man i just don't want these guys to go away
#💬｜general-chat
karmic cedar
I think we are now in an age of direct confrontation between the core ideals of Humanism and the ideals of economic logic.
Which take…a certain teleological form.
A very predictable, mathematical one.
So, it’s the rounded versus the pointed.
shell tendon
man i just don't want these guys to go away
yeaeh i think we are all rooting hard for them
i'm so glad they exist and for what they've given us
sure as hell hope the party can keep going
karmic cedar
Save your work ☺️
fervent thunder
karmic cedar
Has anyone tried to do any 80s britpop on Suno?
“Never gonna release SD3”
j/k
“Never gonna give you Half Life 3”
…not joking there
fervent thunder
With the right promt you can make anything it's how and where you place words in a promt
karmic cedar
It’s true.
shell tendon
yeah i think ordering is a lot more important than parentnheses
the latter is a bandaid
fervent thunder
Vintage stuff is tricky
shell tendon
Vintage stuff is tricky
got an example image?
karmic cedar
Vintage stuff is fun tho
fervent thunder
got an example image?
Yes
shell tendon
post it in #🏞｜general-with-images i'lll see if i can copy the style with the new ipadapter and some controlnet shit i was messing with
strange egret
Hello everyone, I need to do a small task of merging two images. I have to merge the image of a baby with an image of their grandparents. Is this possible? How do I do it? Thank you in advance, best regards.
fervent thunder
:One sets the standard, the weak randomly follow
Space ain’t shit, some kid’ll take your place tomorrow
The cycle never stops cause you can’t recycle styrofoam"
peak orchid
SD3 when?
when they iron out the kinks and determine if their triple encoder experiment works
spiral crest
I really hate "X isn't expensive" meme
it gets mainly propagated by
- people who are rich and thus uncreative (look at my computer I bought for 60000 USD)
- crazy scifi authors like Gwern who smoke to much weed
- people who don't have any [novel] ideas (basically OpenAI)
in reality most energy is wasted for nothing (oh look we have a new model which doesn't reuse the work done for the old one, it's so great that we produced X tons of CO2 for nothing)
and now TileAI wants to tile the planet into fusion reactors and datacenters
spiral crest
man i just don't want these guys to go away
they will go away and it won't matter
cedar salmon
waste, progress, same thing right
karmic cedar
and now TileAI wants to tile the planet into fusion reactors and datacenters
It’s like coal mining never educated anyone
Perhaps it was the fumes.
low moon
Unlike Valve Stablity can count to 3.
broken cave
I really hate "X isn't expensive" meme
it gets mainly propagated by
* people w...
i don't know, name 1 ultra low budget movie you've watched
i can think of a handful that i've watched and liked
low moon
Star Wars was essentially an expensive B movie.
Everyobne laughed at that shit.
Hard.
a talking dog flying a ship.
wtf man
broken cave
star wars was not an ultra low budget movie. it had an $11m budget
low moon
and who is this vader with a scottish accent
wtf man
what is this
a parody?
a kids movie?
spiral crest
i don't know, name 1 ultra low budget movie you've watched
I meant in AI / ML . No need for useless analogies to film production budgets.
low moon
but forget all that my friends
broken cave
I meant in AI / ML . No need for useless analogies to film production budgets.
i don't think there is very much useful stuff that has been made by focusing on low budgets, unfortunately
broken cave
it's a huge distraction
low moon
from our messy bedrooms
broken cave
it matters if you are an open source contributor making a porn generator of some fame, that porn generator end users can use your tools without spending any extra money, only the money they spent on their other addiction, video gaming
it doesn't matter in any other sense
spiral crest
what a narrow world view
broken cave
i think it is a really bad time to be trying to go from $6,000 to $0. It's a great time to be going from $250k to $6,000, aka, to eliminate the need for DGX machines
karmic cedar
Hollywood is fucked either way—because even if they choose not to produce with AI, others who have the ability to enhance / stabilize their creative expression using AI will up the playing field, lower the prestige, and once again remind the world in a big way that artistry is subject to context.
spiral crest
let me guess, your age is under 22 😄
spiral crest
i think it is a really bad time to be trying to go from $6,000 to $0. It's a gre...
that wasnt my point
low moon
I'll make Dune 3 and use a color palette of more thna 2 colors.
broken cave
that wasnt my point
i am trying to read "people who are rich and thus uncreative" very charitably, i think i know what you meant by that. it would definitely help me, the most, to have the best models, and the $20/mo ideogram or midjourney costs or whatever, it's fine.
it's already basically $0
low moon
The amount of media that will flood the inerwebzones in a few years will be so huge no one will be able to surface
astral goblet
within 5 years any of us can outdo Hollywood.
depends really. i dont think ai will ever be a visual director or art director. in 5 years you'll have people cranking out very low effort 2 hour feature length products with absolutely no regard for direction.
broken cave
like i would pay $1,000/mo for chatgpt. i don't know if they know that
low moon
tits will be so huge they will eclipse the finest art
broken cave
a lot of people would
low moon
it will be an ocean
a tsunemi of media
i dont think anyone knows whats about to happen
astral goblet
there'll be a lot of VERY ai looking prducts and it's hard to say anyone is making those. they'll just be a derivative of the model with nothing very transformative at all. legally this may be up for question, but in my court of opinion it'll be an open and shut case. i'm already sure there will be a deluge of veyr low effort ai content
broken cave
stability post emad, their thought is going to be to charge enterprise pricing for SD3 full weights, which is too greedy. whereas the most midjourney can cost you right now is $1,000 or whatever a year
astral goblet
i'll be hesitant to call most of that human created.
low moon
like you cna literarily make a masterpiece and no one will find it
charred mesa
I hope AI will help vfx mostly
low moon
flaminog giraffes
astral goblet
like you cna literarily make a masterpiece and no one will find it
wait till you learn about van gogh
that guy died completely unrecognized
charred mesa
some people want AIs that can make full anime fight scenes
broken cave
I hope AI will help vfx mostly
i am pretty sure that's aspirational / fantastical thinking on the part of guilded creatives. all the resources are being poured into replacing actors and writers right now
charred mesa
all the resources are being poured into replacing actors and writers right now
oh yeah right 💀
low moon
its already been hard to get noticed
charred mesa
ai generated text -> next movie!!! "stupid writers, why would we need them!"
hollywood is fun
low moon
even if you are very honest and good artist tryign to put out good original meaningful work
its been impossible to find an audience already
now the chances will be 0
broken cave
I hope AI will help vfx mostly
VFX is that thing that, from the POV of a guilded creative in LA, it's already being done by north korean slaves or whatever
astral goblet
i am pretty sure that's aspirational / fantastical thinking on the part of guild...
actors won't be replaced. they'll just be moved into different media. if you have some business graduate directing ai to act, it'ls going to be a shitty product. get a trained thespian to use an ai model to direct the acting and you'll have somehting much higher calibre
broken cave
so it makes sense to them, in their worldview, that this technology demoing photoreal, natural moving full motion people 7/8 times, that the 1/8 time there's a VFX shot, "oh, this is going to replace the Other"
astral goblet
you'll still need the skills to pay the bills
low moon
coz any old joe can type in titblast 500 blocuster super movie watching now " and get a decent movie
even with typoes
the muels can get whateve rthey want
the hoardes
astral goblet
low effort content will stay low effort. that'll never change
low moon
already i find AI generated female forms more refined and appealign than real ones
broken cave
actors won't be replaced. they'll just be moved into different media. if you h...
well actors are already being replaced in industrial and advertising projects, with limited technology, and that's already like, the working actor's bread and butter
broken cave
i don't know. they're not wasting time on VFX avatars
it's very very aspirational thinking
astral goblet
thats just standard beauty delusion. same way porn marketing works. porn addicts actually think those outrageous implants are attractive
low moon
you need less and less skills to get good results
heck you suck at inglesh ask chatgpt for a promt
lol
astral goblet
well actors are already being replaced in industrial and advertising projects, w...
actors will move from acting in commercials for pickup gigs, to acting in training sets
astral goblet
if you got applicable skills people will hire you
low moon
In a super meta way a very pholsophical theologic way I think AI is showign us how superficial and fake the world is and how easily our sense are fooled.
astral goblet
ai tools helping all the low skilled people pump more "quality" content out, will just raise the skill floor. thats all
broken cave
if you got applicable skills people will hire you
well i can tell you now, in the video game industry, which pays a lot of visual artists who are not guilded, it's donezo. maybe a year or two from now, the expectation is that most video game art will be made by computer
low moon
Also AI is showing us how dreams work.
broken cave
you could be the most skilled artist in the world, and people will pay very dearly for a technology that means that Mr. Exec does not have to Deal With You
low moon
AI looks remarkably like dreams. Early video AI anyway.
astral goblet
well i can tell you now, in the video game industry, which pays a lot of visual ...
ever played procedural games? that's what happens when you don't have good art direction.
a lot of bland content that isn't very riveting
low moon
AI is showing us reality isn;t as real as we thought.
astral goblet
3 exampls of procedural content done well, minecraft, star citizen, no mans sky. all 3 instances have heavy amounts of art direction and technical skill wrangling it to get the art where it needs to be
broken cave
a lot of bland content that isn't very riveting
yeah but that's more because the average person is more excited about the aesthetics of a game, like quite literally what is the IP, how does my avatar look, etc., than any substantive gameplay. one way to look at it is considered the appeal of movies, the audiences for like counterstrike is still very small. it will really be about the technology's capabilities to learn from some cozy IP in someone's childhood, like monopoly or hotwheels, and then adequately replace the engine of artists you need to make the IAP for a monopoly or hotwheels game
astral goblet
yeah but that's more because the average person is more excited about the aesthe...
i dont think it'll ever be that easy
low moon
Cinema is over. Imax is over Hollywood is over. within the next 5 years itll be over. like the studio system
astral goblet
we'll have augmented intelligence before ai gets to teh point where it can make those kind of stylistic calls
broken cave
we'll have augmented intelligence before ai gets to teh point where it can make ...
yes, there might be an artist who defines the IP, but there won't be the next 19 artists creating all the collateral for it
and anyway, you might want something that can be generated, instead of something that meets other aesthetic goals.
astral goblet
we're more likely to see the entire AI economy bubble collapse before AI generation replaces all artists
just look at stability partnering with block chains and daz 3d. that's not the direction anyone thought it was going
broken cave
we're more likely to see the entire AI economy bubble collapse before AI generat...
okay another POV is that there are talented indie game developers making procedural games not because they can't draw, but because they don't want the interpersonal challenge, no matter how small, of working with other people
low moon
The sad part is that we all have somethign to say but no one is out there to hear it. And scoail media / AI isn;t helping.
astral goblet
it's like corel draw releasing a new verison an people hyping it as the next big thing.. no buddy... it's corel draw
astral goblet
okay another POV is that there are talented indie game developers making procedu...
lol that'll always be an artist problem too. always was always will be. tortured souls
low moon
more alone than ever
broken cave
you may already know this, if you try to develop video games without a huge budget, but most low budget visual artists are flakes lol
yeah
well SDXL is not a flake
astral goblet
remember fez. lol phil fish
broken cave
yeah. who wants to deal with that?
astral goblet
didn't like what his fans were saying so he cancelled fez 2 entirely
broken cave
and anyway, for visual artists who aren't flakes, they have a tool that can program for them!
it's two sided
low moon
I see people use AI models to conjure up gone loved ones and make their own little heavens and ignore earthling else as it is too painful and confusing and just useless and frustrating.
its like the next best thing aftr suicide
thats what i would do
or use it for when it gets powerful enough
astral goblet
tech has long been redefining artists' roles. cartoons used to have armies of inbetweeners wrking on them. people drawing all the frames between key frames. now that's all procedural tech. artists usually only make a few key frames andthen adjust a computer generated product
low moon
recreate my childhood memories and stay in them
astral goblet
it never "replaced" animators though
low moon
maybe use some vr headset
astral goblet
it just raised the skill floor
astral goblet
i think most people are smart enough to know that a digital clone fo someone will be a digital clone and not the actual person
astral goblet
sex dolls already exist but emotions are only bluring those lines for a few very deluded people
it's a very small niche market
low moon
but its growing
astral goblet
is it?
broken cave
tech has long been redefining artists' roles. cartoons used to have armies of i...
programmers have their own insufferabilities. i think it comes down to who holds the capital, now, to build datasets and buy compute or whatever, and who has the knowledge, now, to command armies of programmers. it isn't actors and writers, and it's not vfx artists. if this were 60 years ago, you know, Disney Corp had a little bit of a hope of doing what stability does now. today? no chance
astral goblet
is it though?
low moon
very much so it is
astral goblet
programmers have their own insufferabilities. i think it comes down to who holds...
yeah that capital control is why i'm cheering on star citizen. it's come so far on crowd funding alone
low moon
less people are heavign real kids and families
too expnsive and painfukl
easier to have a soft waifu
broken cave
tech has long been redefining artists' roles. cartoons used to have armies of i...
so even though there are LLMs that program much better today than they can write creatively, and therefore an aspiring visual artist who wants to make a video game is better positioned than ever to fulfill that dream... you know, you need a lot of money and knowledge to run a video game nonetheless
astral goblet
the people you know maybe. world population hasn't capped yet
if people aren't having babies there blood lines will be replaced. that's just nature
low moon
its only the religious ones that keep having kids lol
muslems mostly
every big developed nation is on the decline with birthrates otherwise
no monye
broken cave
stability should be focusing on making tools for programmers, period, full stop. if it's art, it has to be the kind of art programmers want to generate.
low moon
and also who wants to brign into this worls
broken cave
programmers don't hold all the cards, but they hold the most of them
astral goblet
muslim is a religion lol. entirely different topic. forget i brought blood lines up. that'll get the cochroaches all riled up
low moon
most people i know includeing me say well i am miserable and messe dup why would i brign another soul into this
astral goblet
china is the fastest growing population though
low moon
not india?
astral goblet
so even though there are LLMs that program much better today than they can write...
"creatively" is subjective. I find LLM authored content to be VERY LLM authored. They don't really have style or pizazz
perhaps a trained writer with a knack for it can leverage an LLM by knowing what to prompt and what to discard.
stuck gazelle
Any new/updated plugins to save state and remember last parameters used on Stable Forge?
astral goblet
many will just take the first result they're given and call it done and publish it. meh.
astral goblet
Any new/updated plugins to save state and remember last parameters used on Stabl...
the text under the image. all the generation parameters. save that. when all that is pasted into the positive prompt and you click the little blue angled down arrow, it fills all your fields
you can save favorites as a style, then select the style and fill your prompt box with it, then blue arrow it into the fields
bit of a hokey pokey but the functionality is all there
stuck gazelle
Right, I'm just curious if there's a way to load save states, there was a plugin in A1111 but it doesn't work on SD forge
astral goblet
i just described a way. there's no extension. i don't trust UI extensions since they break all the time and often break other extensions.
low moon
Help us reload Styles in fooocus without having to restart fooocus to help you all. 😄
low moon
When an active Discord server goes quiet I imagine everyone had to go to the loo at once.
😁
woven charm
Lol
Actually, I had a sorta general question about training a lora in stable diffusion
woven charm
So I have seen the guides about training them to make characters, art styles, etc, but they dont quite cover what I was thinking
low moon
which is what
woven charm
What images do you use for traning, if you want to make a lora to help you generate images of people wearing a specific item
low moon
well you want idealy images of that item from different angles and not much else like no0 faces or background as much as possible
woven charm
The item being worn? or the item just by itself
low moon
doeasnt matter really as long a sits the item from different angles
sharp clear images also
remove faces and backgrounds as much as you can
becaus eif you leave them in whe nyou try to generate it will try to generate those faces and backgrounds also
woven charm
Like take a corset as an example. A corset is usually worn by a female character
if I take just a bunch of images of corsets and try to train the lora, wont it end up just generating corsets somewhere in the vicinity of a female character?
low moon
yes it will
woven charm
So then I would have to train the lora with images of people wearing corsets, but take out faces, backgrounds, extra details?
woven charm
Alright, so then I guess the final bit would be the two image types you need to train the lora.
Not sure if I am explaning it right, but for this example you would first have images of people wearing corsets (background, face, extra details removed). Then you would need mutiple refrence images for each example image
low moon
I would try with just the item from multiple angles.
If you add faces backgrounds etc it will affect the lora big time I know tat for sure
woven charm
Right. But like in the tutorial i was watching, she was training a LORA to make herself into a character. So first was a bunch of pictures of her face, hair, etc. Then there were multiple refrence images of generic women needed, like 5 times as many of them as the specific images of her face
low moon
Oh I think you mean those secondary images , forgot the name yeah
I didn;t use those
Regularization images yes?
I didn;t use those
pearl ocean
Right. But like in the tutorial i was watching, she was training a LORA to make ...
Have you even met Lora?
woven charm
Nah, shes too intimidating
distant barn
Hey guys. Im pretty new to stable diffusion. Im trying to get pictures of my pet dog in different styles. My dog being the reference pic. I tried image to image, control net and failed miserably for some reason. Is there anyone who knows what theyre doing willing to help me out and hop on a discord call with me? Id be willing to pay for your time (if youre actually knowledgeable)
placid comet
h
swift nest
fervent thunder
karmic cedar
I’m still waiting for a really good music album cover finetune. i hope they call it Joy Diffusion
?
verbal osprey
I think once stable diffusion 3 is out, I will try to train my own loras
pearl ocean
I think once stable diffusion 3 is out, I will try to train my own loras
train it on Nintendo charaters 🙂
verbal osprey
train it on Nintendo charaters 🙂
I will train it ONLY on F-Zero drivers
Pico for life
pearl ocean
I will train it ONLY on F-Zero drivers
😮
karmic cedar
Samurai Goroh or bust
nova zodiac
Or go the Pony route lol
pearl ocean
Or go the Pony route lol
generate pony images?
nova zodiac
generate pony images?
Train loras on MLP
woven rune
A noob question. Checkpoints can be used in comfyUI and Fooocus, right? Just moving them to modules>checkcpoints
pearl ocean
noob question, if I add a 4th monitor, would it slow my GPU down?
nova zodiac
noob question, if I add a 4th monitor, would it slow my GPU down?
Yes, but not by much
Will also reduce vram available for inference
pearl ocean
Yes, but not by much
🙂
bold mortar
how do you add libxcrypt-compat as a package in stability matrix so i can get my comfyui working
meager heart
How generate image from discord?
stable pawn
I think the BOTs for that are down, look in #1047610792226340935
meager heart
Thx you. What is Dreamstudio ?
bleak matrix
wild thistle
Halo Everyone am trying to upscale pictures sports pictures to higher resolution but i dont get more details i watched lot of tutorial in youtube but still struggling and please dm to me too because the thread is lost new to discord too.Any help would be appreciated.
@shell tendon can you teach me to upscale please needing help doing upscale with stable diffusion but not getting sharp results..
willow crescent
How about a donation link for SA here on the server?
nova zodiac
Halo Everyone am trying to upscale pictures sports pictures to higher resolution...
What UI are you using??
Either way you are looking for the Ultimate SD Upscaler - its a custom node in comfy, and an extension in a1111 that appears in scripts on the img2img tab
wild thistle
@nova zodiac am using Web UI i did installed the ultimate sd upscaler and also controlnet but i couldnt get a lot of details as shown in youtube i was using a real time sports video and i wanted to add more details to those video.
stone hamlet
hi, i need help. how do I download stable diffusion?
worn aspen
What is the process or concept I need to educate myself on when I have generated images I like, and want to extract further detail or fine tune those exact images. I'm pretty sure if I have and regenerate with the exact seed and node settings that I reproduce an exact image. And I believe dragging that image in to ComfyUI loads a workflow to duplicate that image.
furthermore: if I have the exact image I want, say a landscape image, and i want to add a sprinkling of a certain flower in the foreground without completely destroying the original (I understand that may be impossible), what processes do I investigate?
I'm using ComfyUI. Thank you!
digital eagle
so where i can generate images right now ? I haven't been in for months
rugged vale
which tools can you guys recommend for generating ai voices?
trail lion
so where i can generate images right now ? I haven't been in for months
you would have to install something on your pc or use a cloud-based service
fervent thunder
澡
karmic cedar
How do I find world peace?
trim magnet
generate world peace with SD3
karmic cedar
🫥
short flume
HOW TO USE IT?
karmic cedar
HOW TO USE IT?
the defining question
trail lion
without an auxilary verb it's almost more of a statement than a question
olive juniper
Hi everyone! I just moved from Fooocus to A1111 and I'm having a problem when inpaiting a dress mask. On Fooocus I upload the image of the dress and the mask and it creates a model wearing that dress with very good results, but A1111 makes images that make no sense, like ignoring the dress. I think I might need to download a checkpoint model for this... if that's the case, which one? Thank you!
karmic cedar
yeah, Fooocus crosses a lot of t’s and dots a lot of i’s
i wish the dev had more ability to keep up with the pace of model development tho
low moon
SD3 will certainly cure world hunger.
low moon
Stable Cascade was just a tease.
peak orchid
Given Stability AI's current situation and the details coming out about SD3, its looking like while it may be better than SDXL, it will not be competitive with Dalle 3
charred mesa
if anyone can give me a guide to make good cascade images then I'd appreciate that. People keep saying that cascade is underrated or whatever but I never see it myself
charred mesa
Given Stability AI's current situation and the details coming out about SD3, its...
yeah probably
text is 100% going to be good
prompt adherence is a massive improvement, but I don't know if it's gonna reach Ideogram or DALLE3
I've seen examples and stuff, it's just that we haven't tested it for ourselves
only then can we be so sure
I did try it twice (asking from people who have access on social media) and I think that for the images I would try to make it would be perfect
low moon
cascade is prettier than sdxl but less refined
makes sense? 😄
i use ideogram for letters and words
charred mesa
I used ideogram to test all my SD3 prompts on my prompt list
it didn't go with flying colours as it didn't recognize a bunch of pop culture stuff and celebrities had terrible faces
but it was intelligent enough to make a bunch of stupid funny stuff
I hope SD3 Turbo will do the same
low moon
Ai is brilliant and stupid at the same time.
charred mesa
cascade is prettier than sdxl but less refined
can it do good skin texture? most of the time I generated super smooth skin no matter the cfg or scheduler
low moon
yeah
charred mesa
do you have any good cfg/samlper/scheduler setting
low moon
one of my gripes too
charred mesa
ah
low moon
while the colors co,mposition lighting is really nice and follows promts good too but its too smooth
the grit is not there yet
charred mesa
yeah its relatively intelligent
low moon
and sicne eveyrone is waiting for SD3 no one is making refined checkpoints for it
charred mesa
when I ask for two ninjas facing eachother it gets it right, even if they don't hold the weapons I asked for
charred mesa
and sicne eveyrone is waiting for SD3 no one is making refined checkpoints for i...
yeah its unfortunate
even SD2.X got finetunes
and a bunch of people hated it
low moon
haha
charred mesa
and this one isn't evne hated, it just not cared about
low moon
yeah its sad
cedar salmon
What is the process or concept I need to educate myself on when I have generated...
inpainting might be what you want
charred mesa
and a bunch of people hated it
I was one of the people who used it lol
out of the 20 that is
I hope we can train loras for 8B with 24GB of vram
or some new different method which would work for all SD3 versions
low moon
if SD2 doesnt have higher requirements than SDXL in every department I am good
SD3 that is
charred mesa
well
2B + quantized T5 will probably run similar to SDXL if not better
I wonder about 6B
its such an unfortunate number
cause its a big gap after 2B, but a small gap before 8B
cedar salmon
2B or not 2B that is the question
rugged mirage
looking at this again https://twitter.com/chrlaf/status/1772228848387522728
has there been any news on the API portion?
charred mesa
looking at this again https://twitter.com/chrlaf/status/1772228848387522728
has ...
emad keeps saying that the API will be coming quite soon
and the model in a couple of weeks (3-5 according to CTO)
almost 2-4 weeks now!
rugged mirage
the API was supposed to be quite soon almost 2 weeks ago, and they need that to finetune the model based on the usage
and the API seems delayed, so Im guessing everything is delayed
with sdxl they collected human preference data to finetune for like a month I think
charred mesa
yeah
I remember 0.9 phase and everything
its scary how close the API thing is and the release
what is it even for when its like 1-2 weeks difference
and what was the waitlist for 
I mean it's really epic if they can fine tune it with our feedback so quickly and everything
rugged mirage
feels like that part will either be rushed, delayed, or done continuously both before and after initial release
you'd expect to need more finetuning data as everything else about the model grows (+ there's multiple of them) tho tbf Im not very sure how important the rlhf stuff is for SD
I haven't seen anyone doing RLHF to improve their finetunes for one
gaunt pulsar
I wonder if 16gb of vram can run SD3 8b. I can run 14b LLMs on my 16gb vram easily, so hopefully it should be possible
rugged mirage
at worst I imagine you can run it in stages with 16gb - do the text stuff first and then only load the image part
in the whitepaper they mention the full thing includes the 4.7B T5 encoder which you can do without (and also I'd assume someone will make it so you can do it in stages but slower), and if 8B fits on 24GB and you can shed 4+B then yeh 16GB should have workflows
probably even 12GB will be able to with workarounds
gaunt pulsar
Nice! If it can run on average consumer GPUs I think they are 100% going to open source it
charred mesa
yoinked has stated once or twice that the 8B model at fp16 + T5 at int4 will run on 12GB Vram (if not 8GB)
you just need plenty of normal system ram because of memory management
quartz seal
hey can tell me what happened to that SD thing froma couple months back where images were being generated in like 2 seconds? or 2 iterations or something?
what was that called?
rugged mirage
lightning
quartz seal
i dont recall it being called that
charred mesa
turbo?
rugged mirage
turbo
rugged mirage
lightning models are pretty popular though and fill the same niche
charred mesa
SDXL Lightning is 1024px and generates high quality images
whilst requiring like 6-8 steps if not lower
SD3 Turbo will act like Lightning as well, being native resolution as well
quartz seal
appologies i have been isolated from my pc for over 3 months, i need to catch up so much has changed already
dry trellis
I can't wait until consistent lighting model where we can have it redraw game art in real time
charred mesa
yeah SD3 is coming possibly (massive cope) this month
did you hear about SD3, Metavis
quartz seal
i literally just saw it mentioned in the announcements channel i think
quartz seal
i have no clue
charred mesa
its basically much smarter, its close to Ideogram and DALLE3
quartz seal
i was hoping there would be improvments at least to help my 3090ti cope
fervent thunder
Philosophy twist sell out or are you a sell out
quartz seal
I tried Comfy but it was bit of a steep learning curve
rugged mirage
it shouldnt run out of ram on 24gb on any of them
charred mesa
^^^^
quartz seal
lemmi check out forge
rugged mirage
when I dont touch it for more than a month I just delete everything and setup from scratch based on whatever people are doing currently
twin locust
Hey how do I download stable diffusion 3? Is that not open source yet? Can’t find it on huggingfaces or anywhere
charred mesa
also 24GB will 100% run SD3 8B + T5
charred mesa
Hey how do I download stable diffusion 3? Is that not open source yet? Can’t fin...
no
rugged mirage
Hey how do I download stable diffusion 3? Is that not open source yet? Can’t fin...
by waiting a month or so
charred mesa
no SD3 download yet 👍
twin locust
Ahhh too bad
twin locust
Is 1.5 the best one or what’s the best general one you guys use?
charred mesa
idk I use SDXL, but it's not like im gonna claim its the "BEST"
rugged mirage
given that you have 24gb of vram go for sdxl
twin locust
Maybe use SDXL as base and 1.5 as a refiner?
quartz seal
i use SDXL im saying when i use it it runs out of VRAM if i do anything complex
quartz seal
is there some kind of profiler tool i can use?
quartz seal
I could generate huge highresfix images with 12GB lol (comfyui though)
ha yeah right if i so much as look at hi-res fix with intent to use it it will crash
charred mesa
lmao
quartz seal
i use control net a lot
quartz seal
the suggestion of just deleting and then re-installing... i mean will i even get a newer version? I thought my SDXL was updating on load
version: v1.6.0-2-g4afaaf8a • python: 3.10.6 • torch: 2.0.1+cu118 • xformers: 0.0.20 • gradio: 3.41.2 • checkpoint: 0f1b80cfe8
is this out of date?
loud solar
I have seen people running Python 3.11 ... but still to scared to try 🙂
quartz seal
wait a second if SD3 is coming out, then what happened to SD2?
my version says 1.6.0?
loud solar
You are using a User Interface ... SDXL is only a model ...
quartz seal
ah model yeah - right!
loud solar
You could try "git pull" in your SD directory ...
quartz seal
i risk loosing my installed ext. though right?
loud solar
You never really know ... should be safe but sometimes one module changes and other won't be able to handle it and need update, too
quartz seal
yeah ok
im really struggling right now actually, just using txt2img and its doing nearly 4s/it
painfully slow
loud solar
95% it works fine ...
pure orbit
does the order of the words in the prompt have an impact?
are words earlier in the prompt more emphasized, for example?
proper karma
SD3 Turbo will act like Lightning as well, being native resolution as well
you can load models in chunks to process them in VRAM
quartz seal
testing variations of that never really occcured to me (for speed testing ii mean)
charred mesa
loud solar
testing variations of that never really occcured to me (for speed testing ii mea...
Maybe try to install Stable Diffusion Forge first. It can run next to an A1111 installation and handles resources a bit better
quartz seal
is that from a website, is there a link ref here?
quartz seal
thanks gonna pull now and try
loud solar
Good luck!
You can use your models from A1111 folder in Forge so no need for another download ...
quartz seal
You can use your models from A1111 folder in Forge so no need for another downlo...
thank you ill report back on how it goes
cobalt turret
interesting. identifying sdxl images as ai generated is much harder in the older 0.9vae compared to the new 1.0vae: https://hforsten.com/identifying-stable-diffusion-xl-10-images-from-vae-artifacts.html
charred mesa
I think
it would be a great experiment to put images through the vae of SD3
to see how much it has improved
VAE encode -> VAE decode
rugged mirage
does anyone even use the VAEs on their own anymore? isnt pretty much everyone using finetunes with baked in VAEs
hasty hornet
yea baked in - are just so much easier to use, but when I was using lots of models had to keep track of it, not all models did it at a time
had to name each one with vae in it's name or with "baked" as part of model name lol
rich kestrel
theres only 2 vaes, really
an anime one and a generic all-purpose one
for 1.5. so it's essentially a non-issue
pastel turtle
also 24GB will 100% run SD3 8B + T5
So 20GB of VRAM (A4500) w/o T5 is out of the picture. Too unfortunate.
charred mesa
whaaat
it will run on 16GB with T5
yoinked would probably prove it to you, but I'mma look for a stability dev's reply so its a 100% trustable source
karmic cedar
When will I be able to ask one of Devin’s open source cousins to design for me a meta-theatrical 16-bit RPG
charred mesa
ehhh
idk like 4-5 years if we're lucky
4-5 years if you want something that's actually decent
no wait, actually like 2-3 years tbh
loud solar
The Matrix is coming 🙂
pearl ocean
I’ve noticed that models tend to leak into other models which switching sometimes
Am I missing some argument on startup?
charred mesa
?
loud solar
I’ve noticed that models tend to leak into other models which switching sometime...
AFAIK A1111 has/had a problem with switching between models
One reason why people changed to SD Forge
pearl ocean
AFAIK A1111 has/had a problem with switching between models
I’m using forge
loud solar
I’m using forge
OK, should have been fixed there... no idea ^^
spiral crest
When will I be able to ask one of Devin’s open source cousins to design for me a...
2030
pearl ocean
Guys, I feel I might be starting to hoard models like, I already have a bunch lol
nova zodiac
Guys, I feel I might be starting to hoard models like, I already have a bunch lo...
If you arent upgrading your ssd to a 4tb or bigger, you arent hoarding!!
sonic token
Hey guys, this is my first message. I want to transform a video into a different one using stable diffusion, but the 'Lora' option I've chosen doesn't work or I can't do it. I don't have much knowledge about AI, so I need someone who can set up the model for me. I can pay for it, if someone want. If you can help, please dm me. Thank you. 'I don't have stable diffusion knowledge'
quartz seal
If you arent upgrading your ssd to a 4tb or bigger, you arent hoarding!!
damn i just upgraded to 2TB NVME and have 518GB models on it... I thought that was already overkill... and then you dropped that 4TB story lol
oooh i just got Forge running
woah 12it/s on 512/512 and 4it/s on 1024/1024
loud solar
Congrats
quartz seal
Just Switched to DPM++ SDE Karras from DPM++ 2M Karras and im getting even better performance more than 2x
loud solar
You should have some new samplers now, too
quartz seal
i see some turbo ones also
loud solar
A whole new universe full with possibilities 😄
quartz seal
anyone know a good model to use for making logos?
fair crescent
@quartz seal probably like adobe
quartz seal
<@901388916496687155> probably like adobe
is there a SD Model for that, or do u mean just use Adobe?
quartz seal
Have a look at Civit.ai ....
yeah havnt checked civit since christmas
karmic cedar
no wait, actually like 2-3 years tbh
So, one
quartz seal
i got 3090ti i aint paying no-one else to make pictures
fair crescent
@quartz seal maybe ask chat gpt for websites or apps
quartz seal
yeah was just about to
tepid finch
hello, so i've been trying to install stabilityai/control-lora-depth-rank256.safetensors control-lora-depth-rank256.safetensors Control-LoRA: depth rank256 but it keeps giving me a download error when it pulls from hugginface
fair crescent
? @tepid finch
nova zodiac
Just Switched to DPM++ SDE Karras from DPM++ 2M Karras and im getting even bette...
Euler A is usually used for benchmarking
nova zodiac
anyone know a good model to use for making logos?
Harrlogos
karmic cedar
These samplers really need a good drag through a marketing department. Their names are too silly
Like DPM 2++ Alpha Turbo Championship Edition
nova zodiac
Any good models to download?
Realism or anime?
loud solar
Working with the Mohawk at the moment
nova zodiac
Realism - icbinp xl
karmic cedar
Or bad 2003 digital camera?
pearl ocean
Realism or anime?
Anything.
But I think I got quite a few realism ones
karmic cedar
Does anyone use Zavychroma XL with DPM 3M?
pearl ocean
Realism - icbinp xl
I got that
nova zodiac
I approve 😉
karmic cedar
I usually stick with Juggernaut XL v9
nova zodiac
I usually stick with Juggernaut XL v9
You missing out!!
pearl ocean
I find epicrealism XL pretty good
karmic cedar
You missing out!!
Depends on the application
nova zodiac
Depends on the application
True, but I can still get a decent anime out of ICBINP XL, juggernaut slightly overtrained
karmic cedar
I’m not interested in anime at this time
pearl ocean
This one looked fun
nova zodiac
Looked fun, never gave decent results
karmic cedar
But I do take note of syntactic complexity whenever possible. I try to look for aesthetic nuance as much as my eye has been trained to see it.
pearl ocean
I find I get results with XL models
nova zodiac
But I do take note of syntactic complexity whenever possible. I try to look for ...
Try ICBINP v4 with the midjourney mimic lora at 0.2 and the juggernaut lora at -0.9 then 🙂
nova zodiac
Can do
The idea is that the lora substitutes/adjusts weights in the model to give you the desired result
loud solar
Time to sleep. Nini!
pearl ocean
just used a logo lora on a sd_xl base model, and it worked pretty well
I always wanted to make SVG logos from SD
quartz seal
I always wanted to make SVG logos from SD
just make normal ones in sd and then use inkscpe to svg it, waaaay easier
but yeah if there was an extention for svg production i mean that would be pretty cool too
pearl ocean
😮
quartz seal
yeah just double checked its called inkscape
i mean if u make a cool SD logo, then use the extention called bg remover, then upscale save to png and send to inkscape save as svg
pearl ocean
Sounds like a lot of work
quartz seal
it couldnt be quicker, i mean from SD to a finished SVG file from inkscape u looking at 1minute max
rich kestrel
If you arent upgrading your ssd to a 4tb or bigger, you arent hoarding!!
not wise
on a tb per tb price, it no longer scales in your favor after 2tb
the key is to get a motherboard with 3 or more SSD slots and buy multiple 2tb ones
plus a big ol 6tb+ HDD for vault storage. I got like 14TB total space using this strat
quartz seal
my mobo has 3
rich kestrel
yeh so just fill them out instead of splurging on a large drive
quartz seal
x570 taichi
i bought this 2TB called Crucial P5 Plus, was a great deal on Amazon and performs increadibly well.
nova zodiac
not wise
Not familiar with current ssd pricing tbh, but I do remember having 2x 3tb hdd worth of music and movies and tv series like 10 years ago
rich kestrel
hdds are dirt cheap
but must only be used for long term storage, not daily storage
for me, it's basically an inbetween before the trash bin. I just dump stuff in there and later come back to organize since theres so much room
pearl ocean
i bought this 2TB called Crucial P5 Plus, was a great deal on Amazon and perform...
8TB when
quartz seal
i do 1-2TB NVME Drives for OS and Running Apps(incl models), then bigger SSDs into the SATA slots for general storage and then I have a 12TB Western Digital Enterprise NAS Disk for long term storage
quartz seal
8TB when
i dont think ill buy an 8TB nvme, i just cant find a good value/money proposition (at least for now)
quartz seal
models are expensive but also video production and virtual machines, they take lots of space and really work best when they run on faster IO disks
tepid finch
can someone help me in gen with images? I posted a question
pearl ocean
to think, on SD local, you can generate images till the cows come home
sweet flint
Hi. is there a way to see how much time it took for a paticular image to generate?
cuz sometimes I leave big images that take 10+ minutes. but when I come back I have no idea when it was completed or whatever. if that makes sense.
so is there a way to see how much time it took for a particular image to get done on stable diffusion.
dry trellis
so is there a way to see how much time it took for a particular image to get don...
It shows in the command prompt
sweet flint
It shows in the command prompt
ah. also every image I make gets saved on its own. is there a way that doesn't happen. cuz I don't really need all of them.
it becomes cluttery.
quartz seal
it becomes cluttery.
it would be cool if they could add a button to the Image Browser extension that lets you select a ranking number and then click a trash icon to delete all images associated with that rank number.
Also to automatically rank all images as #Unranked until it is #1 or #2 or #3 ranked etc...
So then you can rank all your favorite ones with a #1 or #2 or #3 etc and then click Unranked and click the trash icon.
this will delete all unranked images.
something like that would be pretty cool 😎
karmic cedar
Try ICBINP v4 with the midjourney mimic lora at 0.2 and the juggernaut lora at -...
What does that do better?
nova zodiac
What does that do better?
More details, more style, same good realism
floral nimbus
when it comes to the version of stable diffusion on github which has the least amount of issues
its really annoying because certain extensions just say nah or just freeze
besides forge cuz screw the controlnet in forge
i went with 1.7.0
sour needle
warped herald
Hello friends, I did a podcast conversation with Simone and Malcolm A Conversation with Malcolm and Simone Collins (Supercut)
https://youtu.be/QIqoMSf4P88
sly eagle
do you know if it exists a stable audio for generating voices?
coarse marsh
suh
trim magnet
zucc
pearl ocean
so yeah, I have had SD on my PC for a month or 2 now
now im just wondering what I actually can do with it, other than just having fun lol
karmic cedar
It’s a ritual
karmic cedar
Unless you’re generating function, you’re generating reflection
granite solar
hi
quartz seal
hey guys, this SD Forge is 1000x more stable \ reliable \ smooth than the standard A1111. Thanks so much fo rthe headsup to run this last night. @loud solar
loud solar
hey guys, this SD Forge is 1000x more stable \ reliable \ smooth than the standa...
YaY! Have a lot of fun with it! That's why we are here. To share experience and have fun!
olive tapir
Where can i create images ?
loud solar
Where can i create images ?
Hi! No bot here at the moment. We do local but there are some webservices like Leonardo.ai
quartz seal
Where can i create images ?
hey Sady how u got the clown pfp thing?
polar pike
Hello everyone! I sincerely request you to fill out this short and quick survey which comprises of 5 questions that compare realistic face generations from SD 1.5, 2.0, SDXL, Dall - E - 3 and a fine-tuned SD1.5. It is for a research to check whether it is possible to increase the accuracy of real face generations of Stable Diffusion based on given prompts of facial descriptions.
Here's the url: https://forms.gle/avNEgKWp8nj3UAEg9
wet dawn
hi
wispy nacelle
Found this free app that finds me the right Ai tool for each task instantly. You guys should check it out! Look for “Ai Index”. It’s the yellow one!
quaint walrus
What are the chances of Stable Diffusion 3 being released within the next 1-2 weeks? Also, I'm curious about how large of a model I can run smoothly with a 3080 10G graphics card. (Translated to gpt4)
low moon
SD3 will be released by the solar eclipse.
quaint walrus
I searched the internet, and it says, "On April 8, 2024, there will be a solar eclipse crossing the continental United States from Texas to Maine." I'm not sure if it's a joke or serious, but I would really be happy if SD3 is released around that time.
deft horizon
So... In 2 days ?
honest spear
it'll release at the next full planetary alignment
charred mesa
If SD3 weights actually come by the end of this month I'd be super happy
But we can't be so optimistic, we have to consider the time it takes for them to optimize and make controlnet and fine tuning tool(s), which Emad promised
and the API feedback stuff hasn't even started yet
inner summit
arctic sedge
and the API feedback stuff hasn't even started yet
"and the API feedback stuff hasn't even started yet"
Wait... So does SD3 have like a personalized image rating system for reinforced personal feedback or something?
I didn't know about this.
charred mesa
oh I mean like the waitlist part
there have been beta testers whom had been invited to the secret discord server
I just don't know if there's enough time to suddenly invite a couple thousand people from the waitlist AS WELL and process the massive amounts of feedback that would be awaiting
I mean its all automated anyway ig, just like sdxl 0.9 -> sdxl 1.0
arctic sedge
I mean its all automated anyway ig, just like sdxl 0.9 -> sdxl 1.0
Ohhhh! Yeah. Sorry. Yh.
Still can't wait to have more complex prompts with our new waifu generator. xD
tranquil steeple
bot generate
shell tendon
bot generate
Here is the image you requested. #🏞｜general-with-images message
tepid finch
Gm
So I want to try out animation. What's a good module to use to animate psychedelic style images like Alex Grey?
karmic cedar
ooo someone throw some controlnets on some supermarket buildings and give them alex grey paint
cinder cloak
We used to be able to create images in Discord. Is this always the case?
charred mesa
I wish I could try pixart-sigma, it's a DiT model
but the python dependencies keep messing up
it even has a conda environment .yaml and that has an error
rugged vale
is there a tool in kohya ss that allows users to analyze and remove certain tags in a data set?
as in bulk remove
karmic cedar
keep an eye on this one kids, it’s gonna be a doozy once it’s scaled up 😮
a colab as well
autumn frost
hi, i just join a minute ago, would like to know if there is any mobile app for ios to learn to use stable diffusion?
oblique elk
hi, i just join a minute ago, would like to know if there is any mobile app for ...
not quite sure, Draw Things mentions that it need "only" an A12 Chipset which... so maybe possible. Will not be the fastest i guess but maybe it works....
Might take a minute and needs to download 2 gbyte Model file as far as i read
autumn frost
not quite sure, Draw Things mentions that it need "only" an A12 Chipset which......
thanks so much for your reply. I am looking for something that is better quality than whatever the best claimed to be. Thanks again
rugged vale
hi, i just join a minute ago, would like to know if there is any mobile app for ...
wait learn to use?
oblique elk
thanks so much for your reply. I am looking for something that is better quality...
Better then best, sure why bother with bad quality - i feel you. Guess maybe you should rethink about your strategy using the mobile device approach 🙂
rugged vale
hi, i just join a minute ago, would like to know if there is any mobile app for ...
youre best off just watching some youtube
autumn frost
wait learn to use?
it might be a little bit for me to learn SD, as i am fairly old ..haha
oblique elk
Uh stop, you are right, it is about learning not about using. Draw things was meant for creating images not learning about stable diffusion... sorry my fault
autumn frost
Better then best, sure why bother with bad quality - i feel you. Guess maybe you...
thanks ☺️
autumn frost
it might be a little bit for me to learn SD, as i am fairly old ..haha
i meant hard to learn to use SD
autumn frost
Uh stop, you are right, it is about learning not about using. Draw things was me...
i see
rugged vale
i see
https://www.youtube.com/watch?v=kqXpAKVQDNU&list=PLXS4AwfYDUi5sbsxZmDQWxOQTml9Uqyd2
here is a playlist of a guy explaining a) installation of SD webui on windows and b) how to use it
oblique elk
i see
i would suggest the way of learning by doing. You nearly can't make errors so after installation try prompting, learn about styles and models. After that maybe Controlnets and Loras and followed by the training merging of own models....
autumn frost
https://www.youtube.com/watch?v=kqXpAKVQDNU&list=PLXS4AwfYDUi5sbsxZmDQWxOQTml9Uq...
thanks for the YT link, unfortunately i don't have a computer. long story short, i am desperately looking for an alternative to replace what i used 2 days ago
autumn frost
i would suggest the way of learning by doing. You nearly can't make errors so af...
i assumed has to use a desktop computer not using a phone or tablet?
rugged vale
i assumed has to use a desktop computer not using a phone or tablet?
you could try using colab, https://www.youtube.com/watch?v=8_V8CO_Dbdw
-its less user friendly though
autumn frost
you could try using colab, https://www.youtube.com/watch?v=8_V8CO_Dbdw
-its less...
thanks again so much 🙏🏻
oblique elk
As said at least prompting and styles can be done locally on a phone (will take some time when generation takes 1 minutes and trail and error is usally the way to become better at prompting).
autumn frost
As said at least prompting and styles can be done locally on a phone (will take ...
ok, great guide from you. i suppose it is all totally different promoting patterns vs Dalle and MJ , right? I did check out months ago. looks difficult prompting parameters in SD
thin oyster
Hi everyone 👋
low moon
Hi.
I am human and not AI I can assure you.
Running stable diffusion on phone is like watching imax ultra high def movie on phone.
forest trout
it might be a little bit for me to learn SD, as i am fairly old ..haha
Just start with default settings. When you feel comfortable, only change one thing at a time to learn how it works.
low moon
David Lynch will come and arrets you.
autumn frost
Just start with default settings. When you feel comfortable, only change one thi...
thank you
forest trout
David Lynch will come and arrets you.
Here is the image you requested. #🏞｜general-with-images message
rugged vale
how can i best manage tags in a data set?
i want to remove a some unecessary one, but the dataset is on the larger size and doing it manually would take some time
forest trout
You training a model?
pearl ocean
A hot model?
forest trout
i want to remove a some unecessary one, but the dataset is on the larger size an...
If you are comfortable in bash / terminal I'd probably just create a loop and use grep / sed commands.
rugged vale
If you are comfortable in bash / terminal I'd probably just create a loop and us...
too bad im using windows
im only familiar with linux
forest trout
You could hack one out with ChatGPT helping you.
rugged vale
You could hack one out with ChatGPT helping you.
its worth a shot i guess
forest trout
Just watch for your edge cases and that's likely the best method I can suggest.
Don't rm -r system32, etc.
rugged vale
Don't rm -r system32, etc.
lmao hope not
rugged vale
What about system 33
its a virus delete now rm -r
can anyone recommend a good system monitoring tool?
task manager telling me im using 5 % gpu while training
pearl ocean
😮
ancient osprey
I miss the bot. Nothing was picking up what I was trying to do as well as the bot. 😦
coral granite
Hi, whats is the best setup for realistic images? And is there any universal prompt?
pearl ocean
Hi, whats is the best setup for realistic images? And is there any universal pro...
I guess the best realistic method would be take a photo of something you want, and let AI manipulate it
coral granite
I guess the best realistic method would be take a photo of something you want, a...
But i wanted to generate something that doesnt exist
floral nimbus
pearl ocean
i just want to connect A1111 to Civitai
Forge is better
nova fable
Anyone have recommendations for tools for removing watermarks/logos from images? I have a few million photos with watermarks in the corner, and I'm wanting some sort of tool that removes the watermarks and uses the surrounding colors to intelligently guess how to fill in the space. I don't care if the filled in space is a bit off/blurry/distorted, but just want it to be relatively close to surrounding image so it's not glaringly obvious right away.
forest trout
Photoshop...
karmic cedar
Anyone have recommendations for tools for removing watermarks/logos from images?...
I suggest reviewing a cursory history of digital image editing on YouTube, starting with Photoshop 7 and the introduction of the Healing Brush algorithm.
vague pond
Forge is better
forge would be better if it wasn't a month behind on updates
by this point i have a feeling that mainline stable-diffusion-webui will have implemented the unet patcher and all other worthwhile features of forge before there is a single update to it
charred mesa
I finally got pixart-sigma working
but its super slow
slower than whatever SD3-8B might even be
pulsar lintel
Forge is better
That’s the problem it’s not connecting properly
To it sense the new update and pay for buzz has happened
nova fable
I suggest reviewing a cursory history of digital image editing on YouTube, start...
No, i need to do this to >2.5 million images. I don't want photoshop tips lol - i need an AI based automated / batch editing solution
i already know how to do it by hand in GIMP
pearl ocean
That’s the problem it’s not connecting properly
😮
pine path
I finally got pixart-sigma working
I think you can quantize it. I'm not sure how to do that myself though
I've also heard using cpu for t5
civic schooner
Daily question: Any news about the release of SD3? It's been 6 weeks since the announcement.
tacit tiger
Daily question: Any news about the release of SD3? It's been 6 weeks since the a...
I saw a reddit post that said it will take 4-6 more weeks, so until around first week of may :/
civic schooner
I saw a reddit post that said it will take 4-6 more weeks, so until around fir...
Thanks
pearl ocean
I saw a reddit post that said it will take 4-6 more weeks, so until around fir...
may the 4th be with you
civic schooner
Knowing the existence of SD3 makes playing SDXL to be not enough
pearl ocean
Knowing the existence of SD3 makes playing SDXL to be not enough
wait till SD3-XL comes out
civic schooner
fervent thunder
Psa vegetarians don't need teeth
If human was ment to be vegan u would have no teeth
fervent thunder
Pretty sure cows have teeth
Cows are domesticated and genetically bred to how we want them
fervent thunder
Ok so it's unnature man made
Samme with died of a cow
Diet
Also global suppression of silence by YouTube showing you the top comonets first also Instagram showing you top comments rather than timeliness order
And postings
Shadow ban AL these are connected some how
My degen house mates don't even take trash out but cry and report me bc I offerd a 28 year old tequila grow a pair and stop degening
crude notch
theres a 8b (largest) one and a 2b (sdxl sized) one
dusk mesa
Any ideas on when SD 3.0 goes live?
pearl ocean
Any ideas on when SD 3.0 goes live?
Your guess is as good as a Stock pick
desert swan
I created a close-up photo of a man with Dreamshaper XL model that I really like. When I was looking at it again the next day I noticed
that one of the ears is about half the size that it should be and it has been really bothering me ever since then.
What are some possible solutions to fix this?
I appreciate any suggestions
dusk mesa
Need some help. I want to build a pipeline that takes picture of my lawn and enhances it, like adding patio area, lights, flower bed etc. on it. Any one who has achieved this before? Or maybe any wisdom you could share?
peak orchid
Anyone have recommendations for tools for removing watermarks/logos from images?...
build a detector model and then use SD for inpainting
nova fable
build a detector model and then use SD for inpainting
Maybe that's what I'll need to do ... I was hoping that I didn't have to roll my own, because I figured somebody must have already created a batch watermark removal tool 😦
pearl ocean
do you guys just generate images all day?
dusk mesa
do you guys just generate images all day?
Wouldn't have to if it worked on the first prompt.
karmic cedar
help does anyone know what shampoo hunter gatherers used
karmic cedar
do you guys just generate images all day?
i’m down to a few short sessions a week. i like to try and bake my ideas in my own head case as long as i can before i take it to the bridge
karmic cedar
😮
i find it keeps me fresh. when i produce too much stuff, i burn myself out. on myself. and it sucks.
pearl ocean
what do ya use the images for?
karmic cedar
usually i’m either on a restoration workflow of some kind, or i want to animate something. i enjoy making fake concept trailers for shows and movies that never existed
pearl ocean
😮
foggy halo
guys SD3 is actually good
karmic cedar
yep
foggy halo
Daily question: Any news about the release of SD3? It's been 6 weeks since the a...
SAME QUESTION 😭 😔
karmic cedar
it ain’t the instant mashed potatoes, it’s the fresh stuff with the roasted garlic
foggy halo
guys SD3 is actually good
wish i had access
yes yes
also stable cascade is amazing tbh
foggy halo
lol yes..i generated a few images and was not disappointed
karmic cedar
foggy halo
wow this is good thanks a lot!
tiny vortex
Hi guys, I'm sohan and I just integrated Groq and OpenAI APIs inside comfyui, would appreciate it if you guys could check it out. https://github.com/ronniebasak/ComfyUI-Tara-LLM-Integration/blob/main/README.md
karmic cedar
one thing Stable Cascade can’t do quite right is “photograph of thomas the tank engine as a human male”
it couldn’t do that one
mainly because thomas the tank engine is already a human male turned into a tank engine, so to turn a tank engine back into a human male, well, that’s a tall order mmmk
pseudo wadi
hey im a hobby artist trying to get with the times and know fuck all about programming. are SD models or checkpoints just code trained variably using filters?
low moon
They're made with Unicorn dust.
A gold parrot shakes the dust real hard and shouts our prompts at it so it settles into our images.
pseudo wadi
A gold parrot shakes the dust real hard and shouts our prompts at it so it settl...
is that a golden parrot or solid gold parrot
quaint walrus
With a 3080 10GB graphics card, the limit would probably be the SD3 2B weight, right?
fervent thunder
WS
low moon
is that a golden parrot or solid gold parrot
Like the T-1000
unkempt hatch
I'm trying to make more gender balanced models, as I find the models still balance to female
if anyone is interested in sharing datasets, specifically featuring 50% or more males, I'd be very interested
sleek gazelle
OMG
i had not use this channel for a long time
if somebody will tell me where is the bot , i would greateful to this guy
loud solar
if somebody will tell me where is the bot , i would greateful to this guy
Bot free zone ...
stone latch
if somebody will tell me where is the bot , i would greateful to this guy
There is no bot
sleek gazelle
There is no bot
so how can i find bot in this channel or other channel
stone latch
so how can i find bot in this channel or other channel
There are no channels where a bot can be found because it's currently down
#1047610792226340935 - the red circle indicates it's not working
stone latch
i dunt know how to finish it now
Someone might be able to generate an image for you, you can ask at #🏞｜general-with-images - don't write only a prompt though
I'd do it but I don't have access to a computer at the moment
sleek gazelle
I'd do it but I don't have access to a computer at the moment
thank u for answer my question
sleek gazelle
Someone might be able to generate an image for you, you can ask at <#10041591223...
ok. i will trying to do
fervent thunder
'ơ
true canopy
im sure someone can help
rugged vale
hey anyone know what to use as regularization images when training a style?
wild thistle
Someone wrote a message to me how i search the thread for mentions #noob
frosty condor
Have you tried applying for another job, rather than eBegging
One message removed from a suspended account.
pearl ocean
One message removed from a suspended account.
Obviously not trying hard enough, your eBegging on discord
true canopy
well
outer crescent
parasite is doing it on other servers too. fake dead dad.
true canopy
u can always try spamming more channels, keef
true canopy
im no expert, but i wouldnt ware a choker to job interviews either
pearl ocean
im no expert, but i wouldnt ware a choker to job interviews either
Wear a pink dress
true canopy
he might as well
amber bloom
im no expert, but i wouldnt ware a choker to job interviews either
That depends on the job I think
true canopy
thats true indeed 👍
pearl ocean
thats true indeed 👍
.com
shy lotus
hi
gentle plume
Hello community, is anyone able to use stable diffusion for free (as was possible before) or is it finished via discord? Thank you 🙂
true canopy
not sure what ur asking, but yes u can freely use it for free
outer crescent
do you mean as a bot?
sonic token
guys help, my img2img work not looking same as the original one. i need good controlnet any suggestion?
gentle plume
not sure what ur asking, but yes u can freely use it for free
thank you for your feedback... before I could access the sdxl bots... but for 5 months, no bot has appeared... I don't know if I'm any clearer, sorry
gentle plume
do you mean as a bot?
thank you for your feedback 🙂 before I went through the yes sdxl bots but I no longer see any... maybe there is another way now to use stable diffusion in free version via discord? I'm lost :/
amber bloom
The discord bots here are turned off, possibly permanently. I don't know of any other discord bots to generate Stable Diffusion images, but you can generate some images for free with services like https://app.leonardo.ai/
gentle plume
The discord bots here are turned off, possibly permanently. I don't know of any ...
I think I expressed myself incorrectly...before I created images with stable distribution for free using an SDXL bot...but for several months...these SDXL bots no longer appear...and I don't know how to still use stable broadcasting for free like before...
amber bloom
I think I expressed myself incorrectly...before I created images with stable dis...
you can't. not on discord. You have to either use a web service or run stable diffusion on your own PC
gentle plume
you can't. not on discord. You have to either use a web service or run stable di...
ok thank you for your feedback. I am very new, what can I do in a simple way to use stable diffusion for free? ..I'm sorry but I don't understand "run stable diffusion on your own PC"
true canopy
best would be to follow a guide and run a web ui
get stability matrix and run it, its the most simple
sharp hemlock
Does anyone know how LORA training works well? I have a question about one I'm doing. It's to maintain coherence in the faces of a girl. Is there a way to improve LORA by adding more photos? Or do you have to start from scratch if you want to add more photos?
true canopy
it would also help if you have a nvidia graphic card
gentle plume
it would also help if you have a nvidia graphic card
It's really very kind of you "Brave" to take the time to respond...but I don't understand what you're telling me to do, sorry
gentle plume
it would also help if you have a nvidia graphic card
i'm not nvidia graphic card, no :/
true canopy
amber bloom
i'm not nvidia graphic card, no :/
Then your simplest option is to go to https://app.leonardo.ai/ and sign up. you can generate some Stable Diffusion images there for free. Not unlimited, but you get more free images every month.
fervent thunder
it would also help if you have a nvidia graphic card
I've got a GeForce 1050 with 3gb yet it keeps running out of memory even with 500x500 images.
--lowvram helps but damn
warm junco
I've got a GeForce 1050 with 3gb yet it keeps running out of memory even with 50...
Make sure you use --xformers
true canopy
perhaps you could try forge ui, some people have had success with it
with low ram
vram*
true canopy
i'm not nvidia graphic card, no :/
https://www.youtube.com/watch?v=gEwPGyWjK70
use directml
fervent thunder
Make sure you use --xformers
yea that too. not sure what it does but i use those two and those two only. are there any other args that could help?
warm junco
i'm not nvidia graphic card, no :/
What's your gpu?
fervent thunder
you could always use cpu but that's gon be super slow i reckon
warm junco
yea that too. not sure what it does but i use those two and those two only. are ...
Nope, but as Brave said you could try the Forge webui as its better for low vram gpus
true canopy
i think, the worse hardware u have, the better forge is, with highend, its a lot less performance boost
gentle plume
Then your simplest option is to go to https://app.leonardo.ai/ and sign up. you ...
thank you very much...; I'm going to do that... what a shame because I loved working with stable broadcast via discord... because the results were always great
fervent thunder
Nope, but as Brave said you could try the Forge webui as its better for low vram...
intersting. i always thought the webui was just an interface so it should have 0 impact on perf. the backend is what powers the webui with python.
warm junco
intersting. i always thought the webui was just an interface so it should have 0...
There are different "webuis" they also use different Backend Settings
Gradio is the browser ui for them
gentle plume
https://github.com/LykosAI/StabilityMatrix
thank you again for the link...but if I understand...if I don't have an NVIDIA card and a small laptop...it's better not for me to install this software "stability matrix"?
true canopy
some seem to be getting 30-40% performance boost
warm junco
thank you again for the link...but if I understand...if I don't have an NVIDIA c...
What's your gpu ? Because it also works with AMD cards
true canopy
thank you again for the link...but if I understand...if I don't have an NVIDIA c...
im afraid its going to be hard to run it on a laptop yes
gentle plume
What's your gpu?
I don't know...I'll ask the friend from whom I bought the PC and I'll tell you 🙂
true canopy
i think i might have a solution for u @gentle plume
warm junco
I don't know...I'll ask the friend from whom I bought the PC and I'll tell you �...
You can also open up the Taskmanager, and under the Performance tab check the GPU
true canopy
@gentle plume use this
https://huggingface.co/spaces/PixArt-alpha/PixArt-LCM
gentle plume
i think i might have a solution for u <@1160489634195849416>
oh that would be really awesome!! :))
gentle plume
<@1160489634195849416> use this
https://huggingface.co/spaces/PixArt-alpha/PixAr...
ok i see...
true canopy
u can do endless images there
its very fast
u can go to advanced tab and change style as well
gentle plume
ok i see...
I just tested...thank you very much...unfortunately it's far from the quality I had before with stable broadcasting :/ but than you very much ...its cut
gentle plume
sorry i don't speack very well english... I mean it's nice of you to take the time to answer me
true canopy
i find it hard to believe its worse off then,,,,stable broadcasting, what is stable broadcasting even?
gentle plume
i find it hard to believe its worse off then,,,,stable broadcasting, what is sta...
I told you... thank you very much for your link to pixArt (huggingface)... I tried... but I find that the rendering is less good than on stable diffusion sdxl... but thank you very much for having me showed your link because I will use it anyway, I think
true canopy
@gentle plume feel free to use sdxl here
https://huggingface.co/spaces/ByteDance/SDXL-Lightning
if you like that more
gentle plume
<@1160489634195849416> feel free to use sdxl here
https://huggingface.co/spaces/...
thank you very much "brave' 🙂 i try 🙂
placid hatch
anyone know a good upscaler to use with hires fix for sdxl?
gentle plume
<@1160489634195849416> feel free to use sdxl here
https://huggingface.co/spaces/...
i try...but the quality is not very good...but thank you to your link 🙂
warm junco
anyone know a good upscaler to use with hires fix for sdxl?
4x Ultrasharp is good for realistic stuff
placid hatch
Are upscalers that came out during when it was just sd 1.5 compatible with later sd model types?
low moon
supir is the only uspacaler worth looking at
red raptor
Hello, is there a way to retrieve the prompt used to generate an image few days ago ? is there some sort of log for those things like the one in Fooocus ? Thanks!
grizzled harness
Hello, is there a way to retrieve the prompt used to generate an image few days ...
was the metadata embedded in the image?
red raptor
was the metadata embedded in the image?
i'm new to all this. i have no idea if it was or not.
grizzled harness
i'm new to all this. i have no idea if it was or not.
can you upload (not copy/paste) the image here?
or, well, maybe in general-with-images 🙂
red raptor
can you upload (not copy/paste) the image here?
sure
red raptor
can you upload (not copy/paste) the image here?
did upload it in general-with-images
grizzled harness
did upload it in general-with-images
yup
`anime, 1woman, worrying, alone in the middle of (empty road in a city:1.2)., Cartoon caricature realistic, cute as a pixar disney character caricature cute full body from up 2009 unreal engine octane render 3d render photorealisticlora:3DMM_V12:1
Steps: 20,
Sampler: Euler a,
CFG scale: 5,
Seed: 4242146666,
Size: 512x960,
Model hash: 7ed8da12d9,
Model: animaPencilXL_v100, Lora hashes: "3DMM_V12: 7d5c705f0ff8", Version: v1.8.0`
red raptor
awesome, how can i see these ?
grizzled harness
Most of the image generation tools embed the metadata in the image. Including this one.
The tools themselves also usually have a way to view back metadata. What image generation tool are you using?
Alternatively, there are various programs that will help you view and manage images.
Shameless plug for two tools I created..... 🙂
https://github.com/GarlicCookie/PNG-SD-Info-Viewer
https://github.com/GarlicCookie/SD-Quick-View
red raptor
The tools themselves also usually have a way to view back metadata. What image ...
Automatic1111
grizzled harness
People also seem to like the "Infinite Image Browsing" program. It's a larger feature rich browser that I believe runs in a browser.
grizzled harness
Automatic1111
In Auto1111, go to the "PNG Info" tab and select your image.
red raptor
In Auto1111, go to the "PNG Info" tab and select your image.
Thank you so much for the help 🫡
grizzled harness
No prob, Bob
regal snow
Sharing this again in case anyone missed it… I forked the best extension I could find for aspect ratio buttons, then added a number of improvements. https://github.com/altoiddealer/--sd-webui-ar-plusplus
fervent thunder
12 gb vram ok or 24gb needed
regal snow
I imagine when you have 24 you’d wish you had 48. 12 is fine for most things
wicked crater
Hello community, is anyone having problems using stable audio? because from the first stable audio model I want to start using it and it gives me a sign saying that I was blocked and it won't let me enter
fervent thunder
I imagine more vram will be required in the future so , better just go with 24
karmic cedar
supir is the only uspacaler worth looking at
SUPIR represents the next iterative phase of super resolution models. When you’re upscaling something that barely has any fidelity, you need to rely on more imagination—that’s essentially how SUPIR is making use of LLAVA prompting
fervent thunder
SUPIR represents the next iterative phase of super resolution models. When you’r...
I really dislike SUPIR, ok for adding extra details and keeping the image the same, but i think reimagining the image is way better, upscale via ai, not just to add details but to create them as well.
bold crag
Anybody knew why dreamstudio draws better than the sd modle I installed in my computer?
karmic cedar
I really dislike SUPIR, ok for adding extra details and keeping the image the sa...
SUPIR is anchored around preserving fidelity; if that’s not a concern, then I like to use stuff like Clarity Upscaler—have you tried that one?
also good morning to the community from PST
tidal trellis
prime fulcrum
I am very new to all this so do bear with me please. I run a silver jewellery store and I just want to know if it’s possible to generate models wearing my exact jewellery. Not like it, but exactly the same. I have high resolution images of my jewellery pieces and I want to generate models wearing it for online marketing on Instagram
fervent thunder
SUPIR is anchored around preserving fidelity; if that’s not a concern, then I li...
seems good, just not for people. I'm sticking with krea ai atm
fervent thunder
I am very new to all this so do bear with me please. I run a silver jewellery st...
It's possible but definitely not easy. It would depend on how many photos you have of the jewellery.
I would create 3d models of them first, using ai, then put them in images and apply cleanup with stable diffusion in the end to remove noise + higher quality.
I can give it a go if you want : )
vast marsh
Hi, I have an RTX2060, will sd-webui-forge significantly speed up performance for me compared to using A1111?
prime fulcrum
I would create 3d models of them first, using ai, then put them in images and ap...
Thanks, really appreciate your help 🙏
spring herald
Hello people! Are there any image making enthusiasts from China? We are having a small contest and need one of the judges to be from China.
If you want to be a selection of the best neuronics papers write in private
fervent thunder
Nope, but as Brave said you could try the Forge webui as its better for low vram...
damit it's only available for windows not for linux. too bad
mortal delta
Dose stable diffusion and other ai tools, have a use case yet besides for memes? like what can the everyday joe use ai for anyways if i might ask?
amber bloom
Anything you need an illustration for. I've seen ads with AI generated pictures on the subway, for example
fervent thunder
SUPIR is anchored around preserving fidelity; if that’s not a concern, then I li...
After playing around with it for a while, it seems amazing! Best open source upscaler i've seen.
trail lion
is there a link for that on auto1111?
proper karma
anyone heard any news on SD3?
charred mesa
not much
I did get to know that 2B and other smaller models beside 6B are 512px
they look good despite that, thanks to the 16 channel vae
and they will be getting 1024px finetuning officially
wary belfry
and they will be getting 1024px finetuning officially
but you don't mean a refiner for 1024, but the model itself outputs 1024? The refiner workflow was pretty unpopular in the end 😄
charred mesa
no refiners for SD3
8B will generate 1024px natively with no special fine tuning
2B (and maybe other smaller models) generate at 512px though and they will be getting finetunes so that they can generate at 1024px as well
its not like they might need this as the 16 channel vae is a massive improvement
vocal leaf
hello
vestal dirge
When I'm writing color in a promt, for eks: white and pink Spiderman outfit, the hair also changes color. is there a way too fix this without prompting in that the character is supose too have (black hair) for eks? too leave the hair too somewhat random other then it gets pink?
honest spear
When I'm writing color in a promt, for eks: white and pink Spiderman outfit, the...
You can use masks that take different prompts in comfyui and this usually mitigate it quite a lot, there is also a cutoff extension that does something similar. For a1111 I don't know it's a lot that I don't use it
crude notch
2B (and maybe other smaller models) generate at 512px though and they will be ge...
ehhh, ive seen 2b do 1024 native
prime trail
i like cioccia
regal snow
When I'm writing color in a promt, for eks: white and pink Spiderman outfit, the...
Here’s two thoughts: 1. You could use dynamic prompting extension to choose a random hair color for the prompt. 2. Or You could just inpaint the hair whatever color you want after
autumn frost
hi, sorry to ask again, i am still looking for any app that use SD can run on tablet, as i don‘t have a computer. thanks
fallow garden
hi, sorry to ask again, i am still looking for any app that use SD can run on ta...
You can use thinkdiffusion, it has a1111 and comfyui already setup. Pay as you go, they can’t really make it anymore enduser friendly then that
Runs on any browser
autumn frost
You can use thinkdiffusion, it has a1111 and comfyui already setup. Pay as you g...
thank you
fallow garden
Youre welcome
cinder cloak
Hi everyone, I had a question. before we could create the images in the discord it's always possible?
charred mesa
ehhh, ive seen 2b do 1024 native
yeah, it's just that mcmonkey4eva said that it requires more fine tuning
untold latch
Where did the ❤️ UI button in civitai went? it seems as it was replaced with a cloud with a lock "vault"
wheat hazel
shared an image somewhere based on an old sketch. Dude (self proclaimed AI expert) did not believe me when I explained that it's not img2img.
wet olive
True
untold latch
steps:
- 3 dots next to the message
- "more"
- "report message"
- "Spam"
floral nimbus
when it comes to training loras, is it best to use kohya ss or a website like civiai
floral nimbus
steps:
1. 3 dots next to the message
2. "more"
3. "report message"
4. "Spam"
well said
warm junco
steps:
1. 3 dots next to the message
2. "more"
3. "report message"
4. "Spam"
Yes and you can react with a ⚠️ to the message so it gets send to our report bot
still glacier
"Report message" will report it to discord itself, and ⚠️ like cs1o will report it to our bot. So if you want us to see something and take action fast ⚠️ is probably the best course of action. And if it's really something outrageous that must be acted upon quickly, maybe ping some moderator (or community guide if there's no mod online). Just don't spam them ok :p.
shell tendon
"Report message" will report it to discord itself, and ⚠️ like cs1o will report ...
what's the definition of "outrageous that must be acted upon quickly"? a while ago i pinged mods cuz some pretty nasty hate speech was getting posted on here, got told not to do that
still glacier
what's the definition of "outrageous that must be acted upon quickly"? a while a...
True, the definition probably from member to member....
Let's just say that it should be the last resort.
it could also because they got pinged 5 times in under a minute about an incident they were already handling in the backscene.
shell tendon
to me, "very urgent" is hate speech, spamming porn/violent images in a manner that it makes it clear the intent is to antagonize, doxxing ppl, etc
spam wouldn't be on that list
yeah, true
but anything where someone on here could feel targeted and attacked in a serious way
prisma aspen
👍
karmic cedar
SD3 who: old team
SD3 what: a stable diffuser
SD3 where: the internet
SD3 why: because anime boobs apparently
SD3 when: …
trail lion
🧐
trail lion
why do we need video period