#🏞|general-with-images
1 messages · Page 58 of 1
what does this do exactly?
ill have to check what i normally run it on
doggo
ive tried 80 decent ones and another with 10 but really high quality, the 10 came out better
how
hes lying
reload it
i did
reload ui
Floof
and now we wait
if that traffic light didn't have like 5 too many positions i'd buy that as a real image
this
tbh i think mine is mostly better
i like the photorealism a bit more for animals due to the fur texture
this is isometric dreams i ran the img through btw
very nice

the image i upscaled of your kitty
"but i thought it was just a video game"
- Every child from Ender's Game
when i send that image can you upscale that one? I wanna see the differences between the final results
i invited you to the Discord server a while back Snoop, did you not get that invite?
oh?
i upscale everything. try and stop me
why did you upscale the real cat 💀
because i want him higher fidelity 🐈
reality isn't good enough for ol' shrekkie boy anymore
if i could wear Google Glasses with CTU upscaling everything i would do that
what's interesting is that's just CTU with the prompt "best quality" and a series of good negatives
i don't bother passing the original prompt in
you can change the prompt to be like "8bit luxury" and it'll make it look like an old DOS video game character
I don't really see many words/prompts that give consistent effects when it comes to quality. Well, depending on the model. But it seems most newer models have already done a lot on that front as standard :P
i'm using dreamlike-photoreal and Realistic Vision 2 and both experience major benefits when CTU'ing their output but if your output is already very detailed and higher res than 1024 on one of its smaller dimensions, chances are high that CTU will nuke details from it instead of fixing it
it changed his face but didn't make him 8bit 😦 i think HARDblend doesn't know 8bit
bro stole my prompt
im the worst
literally just complained about it, it's HARDblend 😄
sounded like a vae
ah i am stuck with VAEs that work with diffusers so i stick to the vanilla MSE or EMAs and mostly MSE now.
intuitiveye/HARDblend on huggingface hub
which one
i can't grab models from CivitAI because they don't really have the diffusers style supported yet
i have no idea
I can't say anything other than from my own experience. Things change too much and as I haven't really made any images for months, I might be incorrect, but as always, at the end of the day. The answer will always be: it depends :P
I've had great images from 2 word prompts on bad, and good models, as well as bad and good results from throwing both the thesaurus and dictionary on a bad/good/old/new model :P
I'm just happy that people are having fun creating awesome art :D
probably the HARDblend.safetensors but i do whatever hf wants me to
real
whispers wake up
wakes up and breaks your nose by accident
that's what really should have happened when Zoolander hit his face on the bunk bed
i know this cuz i took an MRI once and fell asleep and woke up all of a sudden and RIP nose
alien by old master
skyviding cosmonaut?
i think CTU is amazing because it's the first time in my life i've preferred an upscaler. though my example i just shared is one where CTU downgrades the resolution without using Ultimate Upscaler technique
it added a pixellated thingy
the model keeps saving at 5gb 
I think this model just might be undreamboothable o:
or maybe it makes no difference?
nice detail
Hey, guys. What would you say this style is? It's a mix of disney style and anime, but I can't reproduce this style.
Any Ideas?
the consistency is really impressive, lets see some full body poses
other than "long necks" still being a thing, I believe there's been at least some upgrades over the last 4 months ;P
This was created with Deliberate 1.1 model, but some Prompt Magic was applied and I can't see what's behind.
half body doesnt count
still good, you try controlnet?
does it simply not produce full body results
one of my models is like that
controlnet should produce fullbody
it does but ruins quality and distorts face, could be my model tho ill test it on a base model later
you tryn make fake news articles?
< pretty noob, will check that out
it can with some luck
sure
used mostly close ups for training im assuming
when it does give full body is it high quality face?
ket
nope, face tends to get worse

yea ive found that as well, which is why im trying to see what people are doing to get around that
i tried generating low res full body images with another model and upscaling them with deliberate and the face prompt, works sometimes
i’ve thought about upscaling then inpainting the face, have you tried anything like that before?
Not yet
i have watermark in negative prompt tho
why does this look so compressed
Now this is a masterpiece. 🥲
yaaaaaaaaaaaaaaaaaaaaaas
hello sen
what are your thoughts on cats? 
I have one
too big to be cats
so the maine coon isnt a cat then?
maybe its just me. they look more like bobcats
try making a siamese. thats the one I have
the fur details are pretty dope
alrighty then hold on
omg... nvm either this model is just not dreamboothable or dreambooth is yet again broken in a1111
what model?
zovya
I never liked the DB in a1111 the extension
yeah well thats because its kinda got like 11gb of junk files
ik. I just hate having to go for kohya
or its likely because of that
but my model is 5gb
oh, screw Kohya
I think its because the creator of this model did some black magic and turned a 1024 model into 512
so I think hopefully if I train this as a non-512 model this will solve the issue...?
if not then I give up and I'm sticking to SB1.6.3
this is what i got for siamese cat 💀
looks plastic
:O
dr. who asked
doctor i asked >:(
Emojis
yeah they're freaky to run into in person but they're deffo cats, personality-wise haha
i could imagine a poor maine coon being very scared by a human who thinks he's a meanie
🥺
@hearty karma try "wood grain" prompt in CTU 😄
strength 1.0. better hope you have a strong stomach
how do i prompt it in ctu?
@hearty karma idk if auto lets you prompt CTU, from what i gather it forces you to reuse the original prompt
i really quite like this outcome though... wow 😄
DAMN! This is incredible. Thanks for sharing it. Are you the creator of this model?
well ctu is only on the img2img i think
yeah but img2img allows inputting a prompt
in the ControlNet example code on GitHub page, it uses best quality as the prompt
which honestly is a pretty sensible default. it makes great work. but when you want something very specific like wood grain, i think it's better to be able to prompt it
https://tripleback.net/public/discord/1683656924.3099692708db69eea5644c78c57e293d23df8e.png
LMAO it has no idea what an emojey is
doctor who asked
beat me to it
so if you do 128x128 it really poops the bed but you can upscale that garbage to 512x512 and then back down to 64x64 and wouldn't ya know it. it still looks like garbage
fuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuck
no matter what it keeps saving a 5gb model
if anyone has any idea how to keep the entire model intact with dreambooth then I'm offering a cash bounty
thts how I made my new icon folders lolol
I made this one to replace the missing NordVPN logo
wow, good job. my model i'm using has no concept of logos/icons somehow so it's probably all the finetuning it's gone through ahahaha
gen them at 512
also keep the prompts simple as well as the neggies
I got so much vram dripping out my buttcheeks I can just set my training batch size to 4 and have enough GPU leftover for overwatch 😏
liquid VRAM chunks
RIP
the fuck
ass juice ram is super slick
🐏
bro that rig would be so badass in real life. can you imagine a quantum anomaly powered CPU?
damn auto spell checkers, lol
raytracing on a quantum pc
BRRRRRRRRRRRR
damn
Quantum singularity powered by Intel.
You know if you have your computer in deep space you can overclock it like a boss, right?
currently the Heron system from IBM is the most powerful, with something like 1133 base Qubits and 4 modular subsystems with 133 Qubits in each module, quite an astonishing number. it is also the first publicly accessible quantum computer. you can rent time on it to do the degaussian boson sampling at a rate of about .5 seconds per task, where a normal supercomputer would finish the same task in more than 9000 years
but boson sampling is currently the only task a quantum computer is good at
I would love to see a pc turned on in deep space with no fans.
it would be undercooled
anyways
yep
LOL
🤔
A6000 = 300w, A100 = 500W, 4090 = 450W.
the A100 uses 80 watts when generating images, the A6000 uses 180-250w depending on the task, and the 4090 loves to kick it up to 450w to beat the pants off the image gen speed of both of those GPUs.
bro is rich
you can limit the power of the 4090 to 150w at the lowest, at which point it's about 75% of the speed of an A6000.
i wanna upgrade to 4070
i want to encourage that, but only if it has 16G VRAM
I need 24gb of vram
why?
i forgot about those things
Can you run 2 gpus like the A4000 without display and one as display?
???????
Does the A4000 even have display 🤔
@wispy nest you can on Linux at least but it'll use some of the VRAM on both (about 256M) for buffer space for the desktop
@wispy nest it's called a graphics card, i'd assume it does 🙂
it has 4x DP 1.4 ports
Thought some server cards were just accelerators without display
which site is this
well
same
but i upgraded from 1080p to uw 1440p
I call BS to that site as I tried it before
4k gaming or nothing bruh
I live life dangerously
Im still saving for a good TV so I can game 4k at 120hz
If you see botthneck just hit it with hammer
Moore's Law has entered the chat
looool
and high latency
tv isnt made for gaming
G-sync
doesnt matter
i have a 4090 but i don't play games on it lmao i just use a Steam Donk
thts one thing I mentioned yesterday
Steam Deck*
im talking about the pixel ms
nvidia has all the good drivers
ghosting
AMD sucks dick at compatibility
real
except for all the work they did in pytorch 2.0 to make it use the same interfaces as CUDA? lmao
oof
i was wondering for years "why haven't they done this yet?"
Wouldn't expect anything different for someone who loves Shrek
well they caught up to my amazing brain
@wispy nest i'm glad my reputation preceeds me
🗿
I thought it was an easter egg statue of shrek
Who is beside he jar, baby Shrek?
easter island* lol
What
Is that umm cotton
What
pc on fire
but can it play Crysis?
wood grain vs cyberpunk wood grain versions 😄
Looks fake
didn't know it's supposed to be an X-Box, i've never seen one before
On that note, we should mention that this review is focused exclusively on gaming performance. We don't care about the benefits the A4000 might bring for video editing, 3D rendering, or CAD work. This might be something we explore in a separate piece, but for this one we have a heap of gaming data to go over, and of course, the entire point of buying the A4000 was to answer the question: is it worth purchasing over the RTX 3070 for gaming?
lol TechSpot
Im ready to blame the new version of torch for the shitty dreambooth results
Ima have to do this all over again with my previous backed up version of A1111
Hi, so i ask for a bunny or rabbit and it gives me an anime girl, how do i fix this? i want a normal rabbit
jeez AMD is better than NVIDIA for gaming now, when you limit the comparison to the 3xxx series? what world is this i live in now
use a different model @molten pecan
i never ever would have thought a Radeon card would beat a GeForce
what do u mean by model?
what
the
actul
hell
of
earth
almost done
idk what to tell you, i'm not a console gamer and no one i know ever has the money to throw around on one 😛
my friend had the very first Xbox, i think that was the year 2002
i wish i weren't that accurate
my gpu fans be running at 3000rpm
i hated that xbox, everything about it felt weird. the controller, the interface
this one looks fake too but at least there's the button now
HOW DID YOU DO THAT
what model is that
controlnet tile upscaler with HARDblend
btw mine is done
what
base model = theintuitiveye/HARDblend
some new shit that got updated or released like 4-5 days ago or a week ago now
what is prompt
i think hires fix helps if you go beyond certain sizes but i do not yet use that part
is hardblend 1.5?
i'm not sure which version, it's whatever huggingface_hub uses by default
yeah if i saw this on marketing materials i'd roll my eyes and laugh at how much fun the photoshop guy had
LOL THAT CONTROLLER
Microsoft: "We made the worst controller ever!"
AI: "Hold my beer."
now they're all circles
me protecting my son when people get too close in public
go ahead, punk. push my buttons 
Amazon people: Write that down!!! WRITE THAT DOWN!!!
but i made you a more chaotic build
looks like the desk it sits on top of is inside its case
this is what happens when i give a better description of the img to CTU
omg i love how it defaults to the Samsung GSX display like you used to find at GameStop / Sam Goody stores
i forgot to change the prompt
it made it an aquarium 😭
whats ctu
man, prompting CTU can go horribly wrong
controlnet tile upscaler
send link
link? 
chainner
links
it's called Tiles
😐
https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main
the *_tile.pth files in here
which one
control_v11f1e_sd15_tile.pth
there's literally just one tile.pth
how tf did they do all this
NeuraLink
- cloning tech
force-multiply? pfft directly clone employees
oh now that that lady lost the court case, i can do:
eleanor from gone in sixty seconds 1967 shelby GT cobra highly detailed gorgeous reflective metallic pearlescent paint job
i thought i'd be able to help, but uh, even though i've been dicking around in the diffusers source code, nope, sorry lmao
ahh what is the prompt?
that looks like fondant now
lmfao cyberpunk cake
what does tile upscale do
i used it on his cake
so basically img2img
yeah with emphasis on de-noising and adding details upon upscale
when i remove the keyword carpet from my prompt and put fluffy chocolate cake instead, much better
man that doesn't look bad. i'd eat both
it looks like a cake i'd make and people would be like "what the fuck" but i don't care
tbf if i were judging by whether it's AI or not i'd wonder why they aren't squished or cooked looking. like someone placed those in there after the cake was made.
well that didn't work out
oo
minion fashion show, has ai peaked 🤔
ew
no?
what
just set it higher
😎

no
yes

mh
hmmmmmmmmmmmmmmm
i create fake news
fake news
i have no idea what that is
yea
idk
@hasty nova any chance you could upload digitaldiffusion-2-1 to HuggingFace for use in diffusers? ❤️ ❤️ ❤️
@stone cipher fwiw, the CEO of StabilityAI said they funded the development of MidJourney
sure
by giving them compute credits
nfts
and yes, i am a developer 🙂
they haven't come out and stated anything too definitively like that, just that they "co-developed" these tools and that's how Stable Diffusion's model became to exist.
if anything Stable Diffusion was developed out of early MidJourney tech
and i see it the way you do, that MJ advances and SD seems to be behind them. probably because they have all the people who knew the fundamentals that led to the creation of these tools to begin with.
i have no frame of reference for that, i don't really look at MJ images at all unless a coworker sends me one
sometimes i see them being used in news articles about Elon Musk. they use MJ images of him for some reason
i guess in that sense i would say MJ out of the box gives better photorealism but SD can be way better than MJ due to its open nature
MidJourney is hella tweaked to prefer a single style of output, like a fine tuning gone wrong
@stone cipher i don't know that i specifically saw anything from MJ, do not recall that
oh, yeah. those are all very photorealistic. you would have to be nit-picking to find issues with most of them.
the one who knocks, of course
why they drinking blood?
the twist we never expected
CTU made her elderly
yes, controlnet tile upscaler
account?
you're using a service to make those?
i fed it back in two more times to make her more elderly and im scared
A few more times and a skeleton astronaught
3090, xformers, efficient memory attention
Attempted a Venom Gif with ControlNet OpenPose + Fine Tuned Model
Any ideas on how to make this better?
more movings
I have no idea what that is but I heard swarms are bad
Swarms should just be the same idea as botnets
Ppl tap into ur card/computer and you can ideally do the same
zombie motorcycle prompt
Jesus, you have no idea how to use that controlnet tile upscaler lmao
Only iffy part is whether or not you relinquish all control when you think you do
me?
yes, its so CRUNCHY
see that Sytan?
more keyframes
see what? 👁️
I pinged you
4060ti 8gb and in July 16gb all the same specs and 128 bit bus
LOL, yes
I got in trouble earlier on YT as a violation for harassment because I said I wanted to kick Jensen in his robotic nut sack. I never knew robots could be harassed.
man, try telling zuckerberg he doesn't know how to drink water. triggered
LOL
The community deemed. Yeah, I know that community it is 1-3 people in a room.
Anyhoo, that 4060ti 16gb is even more of a slap in the face.

What gamer is gonna love that 128bit bus?
this is more my style. Rick Moranis as Harry Potter. added 5-o-clock shadow with CTU
its not actually that big of a deal for games, as games are usually a few bigger calculations, while workstation work is usually an insane amount of little calculations
CTU requires less memory than a normal pipeline, for the same resolution, in my experience. A1111 might not do that, if it doesn't nuke the pipe between stages
We are talking about Auto1111 here.
that forces the image through an img2img pipeline and i think it will keep the original pipeline awake while it does the CTU thing.
where is it at?
it's an extension
I have CN
oh
i can't help anymore with the specifics since i don't use Auto1111
Sytan can tho
You know if all extension were removed from Automatic1111 it wouldn't be worth a damn. Seriously.
General doesn't have enough VRAM to use the CN upscaler

Told ya
At least thats what I remember him saying
are you on 4gb?
6
6
6
lol
sorry, had to lol
well, there are somethings I really want to run in my workflow but 40mb short
40mb for the last step? i thought that was an accurate figure til i upgraded GPUs and it was like "cool now we're 76gb short"
I definitely know how that feels. I am less tahn 50MB short of being able to run vicuna 13b on my GPU
naw, it goes to generate and 0% and BAM, OOM
40mb is just the most recent allocation failure, it could be the end of a layer it's inspecting, and then it needs to load a whole 'nother layer after. that's when the major failure occurs.
oh
you need to enable sequential offload
I don't think that is in a1111
See!
I could have SWORN I remember seeing a setting for offloading
but I cannot find it anymore
I have 48gb of ddr ram but meh
EW WHAT THE HELL
Yes, I think they ditched it
sequential offload became unreliable to set across a number of popular models, after pytorch 2.x
of course, cause pytoch is a pain in the ass
eff PT2
I still use 1.13 lol
I am not even on PT2
when i upgraded my bot, i had to wrap the sequential offload call in a try/except so i can absorb the error and ignore it LMFAO it still works. what a garbage codepath
throwing errors for no damn reason
i can see A1111 not realising it'll still work, and nuking it.
welcome to Stable diffusion in general
yup
whut? on your card?
damn, my 4090 is like a lucky duck then, i thought i had a mistake in the code it finished renders instantly
the upload takes longer than anything else
yes, as the PT devs told me on the ticket it is really made for 4090 and 3090 in that order
A100 etc...
I am excited to see the new 7950xtx
Same
cause they dropped 5.6 of whatever that compatibility layer is
and mmm, that looks good
why? that's 24gb memory
Did you know if I buy an Intel for its AV1 and get a 7950XTX it is still less expensive than a 4090? Big deal I lose DLSS
cause theoretically based off leaks, it could beat a 4090 in a smaller form factor, with less power, and considerably less money
gotta love how the 7950xtx is 7% faster than the 4090
yeah but the W7900 is the same architecture, with 64gb vram. but obviously more money, but not as much as you'd spend to get 64gb from a series of 4090s
hi i am new, how can i start creating some images?
rocm did that PR and all the cards and they showed all the cards about to be released
self-leaking ftw 
yep
great, the FAQ bot is down again
that's done intentionally, fwiw
I know
they throw in red herrings there too
when the Ryzen stuff hit the Linux mainline, they had 4 extra models that never came to exist
😂
might exist internally, but definitely never saw the light of day
If AMD can bring the heat with SD support, I will be willing to get an AMD GPU powered laptop
i think they had 9 extra models in the Threadripper patches.
oh, i... wouldn't... i have an AMD laptop, it is officially explicitly not supported by AMD for ROCm
unfortunately threadripper is dead - rip - er
that said, an AMD iGPU > Intel iGPU
by a landslide
How far back do I need to go to get that streaming CN setting back?
the new AMD APU's that are being leaked have a 1660ti level iGPU IIRC
I am sad that this is the last gen of budget cards as they want buget people on apus. Screw that
Yes
APU's are awesome for some situations, but the lack of affordable GPU's also sucks
yeah i just hope that they're doing it smartly to improve their product lineup as a whole. they're not Intel, chances are AMD will fill the gap like they did with Ryzen when neutering Threadripper
Wait what?
1660/1650 has to use --no-half even with xformers or black boxes I don't even have to do that
the 1660 is faster
@dense tapir i have a 5800H laptop and i don't even bother using the nvidia gpu for a game unless i have to
ahh, I see
Jensen butchered, I mean really butchered, the 1650/1660
The 1660 is considerably faster than the 1060, and the 1660ti even more so
in gaming at least
just like my card it only has emulated fp16
remember they released the A4000 with GDDR6 instead of GDDR6X? 
so workstation, much woaw
Yup, I was still in rendering when that released lol
the before-fore times
yeah truly only useful if you need the high precision shit for CAD
or idk, 16gb nvidia cuda
may god have mercy...
or you are running a special software that leveraged quadro drivers
that's what i referred to 😛
oh, CAD doesn't use quadro drivers
I am talking like medical sims and stuff, where non quadro cards are basically useless
Solidworks did
Also, did you guys see NVIDIA's new DPU's?
unless you had a quadro you were locked out of h/w rendering and shit
Looks like we now have CPU/GPU/DPU/NPU
Yes, getting nuts
but man those DPU's are absolutely awesome lmao
mwh, just more expense and more shit to break
so when i ran a MSP and serviced offices, the CAD dudes would notice if we installed the non-Quadro drivers because they were unstable as hell in comparison, and updated more often
they are necessary for servers to function now, and actually end up replacing a massive amount of hardware with just one card that is smaller, faster, and cheaper
seems like nvidia makes riskier moves to optimize game performance on those
The openAI discord flat out banned for having an NSFW model link... in my fucking profile
this is the level of pussiness you will find only on discord ladies and germs
Oh, tell me about it
I know that I just recently discovered gaming drivers and studio ready drivers are WAY different
i let them know i'd come across people trading stolen API keys and they banned me 
Noticed that after my latest studio driver's took away support for refresh sync and support to encode HEVC lol
If zoomer parents are this fragile
I cannot fathom what the next generation is going to be like
@bold caveI immediately left that discord.
OpenAI can't allow anyone to tarnish their perfect image. because the world loves them so, so, very much. no one wants to sue OpenAI. no one wants to hurt OpenAI. we must keep it that way
Ive actually gone record time here without getting banned
yet over there they will ban you if post cleavage
was it cleavage from a male centaur? they specifically hate that
unstable diffusion is a scam anyways
they banned me for saying that too
centaur moobies
I have that discord still for over half a year but I never bother to go into it
so uhh apparently zovya ultra is filled witth a lot of junk files
Sen, they're living rent free
they and their fragile mods can keep their room
🧌
yeah, zovya needs to be cleaned desperately lmao
If all I have to do is upload the model and config then sure, but I don't really know what diffusers are so if there are extra steps it might take more time.
it thought this was a swan
ah yes, flaminwan
@hasty nova that's an amazing response and you can take your time, no one is in a hurry but i think you're right that it needs a model_config.json and then the safetensors file
If anyone knows the last version to have that streaming setting in it let me know please when you find it.
also, I made my own noise offset mix for Zovya, and its now 19GB lmfaooo
this may actually be true
and apparently the clip stuff for zovya were screwed as well
I thought the config was a .yaml?
given that the 5gb model that comes out of dreambooth doesnt appear any worse
not may actually be true it is actually true
it's json for the hub's diffusers models that get pulled in via transformers.AutoModel()
so then what is all the filler junk data?
who knows..
I am actually gonna do a direct test between zovya v1 and v1 ultra right now
@bold cave EMA model vs checkpoint
YES
maybe
post results
will do
Im also suspecting torch 2.0 is not working with dreambooth
thank god I have an older version backed up ready to go
EMA models are like a cliff notes version of the model aiui
will take a moment cause ultra takes a whole GPU refresh cycle to load 
using an EMA model can improve default output for the same prompt by reducing the surface area of the model to traverse, or some shit
i get so confused by all the acronyms
18GB? you're approaching the size of T5, baby
so adding more is just marring the model
at 18GB you're 1/3 of the way to DeepFloyd. KEEP GOING
interesting, it has shown literally 0 signs of noise offset in my testings
lmao
Ik someone here said it, and my training confirmed adding more made it look worse
I mean it does look darker, but not in a good way
Deep floyd should be about 20GB, based off what the devs said
told you
well i'm an idiot so i managed to use 46gb just for that one model and all three of its stages

T5 alone is 18GB
i think you're talking specifically about the DF stage models 1 and 2, because 3 isn't out yet. it uses the SD x4 upscaler by default in their example workflow.
So I could just change the file into a json?
@hasty nova i can ask on the HF discord for you
cool
dont delete
back up into an old hdd
u never know if u may need them in the future and civitai is volatile
civitai
huggingface hub deletes my data too
they are models I have deemed inferrior
they restored it from backup but yaknow
so like the old realistic vision models, protogen, analog madness, all of that
and some failed merges
alright about to mass test between edge of realism, cyber realistic, photo real v1 and photoreal ultra
RV 2.0 is hella overfitted for Hogwarts
i use both for what they're good at
tbh i think 1.5 is prob going to be the standard sd model people use for a long time
really? i'm leaning toward DeepFloyd. just to piss people off and be weird about it
barely anyone is going to be able to run deepfloyd
it runs in just 14GB VRAM bruh that's like most laptops coming out in the next 20 years
i'm playing 5D chess here
i managed 10GB VRAM for DeepFloyd, and you can do it in 4GB but it's so slow because you have to delete each stage of the pipeline and create them as you go, saving only the PT data in between
I actually wanted a star's face a couple of months ago and went to 1.4 to find them. 1.5 barely 2.0/2.1 gone
ok but that doesnt mean everyone will have more than 14gb vram in a couple years
1.5 is too supported, too documented, too well known, and too diverse to just disappear. It will probably always have a cult following
yeah the Windows 7 of the SD world
I will ditch it in a heart beat when I can, and when there is a good alternative
Yep
exactly
its a whole lot of right, with very little wrong
meanwhile there's fucking DALL-E, the Windows XP of image gen world.
so whats the windows 11 of the sd world?
I still use the windows 7 desktop in 10 and will in 11 too
2.1 IMO, should be better, but is just full of issues lol
@hearty karma i don't know that metaphor, i've never used windows 11 but DeepFloyd is likely Vista or 8.0
a few people REALLY REALLY LIKE IT
hmm thats actually very fitting
no one understands them
With the issues being reported with DF it is more akin to Vista
I wonder if thats why your GPU struggles with SD, cause windows 7 has been unsupported for multiple years now
oh right, like Vista you had to have a shitton of hardware to make it work well
It's a theme
ahhh, nevermind then
yo can you tell me where to get that theme?
did you guys ever mess with Windows Longhorn? that was fun
i miss having my windows 7 look
back in the day when Windows betas were closed and you were super cool if you ran it as a daily driver
alright, beginning comparisons
Sytan, i tried to make a video game controller made from a cinnamon roll earlier and couldn't
Did anybody try Photopea + A1111 plugin?








