requirment for SDXL will be higher than 1.5 or 2.0 ?
No official info on that. they have said it will be around 2.5 times bigger than 2.1, but how this will translate into VRAM requirement is a big question for us all
I do hope for the best, but I also see the "XL" in the name, and I shiver
a digital drawing of a cat octopus creature underwater, corals, palace in ruins, masterpiece, catstravaganza
a realistic photo of a mouse on a beach, somewhere, sunset on the ocean, calm, music
Negative prompt: blender, 3D rendering, unreal engine
Steps: 20, Sampler: UniPC, CFG scale: 4, Seed: 877102347, Face restoration: CodeFormer, Size: 768x768, Model hash: 063e9ed089, Model: SDArt_v2_768
My mouse is getting fatter
a realistic photo of a fat mouse on an armchair, eating cheese
Negative prompt: blender, 3D rendering, unreal engine
let's make it fater
Thanks ! hum.. i tried in private message and it just didnt worked, i saw on google that you maybe need a bot or a webhook, is there a specific way to make the hyperlink ?
I really just do it in text :
it's still experimental, they may not have deployed it everywhere
we are reaching strange level of fatness
oh okay the "<>" where missing thx :)
the <> are for "hiding" the preview
you can put it around any link, even on mobile, to prevent a preview of the link in the chat
TMNT 70 years later
LOL
oh you want cats?
yes
I generated some caats the other day
cat father bless us
In V2 train on more cats
definitely :)
woohoo
there were quite a few already
never enough
anime bad
cat mafia boss
you really should do a non anime version
there was only one anime image in the whole dataset
trust me it spreads
begins to style anime
especially with just one it grabs on to it
yeah, and if not doing pure anime, it doesn't seem to land me any good anime anyway
I have around 15% of my dataset that is anime
precisely
but the output is just crap
Good point. I will remove it from the dataset.
I'll remove those in a future version personally
yeah, why the best is all or none. I found that out in my training experiments
the data infects the rest
it works on the faces quite a lot
I am a style only type and one lousy image will infect everything
So what I am going to do is delete the first 100 images in the dataset, which were used to train a digital art embedding and I just added on for this model. Then I am going to add in maybe 150 more, including photographs tagged with "photograph". Every digital art piece will be tagged with "digital art".
I like this prompt it listens
I might re-add some images but in their original aspect ratios and sizes. They were all 512x512 and were resized to 768 for training automatically by dreambooth.
I can say it doesn't infect and destroy everyting else, I do a training on almost everything submited in PoW, even the bad ones, and it doesn't destroy all
But I'm sure the "director's cut" edition will do better
most of my images in there were pretty... creative. Probably caused some issues.
"realistic" works quite well as a base token for photos too
Well, for styles I saw it slowly infect the other styles many times especially if there is just one of it.
but who tags a photograph as "realistic"? that would get you more a realistic render than an actual photograph right?
it did, yes. it did infect some things
also, I tried some other things, like adding the full name of each PoW to the caption, but I'm not sure that was a good choice
I think it has to do with it repeating back over itself
true, my bad. I do tend to equal realistic and photo in my use of those
What is mind blowing is you have 200 images. You train and it seems to have locked on to one image and you can't remove it from the gens so have to go back and retrain finding the offending image to ditch it
I'm trying to test it extensively before next PoW/CoW and see if I retrain another version then, with the upgrades I imagine
Oh, I have a question, maybe one of you can give me some ideas on what to say.
My friend wants to go to college to be a character designer/artist, I suppose for things like video games, movies/tv, comics, etc. I was talking with her about AI art, and she said she would be interested in learning a bit more about it, but I don't know what resources I could send her or what I should tell her.
I wish we had a way to prevent it from locking in on an image like that. I say style and it suddenly turns it into object/subject
my best guess is she needs to start using it, just to discover it.
Ideas will come when she learns about fine tuning, or she can ask around
VideoGame assets making + Artistic capabilities + knowledge of SD is a very powerful combo, and can lift the work of 10 artist in a conventional setup
True, I know she doesn't have an extremely good computer I don't think, so she wouldn't be able to run it locally. I also don't want her to spend any money on it. I guess I could try something like stable horde? I dunno how confusing that is to set up.
Or I guess just set up a server somehow on my computer, but I don't know how to do that either.
multiple things there.
If she's a student, we did have some credits grands for purpose of research, she could apply to some. but even the 1000 free credits is a good start.
Then, I have one example if you want, that illustrates that this can be the tech of tomorrow for "game developers to be"
I did try to make a "hogleg" dreambooth model, back when I was learning how to make my first models
I used the traillers from hogwarts legacy, the game was not out yet
here are some gens I got
She's in high school, like me (not my grade, and home schooled), so idk about that. And yeah I know about the gamedev uses, as I make games myself and have tried to make a gamedev specific model (it failed, though now that I know how to make a good model I might retry it).
the aesthetic of the game translated quite easily, and I could generate interiors that do look like the game, now that I played it
I'm not sure what to tip towards yeah..
The free uses of SD will be too limited to see the impact it can have, imo. not for herself.
examples may be the best to "show the light", but a way to run it feels like essential to dive deep into it
But what about Stable Horde? Is that too difficult to set up?
ah
I've thought about joining but I don't want to wear down my gpu too much. I only have a 3070. And I need to get a bigger case that actually has airflow.
one thing that can be fun and instructive, but maybe a little hardcore : run SD on colab
not through a webUI, those are blocked
but through code
ohhh yeah that might be a bit challenging
this is still OK and a good learning experience
yeah
but also just taking her under your wing for a session or two, showing it
I can try! I'll see what she thinks later, maybe I can have her like send me some prompts or something and I generate, and I can tell her which models I have and what they do, etc.
not on the best side of persuasion, but if she's just in high school still, there is like no chance this job she wants will still exist without the use of AI. So it may just be something she'll learn along the way
no rush to dive into it, but for sure, learning how to swim now will make it easier on her down the line
that's what I was trying to get to in the conversation but I really didn't want to seem mean or rude or anything
and besides she is really good at drawing anyway
what are her hobbies ? what interesting way could you use things like controlnet to make something that would be of use to her and give her the will to try, for example
take 30 of her drawings, train a model on her style !
Well I know she loves Portal and recently fell in love with The Stanley Parable
Yeah she is very nice. I don't know her that well in person, but we have met. We mostly chat online and play games.
Stanley Portals doesn't mix well in my mind ^^
does she have an online porfolio with her art in any way ?
I haven't played the Stanley Parable, but I do also love portal.
Don't think so.
She has a sketchbook and I think that's it
stanley parable is about breaking the 4th wall and great naration.
Greatest game I played, that I'll most likely never play again
if she'd be willing, I see 2 nice things you could try :
1/ use some shots of her notebook, and controlnet to make video game art in the styles of games she likes or knows well, following her drawing (scribble mode, canny mode)
2/ if enough material, train a little LoRA on some shots of the notebook and enjoy
How do I brighten a gen? seems like all light is almost gone
i've had those thoughts, I will run them by her later.
this is the only drawing of hers I have on my computer currently, I think it's a bat playing baseball
I just realized the irony of that
I added bright streetlamp in the background and it helped.
Not sure how to get that brighter
alrighty, all 150 images I kept now have the tag "digital art"
since they are all that style
now on to getting like 150-200 more images!
ill start with cats
so you just put that in all 150 tag files?
yep
other images will have "photograph"
those wil be the realistic ones
man, I can't get this any brighter.
oh those are the cats from pow!
I need help
no neg
aw here was my submission
nope
kinda terrible looking back on it
I captionned it "a pink saturated photo of a kitten, Catstravaganza"
for the megapow
i've taken three of the best options out of those cats
"campfire lit"
"studio lighting"
glowing lights
yeah, I figured it wouldn't train well
this weeks model will be a doozy
what are you talking about
the eyes are pretty bad
this is so well trained, it's a realistic photo of a cat, saturated, so so pink
I'll blame it on the anime in the model
x)
campfire lit
you know it 😉
and inclusive to all SD :p
first try too,it's good on short texts
almost too bright this time hehe
but it kept the composition !
I was afraid it wouldn't after a point
aye
🥄
thanks, I'll use this on the next person that pushes on the rules
"as SD2.1 without finetuning would say, You want BaAAN !§"
lol
I'm trying to send the next batch of pictures, but somehow Discord says they're explicit, even though there's nothing serious in them.
beautiful
yeah clyde is...
clyde
it does block lots of things to me too
and sometime I'm just there wondering why
lol
you are using even their name in the prompt ?
thus the purple ?
I can speak Stable Diffusionese as well.
best languaaaaaaage ever
Ttytd yíimd hefs ygas¿
some controlnet fun I had, using 1.5, and old memes of "overly attached girlfriend"
I almost forgot that meme existed.
memes remix were my jam when learning controlnet
like, this is Forever Alone Guy
I basically use SD for pretty girls and landscapes.
have some of my best girls as a sorry, mostly 1920x1080
Those are some fine girls.
I would use it if I had a non-potato PC.
the prompts are mostly derived from the example pictures on the model there, still good to use on other models
Do you think it could work in Playground?
Cause I had the most luck with it so far.
The only issue is that the skin looks too perfect.
I have not used playground, not sure how you use custom model in there
but if it runs automatic or something like that, it should be able to
Maybe just the prompts?
yeah, already worth a try
It has SD 1.5, 2.1 and Playground 1.0.
I personally like the latter.
I'm working on an artistic model too right now, I'm testing it under every possible light to see what to improve on next
Good luck.
woah that one is really good
really showing off the horror side of dd
maybe I should add some spooky images to the dataset, what do you think?
perfect writing
at least I think it is
I can't read
Yes. I don't normally do happy type pics.
got it. Ill try to find some spooky things.
the
is so 
that's the kind of red hair I know
I dunno but that one really talks to me I just did as there is nothing but death all around, and despair, and roses.
Literally the Bad Piggies movie.
Well, I smell bacon suddenly.
Now do it with Angry Birds.
Which girl do you prefer?
those are pretty good hands
I am still wondering about those nips in the last one
So I've got an update, I'm starting to get the hang of control net a little bit? but i don't know how to incorporate elements of the image into place. for example, i want her to hold the rifle to the left there. is there like a 2 image to image inpaint or anything?
@hasty nova I am having an issue now where everything wants to be half in the ground and half out.
no. i want her holding the weapon that created with the image
that's an interesting issue I must say
here's an original. here's a bunch of controlnet trying lol
i want to retain the weapon, and have her aiming with that. here is my pose
id probably make an open pose reference picture
yea like that
then you can inpaint the weapon in. or mabey pray that she holds the same gun idk
how do i inpaint the weapon in if i can only inpaint the image i have
well you would create a brand new image of her in that pose
you would then inpaint that new image
i don't understand. if i create image A of her in the pose, how do i inpaint something from image B into image A
lots of tricks to this
the easiest is to just edit the gun in after she's in the pose
run a few img2img to blend it in
still listening, but i see what you mean with that
Is there any way to keep more of the character while turning up the denoise?
ooh ooh okay i can work with this i think
alright i have a few working options here
i am honestly not sure how how to get the costume back. i have a lot of the character back now, just need to clean things up
dude this is crazy, i am actually getting closer D:
Oi
nice journey c:
you can actually achieve this effect by outpainting too
you can extend the bottom part (or left or right or top or all) of a picture and then have stable diffusion draw stuff on it that
ooo
is a technique that allows you to generate a related image outside the limits of the size of your original image. example:
WHOA
outpainting would have been useful for my desktop wallpaper
this is still good
So originally I wanted her down in the bottom left corner but I got her mostly on the screen which is fine because she is the character but I wanted her to be looking up at the Northern lights not taking up the whole picture
you can use SAM to make a mask of her and do that same thing
ooooo learning more stuff \o/
can i download this from sd
yeah
im assuming you mean automatic1111
its an extension for it
youd paste that into the install from URL tab
thats likely what you can do to finish off your sniper chick picture as well
okay
she needs a bit more lighting
but now that I added some
she's giving me the look of a murderer
again
why do all my gens collectively want me dead
that's a bit warmer
this is neat
firefox relay - easiest one to set up and use aside from shady 3rd party browser plugins
8/10 for recreating my character Spice
Now I need to learn about this outpainting that @calm canopy and @brazen anvil were talking about to make her fire the cannon \o/
still cant believe i did this lmao
drew a freaking cape LMAO
okay so i downloaded this. do i need scripts?
this is literally instruction diarrhea
i dont know where to get inpaint global harmonious
hi there ,i do
Character designs. And i love draw lineart
.
.. I want SD to draw the final image according to my detailed fined lineart
.and not fully rendered color composition like this image says. I
Has been searching this.for the whole day.no result. Guys. Is there a extention for this function i
Mentioned. Thanks pls.
nope, shared memory isn't available for that. it's basicaly RAM and not VRAM, but disguides as "backup VRAM" for offloading the main VRAM when full
Laptop?
i see, any other SD related use for it?
nah workstation
dang
laptop with this GPU would cost me my soul 😄
not that much. some tools keep your checkpoints in cache in it sometime, when you change a lot checkpoint
and when I'm training/using the full VRAM, it gets populated by windows and other apps that ran on VRAM
woah that's very useful. thnx.
3090 is ancient and they make mobile soldered in versions that cost way less.
"ancient", heh
it was released in 2021 lel
yep, ancient
bruh i had 1070 till yesterday for last 6 years 😄
rip it out and I like to buy antiques so send to me I pay 200 usd
caress its 3090TI
Don't listen to GeneralAwareness my cuty
still saved it in the box cuz i love it
actually i literally bought it for 200$ 😄
initially dude was selling for 720$ but when i went in radiator had decay
all over the place
now that is sad
dude thought that gpu was dying
mixed metals I bet
but i knew as a matter of fact that board was perfect
cuz no damaged and corroded board can run 2 hrs of furmark on proper temps
😄
so i was like "nah bro i'm not buying this gpu will die in few days" and when i was about to leave dude said wait, TAKE IT FOR 200$
well all this nid is alphacool eiswolf 2 AIO
i know the feeling man, i was looking on my nzxt cooler on my cpu first few months like crazy
that's almost NSFW 
checking it several times a hour per day
😄
was afraid water would pour out and kill all my workstation 😄
but slowly i adjusted and got used to it
I was looking at 4090 and at its price I would jump from 3 cycles to 6 due to cost and AIO last about 5 years
yah fck 4xxxx for a while
a cycle is a gen so 2 years
nvidia can stick it up their ass for the price it has
I agree.
I am afraid of the used market as no one in my city sells used stuff like that they toss it or keep it until busted
3090 is best choice (well actually 3090ti is best choice but well, they're way more rare to find for a good price)
used market is really good if u know what u are doing.
like my gpu deal.
problem is miners, gamers who never clean, and ebay
from outside video looked like it could die in a few minutes, coolers were filled with oxidation
and very noisy
little cleaning and they're silent af
honestly dude who had it just tortured card with mining, probly used nitrous oxide to cool it down
cuz that's what he read in the internet
😄
oh, Lord and precisely why I don't touch used
but board when i opened it was brand new
sure as shit I buy and it is mailed to be dead
so this dude kinda brought it up on himself for never opening the card to maintain or clean it
Yeah
im just waiting for my old 3090 to kick the bucket any day now
Bought it nearly on release, I would just upgrade when it happens but not in any rush atm hah. Thought it was dying sometime back for a bit when I started get nasty artifacting before fresh installing drivers
❤️ 🦉
ahhh
yeah, this is the problem i would avoid like death. if i see artifacts, that's where the fear kicks in. 😄
it was def a full moment of panic for a sec haha
glad you are enjoying early xl haha. Hope you like the next version we have cookin
The Dukes 🙂
I don't like it that I can't run it locally forever.
feels SDXXL
yea #✨|sdxl thread is great for us. We dont always respond but we see and try to read through all of it 🙂
mushrooms tripping
LOL
any recommendations?
whats the cat man and the bat doing
Having tea but he decided to wear his Wonder Woman corsette
just trying to inpaint it should help solve this. go into "img2img" tab, select the Inpaint tab, drop your picture in it, draw a mask, and start trying other hands.
full explaination in automatic wiki
(lol give us a sec for answering, you asked it in 5 channels, don't spam like this)
\o/
Spiky Alien Rabbitopuswormine
lovely ❤️
Thank you 🙂
anyone has collab without webui ?
Unfortunately from what I know, SD on collab is kinda dead right now
have a feeling thats gonna single handedly be a huge reduction in the majority of the community
That is true, but at that point just biographics card, cuz it doesn't take long for it to run up your bill high enough to warrant the purchase of one
Yeah, can't wait until the community can get our hands on SDXL and SD3.0 and make some really impressive stuff
SDXL can do some pretty cool things in beta, but man it's gonna be so much better when we actually get out hands on it
it's just webui is against the rules, non webui should be still fine...
The problem is - I can't find non webui anymore
Yeah, I know. Also. No webui is an absolute nightmare to use. It's an absolute pain to control
whatever
I started using stable diffusion when it first came out and it was all non webui, and my god it was unusable
I'm sure it's not as bad anymore, but non webui is an absolute deal breaker for me personally
Also, didn't they make the free Collab time limit like 90 minutes now?
don't think so
I know they have been shortening it more and more recently
It was lowering down if you use it everyday \ fail afk recaptcha check , maybe they have other systems checking if you're doing something
No idea about that
While google cutting SD and over WebUI services sucks for the commities, I also 100% get why they are
It's just not widely sustainable as it is
🥄
@glossy herald I am surprised you chose the mushroom cap instead of a cat dangling on your cat
I tried both, but cat wasn't that good on my cat imo
I'm getting tired of it anyways
I'll remove it soon
Ah, understandable
Lol
You don't need an expensive PC to make 20MPX images in stable diffusion, but I do get what you mean
when you'll get your eyes on the next beta, they are going to pop out 😉
Unfortunately, paying for Google Collab will quickly eat up the money you could have spent on a full blown PC that could have run it faster from the start
It's a very slippery slope sadly
You're not limited to 512x512...outpainting and upscaling exist too
Lmao
Is old dreamstudio still alive? there's no more link to it
You can go that high or much higher even on a 6GB GPU
5120x1440, made in 1 pass in SD 1.5 fine tuned
modelshoot style, (extremely detailed CG unity 8k wallpaper), photo of the most beautiful artwork in the world, professional majestic impressionism oil painting by Waterhouse, John Constable, Ed Blinkey, Atey Ghailan, Studio Ghibli, by Jeremy Mann, Greg Manchess, Antonio Moro, intricate, High Detail, dramatic, makoto shinkai kyoto, animation key art
Negative prompt: plastic doll, plastic skin, (easynegative), monochrome, (low quality, worst quality:1.4), cg, 3d, render, (multiples views:1.2), wrong hands, deformed fingers, author signature, text, watermark, nsfw, clivage
Steps: 58, Sampler: UniPC, CFG scale: 5, Seed: 1554567698, Size: 1820x512, Model hash: c4947fa2c4, Model: DB_mix_realbiter_v10, Denoising strength: 0.44, Hires upscale: 2.85, Hires steps: 30, Hires upscaler: SwinIR_4x
no, open in browser, it's discord scaling it down, it's 9MB file
also 3072x3072
and like I posted tons of 1920x1080 desktop yesterday
3072x2048
what was the prompt for this one?
A TV screen with the text "SPOON" written on it
4352x4352
Is it "text" word triggering it specifically?
I tried "title", didn't get it
"written" helps too but I'm not 100% sure on all the tokens for it yet
yeah, lol, download original if you want to talk big pictures, discord doesn't let you do that in the channels
low quality pixels
sounds interesting
The detail is most certainly there
Tried to do A TV screen with the text "Guizmus, The Moderator" written on it with SDXL lol
That's best 25 credits did 
don't buy second hand pixels, it can ruin your day
Lol
I can't see what you said , your pixels are low quality
don't burn your credits for me and memes ^^
I hope discord support can read me and give me better pixels
it's new acc just to do quick test, wanted to see if it can do it
This one was interesting...even tho it has nothing to do with the prompt
it feels "encapsulated" like the last PoW
yea
:/
You're being a little unreasonable here
Also, if you are referring to the very slight compression artifacts, that was my fault, not the process's fault
it's one of the crispest thing that come out around on such high res but sure, it surely is low quality pixel artifacts
upscalers are a hell do deal will and keep details and quality
Since when are we comparing actual camera to SD?
I mean, AI just trying to emulate real world, it's not perfect, even tho I can't really se "bad pixels" you're talking about 
Your point was - "I don't want to pay for 0.3 megapixels", my point was - you aren't limited to it and you got a proof
You can go as high res and as high quality as you could want in stable diffusion with the patience
Also, cat managed to get trapped between my inner and outer shower curtain lol @glossy herald
lol too cute 🙂
Why are you paying for dreamstudio in the first place then?
It's even lower res
That's cause my process then was using JPG's, so it enhanced the compression artifacts
Now with PNG, those issues are not there anymore
I downloaded so many models to test but no longer have collab or kaggle 🥺
Hmm, new upscale API from SD themselves