#🏞|general-with-images
1 messages · Page 88 of 1
I also added every wildcard as a finished __wildcard-name__ "style" in auto1111's webui "style" foldout. for quicker access :P
and if you want to be extra lazy:
Add all the wildcards.txt files you want into a folder. Then open the CMD and go to the folder with the wildcards and then write dir /b > files.txt
It will then create a new text file named files.txt with the name of each file in that folder. Each wildcard.txt in this case and then you can, with the help of VSC edit them into their true __wildcard-name__ form :D
too bad a wildcard can't include a wildcard call to run though :(
yeah, I think so, but can't remember the name
anyone has used Controlnet for scannable barcode anyone have any idea about it? https://www.reddit.com/r/StableDiffusion/comments/141hg9x/controlnet_for_qr_code/
why would you need a controlnet? couldn't you just use the qr code as a mask?
Someone else did it without an own controlnet model but i cant replicate what he did so far and from the comments it looks like no one else did either 
https://www.reddit.com/r/StableDiffusion/comments/1436nqv/my_attempt_on_qr_code/
I just put all the text files into that folder
after you install the extension it should be there
ok, so you make a folder called wildcards?
cause there is no wildcards folder for me natively
did you install the dynamic prompts extension?
yeah
there is that folder.
stable-diffusion-webui\extensions\sd-dynamic-prompts
In the tab for the extension you can download some pre-made wildcards, that should make a folder
go back one
create one in that folder
yes
I'm assuming there was a spammer
2 of them, sending scam links, racism, and generally disgusting images and gifs
thank youuuu <3
Oh, goodie!
no problemo
last question, can I have a folder structure? or do I need to have every text file accessible inside the first level of the folder?
in the normal wildcard's one, then yes, it will find wildcards in folders...wait... does it really or am I thinking of something else now 🤔
just try it, if it doesn't work, then you can read in the console that it didn't find a wildcard with that name
yeah, fair haha
__test/ingredient__
alright, lets see if I can do this
I made a name structure instead of a folder structure so I could easier see all the wildcards without having to go back and forth inside folders.
alright yeah, so this wildcard system they give with this model has 217 individual wildcards
That looks like Utah, haha
Also, I have not played with wildcards yet. I should. Probably. I was thinking about it.
do you need to add .txt?
Most of the things I make are made with my own research/knowledge
what i dont like is how it spam cmd, or is it only me when Webui is starting?
spam? What's that? ;P
does it to me too lol
i thought everything died
i mean polute or how in english
pollute
don't look behind the curtain
oh sicckkk, it does a different species each time when you do batch sizes
any way to make it do the same tag for one batch?
I don't use batches so, I dunno
I just closed mine because there are some workers outside setting up scaffolding to fix the roof. How did you know?!
sunny, was that your attempt at red shoes? lol
go to settings -> wildcards
It's the only setting there :P
Maybe he's the one on the roof
as long as he know what he's doing, then it's all fine and good and fine :P
oh, for me its under dynamic prompting, and there are several settings
oh well lol
but if he tries to steal my precious prompts then I will...not do that much actually because I don't mind it nor am I brave enough to fight anyone! :P
nice, I might need to re-enable the dynamic prompt, I'm just using the default one because the dynamic one broke about...6 months ago... hm, maybe that's enough time for a fix? :P
we all know when they finally kiss, it's usually on the roof
the roof that's coming off? ... that sounded differently from what I wanted it to
wow, this model is now generating species I have never even seen!
This took a dynamic turn. Sounds like we are now in a construction romance movie, a Tale Of Two Shingles, where two rival companies finally set aside their differences 🍿
yeah i'd probably watch that
Kobold, aardwolf, kemonoko, old english sheep dog, sisu dragon, rat, potoo owl, spotted hyena, and asriel dreemurr lmao
its a mixture of species and specific characters it seems
Oh heck yeah!
what about a kangodile wearing green sequin overalls
"Hot and ready shingles in your areai, just looking to get nailed down and plastered!"
||I am so sorry, I had to||
Shingles in your area.
I missed the G 
such a gutter joke
And how many obnoxious ads there were in the 1990's
Better get those gutters cleaned!
in the 1990's????? Still 100% a thing today lmao
great horned owl, albatross, lucario, riolu, flamingo, nick wilde, least weasel, nergigante, artic rabbit
There's a certain bright, flashy aesthetic about the 1990's that hurt your soul.
I love the flamingo
Yes, I am but an old, withering lady hahahaha
and I'm older than the Chernobyl accident... y-yay?
I can see a lot of the influences of certain artists.
I remember that on TV
bear, hare, worgen, sheepdog, roxanne wolf, MLP pony, spyro, dusk lycanroc, akita
it knows so many species lol
You know, she looks like she is on a metal band cover, so we are good.
didn't have access to those luxery things then :P
and this is just the cut list. It has a whole seperate wildcard for every single pokemon, and most digimon lol
anybody here old enough to remember when the world was black and white?
Rip. I was very young, and didn't understand what was happening, only that it was huge
yes, most of my old phots are black and white because it was cheaper. And I also remember having a rotary phone...also a landline phone for that matter! :P
super long curly cord for the phone handset
ring-tailed cat, yeti, wooly demon, sawsbuck, alaskan malamute, procyonid, dinosaur, retriever, valstrax
these species are so cool
gonna be honest and say I was just over one year old when it happened, but still it's technically true! :P
Black and white movies are interesting to watch. There's also radio dramas, audio books, etc, in the public domain that can be fascinating, and great sources of inspiration
Rotary phones are too fun
BORINNNGGGG
Now I can AI generate radio dramas :p
which is... kinda crazy to think about-
Omg I love them. And they kinda look like fursuits haha
hamster cyndi laupers
That gets tangled in everything, including yourself, but it's great for your hands.
I remember having one of those 20 seconds shake-backup cd players. I never had cassettes as I didn't start listening to much music before my teen years
"Stop talking grandpa, the SyNtHeSiZeD cOnCiOuS iS sPeAkInG"
I've done that
At least the scripts
Pretty fun
help, my furniture is floating now!
HAMSTER LARPING is the new hamster dance
WAIT
dabidi ba didi dou dou
HAMSTER BOYS
why did I read like like Axel F I'm blue
yo listen up
We did the whole record player thing. My grandma had one, too, so every time we were over, she'd bring out her record collection and play while we were over. I can't say when I got into cassettes, but it was very limited. Sound generation is fascinating, and with Harmon, the ability to create more sounds, and music--It's fascinating for sure.
I made a LoRA for Na'vi lol
Love the clocks and aesthetic
listens up Yessss?
so proud of how it came out
learning all the new things are so hard sometimes. I was like 27 when I used a mobile phone for the first time because I didn't see a point.
AND I also thought that mobile phones would vanish as one of those "trendy" things...yeah...
well, it didn't get the idea, The Caretaker is an album about an illness
but yeah pretty cool
i'm blue
skidwurd
yes, I did the sound effects in my head as well :P
You know, the first mobile phone I had was a big chonky one, with a green screen. It was like a BRICK. Not a Nokia, but this thing was massive. It was my dad's old work phone, which had roaming, so its antennae was from a bygone era. It's kind of hilarious.
https://www.youtube.com/watch?v=eM5DWw3J4BI just in case @smoky oak doesn't know what we are talking about
Eiffel 65 - Blue (Flume Remix)
Read your comments & slapped together this bad boi 4 ur listening pleasure.
LONG LIVE EIFFEL 65.
Subscribe to the Flume YouTube channel: https://flu.me/YouTube
Flume apparel: http://flu.me/Store
Follow Flume
Facebook: http://flu.me/facebook
Instagram: http://flu.me/Instagram
Twitter: http://flu.me/twitter
YouT...
I am stupid, blue is by eiffel 65, not axel f lmao
my grandpa had one of those, or a bigger(?) version you had in your car.
I switched to one of those touch screen phones 2 years ago, never had one before that
The thing I love the most about AI is just tying really old stuff with new stuff. You know, throwing a record player into a space suit kind of stuff.
Ah, illness, my friend
Sounds familiar--the Album
I still remember when he won like triple j unearthed high or something, just a scrawny teenager
I can't unhear that song, y'know
dementia, that album is the most horrific shit i heard
Also, IM LIVING IN A BUBBLE
I can't unhear all the sawing, and drilling, and stomping, and shouting, and tossing before breaking. This, and the next week, will be so, sooo much fun
Dementia is a sad thing--it's hard to watch someone go through it
yeah, it's even more horrific when you have an 80% chance of getting it when you get older..
Those car things were massive. But, you know, now that I think on it, they'rd prolly not as annoying as how they used to make stick shifts.
my sister's first car was an old volvo. that had a car phone
The best thing about forgetting something is that you won't remember it
I know, because I might have a bad memory but at least I don't have a bad memory
It's strange and confusing for everyone watching, at first. My last living older relative was just diagnosed with it. But, that is why we enjoy what we have now, eh? Time is precious.
That's like my ADHD talking right there hahaha
Also, I didn't get a smart phone til like..........2012? I think? Lol
actually in dementia, it's really common when going to the post-awareness stages to have PTSD, most of my family had it.
Yeah, they can.
no one in my closest family have/had it, it's more cancer here... :(
https://www.youtube.com/watch?v=ZLcXIrTwPUM @smoky oak flume covering another aussie goat
Flume covers Bag Raiders 'Shooting Stars' featuring Toro y Moi for triple j's Like A Version. Dancers: Guilherme & Jack from Thunder Down Under and Chestbrah @said_sergeyevich
Subscribe: http://bit.ly/triplejsubscribe
Like A Version Interviews: http://bit.ly/BehindLAV
Like A Version on Music Streaming: http://abcmusic.lnk.to/LAVplaylist
Like A...
I figure if anything happens to me, it is only par for the course
🥂 😎
So, enjoy life--it's beautiful!
for me it's almost ALL my family when getting older get alzhimer's. so for me it's very likely to get it when older =[
yeah, who'd want to live forever anyway? The older one get, the more life turns from going to bed to wake up. While later it's waking up so you can sleep later :P
make sure to keep up your B12, especially if you're vegan/vegetarian
I wake up, shove all the pills in my face, then go back to sleep LOL
Vitamin D as well
Surprisingly easy to be deficitient in
sleeping is one of my greatest skills! Eating is also up there.
lol I have like a 4 course vitamin meal in the morning
sunny your pfp is the best
That's the future, isn't it? Just a meal made of vitamins. Or you take the pill, and the water, and BAM! MEAL!
Also, ty!
does coffee count as vitamins? if yes, then yeah, I also have them. If no... then no... :O
there might be minerals in the water, so I say that counts
in the end, we only drink two types of liquid.
Water, and dirty water
yeah, there is NO such thing as pure H2O, nothing can be 100% pure.
have you had your morning brown?
It might come as a surprise that a song about brown slaps so hard. More comedic bops are now streaming on Aunty Donna's Big Ol' House Of Fun, only on Netflix.
Subscribe: https://bit.ly/2Kncxw6
About Netflix Is A Joke:
The official hub of Netflix stand-up, comedy series, films, and all things funny — curated by the world’s most advanced algorit...
If it's hard water, there's plenty of minerals lol
yep, just the fact that the water is not vacuumed means it has plenty of minerals, even distilled water from the biggest chemistry labs isn't 100% pure.
Indeed. I've lived in a lot of places, with a lot of different...shall we say taste instead of quality? For water? Well water, sanitation, etc.
ooo nom noms!
well time for me to go to bed. the quicker I get to sleep the quicker I can wake up and take more vitamins!
Im trying abstract stuff right now, but its not as crazy as i want it to be yet
yeah she's gonna rest for a long ass time after that
I love the second one
The placement of the plants is great and the pose is just, "Whatchu gonna do about it?"
Deal with it 
pose
Getting somewhere 
Its def abstract 
person in a bubble, fish in a bubble
in the air, who can say!?
love the colors and moon/mood; very powerful. The clouds are great--such sweeping motions.
thank m8
np
got more for you 
Ooooooo! Kind of like a space helmet, but softer aesthetic. I love all the little diffused light/bubbles coming to the surface.
Her expression, though :[
I mean i wouldnt be happy either in this position 
Maybe she has gills.
I also really like this one, came out with a frame too
Ohhh, yes, I like. It's got a vintage but also sleek design to it at the same time. A classic, nostalgic feeling.
imagine adding hat to negative prompt puts a beanie, then beanie to helmet 
because just because the definition of a beanie being also a hat type, the ai might not think the same as the dictionary :P
and headwear is the over-arching term I believe for "stuff on your head that isn't your hair" :P
i added haircut to positive promt and seems to be fine
It is called a style which I like num 1, and num 2 we don't use stupid vaes with 2.1.
because there aren't any for it? :P
We don't need them as it is made with them and why we have a yaml. Superior to the way 1.5 worked it.
yeah, all the 20 people using 2.1 says that. The rest don't ;P
Then it pleases me to be among the elite.
You do that, me, and the rest will enjoy great images that we like :D
GA doesn't know how VAEs work with 2.1. all of the VAEs work for both models.
You do you. Some people prefer McDonald's too which is how we see 1.5. Good for the masses.
idk why the 1.5 vs 2.1 garbage. i give up on those arguments. both are extremely flawed models in their own way
I mostly don't understand how someone think art can be better than something else. Like how 2.1 is better when I've yet to see a single good image that can't be overwhelmingly better using 1.5 models. It's like it's a personal opinion of what looks great or something ;)
Ahem, you do realize art is subjective, right?
yeah, if I could, I'd rather pay an artist than use the ai if I could get an artist to make what I want.
well im not really making excuses but if you put my images in that group of comparison, just don't. my images are attempts at producing a good quality result from a single model without textual inversion, LoRA, text embeds, or anything like that.
so any garbage from me is unlikely to be what you'd get if you actually just wanted a good result no matter how many layers of processing it needs
yes, that's what I just said. "It's like it's a personal opinion of what looks great or something."
and if you do what i'm doing, both 1.5 and 2.1 are just awful at realism. neither can do anything very well, they're generally capable but like, "jack of all trades, master of none"
that's the experiment, to me, is how to retain that general ability of wide range and still make each thing it produces, "good"
Yes, then why do you always have that tone? FFS, you just had it again. We "20" prefer what we prefer. end of story.
i have like 200 checkpoints i've trained and i wish i had disk space to keep them all, or the patience to download them to the place where i do have the disk space. because each one has their own strength
yeah, it's more about what we enjoy to see. My personal opinion is that 2.1, in any image I've seen, here or otherwise, hasn't been above the quality that a model with 1.5 hasn't done better. But that's just that, it's my opinion. I got a billion reason why X isn't better than Y, just like everyone else has different views on it. I'm just happy people are happy making art :D
early on, it can do mountain bikers very well. later on, it cannot do them at all. but it can do portrait faces much better 😦
I mostly joke away when people are rude. Can't really say anything else than to try and make a joke about it. :P
You are also one, you said this, who doesn't like realism with wrinkles, and other real things. For non real there is 1.5 for anything else there is 2.1
curse of the diffusion! :P
yeah, 2.1 is better at realism, or so I thought until yesterday when I saw 1.5 images that were better. So my thoughts have been updated to remove even that from what I think 2.1 is better at. Now 2.1 only has the "possibility" to have more details than 1.5. Rest is worse in my eyes in 2.1.
I honestly feel, Walach, that you come here to troll and bash anything non 1.5. Every single time 2.x is brought up you will chime in defending 1.5 when 1.5 was never even brought up.
2.1-768 is using just a totally different loss strategy internally, called v_prediction. it is capable of things that 1.5 is not
hey, people who want to ride, should ride, even if they don't have all the limbs! ;P
I mostly answer people when I see people write stuff like "we don't use stupid vaes with 2.1."
I just want to balanse it out :)
it's okay Walach, breathe
hehe, now you made me conscious of my own breathing! :P
manual mode active!
vae's was a bandage that 2.0 took care of, therefore not needed and only there due to an oversight in 1.5. therefore stupid.
some of the kodachrome test images are really great, but i did feed it like 7,000 images of that
what oversight? I thought vae's were an extra, like embeddings and loras, etc :O
google maps, 1954
tiny car :3
nope, it was to make up for a deficiency in the original training they said. You can see with and without the devs were right.
so many real photos have motion blur to them
if you have a link to where I can read it I'd take a look :D
screw that you go take a look as they said it when they released the big first one.
it was a HUGE improvement.
if you try to get the old "style" then it's probably just how the ai think some of the "items" look like as movement always got blurry from older cameras before all this new digital tech came around :P
Anyway, I am done playing with the Walach for the day, so have a good one.
I most often just brush things away as a person lying if they don't want to help. Saying something is easy. Showing it is harder. :(
i used about 7,000 images from the Cushman collection of kodachrome slides
he wasn't a fabulous photographer, just took a lot of photos and labeled them well
yeah, I believe labeling will be one of the steps that needs to be looked over again, and again, and again to slowly get better quality. Add more details upon detals, etc
is there a model like 50-100-200gb of trained data?
VAE has been a part of stable diffusion since the beginning
they released an upgraded VAE, which was fine-tuned starting from the original VAE's checkpoint, on a filtered dataset
the way GA just misinforms people agan and again is frustrating, and he does so from an elitist standpoint
yeah, I know that, I've read about them. But I never saw them as "mistake," instead I saw them as enchantments, like embeddings, fine-tuning, and loras
VAE isn't to address a "deficiency in training", he acts like he is a god among men who knows how to train better than anyone else, but as far as i know, he has never produced a model
they're a fundamental part of this process
I've been around ai art, or rather, SD for quite a while and while I do not like to "show off" I do believe I know a lot about it.
when you set VAE to "none" inside A1111, it just uses the one that's inside the model's ckpt
now i'm wondering if i should fine-tune 2.1's
somehow.
well, and if the model wasn't cooked with a VAE, you get jpeg artifacts on generations
no matter how much i tune on real faces they still suck
they get better but at the cost of image quality overall
it's the model's architecture, i say when we begin to train SDXL it would fix most issues.
you mean architecture as in, the way they trained the base layers?
this is what I get from realism from 1.5. And I don't even know how to prompt for realism :P
the base model's
to me that looks like unreal engine 6
it DOES follow prompts very well, but makes somewhat grainy images, like, you can tell they were made with AI
hehe probably true, I dislike most ai art photos because they don't "look right" so might the other way around when I try and make one myself :P
the grainy images are because of the flawed schedulers
this is with the fixes from this morning in diffusers
an oil painting
probably true, A1111 UI isn't originally intended for 2.1.
but with XL it seems to fix that issue
yeah, and it look nothing like realism in my eyes. It's more of a "cut out newspaper image and paste it into something else." I don't mean any disrespect, I just want to say what I honestly see when looking at them.
That's why art is subjective, else I hope I'd be the one with the correct view! ;P
you mean the edges?
yeah
I can't unsee the "cut-off parts" sorry :(
you throw away all of the good progress
that's truly annoying
look at his skin, the hair, his eyes, his fur
1.5 can fool more people into thinking that the image was made by a human, and 2.1 knows how to follow prompts somewhat better.
ignore the weird pin on his lapel lmao
it's like, what's the word, uncanny valley? My eyes just zooms in instantly on it. I can't control or "not think of it" :(
it'll still be used as marketing material on a website.
what about this guy, struggling to figure out how to vape correctly at a coffee shop
i see what you mean, but i'd say that 1.5 is less uncanny than 2.1. and 2.1 can follow prompts better.
they need to teach their customers how it works
Maybe, but everytime I see something extremely well made. It's 99.9% made with 1.5. I've not seen a good 2.1 image this year, if my bad memory hasn't failed me. But again, "a good image for me might not be a good image for someone else and reverse"
and ban this guy
can 1.5 make a skyline of Singapore without adding the friggen Empire State Building and times square?
does this look like unreal engine as well?
SDXL doesn't seem to have any of 1.5's or 2.1's disadvantages
looks like air brush painting
darn! :P
SDXL doesn't have disadvantages? check out the feedback channel ahahaha
i love the deformed stuff it makes but it has prompt coherence issues and i wonder why
like i know it's not finished training yet but i don't know what they're going to have to do to fix it, and i want to know, and i hope they tell us. but i don't know, the papers SAI releases on their process are lacklustre
well, at least for the most part. it fucks up less than previous models
I just want good images. If the technical stuff is better only matters if the images from it are also better. :P
@cyan snow and Artius v2.1 is possibly more coherent than SDXL and was a model merge done by a community member, but SDXL is being done by a team on a huge cluster. it's just interesting to see how even the experts struggle with this stuff. it's not a solved problem, and i worry that statements like yours are over-hyping SDXL the way Deepfloyd IF was overhyped and the way 2.0 was overhyped
both were considered 'failures' by the community because the community has, on-average, smoothbrain takes on things. even if it's groundbreaking and amazing results, look at all the complaints already in the feedback channel, on the most mundane of issues. like the grass in a particular image not being perfect. trying to fix every last little issue in its output will mean it is never released
I think the true answer lies in-between. We need a basic "unit" that we fine-tune, and fine-tune. Then create a new "basic unit v2" from it and redo all the fine-tuning again until perfection :D
we need really good fine-tuning tools but they don't exist yet
everyone is currently guessing, and i don't think SAI has those answers for us there either or they'd have done it already too
they'll come one day. Until then people seem content to use what is. Probably more work, but you have to start somewhere :D
well things like LoRAs have come along to fulfil gaps in the abilities of the base model. and maybe i just need to start using them and being okay with their presence in my workflow. but i still can't drop the dream just yet of making a good base model with generalised fine-tuning that can do dreamlike photoreal
like i've seen amazing things with just 2,000 steps of training on the base 2.1 with all of the SNR fixes! but there's a greedy part of me that can't let that just be how much i train it.
i need it to absorb more data than that, and i'm not sure how to do that effectively
I'd say you should. If it's helpful, then it's great, be it embeddings, stupid vae's, loras, or anything else. If it gets better results, then they are great tools in my opinion :D
small fine-tuning jobs seem to work much better than very large ones
2k steps with fixed SNR schedule allows it to make scenes like this, a man in a park illuminated by floodlights at nighttime
he got off his bike! :P
it just literally cannot do that for 2.0-base, 2.1-base or 1.5
I won't pretend to understand that. All I can see, is the thing in the image. :P
try and make an object on a pure black background or a pure white background
offset noise can kind of make images darker but it can't help making them brighter
it's not the model learning new shades of contrast, it's just forcing it into dark/high-contrast all the time
this new approach instead allows it to make a more varied brightness, rather than fixing it to that one style
Can't say I've noticed any issues with that. Other than the "light" problem in 2.1 that is
rescaling the CFG is a new approach to stop overexposed results
I only see overexposure in 2.1 images more of a default thing than in 1.5 :/
all of the cfg configs for the new 2.1 schedules
I wonder how well can we train it because if we can't train it, or it is as hard to train, or worse, than 2.1 currently is for me, and a lot of others, it will be fun for a month then time to wander back from whence we came.
you can see how rescaling the cfg too far makes the image very neutral
rescaling it not at all, very high contrast, potentially over-exposed
yeah, I know of the CFG, most things with the art ai hasn't really changed that much from when I learned it last year. Other than new stuff like loras that is :P
aesthetically speaking, the researchers discovered that around columns 5 to 6 were the most commonly accepted
this is new
cfg is the old thing, that is along the left-most column, from top at 0 cfg to the bottom at 9 cfg. this was the way we could adjust prior
pick a row from the left most column. and then move right. that is rescaling that chosen CFG value by 0.0 to 1.0 using the new feature
that's why there's kind of a triangle of similarity at the lowest left. but it's quite interesting because i assumed maybe a 9.0 CFG with 0.5 rescale would = a 4.5 CFG. it does not
it has higher quality than either
hm, I don't think I follow you, sorry :(
Sounds like the same thing as I know of it, but with different numbers :/
i don't know how the heck the math works internally or maybe i could explain better 
Might be that I mostly use words like "don't go too high or you get bad results" and not the technical numbers :P
It's not your way to explain it, it's me just wanting to see good art. I don't, as an example, pick my brush based on the thickness of the hairs, I pick the one my brain has been taught makes the correct strokes :D
I like the 1.5 model more, but mostly because of the sharpness of it :P
same, same. 2.1 is like, you can tell it's AI generated
and with 1.5 it looks more like a render or something
If I'm to be honest, I try and think that I'm much better at prompting using 1.5 and probably have a huge bias for it. Those 2.1 images aren't bad, but if it's a "this or that" then the 1.5 would win it in my opinion.
In the end, there's a reason why 1.5 is so, so much larger than 2.1, and it isn't the simple excuse some people make. And it's not a bad thing to be "worse" than something else. It's just that, for me, 2.1 hasn't been able to "be better" than 1.5 thousands of fine-tuned models toward one specific thing. 2.1 might be better if there was enough work on it. But as it is now, there isn't. :(
The most photo real image 2.1 I've seen is not even close to some I've seen people make in 1.5. But those are only the ones I've seen. Not made myself, but when there just aren't any shining 2.1 example I've seen it's so hard to believe in 2.1 future :(
my prediction is, that sdxl would just revive this community, but idk, only time will tell
Well, I don't think they need to have that to "revive" the community. They just need to make it so people want to be here :P
Across the 10 discords I use 4 died 100% so I removed them and the other six or near death. Hell, this one I can see half a day, at times, before any images are posted.
(not as a jab at their work, but in a "this is like MJ, come over and have lots of fun" way)
A lot of the fun people to chill with that used to be here with me I still talk to and they hung up SD
" I only use it occasionally when I have an idea or if I want to make a label for my other work"
That was dm'd me earlier this morning.
He did good work in it too
to be fair, SDXL is somewhat similar to MJ. not only with capabilities, but also in the way the bot functions
Yeah, there's a couple people from last year I haven't seen around here. It's sad, but at least there's some left :D
Yep, I spotted that immediately.
oh well, time for my late shift work. But I don't want to work, I want to sleep >:I
Here's the perfect photo I need from the ai :P
this is actually terrifying
i recognize the string pattern
did chatGPT just try to rickroll me?
@cyan snow
wtf, am i being targeted by chatgpt
I am perplexed
there's a lot of people arguing online whether this is AI generated or not
for context this is the album cover for the next King Gizzard and the Lizard Wizard album. People have referenced a supposed artist painting it on instagram tho i couldn't find any source. The background has a lot of AI artifacts tho, especially those towers if you look at it closely, for me it's a dead seller
what are yall thinking?
I tried to do inpainting to replace eyecolor... can someone give me tips for what I can do to make it not horrible
but I'm trying to make them have red eyes
Oh, lol.
This is the original
Like this
yeah something like that
I'm trying to figure out how to do it right
I did
text->Image to get my image
"eyes"+image-> mask that covers only the eyes using SAM
"red eyes"+image+mask-> inpaint-> goofy looking picture with cross eyes??
SAM?
Never used it
it's not important, all it does is outputs a mask you can use in Inpaint
for example, if you did "person"
you can see it automatically creates an inpaint mask you can use
Is that the same as the background remover?
a mask is just a black and white file. When you use inpaint, it will only paint inside the white part
(or the black part, depending on what you want)
No matter as you see without it I had no issues and I like to keep as many tools away from me as I can.
Yeah, I know what a mask is I am just into doing it in other ways
Yeah not sure why my eyes ended up weird... what prompt did you use?
I just said "red eyes", should I say "person with red eyes" or something?
Ok I'll try a few more things
I'm doing everything from code because I want to do a lot of images, instead of clicking a bunch
Diffusion doesn't take well to automation I found.
Lowering the noise for 55% seems to have helped a lot
Hmm it's a little weird
Mmm, nah still is crosseyed sometimes, but I'll keep working on it
"beautiful woman with red eyes" fixes it
Much better
whut lol why do you always nay-say everything even when it's demonstrated to work pretty well
generating regularization images is highly automated
I feel like the perspective of your snake...frog.....fish matches the destruction of the city. There's a story going on with these three images.
Also, ooo, apple!
Oh yeah I see what you mean 😸 like the collapse of a futuristic society and then the evolution of a new species
You want red eyes? Or you are trying for blue?
Yes, yes, indeed
Or the opposite!
The fish did it--classic kaiju tale!
I started a big run of gens before going to bed. That apple was the first image, and probably the best one out of all of them
I like that dude's style.... Forgot his name--he's very good at trees. Watercolor, I think, and pencil. I have written...somewhere....in the black hole that is/are my drives/file structures. What was your prompt?
Apple of my eye - Eldritch entity holding a monstrous apple, its gaze reflecting in the apple's surface, style influenced by the uncanny and grotesque elements in H. R. Giger's art, dark, decaying orchard with fearful onlookers.
here are two others from a similar prompt
The eye is either....from Mordor, or a Keese 🤣
that creature with the big eye reminds me of something, but I can't place it
That's some good looking bark. The apple placement is kind of hilarious on the left but the gnarled branches are great.
The head shape? I can think of a few silhouettes for all of them
it does some styles really well with great details, and then I try to do something anime really basic and it's all glitchy and pixelated 😄
Trying for arbitrary eye color, in this case red. Mainly because there are tricks for a consistent face (celebrity names) but the eye colors are subject to vary, especially if you change their clothes color
Well, you know, it's just that...low res and jpg artifact you added as a positive, naturally
I have 0 clue what model you be using, but here's my thoughts, lol: if you want a specific eye color, using blue and green as a negative, (and sometimes, as both positives, both separate, and apart) may help you. Consider adding hair color. You can also specify certain types of noses, lip shapes, etc. SD does "memorize everything"--faces aren't 100% symmetrical--so think about things that may not be making things, symmetrical, and consider adding them into your prompting.
Face is solved
Just not eye color
Hair color I can do, you just have to use the right words and hope it doesn't bleed into your clothes (or vice versa)
But eyes are really hard to control
So that's why I'm doing this eyes inpainting thing
man just load it up into like, Gimp, and split the eyes into their own layer and colour-correct
overthinking it like mad
strawberry blonde is a classic one
I might do this (but programming)
and bowl cut
Hair color is pretty easy, if you use the right words you can get good control of exact shades
"strawberry blonde" being dangerous because it adds strawberries 🤣
it handles other two word tokens well, some not so much
had to do that before using words like lemon yellow too
See, someone should make a Lora trained on hex codes
So you don't have to fiddle with this silliness
bowl cut
interesting, because it generates mostly boys for bowl cuts, so it picked up on that bias, but then it's like, you wanna bowl too right?
Let me...restate, then, lol
If the eye color isn't working
You can use inpainting, but if you use certain colors/words, like those stated above, it'll improve it
Maybe I do want a bowl. A bowl of food.
Also, it's close, so very close.
On that hand.
The extra long middle finger is hilarious
i made that one after christmas, with the amazing analog diffusion model. they don't make em like they used to
The color is beautiful and the gradient is fantastic. The hand is great, too
That hair belongs in a commercial
It did a really good job with the face
💀
But honestly
Considering what i prompt
This is innocent one.
Gordon Ramsay
gordon ramsay, in his kitchen, yelling at his food for being too dry
@wispy nest 🤣
in my model, Jack Black is dog
it required a lot of inpainting to get that, the model won't spit something like that out easily 😄
@smoky oak why did you not tell me DDIM is so good
horsecycle
me: Salvador Dali
AI: 
I could use Nfsw models and will get it in 10 minutes of attemps.
Not in exactly your style
Obviosly
@sterile temple By the way, can i ask you something in personal?
Accept friend request.
!Berserk
@smoky oak
I am searching for an unused token word for the last half hour and stumbled on that
"I took them 1 hour before noticing that their friend, jack, was not with them anymore." :P
monk image siiting in himalaya
how is it broken?
But it is not in the dropdown
it's not an upscaler model, it's a script
Oh i see
Its not in scripts either
nvm it is in the img2img one
i found it
thanks for helping
taking a leak on mars
cause I've got both hands in my pocket
And the other one is holding a duffle bag 🎶
ngl but no one gives a crap about the rude stuff you say lol
that's why people have coined the nickname, "inept"
"ngl but no one gives a crap about the rude stuff you say lol"
https://github.com/dome272/Wuerstchen
even smaller internal latent space rather than 64x64 for Stable Diffusion, Wuerstchen uses 12x12
changing from legal to not legal
"mix" probably has "chillout mix" in it, which is always illegal. that model is nasty
yes
Something wrong?
i trained 1.5 on v_prediction and terminal SNR and it's capable of dark images without offset noise or SNR gamma but looks blurry
and it does this weird stuff with red
1.5 vs 2.1
same dataset etc
1.5 is on the struggle bus.
maybe 1.5 needs like 10 times the number of steps, which means it's a LOT more expensive to train
🙂
That's odd, i'll take a look
Is there a commissions or for hire board on here where I can hire an experienced tutor?
Like mentioned previously, we don't have a specific board for hire/commissions. That does fall under the umbrella of #1092446741984444416 though.
I've opened a mod support-ticket for you, we can continue this conversation there privately.
yeeeesss
Well, same actor 😉
Flynn was more recent
Indeed, indeed
this is me training from the beginning again using a huge batch size
i quite like that result there
Yes, it's excellent!
nerd
the first two images are from a new model I downloaded, and the last two is the same prompt but on my own merge from last year. I don't know why it's like this. But I like both results at least :D
Did you have any luck finding them? I've tried looking for it in the open source version of Dream Studio (https://github.com/Stability-AI/StableStudio), but I can't really find it...

What's up losers?
How to not start a conversation 101
Post more satanic renders
why does your name colour change all the time 
Not my fault 
Pffttt, I wish. AI doesn't have much support for occult style renders
I'd it did, I would be generating satanic shit all the time and working it into brand imaging lol
It typically works pretty well :p
we'll see
perspective
apparently occult stuff is easy, it's just candles!
i wonder where that egg thing is supposed to go
Haunted house terrariums. These were fun to make.
some real Eyes Wide Shut stuff happening there
@smoky oak may the force be with you 🤣
Those occult generations look kinda cool, but still pretty poor in execution
Really hoping that AI gets less limited in smaller subject matter as time goes on
I'd really love to do some actual satanic image generations, but that would probably require me to make a LoRA of it from other people's work, but they are small creators with not much income, so that would be pretty scummy
name that celeb
Ribbon wilms
lmfao
i trained 1.5 on terminal SNR btw
that's the occult stuff it comes up with on one try
weird eh
probably will take a fair more tinkering than i've put into it to make that work
devito
An attempt at occult
This doesn't remind me of that movie at all
Have you seen it? The subject matter is decidedly different
the opulence of the last three images is exactly like some of the sets in the film
I guess, but that's not what people mean when they reference that movie
that's what i meant, and now you know 🙂
I guess your take away from that movie is "big rooms" 🤣
yup
For reference
i guess it's more like rocky horror picture show
🙂 I guess some people make different connections to things and that's ok
RIP 
🌅 morning
Provided to YouTube by Universal Music Group
The Fool On The Hill (Remastered 2009) · The Beatles
Magical Mystery Tour
℗ 2009 Calderstone Productions Limited (a division of Universal Music Group)
Released on: 1967-11-27
Producer: George Martin
Composer Lyricist: John Lennon
Composer Lyricist: Paul McCartney
Auto-generated by YouTube.
morning music 
sound's not working through the remote connection 🤔
but the floot
I'll look it up on my work laptop
the remaster is so good
nice and crisp
my dad had a beatles cd collection with the green apple artwork from what I remember. I used to listen to it on the boombox in my room
looks like they've done a few box sets with the apple over the years
oh yeah my mom had that one too
did you have one from Enigma too
always tripped me out
nah, they didn't buy much music on cd. he had an old vinyl collection from when he was younger but sadly we didn't have a way to play them
usually they had the easy listening station on in the car, so nothing too far our there like enigma 😄
although I'm sure it's being played on easy listening radio these days
oh, we had The Wall on a 4 'disc' vinyl set
discovered that on a road trip when my dad put it on and it was like i was hearing music for the first time and i asked what it was and then mom said she'd show us the record when we got home, and boy did she
I didn't really discover music until I had my own computer and had access to emule and kazaa etc and started downloading random stuff on those
oh no what have I done :S
a port-hole 👁️👄👁️
So what can I do about artifacts like these when I am upscaling? I can't seem to figure out how to get a clean, simple upscale for abstract stuff.
This is with CN tiles and SwinIR, using the SD ultimate upscaler. I will try that out.
That looks like a VAE issue. You may have to manually set your VAE rather than having it run "auto".
Awesome, I appreciate it. I'll read the documentation.
time travel pod gone wrong
rip
if you are using the latent upscale,r I'd recommend using the Anime 6b one instead (sorry if thats not the exact name)
you can use a lower denoise, and its much better fir smooth designs like this
Hey guys, question. I've just started with stable diffusion last night, and was able to generated an image like this, right now there's some error as the image character has 6 fingers. How do i use stable diffusion to fine tune the errors? like maintaining this image, but just fix the fingers
why you peeng meee
im pretty sure thats satan appearing in my test gens
tried to summon him on purpose earlier and now he's coming without asking
why is it so difficult to get a guy with proper jean
i think i need a model that not made by weirdos
are you wanting a cartoon guy with jeans?
what model are you using?
tried all
so doesn't have to be a cartoon?
what's the prompt? there's a lot of words that the ai "thinks" of in quite...different ways from us mere mortals :P
character sheet of a mysterious kid, gloves, modern casual dressing, jean, haircut
negative extra fingers, woman, female, anime, facial hair, horror, headwear
and always get weird clothes with weird colors
this was what i want but the clothes..
same prompts by the way
did you try specifying each item of clothing? E.g. denim jeans, white tshirt, black gloves
i get dumb when i think of the prompts lol
might be the jean and dressing that confuses it. As well as "kid" because how many kids (children) are dressed
I just found one of the only paintings I actually like lol
"Making of the perfect Martini" by Guy Buffet lol
The energy is immaculate
How to create AI images with reference images can anyone help ?
I like his hair at the end
oh, I didn't even notice that lmao
Google how to use "reference only" with controlnet in stable diffusion
there is a special workflow for just that
Thanks @smoky oak will have a look
no problem, best of luck!
@smoky oak How's the dynamic prompting with wildcards going?
I have only messed around with it a little, but it has been very cool
I have been having some breakthroughs with music, so its had my attention as of late 😅
go with the flow, whatever is floating your boat atm
just made a new audio rack for myself with a custom UI and everything, and its already revolutionizing the way I process melodies and such
nice
so maby knobs
Small taste of what it can do, without, and with
Its a crystalizer/atmosphere rack, to streamline the very tedious task of layering effects for such a result lol
gotta love node workflowssss
i see, i see pretends to understand 😛
One more example of without vs with
the yellow connections are the flow of audio, and the red connections are the knobs on the surface linking to knobs in the plugins themselves
oh right, cool. I hear it adds some extra to the stuff in there
so its basically different strands of the same audio being processed in different ways, and then combined at the end
I wish FL had the ability to import images into patcher surfaces, cause I would decorate my racks with SD generations haha
what are the odds? I was checking old screenshots and saw this, and then the discord :P
oh my, back when FL looked dated AF
OH WAIT
thats just that old of a version of FL lmao
yes, it was when I used it, in 2007 :P
oh wow haha
yeah, FL looks wayyy different now
the cleanest and sleekest of DAW's IMO
yeah, I still have it and use it, but not as much because of lack of music time vs ai art time :P
ah, those were the days, msn messenger, winamp, kazaa, playing around with FL and PS
but it's a paid version of photoshop, those things are very rare ;P
funnily enough, FL is the only one you can still use
PS is priced out the ass now, and all those other programs are dead
this is the beginning of the song in the screenshot. I lost about 90% of all my stuff when a harddrive crashed about 12 years ago or so, so this version isn't the finished version (the cut-off was made by me because of size limit of discord though) :(
Another example of how it works without and with
why is it reminding me of the old rinnai gas heater at my old parent's house
I have, no. I dea. At. All. :P
And I don't know if it's a good thing, or a bad thing to be compared to a "old gas heater" 🤔
Wait, I AM an old'ish gas heater! 😱
we're all gas heaters at the end of the day ;P
getting an atmospheric space game soundtrack
its made specifically for atmosphere haha
its supposed to be glittery, and airy, and atmospheric
I don't even know what that is from your screenshots, I just wanted to butt in and show off my poor skills (general and artist ones) before being pulled back to the depths I came from :P
its a custom patcher rack I made in FL haha. Its their built in nearly limitless node system for processing
I recognize some of those words, yes... I think I do anyway :P
as cool as this new rack is, my actually big breakthrough in audio has been atonal bass processing. I made a goof with a new friend in a voice call, and it lead to the single biggest jump in my bass production to date lmao
Here are a few new basses I have made in the last few days
(WARNING, LOUD AND INTENSE)
just busted out the midi keyboard, lord have mercy :p
I really wish this keyboard was more than 3 octave 😅
I was hoping to find a video of one of those gas heaters so I can hear the sound again 😦
one would think those things would be everywhere? ;P
found a few videos on how to fix/clean them, but cmon, you gotta show it working with the fan running. nup
go
i attempted an img2img upscale, idk why but it made it into a collage
still looks nice tho just not what i was planning to get
denoising up too high
I kind of love it--The collage/kolaidescope effect is fun.
yeah i also think its kinda nice
ill try reduce it
oh that explains it, i accidentally typed 0.75 instead of 0.35 xd
When you said denoise, though, I suddenly pictured it as "DJ Denoise." 🤣 And he'd have these screens behind him with something like that in the background
It's kind of fun to see what happens when you set it too high, hahahaha!
this is it with the lower denoise, much better
still has some banding but i retried with diff settings
it's a "feeling" thing you'll learn over time :D
i am using lollypop upscaler, it worked the best of the ones i tried for this
the rest made it look blurry and a bit dim
Banding does not seem to go away with different settings so i will stop here
its only noticeable in nthe sky tho so maybe some blur would make it go away
not really the same style, but this is as close to anime I wanted to search for when it comes to upscaling
I also found euler can change the image too much, even at a low denoise strength
this one is slightly closer, maybe :P
what is a good sampling mathod?
method*
try dpm++ 2m karras at 0.4 strength
Cool, what model and prompt did you use?
You need to use a vae file for anime models too.
For better colors
(by artist David Diao), features intricate detail, a still life, naive art, boy amongst flowers, egasumi
Where do i select a VAE in automatic1111?
Here i link you the vae quick guide:
The last step is very helpfull.
#🤝|tech-support message
Thanks
Flavor bomb
Gods of Kross(my book):God of reality(Wonderan),God of change(Pewxanes),Goddess of souls(Azerox),God of time(Dekaran),God of matter(Ganrasden),God of space(Zetrox)
that's too cute!
But also, way, WAY too dangerous, look out! Also: It's against the rules not to have a welding mask when welding! >:I
i know 
that makes it worse! Knowingly doing it is even worse ;P
look at all the PPE i got him to wear
he has everything on EXCEPT the hood
you can't blame me!
sometimes kids just need to weld
now I also have tiny houses...if only the rent was also tiny >:(
this guy will defend your land for a small fee. just a mere 90% of your income
if you don't pay the fee, however
this dinosaur watching the meteor come down that causes the extinction event
sure, I'll just count it on my calculator
0.9 * 0 …
Jim and Steve looked wearily on at the meteor, knowing full well what would come next.
Jim looked down at his human friend, Steve.
"I am so sorry, Steve. I wish I could have done more for you."
"Actually, can you lay down in that oil pit over there? I have an idea."
"Dude, really? That sucks, won't it hurt?"
"You won't feel a thing."
Jim laid down in the oil pit, and took his final breath, as Steve climbed into his cryo-freeze tube and settled in for a long sleep.
hm, didn't work on tiny dinosaur :(

howdy hello, plebians!
||myself included||
@dense tapir Just realized... its been a long time since we saw a pathetically disgused Sen alt

would be an astonishing leap in the field of natural intelligence
Don't let sen hear you, they talked to the best of the best, and they know for suresies that DB are the best cause of a 5 month old youtube video with like 25k views
by the way i was looking at the A1111 code for LoRAs and it definitely merges the weights of the LoRA into the unet of the base model that is being used
thats also why after literally 3 weeks, theys till had no dreambooth to compare against my LoRA I made in less than half an hour 
Diffusers is implementing it in a more drive-by manner so that multiple LoRAs can be loaded and unloaded without reloading the base unet
interesting, I have never had any sort of spill over
i thought it was interesting too but it explains why the regularization images worked for you
i was so curious why that would help, when the LoRA is its own network
but it's due to the merge
https://www.youtube.com/watch?v=3_9LGSex1JY
A rare interview maybe ;P
!! Seizure alert cause too many cuts!!!
Open source
Interview with an Open Source Maintainer with Bob Hopeless - aired on © The OSS.
Part 2: https://lnk.bio/kailentit
Find more OSS opinions under:
https://github.com/zloirock/core-js/blob/master/docs/2023-02-14-so-whats-next.md
Stolen from Ben Awad jokes
https://www.youtube.com/watch?v=MT6M_sqA...
Ahhhh, hmmmm
where did you get this video of me 
"all these emails i get about my model. 'this model is bad.' 'you should feel bad.' 'this model made me cry'.
I do not care about people's opinions"
you don't want to know...by the way, you need to restock your fridge, and also fill up on more coffee for the next time I—I mean no one wants to… erm, nothing! disregard me
go
why do you paste this same image? you think a bot will do something with it?
which path do you take?
for ai images the answer would be: Both ;P
oh my god, I am going to hell for saying this
but she looks like the uhhhh
LMAO
i love this dipshit pirate
It has been. Oh, I got banned from GitHub
We do not allow content or activity on GitHub that:
is off-topic, or interacts with platform features in a way that significantly or repeatedly disrupts the experience of other users
Specifically, the content or activity that was reported included creating multiple off-topic and disruptive issues, pull requests, and/or comments in other users' repositories, which we found to be in violation of our Acceptable Use Policies.
iow, there is one person
the other stuff I never did so it goes back to you know who
blaming vlad again, are you
bold strategy, Cotton. let's see how it plays out for him
I mean, I have seen first and second hand how big of a douche Vlad is
github doesn't necessarily ban people for zero reason
@smoky oak I sent you a pm
well, other than having two free accounts
for sure, but I do know how Vlad could have spun it
he closed a legit concern we had for being "off topic" and being "spam" when we literally just shared an error report with him
if it's spin, appeals should work. but overall i'm not laughing about GA being banned from Github. that's pretty fuckin devastating. like, RIP livelihood for many people when that happens
just screenshotted eactly what it was, asked what was the matter, and he said we were wasting his time by not giving him enough "evidence" or whatever
like we are the ones who made the program or something lol
to be fair to him, he didn't make it either 
like what more do you want than the entire error code
it was specifically one of his changes he did to the UI, IIRC
A him exclusive edit to it
that's fair, and i knew that. but also, it interacts with all of the tower of garbage he built on
forgive me if I am wrong, it was a while ago
very true
just not sure why he needs to be such an ass about it, and act like we are trying to waste his time
at the end of the day, it was an addition he made that didn't work, so we asked him
finding out how A1111 does some singular (simple) thing is so hard. it has like 4 or more ways to do any single thing. and that's through the UI! in the code, it's worse.
developers need to learn to be people persons in this day and age
lol
if you don't want to hand hold users, do not publish the garbage
or close the Issues tab on Github. just disable the issue tracker altogether. problem solved
especially if your work is for the people, and you are putting yourself in a situation on a platform where people are supposed to interact with you lol
maybe he appreciates help from a certain type of person, but not others. like, "users" vs "contributors". a person who contributes code, might be welcomed with open arms.
i get uppity when users sound demanding on me. man, some of the DMs i get from users trying my model on Civit. just makes me want to delete all my crap from there
Too many thin skinned people connected with this SD stuff.
btw, do you know what got me?
Posting off-topic comments
Opening empty or meaningless issues or pull requests
I didn't do either, but Vlad said both. Github saw that I actually file tickets on projects and I guess under the new Microsoft Github that is frowned upon. Good to know going forward.
going through the WayBack Machine to see your comments, lmao. toxic as hell
lots of "f" bombs there
@smoky oak Funny thing is they never told me as I can still use the service and only found out when I tried to use Colab with it. All 3rd party apps are disallowed due to my spam status. Not sure I can post but I can still use my repositories.
I might have been banned since Vlad incident happened.
"banned"
your whole account is 404'ing right now. you are fully banned.
I see a lot of the chat got deleted while I was away lol
eh i decided it wasn't worth it to address those things
What is it with Russian named devs? Vlad and Auto are both, supposedly, from Russia.
likely just two bad examples of a demographic
I hope so
In my gen they were thicker skinned (had to be thanks to the USSR) than either of those and were pretty on the ball.
now you're taking your one experience and applying it to an entire ethnographic group?
are you going to think all Americans are like that because of any poor interactions with an "American named" dev?
@tropic shell is this kind of thing welcome here, disparaging an entire group of people like that?
starting to look like the bad guy from Toy Story
hey, i made this comparison and im not sure wich model did the best job here. thoughts?
If you wish to discuss this topic further, I recommend that you take it to DM's. Let's keep chat positive. 🙂
Right side or the Left side with the right side having a slight lead for me
i think i kinda like the left one the most, but yeah, the V5 of that model is definatly better than V2.
It's interesting how each has its own character. I like each of them, but it depends on what you're going for, I suppose.
The 1st and 4th have the most realistic depth/perspective.
Paragon is a 1.5 model?



