#🏞|general-with-images
1 messages · Page 54 of 1
I am VERY Surprised that it knows how to create SD prompts though
I would imagine there isn't a large body of literature for SD prompts when ChatGPT was originally trained
it just goes by 'condensed keywords and (grouped words)'
what's funny is how much better its images are than mine and how it "doesn't get" how the prompting works, but maybe it's me who "doesn't get" it
Ahh, I see. It actually DOESNT know how to prompt SD, but the "condensed keywords..." is what formats its response to be similar to SD prompts
That is clever
Is your actual occupation related to any of this at all?
i can give it a theme like "cosmic nature" and it will definitely use that in its output
Or is this a hobby
i'm a developer and i'm working on this as a model playground for researching things so we can start to do more with model training/fine-tuning
What do you hope to be able to do?
the goal is to build a highly specific LLM, honestly the image stuff has very little to do with anything
The LLM would just be basic research, or is that for a specific commercialized product?
it's for internal use for coding
there's some talk about building a user-facing instruction tuned model that helps customers customize stuff on our product. i can't say more about that part
I see. What industry are your customers in?
but that's why the bot supports StableLM, Vicuna, Llama, ChatGPT, etc
i don't even know
So it sounds like you want to create a bot that allows customers to ask it to do something that your product isn't specifically designed to do, but become able to do
nope it's all capable of doing it, but the user might not be capable of getting it to do that, on their own
training it on working existing examples so that it can produce the result based on their nagging prompting
it's tempting to build a large language model i can use for attack analysis on incoming network requests
unless i use Hyena or something, it's hard to fit enough context and perform in realtime to do that. so, it don't happen
That would be super interesting. It is cool that you have so many ideas on how to apply a lot of this up and coming technology
I really like SD because it creates such cool images, but I can't really think of many good commercial uses of it
yeah way better than the current work in progress, getting the AI to learn Tic-Tac-Toe
The best I could think of was generating FB/Insta/etc ads
Though I believe many are already working on that
oh companies love generating impossible ads that show their product actually working
It was great chatting with you. Thanks for all the info! I'm gonna head out. Take care!
https://tripleback.net/public/discord/1683246167.9170368f2cd3c154a10a4be7ec100137a336265.png
love how the higher res images make no sense
yeah, native high res gens are a mess lol
sometimes you can find a magic seed
i don't understand this outcome from the prompt, a babbling brook of [censored] in the desert encapsulated in a glass sphere sitting on a beach next to a blue sunset with a pirate ship waiting off-shore in the background, soft focus 8k bokeh f-stop full-frame DSLR Canon Kodachrome 1982
a friend's attempt at the cast of the big bang theory
it's like a pressure gradient
I am seeing some biases lol
yeh where's raj
lmao
a volcanic eruption encapsulated in a sphere of glass, floating in the ocean, under a deep sunset with full view of the milkyway overhead, epic cinematic 8k sharp focus cinematic highly detailed trending on artstation trending on deviantart masterpiece bling tech-space cyberwalker daysun nightmoon
i haven't gotten a single proper horse
I end up with fantastic horses but with 3,5,6 legs. Mostly 5 legs.
Currently, there is no public bot on the server that generates images. There is an experimental bot available for early server members & Stable Society. You can obtain the Stable Society role by winning our weekly events #⭐|pow-info #1087493421209485393! However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝|tech-support for any problems you encounter while installing it locally!
there are so many already. Illustration for content (articles, websites, etc) ; 2d videogame assets, teaching supports... and just wait until video diffusion models reach the same level of quality and the film industry is gonna shit its pants
Enders Game
hi
anything-v4
I keep getting all this noise in my faces, after rebooting and such. Any advice?
@oak osprey looked at paperspace, kinda nice and only 8$ 🤔
yeah it's often out of capacity tho
automatic 1111 making a buff dude with a depth map on control net
nvm, sampler isn't jiving with model.
doing this is slow. :.)
What service can I use where I have more control over the models using the auto1111 thingy
Yes you just doing locally?
ye
if i had a better gpu id probably be blasting through these images lol
damn 6gb vram
Dang, I was wanting to run it on cloud but finding a service is hard
tbh i have no idea
but itd definitely be better than what im using
beefy 3d man
lol, mostly fixed.
how many samples are you doing for these?
same prompt, 7 months apart (DreamBot August vs SDXL/StableDreamer)
best results cherrypicked
buy a 4090 lmao
trying to make a map of the earth where europe is equidistant from every other country is hard
lol SDXL outputs an artists' watermark
i wouldnt ever trust a 4090
unless im using it to cook fried egg
ok then keep wondering how to do what ya want on your 6gb VRAM
why would i be wondering how to do what i want on my 6gb vram
when im already doing what i want on it...


Time for a little red and blue.
just recently started trying img2img to edit existing images (non-ai generated) and jeez it makes it super easy. throw in a bit of paint.net and this took under 5 minutes to go from the first to the second
considering i still consider myself very much a rookie when it comes to img2img, it's obscenely powerful. most of you probably already knew that, but it kind of blew me away
for spiderman meme lol
Man this new upscale process is insane
@smoky oak what can you get out of that thing with the prompt "lisa frank forestry"
wow, photorealistic
hagrid riding a unicorn
im a fookin genius
I just realized I can use controlnet for my reg images now o:
Im gonna be there very best, like no one ever was...
to dreambooth them all is my real test, to gen them is my cause
I will travel across the web, searching far and wide
is this easier on your eyes Sen?
there's marble
was that steve aoki
looks like that guy from Operation Takedown
Skeet Ulrich
hey sytan when you use the CTU with the image loaded in the preprocessor does it also take an eternity for you?
I cant believe one of my reg image grannies got me so rock hard I had to upscale the image 😊
How do?
*so?
Eternity in what aspect?
for example to do a 4k with just UU is like 2mins or 3 mins
but with this tool it takes like 10-13mins
with seams fix
its only fast when not using the preprocessor image but it seems to work better with the image
oh wait, with seams fix?
I never use seams fix, its like 3x as slow
for me, its not much slower to use the CN
lol nice
love this one General 🫡
@bold caveI am messing with all of the settings and I am finding some cool things with this upscaler
the downsample feature is cool
it lowers the resolution, which allows it to maintain the base shape and form, but it can add the higher res details with the prompt
so kinda like a soft guide
thats really just not true anymore
high res fix into CNUU is best right now
I am a believe now
do you guys have any suggestions on how to make generations with hands out of view?
blegh, that looks terrible 😅
and you deleted it yet again
IDK why you delete almost everything you send
1 leg soldier left to be deleted
right lol
only a matter of time before he deletes that too
Walter Gray
yup, called it
right, he started deleting everything and IDK why
SD is getting on my nerves
I cannot get this generate yo not have outstretched hands
upscale then inpaint
my problem is I dont want the hands reaching out at the camera
it keeps doing this with the hands
oh ok thats different
I have hands to side as a 1.9 strength positive
try in pockets, behind back
apparently hands behind back means face away from the camera and let me see your hands
so annoying FFS
or hands in negative
might be the model
every single god damn generation of this prompt has their hands on their chest for literally no reason
the loose sleeve clothing style?
yeah a wizard would be hand focused
that makes 0 sense
most wizards have a staff or some sort, sorcerers and mages are hand based
yeah, it does it for sorcerer too
exactly what i want from hands -4
@static tuskok yeah, this is absolutely stupid
no mention of magic, no mention of wizard, no mention of anything, every single image still has at least one hand on his chest
this model has had no issues before, and I have never had this through like 100k gens
if its still an issue there is always cnet openpose
I try to use controlnet as little as possible
it looks like out of all of the words, the one that is causing issues is "fantasy" fpr god knows why
alright, looks like it is impossible to have an image of somebody surrounded by magic without hands on their chest
guess I have to use open pose, which I really do not wanna have to screw around with
@static tuskJFC
Even WITH controlnet open pose, hands, still!
What the actual hell
play with the cnet settings with weights etc it should follow the pose
evenw ith weight 2, its still managing to fight and lift hands up
I saw a video and it said it used ControlNet tile to make this screenshot of this image into a much more detailed one, yet retaining the composition of the image. I don't know how it was done though. Before image:
click ControlNet is more important, use a higher denoise. assuming youre using img2img
After image:
Usually adding details will change up small details like clothing patterns, but this one retained it and made a simple image into a much more detailed one, while keeping things like eye shape and composition the same too. Anyone know how to do that?
Some things have changed, but generally the same
That's one way of avoiding bad hands...
So you use the tile preprocessor and ControlNet, and a separate upscaler?
Gotcha, I'll experiment with stylized upscaler
upscale
(I was working on my phone, so I didn't see her teeth and inpaint them before)
but as you can see, it works outrageously good
and thats only 1 pass
The quality you can get out of it with good prompting and a good model is insane
Does it make your previous upscale guide obsolete?
that has been obsolete for a while, as it was a problematic process that seemed to degrade the more you used it, with no solution to stop it
grandpa image is a bit noisy
that sort of tonal/local contrast effect is something I have learned to lower with VAE work
Tiling is now probably the best way to upscale then I think?
for sure
there are also some cool things you can do with it
it has a built in downsampler, so you can use a blurry version to guide it
so it guides the shape, but not the fine details
it also allows you to use WAY higher denoise without nearly as many issues
Nice, I'm going to play around with it.
Hi guys. When using DreamStudio there is a Style option which has a few preset styles that guide the output of SD. My question is, if I'm running StableDiffusion locally on my machine, what is the equivalent to the style option? Are the styles in DreamStudio just preset prompts that get added onto my prompt, or is it something else? Where can I find the assets for these preset styles? I've tried searching online and can't find any info about it
it really shuld be just prompt tweaks behind the scenes
like, it probably just adds additional text to your image to create those styles
yeah that's my best guess too
I really like the outputs these styles give and want to replicate them on my local copy
that's why I'm wondering if anyone knows if the exact prompts or whatever other assets are available anywhere
hi all, How are you doing?
good, and you?
pretty good except for hot weather
@stone cipher Dude, why do you keep deleting crap?
Why do you keep deleting all of the messages you send bro
uhhh
ok
a painting of hell and destruction, in the style of fantastic grotesque, light navy, black, dark red and dark gold, realist detail, schizocore, precisionist, gothic, undefined anatomy , high-quality details
I'm out of ideas to improve this
I'm blocking you so I don't keep getting distracted by your popups and deleting things right after
What would you add?
I am assuming you want it higher contrast?
looks like missing VAE
hmm let me check, i just reinstall a1111 and didnt check config lol
the faded white wash look is generally vae related
yeah, cause that looks like a missing VAE
I didn't say you did, but your constant posting and deleting is extremely disctrating
so I am just making sure I don't have to see it as big
even with you blocked, it still shifts my whole discord when you delete something
could just be really really shy, id give them the benefit of the doubt
he said he is deleting his account and needs to make sure all messages on his account are deleted...
or paranoid lol
I am just quoting what he himself said
How to use drembot I'm new to the server.
Currently, there is no public bot on the server that generates images. There is an experimental bot available for early server members & Stable Society. You can obtain the Stable Society role by winning our weekly events #⭐|pow-info #1087493421209485393! However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝|tech-support for any problems you encounter while installing it locally!

trying to reproduce the controlnet demo results and i get this 💀
nice nose but what about the rest
how about a photo real challenge? Romantic cafe, interpret it however you want haha
ha dont think so
looks more like shes in her room than a cafe tho
but quality photoreal, pete!
whaddidudo

apparently putting anything other than 1024, breaks it
managed to get a better output than the demo img
the resize for conditional method that is used in the demo has a default arg of 1024 so that it basically resizes something so that it's got the larger of the two dimensions at 1024 and maintains the aspect ratio
def _resize_for_condition_image(self, input_image: Image, resolution: int):
input_image = input_image.convert("RGB")
W, H = input_image.size
k = float(resolution) / min(H, W)
H *= k
W *= k
H = int(round(H / 64.0)) * 64
W = int(round(W / 64.0)) * 64
img = input_image.resize((W, H), resample=Image.LANCZOS)
return img
if i try to make a 2048x2048 image that way it results in the ghost doggie
I am confused, what variable are you talking about?
resolution
https://huggingface.co/lllyasviel/control_v11f1e_sd15_tile
this is the page i followed
what in the hell is that
I didn't do anything like that
all I did was update controlnet lmfao
😭
i think you're using an automatic1111 script that breaks the image into pieces and then upscales the tiles and then stitches them back together?
i looked at it but it's quite specific to their tool so i didn't think i could make much use of it
that script i believe is using this tile model under the hood
the model is quite something, man. i love it. because the source puppy image is 64x64
so i will work on a method to reproduce their a1111 script results i suppose
oh yeah, ultimate upscale
its a super easy addon in the extensions tab
Its what has had my upscales so good, and now its even better cause of this new tool
yeah i'm just still adding stuff to the discord bot, rather than just use their tool
i figured this was built into SD
i just really like writing code
technically it is
ultimate upscale is a tweaked version of the built in one
easier to understand what is going on, when i've broken it 10 different ways while getting it working
a script
you can go into A1111 and view the scripts. I could send it to you so you can look through it, if you would like
OOPS
MY BAD
both of my legs just broke from that
the problem with these scripts, i can demonstrate:
if upscaler.name != "None":
img = upscaler.scaler.upscale(init_img, scale_factor, upscaler.data_path)
upscaler is something internal to A1111 and it's not like, a common interface available anywhere else. it's a bit frustrating that they do that. but so do i, so, who can really complain
no idea what that is
oh, they are like the most common and rudimentary upscalers
i have the x4 one workin
all of these are ESRGANS
yeah, thats an ESRGAN
you can use any ESRGAN you want for the upscaler
fwiw what helps me are script examples like this: https://huggingface.co/spaces/akhaliq/Real-ESRGAN/blob/main/app.py
I actually have one there that removes sharpness artifacts as well as one that removes banding
sorry, I am very new to scripting
no worries just figure you might be curious about this half of things so just hinting slightly as i go 😛
luckily, the upscaler is a very common format tho, its just referring to ESRGAN's really, which I am sure there is tons of documentation about
i'm new to all that stuff but i'm a multi-decade developer, so, i can implement this stuff in my code usually more quickly than i can figure out how it works in someone else's tools
totally understandable
I’ve been testing pixel art. This image was made completely in A1111.
Thanks
did you ever play Willy Beamish?
is it a standard gen going through a pixel processor, or did it generate that way?
no
It was put through a pixel processor after generation
ah, that explains why its so clean
great work man
much better way to do it than generative pixel art natively. You always have issues that way
lol imagen?
I have a link to a video outlining the basics of ai pixel art. Would you like that?
yeah that could be really nice! I am focus on hyper realism ATM, so a break on the other side of the spectrum could be fun haha
I have a link here: https://youtu.be/pfGUe2Itf3k.
Unlock the potential of AI-powered pixel art creation with stable diffusion! In this in-depth tutorial, we'll guide you through the entire workflow to generate stunning, unique pixel art using the latest advancements in stable diffusion technology. Perfect for game designers, artists, and creatives alike! 🎮🎨🤖
🔸Step-by-step guidance for beginner...
hahahaha it butchered it but it did upscale willy beamish
if ya squint, she's mint
Much appreciation
@smoky oak you try
I don't really have a model that would do that well, and its also really low res
ah i got it a bit better
oh me oh my that extension came with a LOT of errors
time to see how low that number really goes
AHA
IT RESETS AT
58:37
SMH, false advertizing@wispy nest
🧌
ControlNet question (incl. Scribble): There's no way that ControlNet is just using the pre-processed scribble image, I'm getting a distinctly Harrison Ford face here just from this scribble it's supposedly going off from, maybe that also explains why I get so much blue-glowing when I use the 'softedge' pre-processor on this
got some interesting results with trying to stitch together images lmao
lil boi's all dirty now
Ah yes
@brisk holly how do I get permission
im not sure what i did wrong tbh it looks like the math is just wrong as hell
i was never very good at math
I hope you can get it working. It works phenomenally well for A1111 as I am sure you have seen
aye hence all the effort
😛
so i think i'm not resizing the canvas properly
lmao
it calculates a tile size before the upscale and then doesn't recalc it after the upscale
so it just essentially crops each tile and stuffs it into the same location
hence looking that kind of mega-pixellated, it's technically not though
up top we will be rolling out bot access to more and more users over time. For now its possible to vote in #1100484581037195384 and check out winning creations in #🏅|pantheon. Voting gives us useful feedback for improving the models 😄
spider armor
I'm pretty sure it's possible to use this scribble to output an openpose ?
Yes
How would you do ? Because I can't get it to work using openpose and scribble being as preprocessor or model 😕
yo...

what
bit of a weird picture isnt it 🤨
why u generating cat girls anyway
oh
what upscaler is the best tho
CTU
with this one
the results are insaiyan
i mean from those
68%|██████████████████████████████████████████████████████▍ | 102/150 [04:16<01:57, 2.44s/it]
Total progress: 84%|█████████████████████████████████████████████████████▊ | 252/300 [05:01<01:57, 2.44s/it]
🗿
what is up with people in voice why join voice if your going to go out your way to not talk to other people
i dont even see this kinda seperation at the urinals in the mens room
how do i add that one
those are always the same people
most likely bots
or npcs

idk about others but that kinda makes me feel unwelcome to join voice if there gonna afk dont do it in voice when i see that i think there either not wanting to talk or are waiting for freindos
Does anyone use anything for inpaiting and combining outside pictures together than Inpaint upload? I mostly get this type of garbage where it straight up ignores trying to make it consistent
the colors 💀
idk what it is but something about that drink looks off
didnt notice that
but i mean
i think its the almost realistic colors of the drink against the neon ultravilot of the charecter make the 2 not seem right
why are you stuffing cats in drinks
so ive developed a bit of a obsession but its not what you think
so i like to hangout in the nsfw stable diffusion discord
kink
not for the smut though
would a woman with big boobs but fully clothed be ok to post here its not porny

Might have to use another method since that one isn't working out. I would transparency mask the new layer of Snape in, then img2img to blend him in. ideally your method works as well but doesn't seem like SD knows how to make it work
shes actually pretty clothed but i skimmed it down
but yeah i find nothing more amusing that watching SD mess up nsfw concepts in amusing ways
🗿
thanks, I'll play around with blending it better before the Inpainting step to help it along
why dont you just mask out the face then inpaint a prompt for snape perhaps using a model trained on snape
idk how to do that
litterally color in the head you want gone or better yet the whole body then for a prompt do stuff like snape,proffessor snape,harrypotter ,alan rickman snape
although honestly for that id elect more to just photoshop it
she has 3 hands
i used DPM++ SDE karras 💀
wait no she has 3 hands one is some weird elf paw the other has snake finters
also is she holding a robin by the tail and using it as a spoon?
It's more of a test for combining images. I used what I had on hand but maybe I'd want to copy my face or a specific face with a specific expression or specific item etc. Mostly trying to find the limitations of inpainting and workarounds but tbf so far Im not getting good results with inpainting in general
ah yes. The first video I watched to just familiarize myself with settings mentioned that the person prefrs using the same model they use generally so it can do the same style
so I forgot to try with a specific for inpainting but I will now, thanks. Any chance you have one to recommend? Currently using Artius 2.1 for the general generation
I'll be honest I myself haven't had the chance to play with inpainting models yet. This thread was pretty big though. You can make any model that you like an inpainting model
I have heard the benefits of inpainting models though compared to normal ones and it sounds like a no-brainer to use one over a normal model for inpainting
oh that's a cool way to do it. Almost no inpainting models on civitai so maybe I'll try that
Yep, noticed the same thing and came to the same conclusion that this is the most optimal route forward when it comes to inpainting.
nope.
oh wait, I just saw the thread
interesting. I was gonna say inpainting models have been obsolete for a long time
daaaamn. if this works this is yet another powerful tool to my arsenal
and this is why I keep coming back to discord despite getting banned over 9000 times
There is an anime specific chat you can use if you want. Just saying
how the heck do you get that upscaler?
Can xformers be used if running Torch2.0?
let me see if I can find the download link again
it was pretty hard to find but the results are insane
too bad I make all nudes so I cant post examples
no
ignore them, they're a troll
Where on Vlad settings do I find xformers to enable?
yeah thats definitely not better
thats really not going well
what sampler did you use for them
how are you getting it so consistently to be the same character (or is it a specific named character from an actual anime)
lora
resulted in nsfw
was it some she venom stuff? 🤨
anime iron man
Art
Art
?
I am new
ok?
Help
same
I don’t know how to use it
use what
Discord
magic
real
stable diffusion :-)
this was supposed to be a dog
ah yes the demon dog
I tried using Ultimate Upscale with Tile Controlnet with ESRGSAN 4x but it gave the the error "could not find upscaler named SwinIR_4x, using None as a fallback"
hmm nice
yo
what model are you using?
i mean
the image
it looks really good
and almost realistic
could be a movie cover

you sent that already
its ai generated

dw
testing its realistication limits
ah yes ghost riders bicycle
what model
i meant the model name
send
its on civit
no 
i cant search
i cant out of embarrassment for realizing what it is meant for
My first attempt at spiderman on 2.1. No cherrypicking, I just hit generate and this was the first thing I got.
huh
what model
when i generated spider-man photos it gave me cartoony ones
Digital Diffusion, the model I made
Trained it on my 3070, yep. I made the dataset of 300 hand captioned images, and went throught lots of iterations and hardships to make both v1 and v2.
sure
idk... seems like he did
he published it
anyways jungle what prompt did you do to get that
spiderman
I mean if you scroll up far enough in #1045349359044280360 you'll see my process and testing
._.
ah yes hand foot
ew
i think digital diffusion is the only model that has loaded quickly for me :-)
It's only 2.4 gigs because I made it a fp16 model!
where the photo real all-stars? let's see some sanguine sunday photoreal
@smoky oak making progress on my tiling upscaler
i think i have an off by one error 😭
Looks like you de-upscaled lol
Can I send not-so-SFW images here?
If you have to ask, you better not 
What counts as a NSFW image and what does not?
#✍🏼|rules-and-tos should tell you everything you need
If you're questioning it then you probly could just default to it being nsfw lol
The girl has some big ol’ tohongerekoogers.
I believe public discords as a rule can't allow nsfw channels so it's not like anyone is trying to lay the hard fist of the law down just for fun
I assume if they are mostly covered it's fine
I mean, some (not all) of the stuff posted in #🍥|anime is a little close to nsfw, so take a look there for examples.
At least it used to be
Anyway here’s some food.
It's not but discord has rules on what a public server allows. And the people running the public server probably like that it stays that way: a public discord.
Stable Diffusion is really good at food pictures actually.
I say all this but want to clarify that I am not against tig ol bitties.
So am I.
But I’m wary about posting it here.
Still a lot of my generations are SFW.
These two are my latest ones.
I'm sure there are other popular discords where good nsfw art is appreciated. Haven't looked into it tho.
I'd imagine the nsfw subreddits linked on the stablediffusion subreddit have one that's populated, that's where I'd look first.
But there’s something wrong with their hair, maybe you could help me fix that?
The hair does not feel natural for some reason.
Are you talking about the saturation? The blurriness?
Mabey both
I think the saturation is unnatural but I'm not sure what you prompted to get it to be like that.
I think these are my all-time best generations.
At a distance they feel like they’re not made by AI.
Because of the skin?
So how do I make more realistic eyes?
Can't say that they're bad but something definitely seems off about them
they're fine
I agree, but I can’t really tell what exactly.
Here's where sytan comes in with his amazing realistic work
And then 30 more of his recent creations
No, probably more like paintings mixed with realism.
Also somehow AI still can’t make thin rimmed round glasses.
That’s why all my girls are glassless.
AI is for some reason very bad when it comes to thin rimmed round glasses.
Even Midjourney.
I've gotten a few of those on accident, I prefer the rectangular ones tho. Definitely needs cherry picking for glasses, I agree.
Like these kinds of glasses are for some reason very hard for AI at the moment.
yeah right, add the harry potter keyword
Yeah probably needs a lora or text embedding. I haven't cared too much to do so tho. The ones that I've cherry picked have been good enough
Can it be done with ControlNet?
Mabey??? I'd guess a canny processor of glasses. But then you'd have to set the location just right.
killer stuff. noice.
use cnet foo
its like the same dumbass crowd that says omgg ai cant draw hands
Use what now?
cnet foo
controlnet is pain
She looks too old.
huh? too old for what
i mean yeah she looks like a woman
thats a shocker to most of the weirdo shit anime people post
lmfaoo
finally someone else who cannot understand the obsession with alien-eyed anime girls
Im just glad to be attracted to regular, grown, heterosexual women. No furries no lolis no animemes no weird ass fetishes
yeah i'd do pretty much all the moms
keep that "she's too old" weirdo shit in your head, Purple
Project page for 'MultiDiffusion: Fusing Diffusion Paths for Controlled Image Generation.'
nothing makes me feel incompetent quite like SD2
water colour painting of samurai spider-man
Fix his right eye
i cant be bothered photoshopping it
Then why fuck with it then at all?
huh?
Just seems a bit stupid to just press buttons without the drive to fix it, but you do you.
Bit rude.. I'm sure he's just having fun
i redid it
and stopped it before it finished properly sampling
and it fixed the eye problem 
lol get over yourself
I'm confident in who I am, so I don't need to 'get over myself.'
it just seems a bit stupid to just press buttons in discord without the drive to communicate better, but you do you
Lmao I'm not gonna lambast someone for not fixing a tiny detail in a good generation before they post it to a discord
That's a little extreme.
Everyone knows it's easy to fix the small things
Of course but it is ALWAYS the little things that add up.
Seen it far too often with that stupid ass attitude of "I just can't be bothered". Then why in the hell did you do it to begin with?
fix all of my defects here
ok then do you want a better attitude?
i fucking cant because photoshop doesnt even open properly
and crashed like 5 times before i properly get to the menu
Piracy is bad, mmmkay.
sometimes your goals and theirs are not the same. it's weird but people are different and care about different things. like i sit here wondering why you bothered to be rude to them when you could have been constructive. why didn't you just fix the image for them? couldn't be arsed?
yeah calling them alien eyed is generous too, they look like kids its fucked up imo
same lmao and they call it cute names like loli
is it just me or does sd sometimes get better results when you stop the gen in the middle of the process
that's just an earlier step, hoss
Fix it for them? I can't be bothered. Hahaha. Besides it isn't mine to fix but if they can't be bothered that is a far cry from "my shit doesn't work right" or "I don't know how to using SD", etc...
😐
are you in need of a counselor? you seem like you're not doing too well right now.
whatever you're going through, you don't have to do it alone.
@tropic shell
45 steps vs 50 steps. the 45th one is definitely better in this case, but not always @hearty karma (look at the eyes)
???????????
Hey, ya'll, let's take a moment to remember to be respectful of everyone in chat, alright?
yeah, chill, guys, usually this is a cool place to see some cool photoreals and you all deliver
that super reaction is tight lol
I think that is the problem as those cool days are gone and this has turned mostly into just talking like general-chat room 2.
well if anyones got the skills to get it back on track its you guys
when did he have a cape?
never
i just wanted to give spidey a cape
all though technically spider-verse spidey had a cape 🤓
I was wondering, I see a lot of great work by people, but how do you learn the prompts? Is it by reverse engineering someone elses prompt and seeing what works? Trial and error, or is there some generalized knowledge like a wiki that goes through it. So far I've just been ripping off other prompts and using stuff that seemed to produce good quality stuff.
Youtube is your friend for that.
try the all-female indonesian spiderman reboot cover art movie cinematic kanji text poster
wonder if indonesian spiderman still comes all mutated
Ah gotcha. I'll see what stable diffusion content on youtube is. See if I can't learn how someone elses workflow goes.
much appreciated
welp we're going to find out
tons of vids now unlike 8 months ago. Good luck it is fun.
uhh nope..
i dont think its going to be malformed
well not that much atleast
i take it back
threw it into the upscaler and it took 57 minutes to make this and then i realised i did it in the wrong aspect
I like it lol
i have no idea why indonesian spiderman females are like that. in every single model
Oh, crap, lol. I really liked that one too.
greatest film ever "SPPIRDMAM'"
ah yes Ellamo atemis
theres too many elmos
i love Telleo Symomoi's movies though
you say too many elmos? that's actually a central plot point of the movie.
https://i.imgur.com/VYu9SLzh.jpg
still my jar of shrek prompt is the funniest image ive ever seen that i can share in this server
that toy would definitely turn on you in the night
hehe
add 'little big planet' to the keywords
this is normal for plms right?
i do like the peruvian fabric on the shirt
that was a mistake cause I still don't get what variation strength is or why there is even a variation seed as well?
strength I sort of get, don't under how it works under the hood, but the seed throws me
seed is the same as everywhere else. source of random numbers
if you put same img, prompt, seed. you get same variants
it's not 100% effective tbh but it's good enough for most use
well it's the same thing for both.
they have a seed that causes the random numbers to become the same
isn't that redundant though?
i have no idea why they would do it in two different places in whatever tool you're using, in my discord bot i work on, it just has one seed parameter for everything a person does
it is in automatic1111
those generalised tools trend toward overly configurable and expose all sorts of knobs no one needs to fiddle with
with vlad's it was on by default so was really messing with me.
yeah i'm assuming someone asked for that workflow as a possibility then
sometimes it is hard to say no to a feature request
probably but it is crazy. change the master one there is no need now strength is awesome
what models do yall use?
digital diffusion v2
lol noice. i like this guy haha
i think she was actually in Samurai Champloo
blades almost there
I can't darken the mood with those little guys they are too cheerful
some of those are just there to walk through a museum of torment eg. sd14
oh is that Christopher Walken?
or atleast an attempt of him
oh the british fellow from Top Gear, Richard Hammond
lol @ dreamlike diffusion's idea of tickle me elmo
l0l
I once wanted MJ to make the VRAM monster
essentially a green cookie monster but instead of eating cookies it eats memory chips
oh wow, ELMO SAMURAI MOVIE
multiple 4090s are likely better. they're definitely faster than the A100, especially with pytorch.compile()
Ampere vs Ada
"How stylized do you want your Elmo Samurai?"
"Yes."
the results are so much better when you go widescreen
daddy?
Elmo on the rough side of town
elmo in a cosmic episode of sesame street manifesting a golden sunburst orb of black hole antimatter while emma stands to the side looking horrified as her own cosmic sunburst orb manifests unknowingly
i really like these
ah so i managed to finally recreate the tree of life from The Fountain movie, kinda
nice work. the elmos are hilarious
why is Jack Black so difficult
ty 
yeah just a portrait is easy but put him in an action scene
get him riding a bear in the swiss alps and i will be impressed
hmm
mine makes his face weird
up close he's ok
it's a lot of work
this is hilarious the longer i look the funnier it is
lmfao hes so cute
weird al does well in 1885 as a sheriff but i love the friggen guitar rifle thing
i had one briefly show a 1980s ski jacket style saddle on the bear, in the snow. it's hard to reproduce that
I am sure I could get it to look a lot better with more time, but I wanted to do it really fast lol
it did give JB 3 legs
thats the bear, he decided to wear some uggs on the trip lol
Now that I am sure I could do pretty well
ooh nice
is that via inpainting?
or just YOLO

this one decided to make it more 1800s style once i'm in 512x512 lol
does your model know Robin Williams too?
I already have one but I'm standardizing my dreambooths to either be 100 sample images or 350
@oak osprey








