OH I REMEMBER
#🏞｜general-with-images
fallow egret
minato?
smoky oak
no, not from that series
I got a major mix up
I was thinking of Slave 23 from FMA lmao
fallow egret
ehh, I don't see it
smoky oak
I have a severely strange taste in men
fallow egret
I never liked pony tails on guys
smoky oak
well, I should say I have no consistent interest in men
I just like men
its like a DND dice roll lol
fallow egret
I got so bored watching FMA that the person who introduced me to it unfriended me
fallow egret
IDK why but this character you like is unattractive to me and I don't like his hair
smoky oak
understandable
its not the best picture of him either
only one I could find
Oh also
actually, nevermind
fallow egret
I searched him on google images, they all look the same to me
smoky oak
hes just the guy I was thinking of instead of Naruto, I am not like super into him or anything lol
I think greed is far more attractive lol
static tusk
finished what I was working so thought I would give the redhead a test. Using realismengine
smoky oak
if there was enough good rendered fanart of greed, I would totally make a LoRA for him
fallow egret
I think greed is far more attractive lol
He looks good but I still think he'd look better without the ponytail
How come he doesn't have it in this image?
smoky oak
He looks good but I still think he'd look better without the ponytail
greed doesn't have a ponytail
OH
Later greed
no, I mean first greed
fallow egret
why is he wearing reverse blackface?
fallow egret
he looks like he has been rolling in something
smoky oak
His body hardens by pulling the carbon out of his cells and turning it into basically a black diamond shield, thats what you see
fallow egret
Why does he cry when he turns black? that's racist
smoky oak
bro, you are unhinged lmao
fallow egret
that's literally what's happening in this image isn't it?
fallow egret
I guess they just made the white of his eye look like a tear
smoky oak
and hes not turning black lmao
fallow egret
but he does turn black
smoky oak
technically he turns blue, but the fan art is in black and white lol
dense tapir
@smoky oak Yep, as I figured and Jensen just said it. https://www.youtube.com/watch?v=zUc_rcoFmjU All of this push for smaller and smaller die sizes is why. I knew what TSMC was saying for costs as 5 to 4 was almost double the price and down to 3nm isalmost impossible and is causing a lot of delays plus HUGE cost increases. 40k per wafer which I forget how many were on it but it came to about (not counting any failures that can't be used for lesser card versions) 200 per pop.
If they can't lower the cost at the manufacturing end then they are going to have to figure something out as consumers are not having it. Bad place to be in.
brazen anvil
How do you like the Tiled model from Controlnet? I love that you don't lose the general structure of the image as easily as when I make a classic img2img, I also tried it in ComfyUi and it works great.
1.0 denoise
dense tapir
No idea as I don't use it. Probably will one day if they ever make all of them for something beyond 1.5
smoky oak
1.0 denoise
I have messed with tiled diffusion on several occasions, but I have found that ultimate upscale works better and faster for most things, personally
past pelican
full marsh
it's only taken me 2 weeks of butchery and mistakes to learn enough to get something I'm willing to share. prompt at https://civitai.com/images/669794. It really chopped up her hand and I had a hell of a time correcting it with inpaint. I know it was caused by the controlnet image having 2 sleeve openings (like this character's right arm), extra hands were trying to come out all over the place. I did also try to get a big chunk of broken glass in front of it to cover it but it kept melding with her arm lol, I settled for just cropping it. I think the only other thing I'm not thrilled with is the detail in eyes, but I don't have any idea how to fix it, they all seem to be a muddy mess no matter what I do. These aren't terrible but I've run all the way up to 150 steps and still have trouble with them. if anyone has noob tips.
ebon loom
this is Beautiful 🫶 . could you share the metadata for this one? and the model u used?
dense tapir
this is Beautiful 🫶 . could you share the metadata for this one? and the model ...
It's 2.1 SD
the model I used is perpetualDiffusion10_sun
ebon loom
It's 2.1 SD
i use 1.5 though, but anyways, loved your output. the lighting in 2.1 is so much more sophisticated
dense tapir
One of the reasons I use 2.x since 2.0 hit
I do miss the camera controls in 1.5 though
ebon loom
I do miss the camera controls in 1.5 though
woah, i had no idea about that
dense tapir
yes, that was a hard pill to swallow
dense tapir
dense tapir
dense tapir
bold cave
it's only taken me 2 weeks of butchery and mistakes to learn enough to get somet...
the real answer is that noone cares if ur character has shit hands
as long as the rest of the images rocks then who cares what the hands look like
bold cave
Ladies and goontloomoon @smoky oak
The official SenBlend 1.6.3 general purpose model is out now on CivitAI! https://civitai.com/models/49744?modelVersionId=54293
The challenge is for someone cough cough @smoky oak to match or even beat (ha!) this model for general purpose realism. For everyone else, go snatch yourself the model and start diffusing. You won't need any other model until XL comes out.
Warning: NSFW Model
aaand theres already 5 downloads on it. Damn it feels good to contribute
glossy reef
nature 🙂
wispy nest
noice
dense tapir
mortal cape
Hi, do people know how to take ultra-realistic pictures of feet here? Only feet not the all body
mortal cape
@glossy reef
glossy reef
I do not know any foot fetishists just here, in general, the upper and lower limbs are a difficult thing in ai 😂
always if there is any problem it is with hands or feet
but you could try a simple prompt like close up view on feet
civic moth
#1072220168534642768 create a image of cat
glossy reef
and see what comes out of it
mortal cape
and see what comes out of it
Yes, but my customers want them and don't know how to use AI and I have a problem with feet. For the hands I have badhand_v4
mortal cape
but you could try a simple prompt like close up view on feet
Already done
And got a horrible pic
of a destroyed foot
I can't call that a foot
glossy reef
tricky, never been focusing on feets in my creation
wintry flume
guys hello
glossy reef
hello
mortal cape
hello
wintry flume
where i can generate images as im new to this server
mortal cape
tricky, never been focusing on feets in my creation
Np thx bro
glossy reef
maybe some more nsfw photorealistic models will be better with that?
wintry flume
where i can generate images as im new to this server
.............................
glossy reef
where i can generate images as im new to this server
on dreamstudio or dreambot
mortal cape
The problem is that they already have characters created so if I make a whole image with my basic prompts I don't know why but adding feet will make the image look horrible
wintry flume
on dreamstudio or dreambot
no permissions
mortal cape
I have to go, see you soon
topaz trench
Hey folks. So Im currently trying to get behind how this was made to better understand SD.
My guess would be Deforum and ControlNet for consistency?
However currently I seem to struggle mostly with model+prompt for this kind of work. What model would be best for that style? Also does anyone have some examples on how one could archive these stylized effects?
dense tapir
Hey folks. So Im currently trying to get behind how this was made to better unde...
Not so fast
topaz trench
https://www.youtube.com/watch?v=rlfhv0gRAF4
thought about that as well. but how would you get those smooth transitions between styles in Temporal?
dense tapir
thought about that as well. but how would you get those smooth transitions betwe...
Not sure but it is done.
I don't use any of that as I am on 2.1 so controlnet isn't fully implemented.
topaz trench
do u only use 2.1 model as well?
dense tapir
do u only use 2.1 model as well?
Yes
static tusk
I don't use any of that as I am on 2.1 so controlnet isn't fully implemented.
just watched the video. He uses Canny and HED controlnet models. both are available for 2.1
dense tapir
just watched the video. He uses Canny and HED controlnet models. both are availa...
VERY good to know as I know nothing about control net. Just need a faster card to do it.
dense tapir
just watched the video. He uses Canny and HED controlnet models. both are availa...
Why are there almost 10 gig ckpts and the same file in safetensor is 728megs?
hasty nova
Why are there almost 10 gig ckpts and the same file in safetensor is 728megs?
The smaller ones are extracted. There is very little quality loss between the big and small ones, just use the small ones.
dense tapir
Once I start doing video I will use them for sure unless something new comes along by then.
So many things happening at once the expenses are too much. Need a new case, my gpu is not doing well, my keyboard is dying on certain keys (been cleaned) as is my mouse acting up at times (they served me very well so no complaints). Now I need a new SSD sata (not going second slot on mobo) because when I load up webui it is 100% on my HDD and takes so long to load a model or really anything pointing at it (such as my main browser).
static tusk
yep. when the models are first created they are 10gb, then pruned down to 700mb later
dense tapir
that is so much tossed, lol
static tusk
when he first made these models he wasnt able to prune them right away either
dense tapir
I know my video so I have no choice it will be a 4090 as I slowly save for it.
24/25fps needs a good card to gen fast and having nvenc x2 is nice as well.
thick iron
Could someone tell me what prompt to use when I want the tiny lights/dust in the background of this picture.
hasty nova
Could someone tell me what prompt to use when I want the tiny lights/dust in the...
"particles, cinematic"
thick iron
@hasty nova Thank you. 🙂
smoky oak
The challenge is for someone cough cough <@296720689468145664> to match or even ...
Personally... I don't think this model looks realistic at all-
It looks like wax figures to me, personally
hasty nova
You might want to include more sample images in the civit page, just one isn't good enough to get a grasp on what a model can do, especially for general purpose
smoky oak
I just don't think left looks anywhere near as realistic as right-
dense tapir
@smoky oakI went to their link and saw the pic, saw 1.5, closed it out and say that is the worse trans Taylor I have yet seen.
smoky oak
I'm noy gonna go as far as to insinuate a "bad" image means she is trans, but yes, this model is definitely not realistic-
dense tapir
Well, let me for you cause I do.
Bad model is bad
Its like light beer it is a wannbe but just really isn't.
Now the model could be tweaked I think
smoky oak
I, personally, much prefer to not drag a highly marginalized group of people into the arguments as to why something looks bad
😶
oh, and their model is a checkpoint too, not even safe tensors
:/
running a comparison right now
ok, interesting
so this model is actually pretty good...
Looks like they are just realyl bad at prompting?
dense tapir
Who knows but that image looks like something 2.1 does using a body double as I call them. Almost but not there.
smoky oak
really confused why they would use such a non realistic image to show off realism...
bold cave
You might want to include more sample images in the civit page, just one isn't g...
theres about 8 I think
smoky oak
there is 1 SFW, and 4NSFW
bold cave
Everybody's a critic. Let me see your ultrarealistic model
smoky oak
and they all look like plastic
bold cave
Everybody's a critic. Let me see your ultrarealistic model
Also, these are one-and-go gens. No inpainting no upscaling etc
smoky oak
The weird thing is, your model is doing damn good in my testing, so it means it was just your prompting that made them look like them
smoky oak
Also, these are one-and-go gens. No inpainting no upscaling etc
same with mine
bold cave
As soon as I'm done with my 4k wallpaper you'll see exactly what I mean
If you have good results, you can post them under the civitai comments
That helps showcase the model better
It's also 1.6.3 because it's the fourth iteration out of 3 inferior ones. Trust me this combo is as good as it gets
smoky oak
Here, its clear my model looks more realistic, but your model does just fine when prompted properly
bold cave
The issue is that the one on the right doesn't look like tay
smoky oak
I am doing 1 to 1 comparisons right now
bold cave
at least not a whole lot
but you can always inpaint
also, the one on the left has "cyberpunk background" which gives it that iconic highlight look
It's just a personal preference
Idk. I love the taylor on the left. The one on the right would be good if it looked more like tay
smoky oak
as you can se here, she has a much less masculine jaw than was depicted in your generation, and her nose is considerably smaller too
smoky oak
and mine has her mouth shape more accurately
weird thing is, they look the same when prompted the same
smoky oak
You made that with my model? no way. prompt?
not that image
bold cave
Also when are you uploading urs or when will it be ready
smoky oak
This is your model tho
bold cave
I wanna take it out for a spin with my prompts although I'm telling you we are approaching the limit of what is possible
bold cave
This *is* your model tho
yaaaaaas
Im telling u its flexible
how is it with non-humans
smoky oak
Also when are you uploading urs or when will it be ready
I am probably not gonna release it, after seeing that the base model I am trying to mix is better than any of the outputs
smoky oak
how is it with non-humans
let me see
bold cave
that happened to me too
until I found out which model was the culprit
I originally wanted this to be a blend of 5 models
but more is not always better. some of them shitify the final result
bold cave
This *is* your model tho
hey do you usually generate at this base resolution?
smoky oak
yes
bold cave
I usually dont go over 768 because faces start getting distorted
but I guess for a model as abundant as tay theres so many pictures it doesnt matter
smoky oak
that is 512x768 high res fixed to 1024x1536
smoky oak
I have a feeling the reason your example images look like plastic/fake is cause of your prompting
bold cave
What positives or negatives do you use for realism. It also depends on who the subject is
smoky oak
my prompts change drastically from gen to gen
bold cave
Because of the regularization images I use on dreambooth, I usually dont have that problem bc the reggies have all detailed faces
bold cave
my prompts change drastically from gen to gen
yeh there is no one size fits all
u also might wanna post that tay bc it has no exif and it needs to have generation data to be posted under civitai
hasty nova
there is 1 SFW, and 4NSFW
Oh that's why, I have nsfw disabled
bold cave
but im telling u. Once I show off my 4k there will be no competition. I will have to change her face a little because it is unfortunately based on a real person
smoky oak
Oh that's why, I have nsfw disabled
same here on mobile
smoky oak
but im telling u. Once I show off my 4k there will be no competition. I will hav...
4k is nothing new at all, but I'd be interested to see if it looks realistic, or like plastic like your other gens
hasty nova
Man I just wish digital diffusion worked out for you sytan :/
bold cave
Oh shit I think its the photo studio/photoshoot part that's making them look like that
Cuz I think most females under that tag will have makeup
I'll do some more experimenting although Im happy with how they look
smoky oak
@bold caveI must say, your model is doing really damn good right now. Its not quite as "realistic" as the base I am using, but the composition is on average just a little nicer
bold cave
im telling u. and I didnt even need to make a new one for months because its predecessor sb1.5 was just as good
When I say I hit the spot I really mean it lol
smoky oak
Your model looks good, but also doesn't listen quite as good as most of the others
but by not listening, it tends to do something a little more visually appealing
like in this case, I wanted the result I got from the other model, but the less accurate image your model made looks better IMO
bold cave
Ima try increasing the rez, Im used to small images. that could also help
smoky oak
your prompts seem to be your downfall
way more realistic, and the skin doesn't look like rubber 😅
smoky oak
I think also the hires fix is whats doing something
on which image?
bold cave
I just became acostumed to never using hires fix since it was ass in the old-school SD
the asian girl
hasty nova
Well I've decided to download digital diffusion on my laptop to mess around at school, but since it's on school wi-fi it's going to take ages
bold cave
ima try to recreate her with my own dreamboothed girl
smoky oak
I am messing with some direct comparisons
your model is bothering me in one way
its huge, and is very slow to change over
its by far the slowest of all the ones I have
bold cave
you mean slow as in how long it takes to generate?
smoky oak
ok, your model is about 10% behind on tigers from what i am seeing
bold cave
it is pretty big at 8gb lol
smoky oak
Oh! I forgot to add your config file
bold cave
yeh
smoky oak
lets see what that does
hasty nova
it is pretty big at 8gb lol
Dang
bold cave
probably nothing, I dont even know why it came with a yaml
smoky oak
you mean slow as in how long it takes to generate?
no, the same speed to gen, but when switching between models, yours takes like 2x longer to switch to between all the others
hasty nova
Well cause it has 8 gigs of data to load
bold cave
DUDE
no way
hi res fix makes ALL the difference
this tool isn't what it was back in my day
hasty nova
My model is somehow 2.4 gigs, never pruned anything
bold cave
Not to mention it did that with just 15 steps. Im used to doing 90
lmfao its literally ur same prompt
but let me try it with my db girl
hasty nova
Well it finished downloading, faster than I expected
hasty nova
I feel like jumping on this realism train once I have everything loaded up
bold cave
Does lanczos preserve details better when upscaling that rsergan?
Cuz im pretty sure it does
smoky oak
depends on the use
hasty nova
Ugh maybe I should just install Vlad's UI
Auto is being annoying
(i haven't used my laptop for SD in a long while)
smoky oak
ok, I am testing more edge cases with your model and its starting to fall behind a little @bold cave
smoky oak
Ugh maybe I should just install Vlad's UI
Vlad is a whole ball of its own issues. I have had more problems with Vlad than Auto
bold cave
what issues are u having with a1111
smoky oak
what issues are u having with a1111
Does your model have noise offset?
bold cave
not unless the models I used had it
smoky oak
oops, meant to ping, not reply
smoky oak
not unless the models I used had it
weird... It is having some touchy exposure issues
bold cave
Ok I figured it out
I think ur prompts work well with general people
but as soon as I use a custom model it goes kaput
smoky oak
I really wasn't trying to just "hand over" my prompt to you BTW, but thanks for ripping it from the data
your model seems to have exposure issues
another thing some of the models I was testing had as well
hasty nova
Vlad is a whole ball of its own issues. I have had more problems with Vlad than ...
I prefer it for some reason, once I got it set up I've never had issues.
hasty nova
Yep, everyone has their own problems and their own solutions, gotta just respect what others are doing.
smoky oak
alright, the more I use it, the more issues I am seeing with this model
this looks super weird
bold cave
what do you mean exposure issues
smoky oak
compared to the other model
smoky oak
what do you mean exposure issues
it makes white and tan people too bright
bold cave
Ive found that it works better with low cfg or using the cfg fix extension
yeh the CFG can fix it, I usually go for 4cfg for my dreamboothies to fix the oversaturation
also I always have to keep "oversaturated" in the neggies
smoky oak
I am testing all these models at the same settings for fairness
its not as bad as with some other models, but I am seeing it
your model struggles a LOT with bokeh
bold cave
what is that
smoky oak
consistently thats the biggest thing I keep noticing
smoky oak
what is that
background blur
smoky oak
yours
base model I am using
yours keeps like, being in focus and out of focus
let me test some more
bold cave
holy crap
check this out
tell me this looks plastic lmfaoo
yes its a fail because hires fix doesnt use cnet
smoky oak
it does lol
smoky oak
although, that is the best you have shown so far
bold cave
yeh but also controlnet > hires fix, even tho hires fix is much better than before
other than speed I still think cnet is a better choice
smoky oak
yeah, there is fine texture, but the underlying skin texture isn't realistic, it looks like rubber still
smoky oak
so far I would say your model is top 3, behind the two that I was messing with for my base, but it has some glaring issues I am not happy about
bold cave
that dude cherrypicks his gens
try the cfg fix for the glares trust it works. also add overexposed, oversaturated, deepfried meme, etc. to the neggie list
smoky oak
I am not gonna cherry pick tags for a single model, they all get the same prompt to work with
I am done with testing this model
bold cave
I hope this girl doesnt find out I will use her as the poster girl for my model looool
smoky oak
it did surprise me for some compositions, but edge cases are not there IMO
bold cave
If I cnet upscale that it will gain all those tiny wrinkles
hasty nova
that dude cherrypicks his gens
I mean, he's also using base SD 2.1 which is hard to get those kinds of results for
bold cave
oh is he
what a clever little rat
once xl drops out its gonna be fair game for everyone
these models are just to keep us busy in the meantime lol
smoky oak
Wait a minute
bold cave
I am done with testing this model
once u do end up uploading urs I'll help post some examples under its civitai comments
smoky oak
once u do end up uploading urs I'll help post some examples under its civitai co...
As of now, I am not making a merge
bold cave
I actually want to see if blending this with another megamodel will work better or worse
smoky oak
this model I found is just better than everything else I have been able to blend
@bold caveWait a minite
are you the 80GB God?
bold cave
Im not breaking any rules anymore tho
smoky oak
No wonder your generations looks so familiar in texture
bold cave
If they do end up banning me again just out of spite then they'll force me to come back with a brand new identity completely anonymous
I like hanging round here cuz people post useful stuff pretty often
bold cave
Me, that was a sarcastic name though. Just like my current name, obviously nobody has a 7ghz processor unless ur constantly pouring liquid nitrogen on it lol
smoky oak
nothing, no need to worry about it
bold cave
smoky oak
I just realized it was them
bold cave
Well I aside from my new kewl public model I have a little extension which I use to this day credited to my name
always looking for python/gradio devs
hasty nova
Finally got everything set up, but my internet is extremely slow because I am tethering my phone to my computer over wifi to use my vpn
I was setting stuff up on my laptop
I haven't used it for SD in a long time
smoky oak
Do you have a remote client running on your home PC?
hasty nova
No
smoky oak
or are you running it on your laptop?
hasty nova
My laptop has a 3070, it can run SD just not as well as my PC.
smoky oak
damn, very nice
I just run a persistent web URL I can access from anywhere in the world to access my home PC like a web service
password secured too
hasty nova
I should probably set that up
smoky oak
much more annoying than just running it as a website
cause then when you load it on mobile, it renders the UI for a mobile device
bold cave
daaaaaaaaamn
well the details look insane but she's no longer the same girl
ur using 2.1 orange u u lil sneak
smoky oak
Peter and I are the realism buffs lol
smoky oak
I have a recommendation for a model I found that beats real vis quite a bit in realism, if you want, I can send in DM's
hasty nova
These are eh, but could probably be fixed with a higher denoising on hires fix (currently at .3)
smoky oak
Real vis vs the new model I found
1.5
its the best model I have ever used
no, that model does pretty terrible for realism
second worse, behind protogen 3.4
hasty nova
Is this decent realism?
smoky oak
Is this decent realism?
ehhh, not super realistic, but it does look really good
I have no idea
hasty nova
ehhh, not super realistic, but it does look really good
I agree but I'm not sure what could be improved with it?
smoky oak
0 clue, cause I have never used a reliably realistic 2.x model
thats why I don't use 2.x
hasty nova
But what's making it look not realistic? Is the lighting too perfect or something, because it doesn't look plasticy to me.
smoky oak
It looks more fantasy like IMO, like the lighting is a little too perfect, and the textures are a little too soft
Peters remake looks much more realistic, with the way the light falls off, the background blur, skin texture, and ligting
hasty nova
Well that was the goal, but yeah I guess everything is "too perfect"
smoky oak
this one looks much more realistic, awesome job
hasty nova
I can't see his, my Internet is too slow to load it 😆
smoky oak
@hasty novadon't get discouraged, realism always best, your image has a great aesthetic to it!
hasty nova
I loaded the first one, yeah it's probably better
smoky oak
the second is very similar to yours, just more realistic
bold cave
But what's making it look not realistic? Is the lighting too perfect or somethin...
these guys are nazis lol no matter what you post its not realistic enough
As long as you like it thats what matters bud 👍 plus we're here to help improve our prompting not to have a dick measuring contest
hasty nova
the second is very similar to yours, just more realistic
Yeah I guess there are more imperfections on the face, but honestly it seems like too many that it was forced.
If you were to zoom in
smoky oak
he was also having to work with all the artifacts baked in to your 2.x model as well
hasty nova
I don't think that was with my model
Alrighty, but I can't do anything anymore I'm in a different class now
So I actually gtg
smoky oak
yes, they are
wispy nest
made that on a 2 gb gpu
smoky oak
nice try
smoky oak
I have messed with a 4GB GPU, and it was able to do 1024x1024 with high res fix
prettyy crazy how far this tech has come
bold cave
damn
hushed stump
Hello! Is there anyone experienced with inpainting or just SD models in general that can help me with a small mother's day project?
Right now I'm trying to remove confetti from a picture of my nephew (some of the confetti covers parts of his face and body)
bold cave
face could do some work
bold cave
Hello! Is there anyone experienced with inpainting or just SD models in general ...
do you have more pictures of the nephew
depending on the amount of confetti you can either photoshop, dreambooth the nephew and inpaint him or otherwise mask it
hushed stump
do you have more pictures of the nephew
yes! May I DM you? I don't feel comfortable uploading them here at the moment
smoky oak
I am not really doing challenges anymore. We are both very good at different aspects of realism
bold cave
yeh
hushed stump
friend request sent!
bold cave
I am recreating a girl I know from memory and that looks just like her
Im testing the theory to see if you can do a convincing dreambooth model of someone with just 1 picture.
The results might surprise you
Do realistic metalocalypse like I did above
wispy nest
gpu abuse is fun lol
bold cave
ayyy he looks like he loves a good beer lamo
remember when people were amazed when someone made realistic simpsons
wispy nest
he can be my friend
bold cave
funny to think we can make realistic anything at this point with a few clicks
wispy nest
quite impressive if i say so myself
what gpu yall got
wait what lol
what phone
;-;
neon crystal
So I explored photorealism with 2.1 and agree with the folks who pointed about it generating stock photoish stuff. It had this strong analog-coffee-discount-Wes-Anderson-bokeh-dreamy-plastic thing going on which was kinda hard to get rid of. But again it wasn’t that bad either. After a bit of prompt hacking, I was able to generate these. These are obviously not as good as the ones that were shared here to recreate with 2.1 but hey if someone who isn’t even a photorealism expert with a grand total of 3 days experience with 2.1 can do this, I’m sure this can be bettered. Also, I don’t understand the hate for generating stock/professional photo type stuff ( at which 2.1 really shines ), I’m sure a lot of folks are interested in that.
PS: Yesterday I posted about 2.1 being really good for art but somehow it devolved into a pissing contest about photorealism 🤷
bold cave
So I explored photorealism with 2.1 and agree with the folks who pointed about ...
2.1 is definitely more detailed, noone argues that. The issue is flexibility. General purpose models can do celebs, nudes, crazy poses, all kinds of unique scenarios which are extremely hard to replicate in 2.1.
I would rather be able to do for example a lesbian kiss with slightly less quality in 1.5 than a horrible attempt at one in 2.1
well for custom models at least lol
its 768x768 so naturally it carries more detail
neon crystal
Nope. Realism Engine
What model do you use btw? Something of your own? Insane realism that without ever hitting that uncanny valley. Mind blown
Ah ok cool.
Thanks for the tip. Appreciate it. Will play around and see if I can come any close to your level of quality and detail. Cheers
hasty nova
Same settings but with "3D render" in the negative.
@smoky oak is it any better now?
I might try messing with the prompt some to get it to look better
Felt like it was too mossy, this is probably better
Meh it looks too photoshopped
hasty nova
Well I am just going to start messing around with a simple prompt generator tool I have and plugging it in to the model
wispy nest
noice
hasty nova
First go with the prompt generator, I think the prompt I got is pretty cool tbh
Here is a "digital art" one! I think it's really cool
Both are generated with the same model on 2.1.
And here is the prompt, generated with a standalone program I helped develop a while ago that I can link for anyone that is interested.
(digital art) of a gorgeous librarian with a rugged pouch in a sooty jungle, dystopian, gritty, ominous, incredible, colored, brutalism, ambient occlusion, highres, raytracing, masterpiece
wispy ether
hasty nova
wispy ether
hasty nova
<@800501037793280030>can you generate dean winchester
dunno who that is but maybe? It's 2.1, which doesn't have a great grasp on celebrities
And here is another with that prompt
wispy ether
dunno who that is but maybe? It's 2.1, which doesn't have a great grasp on celeb...
smash or pass
hasty nova
This close enough?
smoky oak
I am absolutely eating up this model. My realism is getting next level IMO
wispy ether
too big forehead
hasty nova
I've never watched the show, I just know the name
wispy ether
nose too long
wispy ether
I've never watched the show, I just know the name
formal geode
cat and dog fighting kung fu in a monk
wispy ether
cat and dog fighting kung fu in a monk
what
crystal deltaBOT
@formal geode
formal geode
the person is trying to prompt I assume
yeah I'm knew 😅
hasty nova
<@800501037793280030>are u familiar with highres fix
yep
im using it for those images
hasty nova
Some models don't do well with it, and also you can try different resolutions that are more native to sd
hasty nova
Stable Diffusion I meant, not standard definition
Like 900 is not divisible by 8
Try like 912
wispy ether
Like 900 is not divisible by 8
why does it have to be 8
wispy ether
It's a 768 2.1 model
wispy ether
how to filter models on civitai
hasty nova
Look at the base model, not then name. The name usually doesn't have to do with what version of Stable Diffusion it was trained on.
wispy ether
i want to try another sd 2.1 modewls
hasty nova
hasty nova
Oh cause it's a jpeg
neon crystal
wispy ether
the fuck
fluid walrus
Smash.
neon crystal
I am absolutely eating up this model. My realism is getting next level IMO
You can never beat @stone cipher STOP TRYING!
wispy ether
ew
smoky oak
You can never beat <@893635795657818134> STOP TRYING!
I am not trying to beat them lol
we are better at different types of realism
was just a general term
we have different strengths to our realism
neon crystal
was just a general term
Ofcourse I was just kidding lol 🙂
wispy nest
behold the power of my potato
wispy nest
senk go brrrrrr
wispy nest
with what model can i create something like this?
smoky oak
I use DDIM always
sinful mirage
hmm ill try it 1 sec
smoky oak
never found a reason to use anything else. Its the fastest, and only needs 15 steps to look good
sinful mirage
i usually do about 40 steps either way, although i usually go for abstract concept art type stuff, it does seem to mess with the "realism" of the img
ddim does make some cool stuff tho, it seems to combine some [ | ] prompts a lil more, this is a straight up illusion lol
wispy ether
sinful mirage
prompt was essentially [gross goo | coffee cup] so ddim did do much better, ty :)
sinful mirage
eyeball + summer home
pine briar
a painting of a rocky mountain range, a detailed matte painting, raytraced, light rays, glowing highlights, anime aesthetic, matte painting, clean, crisp, natural
have fun
sinful mirage
crystal bloom
sinful mirage
wispy nest
can never pick between these 2
++ SDE Karras
but eh, everything works, it's just my favorite thing to play with
smoky oak
You know, the sheer amount of articles of people gawking over mid journey never ceases to amaze me
Like, I would love for people to write actual articles about how much more creative freedom and capability stable diffusion has over mid journey. Like sure, mid journey can do some really cool stuff, but it's just not a competition
Like I'm seeing articles about mid journey 5.1 releasing and how it's a "revolution to AI image generation"
And then I look at their examples and I'm like bro, these look like batch images out of stable diffusion lmao
ON A SIDE NOTE
would somebody mind helping me experiment with achieving a specific sort of facial emotion?
static tusk
because the tools we use goes over the head of the average person, plus access to hardware. MJ is ease of access and use
smoky oak
I am really trying to get that sort of head down, aggressive/determined look under eyebrows sort of thing
like this sort of emotion/facial expression
I keep getting a pitiful glare, rather than a determined one
ripe cedar
Anyone know if Vlad install will work for the Al-Faraj SD Photoshop plug-in? I don't want to have to install another UI if I don't need to, ideally
smoky oak
Anyone know if Vlad install will work for the Al-Faraj SD Photoshop plug-in? I d...
I have no idea, all I know is that I personally do not use Vlad, mainly because it likes to cause issues with plugins, so that may not bode well for if it will work or not
wispy nest
Ok bye
smoky oak
I regret to see how your lack of understanding for how this works is our problem, or the tools problem
nah, no need to worry lol
you are at least being educated with your conversation, rather than resulting to childish insults
brazen anvil
lol
smoky oak
let them go back to Dalle if they need, no need to have their negativity in this community
yeah, no need to waste time on such childish people
wispy spindle
calm canopy
legend
smoky oak
Thanks fruit, much appreciated
pure reef
Imagem
#🏞｜general-with-images de um búfalo com armadura e olhos com raio azul nos olhos brabo
Aonde faço?
Obg
crystal bloom
smoky oak
image prompt courtesy of my immature friend haha
dense tapir
@smoky oakI sure hope this rumor is right but it appears the 4090 competitor 7950XTX is coming. AMD already said they were going to release it, and rocm 5.6.0 will work with it. https://www.youtube.com/watch?v=8LYbN-6CdwY
sinful mirage
tree
brazen anvil
I remember that in the space of ControlNet 1.1, Openpose generates the full version with hands and face, or did I dream it? . I was about to start doing the POW for this week and I came across this : c
dense tapir
youtube vids about CN doing hands
dense tapir
neon crystal
@stone cipher Not there yet to your brand of realism but yeah I’m happy with the direction this is heading.
opal rock
hi
calm trout
Not there yet to your brand of realism but yeah I’m happy with the direction this is heading.
fringe plinth
@stone cipher middle school
@neon crystal middle school
@wispy nest middle school
bold cave
lmao I come here to find another flame war and why am I not surprised of who the participants are
fringe plinth
i see
bold cave
dont feed those trolls
they do nothing but criticize other people cuz they think they're above everyone else
although I respect them as fellow diffusers and would like to collaborate instead of measuring dicks
dry crow
i see
You can check #1072220168534642768 and #1080946152318443610 to get more Information on how to use SD
bold cave
OOOOOOOOOH I forgot. I found this post on Aitrep's discord
If this is true then this is a MASSIVE game changer. Essentially ultimate upscaler with CNET omggggggg
I cant wait to try this shit. I assume it's new cuz that post is only 2 days old
Could it finally be that I no longer have to photoshop my prompts? Will post results soon
boreal falcon
Like, I would love for people to write actual articles about how much more creat...
Ofc is MJ a competition
Having advantage as well as big disadvantages over Stable Diffusion
bold cave
Sytan is 100% inflexible when it comes to changing his mind
he insists that 1.5 > MJv5 even though that is factually incorrect
ofc you'll never catch me paying for MJ. My motto is, if it doesn't have dreambooth AND cnet, it's garbage
boreal falcon
You dont need those
bold cave
Once XL comes out hopefully we can all rejoice and hold hands and sing kumbayah around the campfire since XL seems to me like endgame for image generation, and we can finally move to the txt2video era
boreal falcon
I mean okay for Stable Diffusion its almost a must have
But for such like Adobe Firefly those two are obsolete
boreal falcon
XL wont be the end game
bold cave
what if I want to portray myself as I did with my pfp
well we don't know if its 768 or 1024
boreal falcon
Had to correct myself above
bold cave
my final opinion is that as of now, MJ wins in terms of model quality but SD wins in terms of adaptability
boreal falcon
SD is superior in terms of flexibility and controlability by far
bold cave
thats an understatement. MJ baaarely has an img2img
and its interface is discord r0fl
let them be
some ppl prefer macs even though they are factually inferior. so let them
boreal falcon
But Firefly is here too although in early beta still and Picasso from Nvidia might be as well a big player and very likely will
bold cave
well it all boils down to two things. will it have cnet and will it have dreambooth. if the answer is no to either of those then it's trash.
boreal falcon
and its interface is discord r0fl
They work on web UI
bold cave
MJ doesnt
boreal falcon
Not yet
boreal falcon
well it all boils down to two things. will it have cnet and will it have dreambo...
For Firefly i dont need exactly Dreambooth or Cnet anyway
If i used SD i would
It also comes down to some other factors too
bold cave
so you're telling me u can finetune models in firefly and get the exact pose you want without cnet or db...
bold cave
what factors
there is literally nothing about firefly that puts it ahead of anything
aside maybe ease of use
boreal falcon
I prefer even more control which requires me tho to have some artistic and editing skills. I dont depend 100% on AI
bold cave
im a photoshop vet
boreal falcon
FF is in early beta
bold cave
if I can get AI to do the hard work for me, then im game
boreal falcon
Where was Midjourney in early beta
bold cave
its kool if u prefer firefly, but I guarantee 0 people out of the 100k in here will agree with you. it is quite simply normie garbage
boreal falcon
Plus...we speak about Adobe and not some trashy individual dev
bold cave
well a1111 seems like an individual dev to me and he is a god
his interface is literally what made sd possible
boreal falcon
its kool if u prefer firefly, but I guarantee 0 people out of the 100k in here w...
I dont have to listen to every trashy peasant tho
I orient towards other people
bold cave
shut sinew
If this is true then this is a MASSIVE game changer. Essentially ultimate upscal...
Does work well but cant upscale to very big resolutions with it, 4K is fine but at 8K i run out of VRAM on the 4090 
bold cave
who the hell needs 8k lamo
shut sinew
I do 
shut sinew
Combining AI Backgrounds with actual shot photos, and those tend to be high res
boreal falcon
artists, especially advanced and professional ones amongst all.
shut sinew
It can, ive done and seen some stuff
boreal falcon
It can, but not yet on level of Midjourney which is the king in that area now
not as consistent as MJ in that regard
bold cave
oh boy
can we just please agree mj model is better but SD is better at everything else and call it a day
better in details and thats it
boreal falcon
For an artist i would prefer SD over MJ anyway if i used it
Because control and flexibility is far more important for me and if you are artistic and editing skilled you compensate those mistakes on images
I dont need a ready to go picture
bold cave
also last time I checked MJ has no inpainting m8.
boreal falcon
It doesnt have them, yes
bold cave
so literally any serious artist is gonna prompt on MJ but then go straight to SD for help lmfao
boreal falcon
SD and especially in future Firefly are clear winners and Nvidia depending on what they do with theirs
boreal falcon
If they implement the planned features the right way definitelly
Some are showcased a bit, others yet to see
bold cave
also u gotta understand, big companies care way too much to stay on the safe side
if they realize ppl are gonna start deepfaking when they implement dreambooth, they may not ever implement it at all
boreal falcon
I almost forgot that SD is available as plugin for PS too
bold cave
I almost forgot that SD is available as plugin for PS too
wut
I gotta try that
although I dont see why u cant just use them separately
boreal falcon
also u gotta understand, big companies care way too much to stay on the safe sid...
Good for me with a cost tho
boreal falcon
But then again for artists not a deal breaker
shut sinew
I think it really depends on what youre planning to do and how deep you want to get into it
boreal falcon
although I dont see why u cant just use them separately <:thinking:1045136500393...
Integration is always nice to have
You inpaint within PS etc
bold cave
I think it really depends on what youre planning to do and how deep you want to ...
the way I see it, the more flexibility, the better
even if its something u dont plan on using
why restrict a model when u can have more
boreal falcon
I think it really depends on what youre planning to do and how deep you want to ...
Severar things are "deciding" which one to use
Several
shut sinew
Yeah but some people just want to put in 3 words, ready, go
bold cave
exactly, thats what MJ is for
but to say its "better" or that it doesnt need dreambooth is an insult to the hardcore hobbyists
boreal falcon
As im skilling up in art fundamentals and editing i dont care about ready to go images and thats the case for many other artists, let alone pros
Plus i dont want to pay anymore for MJ anyway
30-60$ + VAT
Not yet, it does photorealistic stuff but not quite yet there compared to MJ or some SD models
bold cave
I hate to agree with peter but he is right in this particular case
MJ cant do photos that look like photos, they ALL have some weird filters that make it look like a videogame or comic book
or a 3D render at best
finite plover
What are the best sites for using stable diffusion models? I already know prodia , openart , leonardo , randomserd ,pixelai and another one i forgot the name of but most of them have very less models and require ingame currencies... Can someine please give me more sites with some anime models? (Excluding google Collab)
bold cave
Now if MJ had a way to finetune the model... then the results would be absolutely mindblowing
but seriously can we just shut the fuark up about midjourney. This is an SD discord go kiss their boots in their own discord
boreal falcon
They work on a web UI
finite plover
I want websites 😦
boreal falcon
But who knows how long they will need to implement that
They have like 10 people in their team
bold cave
also can we stop talking about firefly too. im sure they have their own discord
finite plover
Sir playground has slim to none models , leonardo requires ingame currencies and dreamstudio also has slim to none models
bold cave
everyone here is just going to argue with u anyways
shut sinew
The problem is the people all have dead eyes, no matter which AI
boreal falcon
I want websites 😦
Ask GPT 
shut sinew
Yeah they all have that empty look with no real emotions
Just cant be compared to real photos atm, but thats just my humble opinion
bold cave
oh thats what u meant
nah ur info is outdated
with proper upscaling u can get some seriously impressive eyes
shut sinew
Im ready to be teached other ways, show me
boreal falcon
@shut sinew i mean AI images pretty much always have the weirdness in them that makes you question their realism
finite plover
@bold cave sir can you give me some sites?
finite plover
Well I'm a droid person 😭😭
bold cave
Im ready to be teached other ways, show me
I keep getting sidetracked, Im about to upscale a dreamboothed girl with the new tool and you'll see how good those peepers look lol
especially since she's chinese, asians have the best looking eyes period
no blue eyes no matter how blue can ever compete with those beautiful slanted eyes
beauty is always a matter of opinion
boreal falcon
Look at that abomination behind
bold cave
bru
I could show these cats my 4k wallpaper and they'd still criticize it
especially after they see the before and after
finite plover
What's the best prompt guide now? The openart one is outdated. Can any sir/ma'am please help me out im a graphic designer and im trying to improve using ai
bold cave
but it has tiddies so I cant post it here lamo although I could censor it
bold cave
Look at that abomination behind <:kek:891355818224664636>
this dude
boreal falcon
Do you jerk on those images?
bold cave
porn is what keeps sd alive my friend
boreal falcon
Lmao
bold cave
why rely on corporate porn when u can make ur own?
finite plover
What's the best prompt guide now? The openart one is outdated. Can any sir/ma'am please help me out im a graphic designer and im trying to improve using ai
shut sinew
Yeah imo it still looks rather fake
bold cave
I wasnt talking about u two this time
shut sinew
And the mobile filter to hide the imperfections doesnt make it better
bold cave
im saying that everytime theres an argument here, no one will ever take ur side even if the proof is right in front of their eyes
I can pull the most realistic google images wallpaper and say I did it with SD and they will still call it fake and plastic
thats why I dont even bother proving ppl wrong here lamo
shut sinew
thats why I dont even bother proving ppl wrong here lamo
There is a difference in telling people they are wrong and proving it
Im fine with learning new stuff
bold cave
aight u wanna see what ive made so far
boreal falcon
What is the stable society role
bold cave
the background isnt finished, but the rest of the picture is bomb. I just need to censor the tiddies and bussy
shut sinew
What is the stable society role
Its from winning one of the currently two weekly contests
boreal falcon
Its from winning one of the currently two weekly contests
Aaah okay. Thx for clarification 🙂
shut sinew
the background isnt finished, but the rest of the picture is bomb. I just need t...
I think ill pass then
bold cave
man we really need a dedicated ai porn server
upscale that joint peter
o wait u posted one that really blew me away
I think I saved it
boreal falcon
man we really need a dedicated ai porn server
Good idea, so all the sick perverts get into one place 
bold cave
this one is my fav from ur gens
yeah if u think those eyes look dead ur hallucinating
the grain on that one makes it look so cool
like a real mugshot
finite plover
What's the best prompt guide now? The openart one is outdated. Can any sir/ma'am please help me out im a graphic designer and im trying to improve using ai
bold cave
u can ask on #📝｜prompting-help
or I can shoot u my go-to prompts
do u put grain in the prompts?
I usually keep noise and noisy in my neggies but might remove them altogether
cuz grain does add a bit of realism
ooh
aight I gotta try dat
what about mugshot
damn thats clever
(county jail)
lololol
kodak
see the thoughts are flowing now
nope. I wasnt the one who said that tho
thats a hot babe
theres something so quaint about these pics
they have that warm homey feel
boreal falcon
bold cave
they also have this blurriness about it tho
niiiice
daaaaaaaamn
I think SD has given me a serious skin detail fetish lmfao
damn I love it
I still havent gotten to the new upscaler
but thats the first one ima try to upscale now
static tusk
you need to upscale to get eye details
Illuminati
only from here, not personally
bold cave
I wonder if we can get the best of both worlds and generate an image with 1.5 and then upscale with generic prompts on 2.1
smoky oak
he insists that 1.5 > MJv5 even though that is factually incorrect
(@boreal falcon just adding you to the combo)
If you can get MJ to replicate the level of realism and detail that @stone cipher and I can get out of SD 1.5 and 2.1, I'd love to see if. But as of right now. MJ is lower res, slower, and lower detail cause you can't continue to refine it after you gen
bold cave
I mentioned it in the thread, MJv5 has more detail but it just cant do pictures that look like real photos because of weird ass filters they add
its like an AI watermark they give to all their pics
smoky oak
Then my point stands, 1.5 and 2.1 are better than MJ for realism
bold cave
ok I cant figure out the damn cnet tile resampler https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main
if anyone can show how u get it to work plz do, im toying around with it but as with most cnet tools its not very intuitive
Ok I think it does it works behind the scenes...? check out before and after. I havent done playing around with it but I think its actually doing what its supposed to
ah shit. the bot tagged it and its not even nsfw
the eyes look dope af tho. let me try a different pose
smoky oak
smoky oak
The actual images are way higher quality, just sending JPG versions for now
Some anime is good, why?
loud shuttle
Iranian people after freedom
bold cave
anyone have any idea what down sampling rate could do?
smoky oak
Here's the actual quality of my images I just sent when they are not down sampled
bold cave
niiiiice
smoky oak
2048x3072
bold cave
bro my 3090 is finally shining. its gonna take 7 mins to upscale with karras, but its looking promising so far
and im only upscaling to 1920
fucking luv that vram g
Okay first attempt - not bad, but I know this tool can do better
still promising for literally a model that runs itself
and obviously no more having to deal with the ugly ass seams from ultimate upscaler
ima play with that down sampling rate to see if it makes a difference
the difference in detail in the eyes, hair and b👀bies is insane tho
smoky oak
RIP that hand tho lol
bold cave
lmao i didnt even notice. us ai artists dont even look at hands anymore
smoky oak
I'm pretty proud of myself. I found a way to force my phones orientation to flip 180 degrees when I'm charging it. That way I can still use it sitting on my chest to do SD gens in bed lol
Now I don't have to put strain on the charging cable or port
bold cave
i hate how it only lets you lock it in portrait
smoky oak
I am using a small addon app that allows for tons of additional locks and stuff
bold cave
i have an ipad but even then I hate typing on that shit
smoky oak
Triggers, routines, black out zones, rotation quick toggles
bold cave
looking at 16 muhfuggin minutes for a 4k rescale on euler a. lets hope this is worth it boys
im already seeing the striations on her boobies popping out on the preview lmfao
smoky oak
looking at 16 muhfuggin minutes for a 4k rescale on euler a. lets hope this is w...
What are you doing that's making it take like 10x longer than it should?
bold cave
its the tool. i have no idea wtf its doing because it does its work behind the scenes
all I see is the preview unlike the individual tiles that uu shows
good thing is that since its only one square at a time it doesnt hog up all my vram. that means I can play overwatch while this shit works in the background :kek:
shut sinew
Never said mine are any better nor its your fault. Yall need to chill
boreal falcon
(<@232896089299091460> just adding you to the combo)
If you can get MJ to repli...
i dont use MJ anymore 😄
since mid december actually lmao
shut sinew
But here you go, thats my favs from the past few days
smoky oak
Good lol
boreal falcon
it serves its purpose, but not what im looking for
smoky oak
Stable Diffusion can really hand MJ it's ass in detail and realism, but I do get that it serves a perpose of being good enough for.those who aren't as techy
boreal falcon
i prefer Firefly at this stage already over Midjourney for several reasons. Or people on this server Stable Diffusion over Midjourney
use what you want and need for your projects 😄
photorealism is not my primary area anyway
not that i dont use it tho lol
bold cave
whoa....
this is insane, but this is using a public unrefined subject. now imagine if we used a finetuned model. Oh wait, I do have one of melora hardin. I guarantee that one is gonna look killer
get ready to get your socks rocked off in 16 mins
her hair looks like she just got out of the shower :kek:
boreal falcon
i dunno, for actual photorealistic stuff i use right now free Stock medium (real photos and/or 3D models or 2D artwork stuff) and plan to subscribe to a Stock site and do the photorealistic stuff myself with 2D and 3D software. But generative AI stuff might still help me out
rancid quiver
made with Perpetual Diffusion!
bold cave
like I said, its a public unrefined model
my finetuned subject is about to come out of the oven and boy she looks tasty 🤤
neon crystal
@stone cipher How would you rate these?
vast wigeon
Where are supposed to generate prompts @neon crystal
Answer me now boy
Pretty please
bold cave
we dont generate prompts in the discord anymore
vast wigeon
😦
neon crystal
More outputs from 2.1 prompt experiments
crystal deltaBOT
bold cave
7 mins left and the preview already looks like a masterpiece
Ima post it on aitrepreneur's thread and hopefully he likes it enough to make a vid on this tool
Im making 2 flavors, one is CNET preferred and the other will be prompt-preferred. The prompt preferred adds more detail but tends to deviate more from the original
the only issue is that this thing takes an eternity. hopefully some wizard can optimize it since even UU is faster than this
neon crystal
Cool thanks for the tip.
bold cave
o shit I know how it can be optimized. I bet it does one square at a time. if we can increase the batch size to 8 tiles at a time it will speed it up like a mf
smoky oak
man, my skills with realism in SD are proving extremely useful for character visual development, and visually representing my friends character
smoky oak
I am sure you could easily train a LoRA on something like this
bold cave
fuuuuuuuuuuuuuck yeeeeeeeees.........sort of????
this tool is complete randomness
sometimes it comes off blurry, sometimes hyperdetailed but oversaturated, sometimes just a completely different image
it's on the same level as UU, but you're not gonna see me praise it as the next big thing until it can prove to be consistent
I suspect its doing the same seams fix trickery that UU uses, just behind the scenes, since some areas appear more detailed than others. its weird
smoky oak
fuuuuuuuuuuuuuck yeeeeeeeees.........sort of????
This looks like a water color messy smear of vaseline-
bold cave
lolololol
Ive gotten some crisp results, but they deviate so much from the original she looks like elon musk
Im trying all different variables and all give wildly different results. for example txt2img vs img2img are waaaaaay different
if the gens didnt take an eternity I'd be able to more accurately test all methods
smoky oak
don't you have a 3090?
bold cave
the tool is highly unoptimized
smoky oak
sounds like a great time :/
bold cave
I might start digging into its source development and aid in improving it
I know its doing 1 square at a time which is stupid
smoky oak
Thats how ultimate upscale does it
bold cave
Ik, even it is pretty slow, but this one is way worse
smoky oak
getting to 4k with UU is like 2 minutes on my 3060ti
bold cave
oh my word
now the details and crispness are there but it feels like something is off when you look at it expanded. she gives me wicked witch of the west vibes lmao
I just got an idea. I can double up the cnet with both the tile and canny at the same time. Im hoping it works
smoky oak
oh god, now she looks like she's taxidermy
bold cave
lmaooo
For a concept as simple as UU with controlnet compatibility, they sure overcomplicate it
blazing crow
Maybe look up some Lora for vectorizing pictures?
smoky oak
For a concept as simple as UU with controlnet compatibility, they sure overcompl...
yeah, and it looks way worse than just normal UU, so I do not see the point ATM
bold cave
no
I mean ur right
Some of these controls make no sense
However, no one has capitalized on my idea of having a clip interrogator for each square
If someone knows where to find the source code or any hint at development of controlnet models, I'd gladly take a look now that I have time since I'll be unemployed for a few weeks
wispy ether
wispy ether
If someone knows where to find the source code or any hint at development of con...
Did you almost kill someone
valid isle
can anyone advice me,, if this is stable diffusion model or disco diffusion?
smoky oak
can anyone advice me,, if this is stable diffusion model or disco diffusion?
could easily be either
valid isle
could easily be either
😲 a little too low on knowledge.. forgive me to asking this but how do one even train a model to do that!??
smoky oak
I am not sure how you do, but I know its been done several time
wispy nest
#Write news 4 may, 2023 for crypto currency
bold cave
can anyone advice me,, if this is stable diffusion model or disco diffusion?
that looks like midjourney
I'll be damned if it isnt mj
valid isle
I'll be damned if it isnt mj
pretty sure that isnt midjourney.. that looks like somebody put way too much focussed work on it for it to be chaotic but stable
bold cave
I doubt it's inpainted otherwise the inpaint locations dont make much sense, but who knows
it just exudes midjourney to me
hardy comet
guys i have no idea, how do i run the webui-user.bat? I get an error when i run any of these
bold cave
what error
ur probably better of asking #🤝｜tech-support , home of the ancient wizard CS1o
hardy comet
ok ty
bold cave
hooooly shit. Ok I feel like a total moron because apparently I was doing this all wrong 
apparently ur supposed to use that model WITHOUT a preprocessor image loaded. I guess this is what I get for skimming over the instructions (as if not loading an image made any sense in the first place)
difference is night and day. jesus what a fox
so apparently u just enable it w/o an image, and use it in conjuction with SD upscaler, not on txt2img. And it no longer takes half an hour lamo
the fact that I got something like that on the first try is insane
apparently if ur not a moron and you read the instructions right this really is a gamechanger
for reference, this is the original
and as the guy from the original thread explains, how the model manages to not fuck up the image despite keeping the original prompt is a mystery/black magic. 🤷
boreal falcon
do you use this only for...such stuff? O_O