#🏞|general-with-images
1 messages · Page 32 of 1
^^ maybe what exactly?
that previous pizza is what :p
masterchef nice one
got a little different UI it seems
3 people here including me installed it and concluded it was slightly slower
it may depend on the card though
my computer is very old
omg....i choosed the wrong modell, im the dumb guy XD not the modell^^
32gb 1333, 8320fx, 1050ti 4gb
will switch and post
^^
i use the extension
Kitchen theme for stable diffusion webui https://github.com/AUTOMATIC1111/stable-diffusion-webui - GitHub - canisminor1990/sd-web-ui-kitchen-theme: Kitchen theme for stable diffusion webui https://...
is tehere a trick for upscaling? every time i use it my SD crashes cause of vram (8gb)
thx
in the shortcut.bat?
webui-user.bat
k

ok the modell is really good
i was dumb first , but its not the model whic hhas failed XD, really this will be my new love till sdxl is avaible for everyone in stable
an screaming crying italian look at a pizza with pinapple XD u will get this? like this meme?
have tryed it but no good result
SDXL makes insanely good pizzas
yes me too
caus then the people start to train their models XD , if the base is excellent - the community bring it to next lvl
ok next one will be a pizza related
You only need --xformers
now it works thx to the community 🙂
sure^^
@summer viper what model did you use?
try the modell i used, its awesome
delibirate 2
Looking at examples, seems like they made it less realistic , looks like a mix of fantasy + photo now
maybe cause they didn't have protogen and dreamlike in the mix before idk
maybe, maybe its the examples people uploaded, im very satisfied for now with the results
Jimmy McGill, Bob Odenkirk, Saul goodman from better call saul, realistic, business man suit, lawyer
lora?
Lol
I decided to remake this photo, it's strange, but my neural network caught the notes of anime and started making anime too, although I didn't write about anime
what main model do you use for that?
delibirate 2
thats pewdiepie
how do you get it look that much like he should look like?
use scale fix
Dynamic Thresholding (CFG Scale Fix) for SD Auto WebUI - GitHub - mcmonkeyprojects/sd-dynamic-thresholding: Dynamic Thresholding (CFG Scale Fix) for SD Auto WebUI
how do I download this? i'm too new for this
I just wrote "Saul Goodman" and he immediately knew who it was
Thats because he used an other model
50% Off !!
yes
Scene from S05E02 (50% Off)
Music: Chai - This is Chai
Evil Saul Goodman
I've seen this style somewhere before
its the model i posted before
so yes ^^
sometimes sd has to much saturation and contrast, but its not a big deal, only for perfection needed
put a scale fix and make the scale itself quieter and it is advisable to use different sids
also you have too many steps
ok? will try lower
a bit better
i really wish one day html color codes and hex colors are supported ... 🙂
/大海
Generate
in an hour or so there will be a new daily :-), cant w8^^
Can somebody show me an example of a SD 2.1 image and its config/params? I am struggling a ton to get this to actual make something none neon, beginning to feel like 1.5 > 2.1?
2.1 needs a lot of negative prompts
Can somebody show me an image I can throw in png info so I can see how this works?
or just one negative embed
Mask
how true is
we're in a blip of an age of artificial intelligence. we may never develop artificial general intelligence. What wer'e going to see is augmented intelligence and as soon as that egg is cracked, there's no reason to call it artifical anymore. If it's enhancing our natural minds, then it's just human intelligence augmented
I found this drawing i made when i was 10 yrs old and decided to use img2img to reimagine it
Hi guys. I'm a student from Italy and i'm doing a bachelor thesis about "lettering" using AI. The lettering i want to create is a mix of letters with adds effects (i give you an example). My objective is to educate the tool by providing starting images and obtain a cool images. The letterings don't need to be efficently read. The starting images can be 2,3,4 ecc. So the big questions is, in your opinion, what tool i have to use? I'm tryng to use Stable diffusion 1.5 without extensions but the results are very bed, because SD can't reads letters. I want a good result between effect and text. So guys, I hope in your advice. i started to use AI on week ago so i'm not a pro. Thanks for your time
You can feed letter image in controlnet and propmt what you want to do with it.
You can find some youtube videos about controlnet + stable diffusion on youtube, it's too complex to explain it with text.
if you train enough typographical styles with good captioning, i believe controlnet could very well guide that model to some awesome font faces
name the model "Keming international" as a joke i think it's hilarious
who has some XL images lol
@deft falcon
is that directed at me? 😭
i dont think yiou were around at all when def someone posted the meme first
just a meme remix
not sure though

Finally starting to get a hang on how to not over sharpen the images and still keeping them crisp in HiRes rendering.
this joke never lands well but man, i really think its so hilarious
demon girls squad 
cursed adam sandler shazam
have you been trying to use sd-ultimate-upscaler ?
No 4x Ultrasharp a version of ESRGAN

i can tell lol. butttt it looks really good. Ultimate SD Upscale script is just a completely new level of upscaling. Diffusing in new pixels instead of upscaling existing ones
I found out that if I use regular ESRGAN and then use Ultrasharp as the 2nd Upscaler I can basically choose how strong I want to make the sharpness by dailing back on the visibility.
not as "push button" as the extra tab upscalers though
I think you've reached peak upscaler tbh. That sounds like the right process. Finding two slightly great upscalers then blending their results. its exactly why 1 and 2 exist.
tell me if a1111 will update...I can't even change model lol
I had a small patch today on A1111
Well I now can click finished renders back open in a new window that was something it had some problems with for me.
not too happy with greed's face
but the idea was to make it look somewhat like a golden skull
Is there a word in english language for "vertical" pupils, like here?
i think you're talking about slits
vertical slits
am I?
Reptilian Eyes? Snake Eyes?
yea something like that
youre talking about cat's eyes rightt?
@blazing crow i by no means have the ultimate upscaler script wrangled yet. it's more advanced and i'm figuring it out still. i downscaled your image and upscaled it back to show you a comparison. better to show than just talk about how great it is sometimes and you seem receptive to what i'm saying. so here's the upscaled at full zoom and the second is with ultimate upscaler. my prompt was very detailed photograph, hdr color, high resolution, 8k quality as it's acting on patches of the image at a time, rather than the whole image. probably could've been better cause it messed some details, but you can tell, details are just better. I think i messed the dimensions up a bit too, she seems smushed in mine
but maybe there's something more specific
aight, I'll try that,
@wispy nest goats have rectangle pupils like that. hircus eyes they're called but the ai might not know that. their's are sideays instead of verticle too
It's this one right?
interesting, interesting
yup. it shows up in "scripts"
I'll give it a good tomorow then Haven't got it installed yet. Does seems very promising.
I think best way to prompt is by using simple words, unless they have 2 or more meanings...
I'll try "cat eyes" too , maybe that'll do
it has limitations. i wish i could look the grid of images it was going to process, and then define different prompts for different patches
"highly detailed hair" shouldn't be running on the background patches for example
can someone make a picture like this? I don't know how to define it
alternative world where he was hook not peter pan
looks like an aqueduct
He looks like baron samedi
no jajaja
i might've pulled the exact image it was trained with to get that Robin williams look
and that kind of coincidence is such a munchausen thing too.. iugh i hate that synchronicity shit

Ah that Upscaler is an Img2Img Script add-on I was scratchin' me noodle here trying to locate it as it doesn't turn up in the Upscaler list 🙂 Yeah this baby will have to wait until tomorrow thanks for the info @grizzled sage much appriciated!
Edit:* Wow the typo's tell me it's time to take a breather, good luck lads and lasses, keep that Diffusing Stable!
its certainly a different breed of beast
I'd watch this man play anyone ❤️
and one day - I will 😄
COOOOORL- IM FUNNY NOW CORL
I would say more #1025467151206854736 since devs have more chance to find it, I said that in #💬|general-chat
clearing cache solved it it seems
OMG
horsecat
xl?
its on DS for now
768 part of it i dont think is cooked yet
no official source on that, i just have speculations from keepign my ear to the rails
on ds?
d'oh yeah i see
you should ahve a bunch of free credits to try it out
so the legends have foretold /spooky hands
thought maybe they going back to what worked better haha
a Caw
mooooeow
seems pretty good from what ive seen
better than the 2.x stuffs at least 🙏
meoooo
bread to be anti tippable
that one feels wierd
you guys finding XL does anything in particular especially well?
some pretty cool stuffs
maybe stable can get its groove back with xl
big thing of course will be if people can train it easy
🤞
counting on it big time too
hoping at least
but I do have my hopes up tbh
just seeing the energy that is going in there, and the fact that SD is the only open source one around, I'm sure some big bumps in some specific tech things could be enough for the training community to be able to move from 1.5 already
hopefully
I can nothing about memes, I just know this become so wrong in so many ways that I should not post it.
with mj5, adobe firefly about to release, bing image, more and more ai's will come, it is time to solidify their rep (sd) haha
as long as they stay the only one to open source it like they do, even if with some delays sometimes, I don't see how the community would just move to those.
I may be wrong, SD may get out raced so much that it falls into oblivion, but I would be really surprised if this wasn't the brick of the future
like, what good is firefly, even if it has some edge on the UI and results for a while, if people can't build on it ?
more the merrier how i see it. they sharpen each other
that's true
and less people to bitch just about SD destroying society or whatever when adobe and everyone else in the game too lol
and also, the bigger players coming in means legislation could be smoother
yes, we are saying the same thing there i think haha
yeah... I just hope they keep a competitive edge enough to keep SD on their toes, but not outclass them too much.
because I just love SD maybe, I may be one of the most biased around
slits, cat eyes, cat pupils - none work , but SD can randomally generate it still 
"The day AI art destroyed an art gallery"
"The day AI art destroyed an art gallery with help of furries"
the base models are already OK or almost for that.
Tbh, the model training scene lacks a lot of children's book imo. I can see that one but it's about it https://civitai.com/models/5153/coloring-book
don't delete ><
oh ok
srry, just saw this is a image chat
nah, no problem ^^
I thought you wouldn' get my message :p
yeah it's lacking of other cartoons than Anime currently
yeah, I didn't want a anime style
I tried to see if doing a scoobidoo style was possible at some point, or other cartoons
really just play around with them. theres not a lot of difference between most models for that kind of thing yet
but the image availability, with OK copyright, for such things is almost 0
we're still in early days of this ai stuffs
So just use OG stable with some tags?
yeah, using the name of the studios that did cartoon you liked, that works great for style inspiration usually
Id get some of the bigger models on Civ and just see which work best for what you want
here is the cartoon tag, there are some things
not a lot that are really good at first glance, but some options
and if youre doing like a big project, id be tempted to wait a little bit since XL models could be released soon
may as well get latest and greatest you know lol
I found 2 i quite like
FaeTastic https://ko-fi.com/post/The-Double-Headed-Issue-E1E0JO2UV PLEASE FOR THE LOVE OF GOD READ THIS TUTORIAL ABOVE IF YOU GET DOUBLE-HEADED THI...
A
good luck snb. have to update with some images from your book some time lol
Ofc!
mooowing them down haha
love your token in the mix model @devout reef
it's going further from what you did, incorporating lots of stuff, but it keeps that last cats entry feel
https://huggingface.co/Guizmus/Experiments#7mix-sdart-v1 (mix model for the curious)
there has been some bugs on automatic in the last few days
ask in #🤝|tech-support maybe
honestly, anything it shouldnt be
oops i made the D.O.E'S art style
xl?
nop, PoW model #🏞|general-with-images message
just an expirement though
I wouldn't say it's good yet
work in progress
but still fun, it's trained on all the submission from 7 PoW (picture of the week) contest
yeah I get it
cant focus any other when i know that on the horizon haha
I can be a hard ass with my "when 1.5" when people ask for XL, but people ask for it a lot x)
🐮 Did you know that cows are actually the ones who developed and trained the SDXL model for stable diffusion technology? Yes, it's true! In their spare time, cows can often be found training on pictures and finetuning models. It's their main income avenue in fact. 🐮
mooooost interesting
that's a nice prompt
(extremely detailed CG unity 8k wallpaper), full shot photo of the most beautiful artwork of a medieval castle, snow falling, nostalgia, grass hills, professional majestic oil painting by Ed Blinkey, Atey Ghailan, Studio Ghibli, by Jeremy Mann, Greg Manchess, Antonio Moro, trending on ArtStation, trending on CGSociety, Intricate, High Detail, Sharp focus, dramatic, photorealistic painting art by midjourney and greg rutkowski
Negative prompt: canvas frame, cartoon, 3d, ((disfigured)), ((bad art)), ((deformed)),((extra limbs)),((close up)),((b&w)), wierd colors, blurry, (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))), Photoshop, video game, ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad anatomy, 3d render
let's go 1920x1080
that batch was real good
not sure what to upscale there
I went for number 3 and 4
I would like to share a new stable diffusion video generation tool. We need Alpha testers.
I don't think I can post the discord link here but if anyone is interested please DM me
no discord link, but if you wan to share a repository or something, you're welcome to share around here, in #1045770510153302086 , or even contact us mods in #1010934719455707218 to see if we can feature it
2 little 1920x1080
Can I share a link to my Instagram story which has the discord link and a demonstration of the tool
I'd do 3 and 7 
like I said though, #1045770510153302086 will give you more visibility, the channel you currently are on scrolls a lot
We really do need Alpha testers and the service is free while we are testing
I see, I'll put it there too thank you
but do put some description/teasing around it, an insta video that people can't watch without being logged in may not be the most enticing on its own 😉
eh, you need to login on instagram to see anything, naaah, I'm not doing that just to see a post lol
i dont even have instagram lol
yeah, that is quite restrictive in my book too tbh
here is 8 by mistake
and 7
that was a lovely prompt
(8k, best quality, masterpiece:1.2),(best quality:1.0), (ultra highres:1.0), watercolor, a beautiful woman, shoulder, hair ribbons, by agnes cecile, half body portrait, extremely luminous bright design, pastel colors, (ink:1.3), autumn lights
Negative prompt: canvas frame, cartoon, 3d, ((disfigured)), ((bad art)), ((deformed)),((extra limbs)),((close up)),((b&w)), wierd colors, blurry, (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))), Photoshop, video game, ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad anatomy, 3d render
taken from https://civitai.com/models/4384/dreamshaper and used on https://civitai.com/models/16592/realbiter
not a single one of them have ribbons
for fun, this was the corresponding sample on the other model :
but yea, that's a good and clean prompt
I can't stop making demon girls
my old thing was swordwomans, now demongirls 
realbiter
I'm using prompts from other models on it, it works correctly yeah. I like it a lot, it's what I used for my cat sushi chef too
I need to make demongirl with it
hard to share most of those
but from what I seen it handle privately
you should be happy
I mean not nsfw, just like I posted above
demon claws, horns, surrounded by fire and stuffs
it did demon creature, that's interesting too
Negative prompt: (depth of field, blurry, blur:1.2), (greyscale, monochrome:1.1), (morbid, mutilated, deformed, disfigured:1.2), cropped, lowres, text, jpeg artifacts, signature, watermark, username, artist name, trademark, title, cross-eye, multiple view, reference sheet, hands, big tits, big boobs, big breasts, cat ears, bad-hands-5, ng_deepnegative_v1_75t, muscles, nsfw```
- that's one of my demongirl prompts
Negative prompt: (depth of field, blurry, blur:1.2), (greyscale, monochrome:1.1), (morbid, mutilated, deformed, disfigured:1.2), cropped, lowres, text, jpeg artifacts, signature, watermark, username, artist name, trademark, title, cross-eye, multiple view, reference sheet, hands, big tits, big boobs, big breasts, cat ears, bad-hands-5, ng_deepnegative_v1_75t, muscles, ```
That's just a face \ portrait
some weight restribution and could be interesting to play with
even with blood, that's just perfect
I also miss your 2 negative embeddings, so I removed them from prompt
yea, it's fine, like half of it is useless anyway
I'm just to lazy to fix it each time
I actually like face renders
I'll add your random style
fire got weird, but blood , eyes are good
yea I should play with this model , hopefully webui will be fixed tomorrow
at least model choice =.=
you can download models by hand too, no ?
I can download, I just can't choice it inside webui
That's preloader icons, they are "trying" to load something for hours 😄
yea
it's upscaled already
too much of a demon for my liking tho 😄
I was doing something like that:
#🏞|general-with-images message
Human with demon body parts
the demon arm fuses strangely in 4
oh yea didn't notice
upscaling 2
not what I wanted but nice still
yea ^^
like what?
like what I just shared

What's your denoise on upscaling?
weird
yea that's probably a bug
I'll reload
I have a feeling 2 was made cause of bug 😄
wow
and now, it bugs even without highres fix
I don't get the initial image anymore
yea, that's what I thought too
I think it just bugged on gen and made image it shouldn't generate
I'm checking one last thing, if it could be because of batch
it was number 2, but for upscaling I put it alone in a batch, not with 8
nah
just got a complete different board
let's move on from demon girls
demon boys
lol
demon cows 
I just replace "young fit woman" with "young strong man"
yep
2 epic
cows make good demon
yea
eya that's more like an actual cow
ran another batch with your styles
looks like a bossfight lol
really dark souls like
got a good one
I'm upscaling
no joke SD please
jokes again
damn
original
maybe seed number isn't correct on png info?
or, maybe just maybe that time it's my fault
I forgot to copy the seed
nah, I don't know
I must be tired and missing something obvious
file name seed is same as in png info?
yep
hm, idk
goodnight
Im hungry
that knights ruining that pizza with his hand but Im not about to tell him that while hes wearing that armor. probably not while he's out of it either

He continues to eat like that he will need bigger armour
Ahaha
How many models has everyone got? Just curious
124 gigs, but I recently cleaned it a bit, deleted few
38 models that is
I literally can't have more than one at a time
Because loading another model after starting automatic1111 crashes

I still have few models I probably will never use, but keep them just incase
and inpainting model versions for some of models
maybe not enough ram?
yes
there was an arument for it
--lowram I think
there are different ones for ram and vram
I'm doing some bad choices regarding what to create. It always end up being something with details seemingly beyond what SD can do
(The reason why I want tiled inpainting to try to fix it)
Hypernetworks only work with specific models they were made for?
No
I still love HN but it is a PITA to train with as it never received any optimizations.
considered dead now by people but I still like them as they just work.
That cat reminds me of those old ottoman paintings
This is not AI related... but look at all my babiessss
So many baby shrimp lol
they are in the trees
I'd say I have close to 80 in this little 5 gallon aquarium
Nice. So many what do you do with them?
They are itty bitty dwarf shrimp, they are my pets, and I love them
yeah, but too many for such a small bowel
You can use them in a different way to make small bowel
oh no, these shrimp species love extremely high numbers
you can easily have 40+ per gallon, and they will be just fine
They are amazing little creatures
damn.
They are insane haha
cool thing is they need to food, they eat all of the algae and plant matter, so they need basically no maintenance
Algae is a pita
I have low grade shrimp. They are rejects, but I love them :>
All of these shrimp came from just one shrimp. I bought her pregnant, she gave birth to 25, and those 25 have given birth to soooo many more
must say they are very colourful
they come in all sorts of colors
There are even some that cost as much as a downpayment on a car lmfao
These iridescent beauties can go for $1000+ a shrimp
damn
i bet being a breeder can be quite lucrative
German Armor shrimp can go for like $200 a shrimp
do the conditions of their breeding require special chemicals and things?
So the brother-sisters had kids... I guess it's ok for them cuz they're not people...
Thats my goal, though I am aiming for much less expensive/risky shrimp
its very common in their species, and it causes almost no side effects
other than awesome colors
probably good to go for the consistent ones instead of the expensive ones, less overhead and risk
I wanna have a dream of SSS blue velvets
Huh, their genetics must be something special. Gotta look that up
anyone else convinced there are residual tokens left over in these models after a prompt?
glad i'm not the only one. the nerds who made it don't believe me, but i think these things remember
oh i don't know, the bastards kicked me out
as long as they're using any modified form of CLIP it's guaranteed to be garbage
it blows all of the other models so far out of the water its insane... but for all we know it has SD 2.x's shit TE, and will be nearly impossible to train
What I will say is it listens so much better than all of the other models, which would suggest a drastically improved TE, which would be game changing
It also just understands everything magnitudes better
its crazy cause I did a shit ton of direct base comparisons, and SDXL didn't fail a single one out of like 50 I tried
it can do night time with not finetuning, it understands and can replicate text reliably, it has a next level understanding of stylistic images
It's a photograph lol
why? I thought you liked it result last night
I just want that public release lol
I am trying everything I can to not get excited about SDXL, cause there is still a high chance it's very good but also useless to training
take that back, I need that public release. I'm a prompt jockey and junkie lol
I just hope hope hope that it doesn't have the same trash text encoder of 2.0/2.1
you seem like a sharp guy. do you do video on modelscope ?
Please, for the love of accurately interpreted AI image generation results, please let SDXL make up for the failure that was 2.0/2.1
Can't say I know what this is
What folder does textual inversions go into?
embeddings
Cheers
whats that extension supposed to do anyway?
It was working great until this morning, something got screwed up
I recommend ultimate SD upscale over multidiffusion all day
ultimate is faster, more granular, and has a much higher gross output res cap
and I find it also just a lot more intuitive, and reliable
Aside from this current odd issue I'm having, Ive had more success with Multidiffusion than SD Ultimate
I could help you with settings for ultimate SD that blow MD out of the water, if you would like
trust me, its a lot faster in the long run
I did a 100MPX upscale in 17 minutes
The extension has been pushing updates many times a day, maybe they fixed (and broke) a few things since you tried
you can see every single tree on the mountain side
mind you, this is what the full composition looks like
beautiful
and thats only really this section of the image
Should be top image in Ultimate SD Upscaler github
Thanks :>
I am sure I could do better now
my tiger one is still pretty damn shocking, IMO
@floral flickerlow res
ultimate upscale
I fucking love ultimate SD upscale, and this is a low res upscale too lol
compared to the original
Is there a science to the tile size settings? I was simply trying to go from 1024 to 2048, changed tiles to 768, padding to 16... was getting blurry seams everywhere
the denoise is your enemy. The higher the denoise, the more visible the seams
how do you get ultimate? i dont see it in the upscalers
Yeah, how do you get such good results with SD Ultimate Upscale? Is there a general prompt you use for good results in the img2img tab, best denoising factor, etc?
I recommend doing it like this
Well I don't think I've used any higher than .3 denoise
Low res gen, then upscale 1.5 with Ultimate with denoise of like 0.5 so it can get a lot of new detail, then 2x upscale again at .2 denoise to refine the details
Ive used it before and have gotten decent results, but usually there are artifacts or things I don't want, even at a pretty low denois around .25 or .3
So I wrote 3 chapters on my book and now I kinda want to add art to it
(If you wanna read, heres the link: https://docs.google.com/document/d/1acwzvEnO0DD_9n-B67bPAubomHMZ0HYAEJKI8OYUxis/edit?usp=sharing)
Any ideas for art
(took over like 40 hours to get here)
or suggestions?
Im sweating with anticipation
I think the image I posted would make a good emote on this server <#🏞|general-with-images message>
Something very interesting going on with his neck there
alright, so I always use ultimate upscale with DDIM, around 10 steps (go into settings in the stable diffusion tab and enable force img2img to use the steps you say)
I use 0.3 denoise or lower with no seam fix, if I go higher I use seam fix (either one works, depends on the image and how severe you are going)
I use whatever tile size makes sense, the higher the res you are going, the bigger the tile size I would suggest.
I always use upscaler R-ESRGAN 4x+, as I find it adds the most detail without oversharpening to hell like normal ESRGAN_4x
I recommend linear type unless you know what you are doing with checkered
IMPORTANT
I have tested and validated that ultimate SD upscale works better when you remove the original prompt, and keep the quality words. Keep the same negative prompt it generated with. For example:
A pretty woman with red hair smiling at the viewer, portrait, realistic, detailed, 4k, bokeh, photograph, photo realistic
(your negative here)
that result should be upscaled as so:
Portrait, realistic, detailed, 4k, bokeh, photograph, photo realistic
(same negative)
I do not know why it upscales better like that, but I have tested and confirmed that it indeed does
This image was upscaled using the identical pos and neg prompt
and here it is with the modified positive
Have you used Lora with that method?
you can see it is more detailed without the original words, only the quality words
I feel like the skin has more texture in the first one
both are impressive. what model was it?
You could heavily weight skin texture to fix that in both, it works really well on the models I use
could fool me for some actual photo reals
I believe it was realistic vision 1.2
its a crop in of the full image
sorry
1.3
they just released 1.4 and 2.0 tho, and 1.4 seems to be a little better
(wanna check 2.0 right now, so stay tuned)
oh? on civ?
yessir
I am gonna run the same prompt through all 3 to compare
I do not have xformers noise, so they are completely model differences
laddy guhguh
Idk man, I got pretty terrible results... might be because I'm using realistic photo.
Then tried Multidiffusion and got perfect results (I'm only having issues with txt2img atm)
Yeah, maybe it just works better for what you are doing
I just know that with my experience and understanding of ultimate SD upscale, I have been able to get better looking and more reliable results out of ultimate in less time
i have found from my testing that 1.4 has better tonal contrast than 1.3, so maybe 2.0 is even better
interesting, here 1.3 and 2.0 look closer
can you imagine ai in 100 years 😮
Stylized pbr material. Made with RealisticVision for the realistic reference and NuclearAnimeHybrid for the stylized look.
if humans still exists around then, it'll probably be so integrated into peoples lives (like cell phones, television, the internet, automobiles, etc) that no one will care except people from our generations.
Seems like the clothing is changed mainly heh
oh man, I wish you could have seen barium AI before we were bought out by unity. Its this, but like, insanely good haha
true rally. it wont even ai, just a part of life. lol
It was a text to PBR material gen that did all 8 map passes up to 8k with no seams in 30 seconds
man ive been wanting to do stuff like that. What program did you use to make the material? I know of one thats free but never used it to make anything.
Unity bought a material generation program?
It worked so damn well, but Unity bought it out lol
anyone here done modelscope text2video on auto? just trying to figure out how to control length of video
yeah, its part of their new AI tool suite they are working on
Im a big unity user, that would be so cool to use but it will probably be behind a paywall
i sadly have no info on if its paid or if it runs locally yet
Pretty sure I have seen someone leak an unlisted prmotional video of that ai tool suite on reddit
Substance 3D sampler, its easy to use, creates the rest of maps for you.
my partner (the main dev) is under NDA still
I just asked him not long ago lol
I sure hope there is at least an option to run locally
thanks
ill see if I can remember the name of the one I found
He was at GDX showing off demos
I was just an artist, so i wasn't taken with the team
oo sound designer, neat
Thanks, I do all sorts of complex audio work. Working on an album lol
😩
Yeah im not good with sound, I can make a basic lofi track in fl studio and basic sound effects for games but thats it
photo from my wedding
/Cute 3D-style girl
If anybody wants to listen, this is the very basic and bear album intro demo WIP for my album. The whole album is based off of my chromesthesia (audio to visual synesthesia)
I made over 95% of the sounds completely from scratch
Ok last cake 😛
its very early lol
awesome
ooh but i want one more slice haha
Illuminati Diffusion bakes some amazing cakes
I can find some of my basic tracks if anyone wants a listen (my goal is not to be a music producer or anything I just find it kinda fun)
here are the isolated drop layers if anybody wants to geek out on the technicals. Its 9 different layers
cant find that program, must have deleted it. Ill have to check that one out.
oh, its adobe
I don't have adobe :(
and this is the other way more basic demo i have for another song on the album
this sound is 3 layers, 7 notes, 41 effects, and 14 automations of pure fuckery :p
And here are two nasty atonal guttural basses from my last sound design session
(WARNING LOUD)
well hey I found your barium ai site
yep
I can still access the things I made
I have an admin account, but the service itself is down now
http://boundingboxsoftware.com/materialize/
I found the program I was thinking of!
I had a little chat about it in the unity forums just yesterday
Ah yes, materialize is very cool
Our tool worked very different though
you can use this free normal map generator: https://www.smart-page.net/smartnormal/ i haven’t t tried it. The rest of maps you can do in blender.
our dev made our tool several months before SD even introduced built in tiling. I still cannot believe how good he made it in so little time. It was less than 3 weeks from the start of the project to the acquisition by Unity lmao
dang
yeah, man was a beast
I recommend you read that short forum above I linked to where I shared some more information on it
Unity is gobbling up this AI stuff, I wonder if their suite will contain a custom trained language model specifically built for unities two languages and many libraries
Something like GitHub Copilot (though idk how that works tbh)
Supposedly, based off fo their teaser video, and some confirmation from my dev partner... Unity is gonna change the industry
ooo HOPEFULLY for free, or at least non-subscription based so I can use it!
If its only in the pro version I will die
He was tight lipped, but lax enough to tell me that it may or may not be able to make entire characters with custom textures, mapping, rigging and interactive animations from a single text prompt...
Figuratively
Our Barium AI tool is just one of the tools powering what this new suite is gonna be able to do
yeahhh... from that leaked video I saw (didn't include any graphics, just text)... their prompts were quite lavish and it seems like it is using some sort of language model for easy control over it all.
yeah, its in very early beta though, so do be warned
he said the stuff he was testing is stuff that is far ahead of what they may put out any time soon, as is expected
either way I am excited to see where it goes, but cautious of the monetization aspect
Same here
I even voiced my concerns about it in the official barium Ai sever (I am head admin there), but its basically dead as the tool was live for less than a month before being schlorped up by Unity
THOUGH
I will say...
@hasty novaDo not quote me on this
but he DID say not to fear, that Barium will still be usable as previously, but instead for free in unity
thats what he said at the time, I am not sure if its paid now, or of the full suite is paid
"but he DID say not to fear, that Barium will still be usable as previously, but instead for free in unity" - Sytan, 2023

alrighty time to test out materialize
Mulher
Oh, I also did these two renders for us
I wonder if any of my embeddings could make interesting textures
I can send them now cause they are no longer active promotional material
These two banners used to be lower on the site where the sign up and user statistics were
pretty
All materials were made with Barium AI V0.1
noice
im interested in like stylized textures tbh
That was something barium struggled with, as it was built entirely off base SD 1.3 I think
I am pretty sure it was pre 1.4
again, don't quote me 😅
but yeah, stylized textures were something we weren't nailing yet
We were also in the process of working on a second system that could do fully 2D tillable textures for platformers, as well as low res pixel art generations with bit locked color pallets
ooo
epic
oh no way
the built in material viewer still works haha
I was able to get these more stylized bricks
as well as these
fun times, I miss them haha
I missed whole thing, what are we talking about?
Another 3d texturing thing?
I worked on a tool called Barium AI that was a very very good AI PBR material generator
It got bought out by Unity less than a month after we launched, and its now going to be part of their AI suite
hm, interesting
Is ds still broken?
but I want AI sculpting based on image and prompt
||The new Unity AI suite is supposed to be able to make models, texture them, rig them, and animate them off of a prompt, not sure how accurate that is tho||
okay you got me excited
it should also be able to control cameras, terrain, world settings, lighting, and even material animations, from what I have heard
yet to see it for real tho
Their teaser shows people saying thing slik e
"Make me muddy terrain with an overcast sky"
So next missing tools will be music AI gen...prompt based \ key based \ some piece of music
I know we have some, but they aren't great 😄
Im surprised there is not some better ai for music
Same here, though music isn't often something you can "describe"
especially not more experimental and expressionist stuff like what I and my peers make haha
We can describe genre\ key or notes it should go off of
no but i could see more sliders, like this genre, this tempo, this beat sound, this here, this there, etc i dont know
yeah, that would lead to super basic things
Its easy to say "make me a 140bpm electric guitar riff with an energetic feel in g minor"
but you know... i am really not too critical of any of these things yet. we are in such of the early days you know. it is like watching the first wire connect for the telephone, a major change for the world but still just some early days lol
But good luck explaining an atmosphere like the one I made here lol
suffice it to say, with video obviously being possible with AI, we will be getting music, voice, everything to make our own film in short time lol
Or like explaining the spectral and tonal textures of something like this, along with the syncopations and poly rhythmic characteristics
I've seen some videos from youtubes with AI gen music videos and AI doing interesting things with prompts or music pieces, the main problem is - it's not consistent and messy.
It is pretty creative and what's the right word...versatile, I guess, it's just not clean.
i can finally put myself in movie lol
not great tiling tho
Tilen't
bit I suppose it would be used on a road or something
Oh god, I forgot
because it technically tiles
I made a secret clip of my album intro rendered as if its in a school bathroom lol
Its using a real convolution sample from an actual school bathroom lmfao
neds more fighting and people smoking (I speak from experience, and I go to a pretty well renowned high school in a fairly rich community (I am not rich but I go to the rich school))
lmfao, fair haha
oh...and another tool I'd like is to make AI watch youtube video for me and sum it up lol
Some videos are just too long for no reason, but cause author wanted to make it long...
I made a version with a toilet flushing lmao @hasty nova
beautiful
needs to be lower imo
One of my friends challenged me to make my own take on Phonk mixed with my own more atonal and gritty style in 30 minutes, and this was the final result lol
here is the 808 on its own. Really proud of the CrUnCh
I am an FL studio goober
ah
Im "borrowing" someones fl studio license (they gave me permission, dw)
im not making anything to sell I just wanted to see what it was like
Also, friend then asked me to make it way more atonal and ditch the phonk and I ended up with this nasty and squelchy heat
gosh I am trying to find the not worst one
Can someone tell me how to use stable diffusion?🙂
Check out these videos for a great overview of the website and how to use it! https://youtu.be/014J2Yo1aGI | https://youtu.be/M3jCa6qTHpQ
try /faq @modest anvil
I am friends with a few pro producer friends. One of them just preformed at Ultra Miami last night lol
dunno what that is but sounds epic
Its one of the biggest music festivals in the world
I guess this one isnt too bad
I dunno anythign about music theory so pardon my chords
@hasty novathank you
It had 165,000 attendees this year
woah
Oh, this isn't too bad
I was expecting way worse lmao
It sounds kind of off to me
I have met so many people that just... Shouldn't be sharing their work IMO 😅
but yeah, thats not that bad at all
And this was when I was trying to make better melodies
the concept is there
Yeah, sounds like you have a base concept, but you could use some better understanding of composition, mixing, and mastering
say that to every SoundCloud rapper producer and the world will be a better place
I know the basics of mixing, none of those really have any mixing because they werent trying to be full songs, more just me learning how to use fl studio
I was also trying to do better drums
Hey, there are plenty of good people on sound cloud haha
thats where all of my professional friends came from :p
Lemme fix the message
Some of my faves from my bestie
they are the one that was just at ultra
This is the first thing I ever made in FL Studio, it was following a tutorial pretty much exactly
actually
I might have earlier things
That was the first half decent thing I made
god I had some terrible projects
Hey, I have some shit WAY worse than what you sent lmfao
I am not even comfortable sharing it, its that bad lol
oh i havent sent the bad stuff
I might share it just cause its kinda funny
the drop 💀
so weak
also this hot garbage
but apparently the tutorial was the first thing I did, the stuff I did after was just terrible
this is not actually my worst stuff, I don't know if I still have my worst stuff though (I used to use different DAWS that have since been deleted)
Ok, I have a sound design flex for you haha
I made this base sound, and its super bad (cause its not processed)
this is one of my favorite examples of what I do
I took that base sound I made, and turned it into this
wrong file before
fancy
here it is as the demo
I dont make that kind of music so idk what goes into that
I like lofi and orchestral
And rock but thats quite difficult to make
Here are 3 separate demos for the same song
jesus, these demos are almost a year old

this was my first lofi song, I think I followed a tutorial
this is actually pretty nice
wait
did I? I feel like I made this melody by myself, idk
not really lofi but I think it's ok
I am really wanting to figure out of to make sexy crispy and melodically tonal sounds like sharks
😩
Yeah, hes a very prominent artist in one of my target genres
his music is so fucken WET
ok I am pretty sure this is the only song I haven't shared from my fl studio projects that actually has a rendered file in the folder
oh, eliminate is amazing too
