#🏞|general-with-images
1 messages · Page 87 of 1
24gb vram for 850 sounds worth imo
it's probably broken lol
at that I would just pay the few hundred more for a 4090
That's why I never liked bidding because you wait 7-10 days to be sniped the last few seconds.
i'm going to caption BG-20k
@sterile templeThanks a lot, now I have to divert my focus completely to audio until my momentary hyper fixation wears off again lmao
Actually working on that believe it or not lol
i do believe it
goals for my album
you kids shape the future
It would be a subset of colorbass/futurebass on paper I think
when i was young we didn't even have dubstep, we had to pretend
lmfao
back then we just made annoying noises and the adults thought we were weird. well, we showed them
I synthesized these sounds yesterday after a friend distracted me and made me mess up lol
did you try Bark yet?
very weird atonal sounds lol
I have no need honestly, maybe later down the line
Oh, I also made this cursed shit at request of a friend a couple days ago 😅
He said he wanted fart based Jersey Club lmao
it's perfection
I have so much audio on my PC, OMG
i feel the warm blasts radiating from my speaker cone, subwoofer, and it's like i'm there
29.7 GB just raw audio
well, thats my sampled audio, I have wayyyy more in rendered audio
I would like to see something new from the kids, feel like I haven't got anything new yet, just rehashed stuff, no offense 😄 still good music coming out all the time, just doesn't feel like a new genre
oh man, I could show you so many new genres lmao
mainstream stuff is always the same stuff recycled, but the underground is burning hot out of the oven
Like the jazz inspired minimal glitch subset of bass music that just recently popped up
All of the subsets of melodic bass music are only a few years old now
And then the new age wave of next gen Dance Dubstep
Future color bass (my beloved)
Midtempo dance
@sterile temple Curious to see what you think about any of these more modern genres
Mmmm, and a new hybrid genre
Industrial future rock
I have yet to meet somebody that doesn't haha
Its just a damn good song in general
I like the beat, I like industrial, I liked the lyrics since I like old school Gangsta rap
I loved Sonny Moore
Tearout is MEAN air lol
Its as close to noise as you can get before just pure noise, which is what deathstep is lol
(there is a discord notif in there towards the beginning)
Another flavour of tearout lol
This bass I made would actually be considered tearout
its got that nasty crunch and the spectral overtones
i like how when i wrote "it all started with the big bang", the audio transformer added a ding there like an 80s science documentary
contextual intelligence lol
@oak ospreyWhat are the requirements for Bark?
sorry, back was in a work meeting 😄
same as SD, roughly
dope, thanks
my discord bot has a tts called Bark that's kind of wild and makes crazy stuff sometimes
as an experiment i tried having it read a news article to me, it took like 30 minutes to render 
that's one out of like 5 recordings it made
@smoky oak Did you see this? https://github.com/richrobber2/canvas-zoom
I switched the above (no other changes) to RealismEngine
I'll have to grab my headphones, sounds very tinny through my laptop speakers 😄
ah damn, that was not what I was expecting 😅
got my hopes up for something very different lol
Well, I downloaded it as I like inpainting but hate the stupid canvas
Just a taste of some newer genres that I am a fan of, of course there are 100's more
i had it make an episode of seinfeld
rcnzreemeet
I am trying this prompt to see which ones not to use
My old one prmj
We really don't have that many models on 2.1
is this making all that music?
@sterile templeif you are able to, I highly recommend giving this music video a watch and listen. It has the best audio visualizer I have personalyl ever seen for the type of synesthesia I have
https://www.youtube.com/watch?v=lDfSdA5-PRE
This genre is minimalistic color bass, with some old melodic dubstep flourishes
Stream / Download: https://hypeddit.com/intermolecularlullaby
Lyrics:
Am I code on a screen? Or a fleshy machine?
It doesn’t matter now, I’m somewhere in between
A human inability, a lifelong burden
Inconceivable fragility, but I must keep working
I wish I could fly, and breathe through four dimensions
Don’t wanna say goodbye, before I share ...
noo
no, that was just music I was sending lmao
Oh. I wanna make AI music. 😄
This one is kinda cute
Didja see that one-click AI face replacement tool?
I couldn't get it to work right
figured more for 1.5 or something
although the faces I do are not directly looking at me, or human
@smoky oak some good stuff. But still not 'fresh' if you get me 😄 E.g. kind of reminds me of Sophie's work and dubstep
SOPHIE — Faceshopping (Official Video)
Stream & download here: http://futr.cl/faceshopping
Directed by SOPHIE
Co-directed by Aaron Chan
Produced by Imogene Strauss
Production Co.: Industry Plant
Creative Direction: Nicholas Harwood and Eric Wrenn
Animation Design: Esteban Diacono
Additional Character Design: Adam Gerber
Sparkles: Cabbibo
Edite...
some of these genres are just a few months old tho ._.
And yes, SOPHIE is the grandmother of most of future genres in general
lol to think of sophie as the grandmother, things move fast 😄
Here, have this one
this one is CERTAINLY alternative to most bass music lol
I have heard id classified as minimalist analog bass music
LOUD START
That one I just sent is all its own thing
Liked the intro the rest no
It should also be noted that underscores is also part of the original hyper pop movement, along with SOPHIE
2018 is ancient by bass music standards haha
I do love this new trend of more minimalist middle sections to really contrast with the massive first and second drops
that's nice, I like when they have more real instruments mixed in
I am the exact opposite personally, real instruments are way too boring for me, but I do think they do a great job of implementing them
Virtual riot actually has a song that is a 10 year comparison of the differences between 2012 melodic dubstep, and 2022 future color bass
7323```
captioned 7.2k so far 🤓
this dataset is pretty damn good
it has no salient objects, eg. no humans/animals
@sterile templeI highly recommend giving this a listen, cause it is a direct comparison between the old and now defunct melodic dubstep genre back in 2012, an dthe new and thriving future color bass genre
it really shows how we have moved from massive stacked and messy sounds, to much more tonal and organic shaped sounds with complex spectral characteristics and freform shapes
2012 sounds really digital, but modern color bass is starting to put a layer of organic sound back in that I am in love with
the sounds are a lot more wattery and almost... smooth
Again, I have synesthesia, so maybe I am picking up on stuff that is not actualyl there lol
these sounds are what made me hyped for SD, as weird as that sounds lmao
that song actually inspired my work on this demo for my album lol
pretty good, I'd like if the melody was a bit louder and the snares had more tap to them. i'm a sucker for melody though
my album intro demo has more melody to it in the second drop/cross over part, and god am I proud of it lol
This demo is like a year old, so please excuse the shortcomings it has in the sound design and mixing/mastering
the second part with all of the melodic elements layering together is one of my favorite things I have ever made
imagine how much better it would sound with real drums
These are the melodic layers I made put together on their own, there are 6 different melodies ontop of each other
there is an arp, a background atmosphere that is fluttering, the main bass melody, the swing melody, and the additional reverb fills to swell the melodies
Even a year later, I am so happy with those layers lmao
i think that would be better if there was a little less noise, more empty in that back to get that contrast, if there is a lot of noise it's like a grey and black rather than black and white, i'm not making any sense 😄
Even rendered it out like it was in a bathroom cause one of my friends challenged me to do it lol
using an actual convolution mapping if a bathroom lol
that's nicer, but not sure about the bathroom reverb
so its pretty close to physically accurate lol
lol this dataset is about 22gb for just half of it
it nailed the sony logo haha
Aye
I am surprised it didn't have that little tm in a circle at the end
I genuinely wonder if SD could get sued for presenting non trademarked logos lol
I was thinking the same
Hello! I've been working with NovelAI for about a month now, but have seen so many amazing StableDiffusion pieces that I wanted to try my hand at it. I'm trying to fully replace NovelAI by getting better with making anime pics, but every pic is coming out super grainy for some reason. I googled some good tags, have tried a couple different themes, etc. Is there maybe a checkpoint people suggest? I picked one at random so maybe that's my issue.
Hell yeah, I'd love to get you off of a service and onto local software, got any examples of your problems? I will try my best to help
try reducing CFG
Abyss Orange Mix 3 or 4. Check it out.
Here's an example. It's only 512x512, but the blue sky behind her has like, a pattern on it even without zooming
You need a VAE
Where do I get VAE?
since you are doing anime, I can send one I know works well
doing no vae just uses the default ema one that the model comes with i think
you have to use something like mse 840k
Apologies, I just started today. Been googling non stop so my terminology is lacking a lot.
All good, all good haha
and then let me show you how to use it
I am assuming you are using A1111?
Yeah
alright, just a sec
all those candles around that paper 
Did this just now.
you will wanna go into the settings, user interface, and then "Quicksettings list", there you will wanna enable sd_vae
this will allow you to change VAE's easily
it will make another dropdown for you to select it
Let me restart the webui. Just a sec. I put the VAE in
don't yet
add this setting, apply it, then restart
then it will be all good to go :>
Will do. Comp is a bit slow so will post progress update when I can. 🙂
all good, when you are back into the generation UI, just pop dwon that SD VAE menu and select the anime one I gave you. Generations should look MUCH better from then on
lol @ uhd in negative
It definitely came out a lot cleaner. Thank you very much! Took me longer than it should have cause the first I enabled VAE in the quicksettings, I didn't hit apply lol.
niceeee!
lol, bluewillow v4 gave this for a 'female next to a sphere'
so glad to see it working!
make sure to check out other models as well! Anime has to be the single most common type of AI model in the community lol
Yeah for sure lol. I tried a realistic one first and didn't have the blurry lack-of-VAE issue so the anime threw a wrench in the works. Is that common that every checkpoint or medium has it's own oddities that you have to VAE or some other fix away?
Some models need VAE's, some don't. Always make sure to read the description to see if they suggest one or have one included. Generally speaking, I use the same VAE on all models and have had no problems
This is the community go to
I use it on basically all models I use
you can just place it in the same location and press the little reload next to the VAE selector in SD
I am actually downloading an anime model for the first time since I installed SD 5 months ago haha
Also, quick question. For inline drawing (bad hands etc), what settings should I have?
I would recommend only masked, as its faster, and I would use that same mask blur
I kinda get most of them, but might as well check
denoise is how much it pays attention to the content beneath. 0 denoise is identical, 1 denoise is 100% different, so you adjust that depending on how much you need it to change
I'll change it to "only masked" cause yeah, I always saw it doing the whole thing. Wonder why it's not set to that by default.
there are many layers to inpainting
If i Want generate module and images what can i do ?!
many many different uses
@pastel prismlet me know if you have any other questions, I am always down to help, especially when it comes to getting people off of AI subscription services haha
tho NovelAI is a lot less trashy/scummy than MidJourney, at least from what I know
we do support dynamic shapes but haven't performance tuned that - so hopefully sometime in the future (6mo+ ) we will switch to fully dynamic shapes and we wont have to recompile. We support dynamic shapes now but just haven't spent time on performance with it.
oh, do you work on Tensor RT?
Will do. Will keep messing with it, but I think all my questions have been answered now. I need more VRAM I think lol. That might be my biggest issue. Can't resize to 2048 atm.
what GPU do you have? I may be able to help
FANTASTIC news
We do support Lora Training from the WebUI on linux. On Windows we are using Pytorch 2.1 nightly and they haven't landed all the Python 3.11 + Dynamo patches yet
NVIDIA GeForce GTX 1660 lol. It's a hand-me-down. Has been fine so far, up until generating art.
Is there any information on why TensorRT is so limited in dimension?
The max you can do is 840x840 at BS 1 with tensor RT, no upscale support, no controlnet support, nothing.
While it is massively faster, it is also kinda unusably limited
What kind of speed difference happens from gen and train? Oh, and Lora or does it do Locon/Lycoris/Loha as well since those really took over from the base Lora
1024 test worked and I think I see my issue. It totally remade the picture. Is there a way to just have it rescale without doing too much computation or adding/changing anything?
Have you looked into ultimate upscale? It could allow you to get to 2048x2048 no problem on a 6GB GPU
yeah!
can you send me the image in question?
we haven't tested the Locon/Lycors/Lohas -- but it should be possible. Our next focus is to get this speed on mobile
but yeah lots of features to add 😄
No interest of that for me, but best of luck.
ah yes true different customers pulling in different directions
btw this https://github.com/pytorch/pytorch/pull/94243 is the stack of patches you need for Lora
but i hope they land them all soon
"stack"? :/
I got rid of the hand pretzel and the elephant person in the background. Just wanna get it up from 512 now, and I'll have all the tools I had with NovelAI.
you can upscale for sure
I had "Just Resize" chosen, but this is clearly not "Just resize"
are you able to hop in a call by chance? maybe one of the channels? I can help much better if I can talk and walk you through some additional addons that can do exactly what you want super easy
you don't have to speak, its just easier if I am able to
sure thing
alright, i will be right back, brb
I'm in diffuser4
LOL
what model is that?
Anyone know of a good brand of PSU cable to adapt 4 of the old to the new single GPU power connector?
cablemod?
I only know of theirs for the angle piece. Btw, those suckers are catching fire too.
at the pc store I usually go to, looks like all the kolink and some of the phantek stuff is sold out, so I'm guessing they are the more sought after brands
I see a lot of ones I wouldn't trust if they had 8 connectors as they are too thin, and flat, to handle the current
@pastel prismOop, you died haha
shit, i think i got the flu lmao
is that a thing still
i thought we stamped that out
influenza in postage stamp form
when you select the wrong controlnet processor. 😄
Not sure what to make of this one as it is drawn looking and not drawn looking at the same time.
it's like old school catalog vibes
All of the ones I posted are the same only I am changing styles
Another drawn and realism at the same time
Put a filename in the field
No arm, no chocolate. No filename/full path, no conversion.
I found out which language the ai writes in when adding it to some images: Ithkuil
Wow, did you make this image with StableDreamer?
All of the images I created look, ugly, deformed, radically inaccurate
Then generate one on your pc.
Yo guys .. any stable diffusion expert here ?
I want to know how do i add something that is missing
Like the end upper head of someone face ?
inpainting
Yah but in paiting is to add on the pictue
I want to add a missing upper head outside my frame
Like adding the end of this bunny ears
oh sorry, there is outpainting mk2 on script you can work with that
or you can add blank part manually to image and use inpainting also
Fym wdym
What
So the script called mk2 ?
Tf
With the update thats way less likely to happen now, since you can select a group and it puts in both automatically 
Ty
on pcpartpicker those are the only 4090s currently available. It keeps shrinking as the chip shortage hits due to Jensen removing production from a lot of the AD10x models for Hopper.
no one buys them, so why keep making them
you don't click "Buy" literally every day lmao
there's no chip shortage, there's a consumer shortage
A Mad Villain
Guys how can I get result similar to these images... If possible, well these were also generated using SD
I'm having a lot of issues inpainting a hand. It keeps coming out messed up. Are there any tips? Should I maybe inpaint the whole arm so it has more context? I have a lot of negative tags like "fused fingers" etc but it keeps coming out messed up. I'm using the recommended VAE too.
put somewhere in the negative prompt: (perfect anatomy):1.4
or add weight like this ti the tags in the negative prompt
if it's auto1111's webui, then the numbers should also be inside the ()
but in the end, no prompt alone will be enough to cure the finger curse :P
im in a1111, and the multiplication out side of the () works just fine -_-
they do? I need to check it, if they do, then I know what I have confused about... for over 5 months. I hope you're wrong because that's a lot of time being wrong otherwise for me! :P
im on the dev build, architecture might be different.
Tried a couple with (perfect anatomy):1.4 and (perfect anatomy:1.4). Didn't help idk. Just buried it in fabric. That works I guess.
weird, you can try using hires.
it fixes this kind of issues.
I believe I'm right, you need to have them inside the parenthesis. Try using very high numbers and see if you get nonsense in one, both or none of them
Like: (powersaw:2.5)
Hires? Like low wage workers to mass gen hands for me? That's crazy... so crazy it just might work...
no, it's an extension, it fixes duplicates and even hands, and automatically upscales generations when active.
also they only show them inside here
like it gets rid of the conjoined twins problem
yeah, the high res fix is the best one in my opinion as it creates less, but does not remove, the weird hands issue
Also fixes duplicates
yeah, 99.9% of the time. It's so good at it that I forgot the issue even existed! :D
but sadly, it's not 99.9% for hands/fingers 😭
and bad hands can be fixed with proper negative prompting, and IF that doesn't work, then just inpaint.
it does make a dramatic difference tho
it's defiantly better than without it
I've never seen, nor had a prompt, negative or otherwise be able to remove 99% of bad hands/fingers. I've tried some embeddings last year, but they didn't work neither. Haven't tried loras yet though :D
if nothing fixes your issue and you have an image you like, than you can just inpaint and pick the best generations
yeah, but I, 1. want the ai to not make it in the first place, and 2. am too lazy to inpaint ;P
fair enough, the main con about loras is that they effect performance.
but otherwise, loras should be able to make hand more precise. but nothing is 100%, but it is possible to get somewhat close.
just like there is no such thing as an 100% pure matter.
didn't know that about loras :O
But my fix for bad fingers I've learned over the months, is to not have them in the image in the first place. ;P
let me guess, 2.1 model? i can see the watermark
no it's a 1.5 model + bad luck >:I
luck should never be a factor when it comes to computing, i'd call it bad precision.
well, that is still, just an impurity. rarely happens
impurity should never be a factor when it comes to computing, i'd call it bad precision. ;)
bruv
bruv?
aww, discord doesn't have a jump rope emoji
too bad
have cat instead :D
Does a1111 work with an amd gpu, and if not, what should someone be using instead of a1111?
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs
Is all I know, sadly :/
o-o
why you do that :(
@smoky oak huggingface is saying the terminal snr stuff is kinda overhyped
thats my daughter
car
@vocal steppe
10000 frames video from the prompter
"Parallel Worlds"
Here is some data for you, to try with deforum
Don't expand lol, download it
Deforumation can help to target some area, and 'navigate' into a choosen area, never tried yet
you have rtx freaking 5090?

Generated by a friend, when i still was using the rx580. Would take 3 days for me, 3-4h with a 3090
how long it took
I can do it now
i have
meh, many shit can happens during generation, focus on the quality
like zooming in loop into trees, things like that
I just realized that 1k frames would be 10 seconds of video
= wasted time
nah, 1000 / 15 fps, 66 seconds
100fps, omg no
god damn
30 max for u
15 is fine, and use cadence 3 or 4, regarding the prompt
Also add anti-blur (between 0.5 and 1) if zoom is not a 0
I forgor
I had to restart
and all my settings got reset cuz of error
so i forgor cuz of speedrun
to get it back
Some 20sec videos require huge amount of work
I wasn't able to use 3D with the RX580 too
320 € for the card
a 3060 is totally fine, specially with incoming updates
link meee
but 12gb, not 8gb, watch out
yeah, not less !
Or you'll struggle at some point
how about rtx 4080 48gb
48, yeah, for 20K$ 
You still have one, lucky you
you need a 4090 with torch2 and a compiled unet
other websites will make it on browser
and it will take so much less time
or like applications
so u dont have to dowloand all this stuff
It took 2 full days to convert my models to .trt
huh?
glad i use a batch mode lmao, launched in 5 clicks
how many models there were
48 here
If you need a batchmode for the a1111 tensorrt extension, to convert models, ping me lol
Spent 2 days on this sneaky sh!t lol
kekw
second disk is corrupted
brahhhh
686 gb going to as-
931*
ima do a bathing time
ew i use python scripts not automatic1111's stinky code

it's a prompt, a smiling attractive woman, sitting in a chair, on the beach, ocean visible in the background, soft focus, 8K, UHD, bokeh
from top to bottom each row is from 0 CFG to 9 CFG, though i'm not sure if it completed properly
from left to right each column is 0.0 to 1.0 CFG rescaling
each image there is 1152x768
Okkay, what i thought. But thought it was related to the conversion difference lol
https://tripleback.net/public/grid_vanilla/woman_grid.png
https://tripleback.net/public/grid_vanilla/man_grid.png
https://tripleback.net/public/grid_vanilla/wizard_grid.png
https://tripleback.net/public/grid_vanilla/macro_grid.png
https://tripleback.net/public/grid_vanilla/bicycle_grid.png
https://tripleback.net/public/grid_vanilla/alien_grid.png
https://tripleback.net/public/grid_vanilla/menn_grid.png
https://tripleback.net/public/grid_vanilla/target_grid.png
https://tripleback.net/public/grid_vanilla/child_grid.png
https://tripleback.net/public/grid_vanilla/micro_grid.png
https://tripleback.net/public/grid_vanilla/gecko_grid.png
https://tripleback.net/public/grid_vanilla/robot_grid.png
https://tripleback.net/public/grid_vanilla/knight_grid.png
https://tripleback.net/public/grid_vanilla/cosmic_grid.png
https://tripleback.net/public/grid_vanilla/wizarddd_grid.png
I'm not aware about bypassing the limitations, sadly
base 2.1 stable diffusion testing all of the terminal SNR configs
here's a scaled down version of the CFG chart
3d motions are weirds, i only use 2D in deforum
Can I get some upvote for my feature request to be implemented in Automatic's WebUI ? https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/11044
Why did the nerf the Stable Diffusion training data ? output is now worse than the very early versions of stable diffusion
its now miles behind midjourney
2.1 is a base model and midjourney is an amalgamation of fine-tuned models
no duh midjourney is better
StableDiffusion XL is not very capable... I think they had to nerf their training data because of artists whinging about copyright.
it's also not finished
I have a weird question. Can anyone render a horseshoe? First image - real for reference. I can't seem to get any model to come close. The second is an example of what all the generated images look like, no matter what model, steps, guidence...
Use controlnet
Good point
Odd shapes, or specific ones I have to do that.
Swords, and knives, and even firearms
It's just odd to me. You'd think it would be a common object in training data. Who knows.. I'm still learning.
A lot of stuff is discarded. Now if all the images that are usable were not discarded far more of this we would be able to easily get.
@smoky oak i tried training using ddim as a sampler and everything got so whack within like 6 steps
NaNs from the unet
so that sucks 😦
still working on my realism model
@smoky oak SD2.1 is having fun figuring out how vaping works
1.5 can do vaping pretty good, at least with the models I use
sometimes a little too well lol
i'm amazed that 2.1 picked it up within like 400 steps and started doing it pretty damn good
it likes to make most depictions of smoke around men as vaping lol
mine can make cigarette smoking too, i have an image of toddlers lighting up ahahaha
very 1950s work shirts on them, all miniature
the confidence
i remember my first time vaping
Makes sense because I have never walked into a cloud of shit from a vaper that wasn't a male.
I am glad to have no such memory lol. I will never smoke or do anything like that
way too bad for you. I am happy how I am, not under the influence lol
it was a joke, like, "i remember my first time <doing x>" memes where the person is awful at it
ah lol
I am proud that I am not a substance abuser, I know too many friends and family that are, and they are gonna live much shorter lives cause of it ._.
Considering how this world is going that is more of a blessing than a curse. The curse of a vampire is being around for so long.
@smoky oak by the way i've found out with terminal snr on 2.1 and the new cfg rescaling code it works out that the higher the CFG, the better the result. so i'm doing anywhere from 9.2 to 12.0 and a rescale of 0.3, and it seems to create the fidelity of a lower CFG image, with the prompt comprehension of the higher CFG.
btw, @smoky oak Did you know that the 5090 may actually be made by Intel? ffs.
down -> higher CFG, from 0 to 9
to the right -> higher CFG rescale, from 0.0 to 1.0
it looks like depending on the prompt it can work out to use really high rescale values at really high CFGs, it looks like it makes her really small and the beach chair really large, like i've seen at some tourist places
but that doesn't work for every single prompt outcome like using a conservative CFG rescale around 0.3, which is like the 4th column there
https://tripleback.net/public/woman_trailing_true_stacked.png there's a 6,400 x whatever pixel image with the full 1152x768 images in the grid
like, look at how the digital art wizard pans out from "no snr fix" applied, to "all the snr fix" applied
these are all DDIM by the way. and this is when you apply the new fix for DDIM to sample from the last timestep
scantily-clad women in underwear, especially with protruding ta-tas is NSFW, and falls under rule #4. There are servers out there where this is acceptable, but not in our Christian minecraft server 
what ?
Christian minecraft server ?????
Its a joke. Regardless, no nsfw around here 
this is nsfw ?
I certainly wouldn't want to be caught looking at it while I'm at work, so yeah.
isn't nsfw for more yabai stuff
I'm busy writing a masters thesis on the hatsune miku role. I offered my warning, and will be on my way 
damn
bro really said christian minecraft server
bro thinks everyone is christian
plus where the minecraft at lmao
🤓
hey don't make fun of me! 
bro srsly said that 💀
what did I say 
how to generate photo into ghibli anime tyle
anyone
i need except SD any other websits
\
I am in a bit of a silli goofi mood
hello goofy!
damn my internet is slow tonight
I could watch 8 16k streams at the same time if it was as fast as her
modem rebooted, much betta!
5G broadband
ahhh
you ever tried this, setting the resize from seed that is much bigger than the actual image?
oh, I didn't even know that was a thing
not sure if I'm imagining it or it actually produces more detailed images
extra is if you wanna retain a similar composition of the same seed while changing resolutions
nvm im dumb. misunderstood what you were saying
woo! new king gizzard and the lizard wizard video just dropped 🤩
what websites other than lexica can I go look at strictly AI art
instagram? #ai #aiart
thank you but there is another site like lexica that I can't remember the name of and was wondering if there were more like that.
@umbral lantern
for :
prompt: masterpiece, best quality, award winning photo, A 20 yers-old witch under the moon light, witch attributes, magic, surnatural, epic, complex background, hdr, (hyperdetailed, intricate details:1.10), cinematic, muted colors, foggy bg, hyper detailed, hyper realistic, 8k negative_prompt: (deformed, distorted, disfigured:1.3), dark, poorly drawn, ugly, out of the frame, (bad anatomy, wrong anatomy, extra limb, extra legs:1.6), missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation
guidance_scale: 6.0
sampler: DPM++ 2M Karras
styles: Beautiful Portrait
extra_net: detailmaker
facefix: CodeFormer
batch: 2
Then, you can find like hundreds of extensions, to do anything with your pic, but you'll need to mix several extensions to get what you want.
SD is a lot of work, nothing magic. It won't take Artists jobs lmao
*As default, i'm using Lyriel, the last pic
ohh thx for all that information 😄
i think you misunderstand what i wrote in the other channel, i am looking for image2txt models like clip interrogator
anyways, this information is useful for me too hahaha, in the bag of models that you have, do you know any alternatives to RealisticVision for photo realistic images?
use salesforce blip
Me when my wifi doesnt work
Awww man, my bad. I don't use any caption model
I use the default one, and then a text2text gpt2 model for the prompt
https://github.com/imrayya/stable-diffusion-webui-Prompt_Generator
Clone or install this as extension, for a new tab in the web-ui.
.
Can be installed out of the web-ui (still in a standalone web-ui, tho) here :
https://huggingface.co/Gustavosta/MagicPrompt-Stable-Diffusion
It's literally 2 clicks in my discord bot
The first link give the choice between 2 models, from FredZhang7 and Gustavosta
nat geo teachin 2.1 what a leopard gecko is, finally
That creature doesnt exist
when god gives you a shitty guitar, play shitty music
imagine you run into this guy at the grocery store in the back dark corner of the meat section and he's like, "hey, kid" and you're like "what, mister?" and he's like "i have some candy for you" and it's full-size Snickers bars and other good stuff
why do you just post fucked up stuff in here @dry cosmos
?
@oak osprey
not a leopard that shaped like gecko
hmm same issue as base 2.1
thinks a chameleon is a leopard gecko
is that 1.5?
Eublepharis macularius
i love how they look like tiny derposaurus rex
mine makes a bird fish thing when i put that scientific name in
cursed af
training on national geographic is only helping so much
@smoky oak Still SMH at Nvidia. I have to agree that looking at these specs side by side the 4060/ti has to really been for a 4050/ti instead.
Momma
ahhh i'm loving the kodachrome training results here
i think it's starting to be biased toward Singapore for city skylines instead of NYC
interesting direction the prompt for a man playing guitar at night, illuminated by lanterns is taking
teaching it vaping is going about as well as i thought 😁
Is there an easy way to have it up details while it's upping quality through resizing? I started with this, and basically just got a high quality of the same garbage face
Went from 512x512 to 2560x2560 and was hoping it would figure out the face sucked while it was redoing everything.
what upscaler?
god damn
Controlnet Tiles
original 64x64, ended up at 1024x1024
@dark flare can you please explain what generated images mean from this?
does it mean they were artificially generated using the training images fed into the algo?
they are completely different from the ones that were used to train it?
https://github.com/NVlabs/stylegan
why do you have to ping and ask random people about stuff they've never used before
he talked about ffhq before, so I wanted to ask him
ffhq is a Nvidia thing and they are, 1024x1024 used as real world reference images.
@smoky oak https://www.youtube.com/watch?v=RNrXlABU7ME
"Continuum" out now:
VINYL + CD: http://bit.ly/1WbtZnW
iTunes: http://apple.co/1XacWRW
Spotify: http://spoti.fi/1RVeVCF
Credits:
Music: Shook - Infinite from the album "Continuum"
http://shookmusic.com
http://instagram.com/shookmusic
http://facebook.com/shookshookshook
Video produced, animated and directed by Gustaf Holtenas: http://gustaf...
you might dig this song
finally have a "Do Not Create Grid" option for xyz plot. God that makes me happy lol.
Frequently used it to iterate thousands of combinations of things and previously had to end it early so the webui wouldn't sputter and die trying to make a multiple gigapixel image.
you could have just switched to the dev build a few months ago, it exist for a while now
That'd been nice to know when i asked about it a few months ago -_-
Either way. have it now so i'm happy
i was here then, where did you ask, i would have replied
i'm fairly positive i deleted it after no response, but i'm checking lol
here it is
In this video we take a look at some of the incredible variation possible due to recessive alleles within the genome of Astrophytum asterias and use it as an illustration of how natural selection might work to create new species and forms in the environment.
Filmed in the Greenhouse of @madamcacti (IG), we get a great look at some of the incred...
how model training worrks
wow nice one by sdxl
Yall see this?
They actually scan too.
dude hasn't actually released the controlnet yet, assuming he will at all.
@dense tapir glad to see NVIDIA finally pulling their heads out of their asses and sending higher VRAM GPU's, but of course they are probably gonna be much more expensive, cause why would they release a good product
Well, they are going to be 1 quarter late AND they are going to be manufactured by Intel.
I just find that so much of a conflict of interest when you are making your rival's boards.
Wonder why the ditch for Intel. They aren't particularly known for their cutting edge fab, considering it's several processes behind on density
Yeah, for sure
But are they really your rival if they are lining your pockets? Maybe that's the justification
Ahhh, alright, that makes more sense
The point is they are rivals, or supposed to be, when you make the same product and are supposed to be trying to gain market share to total dominance.
With this revelation I have given up all hope on Battlemage, Celestial, and Druid.
Short of Jensen dying they are the leader and not one other company will be able to stop them, and I know Lisa Su is the good little Asian wifey type loving being number 2. If that were not the case then she is inept, which she isn't, as there have been so many missed opportunities to take market share this time.
I said this with Ryzen in 2016 that they like being number 2 as it is a comfy spot to be in.
Intel's issue is they are hurting VERY badly right now so they see the cash offered and decided to lie down with the dogs not caring if they catch fleas or not.
To me, and I hope to others, this sends a signal that they aren't even sure of their own GPU.
I'm flattered that you think I know anything 😄 but pseudoterminalx was right.
Verra nass! That's been in my Spotify faves for a few years now, neat to see it mentioned anywhere ... even neater to see it has a cool video
what in the hell happened in this chat
full moon?
spam and scam links out the ass, WTF
@tropic shell please get this POS out of here
scam links, spam, racism, all sorts of shit
@dry crow
Here are the mods, FFS
@glossy herald
Where in the actual fuck are the mods, seriously??
@quiet current @dire crypt
@brazen vessel@atomic hemlock@raven mango@proud dagger@fallen mulch@solid bridge@soft drum@topaz aspen@prisma iron
Can literally any of you guys take care of this? this is fucking ridiculous, where are the damn mods
it's been dealt with
why are they still messaging then?
One of them was dealt with, and the other?
thank you
jesus christ, 2 hour mod response
next generation Ai server doesn't have a bot to auto mute people who send over 50 images in a minute :/
an old granny
@dense tapir You on, by chance?
idk what happened but for reference it's 2 am in america and 9 am in england so most people are asleep or just waking up
(also most of the people you pinged are developers, not moderators)
I too am in America, but man, there were over 100 scam links in here for more than 2 hours ._.
1/ this is not the channel for bot nor how bots work.
2/ I'll give you the benefits of the doubt.... but please don't prompt for nsfw.
Yeah, cause I reported over 40 messages, and also tried to use the ticket system, but it didn't work lmao
the support ticket bot isn't working, that probably would have been useful here
Guess I'll go apply to be a mod 
I tried months ago, don't worry, they never get back to you lol
The admins have a biased love for me, I'm sure it'll work out.
You pinged everyone but me 
gotem
I had the same with a mod, but its been like 4 months, and not even a word on if theyt have seen it. I was just told that they are "busy right now"
I had no idea what "Stable Star" meant, so I just pinged everybody else
Does this server really have no bot to take care of people sending 100+ messages a minute?
pinged the devs which are beneath the stable star role
If you want better bots ping Kevi, he's in charge of that stuff last I surmised 
AAAAND kevi goes offline
LMAO
if he made the mod support ticket bot, then I am not too impressed 😅
p sure he manages the stabledreamer
I tried to use it like 5 times, and nothing went through
Sometimes they slip through, you really gotta calm down tho, were not your enemy here
Were doing our best 
I am at a loss of words... 100's of messages across like 2 hours in one of the main chats in the server isn't "slipping through" ._.
Support ticket doesn't start auto banning people the moment you submit a ticket ^^. Yes mods are still only human (for now ™️) sometime they're busy, they work, they sleep.
🍿
You did the correct thing by using the report system first and pinging mod only afterwards.
thats not what I mean, it literally said "The application didn't respond" or whatever when I tried to submit
oh, good to know.
I tried like 10+ times, I think maybe one went through, but then it linked to a channel that said "unknown" before disappearing a few second later
I tried like an hour or 2 ago when the first round of over 50 scam links was sent too lmao
reacted to all of them with the ⚠️ too
well, not all of them... but yeah
I just created a ticket without issue
I applied to be a moderator or guide like 4 months ago specifically for this reason, cause I am active at this time. I was even told I should apply by one of the mods here
whoa, I was told by a mod and an admin that I should apply. You tellin me they won't get back to me for 4 months?
That's insane
I honestly have no idea what to say, at the time it kept saying the application was not responding, and then one went through before disappearing a couple second later
thats how long its been for me at least. Even talked to them, to which they said they were "busy", and weren't currently reviewing more staff, while also saying they are understaffed ._.
I even corrected my talk and stuff to be more professional, not cursing, being much nicer, in hopes that that would make a good impression, but I kinda just gave upo after 3 months passed
Ah, see I took the other route. I just said "how to apply 4 mod".
Watch my name turn pink while I'm asleep.
I went through the whole guideline documents, and even submitted my forms and stuff, but 0 response. I talked to other people who have been in the same boat with this
bot stopping to smell the roses
gigabot for the last 2 hours ^ lmao
no idea, I don't even know what a stable star is
thats why I didn't ping, I though it was just like a VIP member
I think it is the community guide of the month or smthn
ahhh, if thats the case then that makes more sense
I'm just glad ti finally got resolved. Glad I don't have to look at that gross shit they were sending anymore
I'll tell in DMs

that, the plan is to have a neutral lighting
like mor eof the the nook and cranny shade
teh current controlnet I already have is canny and open pose
I assume, I need to img to img it, but I dont know a good prompt or setting for it
i made a movie poster with ai!
original pics
Id play around with different controlnets and see what works best, open pose is a must tho id guess
Most of us are in different timezones, hence no replies
Also regarding tickets, most of the time it works, there's just a bug sometimes it doesn't. Reporting a message however will always be successful
and the prompt will be, a girl with brown skin and short hair?
like a description of the image itself?
Depends on the model youre using, if you are using an anime model you should look into boru tags
I use midjourney model and/or the general 768-v-ema
but i'll play around thank you
oh lord, those are both pretty scuffed models. You would see much better results from better finetuned models on 1.5 specifically, but still applicable on 2.1 as well
is it ok if u post the link to those model??
i plan to download it
Just look on civit.ai
There are thousands for free on there
It's the go to site, along with hugging face
I wanted a dark image, but this is just silly :P
ok ok ty ty
Overachieving lol
that's me, not the overachieving part, the lol part ;P
Haha
yes, those are some of the sounds, how did you know?! :O
I was about to say my name is SOUND design, but I changed it to be more satanic lol
I was trying to get one of those detailed wrinkly skinned old man images people seem to enjoy. But with 1.5 of course :P
people do love them wrinkles
but I might have added slightly too many "sharp" prompts :P
you never know what might be hiding in them! Wait...that sounded so weird... :P
😮
but truth be told, I like details on everything, wrinkles aren't everything in life. It's just what you get from life :P
i dunno what to prompt, so mix a bunch of wildcards it is
So, if you don't mind me asking, what exactly are wild cards? @sterile temple
one of the best things ever! :D
you create a text file containing prompts on each line
They are used for dynamic prompts, really useful tool
looks something like this in A1111 with the extension
how do you write something with dunders?
How interestinnggggg
So it's like... A random number generator?
Or well, a random tag selector?
yeah
That sounds extremely useful
like you can have a file with a bunch of hair colours, and write character with the wildcard haircolour and it will pick a random line from the wildcard file with the same name
How coooolllll
The anthro model I have has dozens of wildcard files with it haha
I have just been using them to better understand how to prompt things lol
now you can pop them in your wildcards folder and use them
Ive got a zip here with all my custom made wildcards that I used, mostly made with ChatGPT
don't mind if I do joink 🥰
Amazing, they have almost a full MB of wild cards lol
Feel free, thats what I dropped them for 
1000's of tags
Oh, been meaning to ask
What gets you the stable society tag, and what exactly... Is it?
same prompt, wildcard magic
Last time I checked, the model I use has these wildcards:
if you tick the combinatorial generation checkbox, it will generate an image for every combination of wildcards, but if you have lots be careful if you add multiple wildcards in the same prompt
50 hue consistent colors
851 species/categorical characters
1972 art styles
554 articles of clothing/outfits/armor
368 predefined jobs and statuses
332 personal descriptors
40 quality alters
102 setting prefixes
82 concept suffixes
26 defined surface materials
268 predefined structures
146 predefined countries
26 fantasy worlds
180 direct locations
111 nature biotopes/biomes
195 "themeatic" settings
405 general/generic locations
11 body poses
30 limb poses
8 look poses
21 general poses
62 emoji triggered emotions
31 text triggered emotions
120 action styles
6 image crop adjusters
103 accessories/addons
181 object/character adjectives
335 feral animal species
502 types of food
16 nature objects
51 singular/misc objects
439 general group/misc objects
28 shape modifiers
19 art themes
106 vehicles
17 ambiance enhancers
27 weather ambiance
35 camera modifiers
10 camera brands
21 camera compositions
76 camera models
59 color pairs
And lots more I can't share 😅
@timid oar here's the one I use
a DODECAPETALOUS INSTRENGTHEN of a DOWLAND WAWRO in a UNPREFIGURED UNHEALTHIER REPAGINATES CLAUDE MONET ANDREA KOWCH MARC-AURÈLE DE FOY SUZOR-COTÉ FRANÇOIS BOUCHER
Is how I sometimes "play" with all the wildcards, just add a bunch and then let it go wild. Every word un uppercase is created by a wildcard, so, most of them :P
You get that from winning a contest
with the addition of those wildcards I'm up to almost 400 :D
Ahaaa, I see
Knowing how to use wild cards is going to be amazing. Especially because I try and make interesting and unique anthro characters all the time, so now I can have it randomly select from the 851 species it knows :>
I started using them for rolling random artist names. I had over 10,000 names so testing each of them with having to write them each time was a godsend
what an intense light show
they are, they were the first thing I added to the webui the moment it was possible.
you can also use wildcards in dynamic prompts
Thats what Im doing
All of these terms I have never used, haha
you haven't used dynamic prompts either like {cat|dog|bird} ?
Nope I only use wildcards
I remember it was with [] and not {} :O
Does it do one and then the other?
it randomly picks one
you can even do {2$$dog|bird|cat} and it will randomly pick 2 or whatever number
Ahhhh, so wildcards randomly pick one, and you can use dynamic prompting to have it randomly pick one of the wildcards?
yeah
So THATS what's up with the example prompts of the line up sections of the model I use :0
Is it consistently that format? Or doe the number signs equal the mount you wanna pick?
Like, would it be 3$$$ or 3$$
Oh man, so you can do 1-3 for it to pick and random number randomly?
That's so cool
Ahhhh, so that's why it's not in the base documentation
I was going to say, I swear I thought I read everything to do with prompting in the base documentation
it should be, it's one of the older most useful extensions
just remember to place the wildcards in the dynamic prompt wildcard's folder and not the "wildcards extension" folder :P
Thanks for the heads up, I have like... 100 wild card lists to put in there lmao
I'm not gonna lie, some of my best prompts came from being lazy and using wildcards to "randomly roll" them 😶🌫️




