#💬|general-chat
1 messages · Page 149 of 1
They could do the same, but in a different way, like just adding their support pages or donation pages
yeahd definitely, there are like an infinite amount of comfyui experiments to try now
then you could do that for many images... And label the images... And do some clustering/classification on the latent, and finally realize you most likely have wasted some hours of your life
I added the hahaha thinking that you knew that it wasn't free, laughing with you, not to you, understood?
not the same as Fortnite, you don't need a license to monetize a Fortnite video, for example
i'm not going to argue this lol. it came across sulky and the model is free. if you meant it another way, be better at communicating.
I'mnot going to appologize for getting a sulky vibe from that reply.
I'm kinda more interested in something like "would it be possible to adapt cogvlm2 to 1.5 or XL clip?"
but you do act like you are and you try to make others look bad
others make themselves look the way they do. i just point it out.
Personally, as someone who doesn't care about support for nudity or nsfw content & is purely using these models to build creative tools.... I've been testing side by side outputs with SD3 and SDXL and finding SD3 to often produce worse results than SDXL. So that's really the main issue
But when SD3 produces better results, they are sometimes way better than SDXL. So I'm hoping there is a lot of low hanging fruit to resolve this. Unclear
We'll also have to see whether SD3 is better than SDXL at parameter efficient finetunes
Isn't it also smaller than sdxl
both data and params wise
It's somewhat smaller, but using what was framed as a far superior architecture and sampling method
my thoguhts on being smart can be found here https://www.youtube.com/shorts/iV3lnLE4-nI
It's still closer in size to SDXL than SD though. 2B is still a very large model. I think the issue is more related to their training data filtering methods
If I had to guess
pure speculation though
Yeah, it's mostly because I am dumb and want to build cool stuff. I am in deep red.
SD3 also don't seem to handle many sampling methods
It only works on non ancestral samplers, and sadly those tend to be the least exciting samplers to start with
My main language isn't english, so sorry about that, if you weren't acting high and making others look bad (because I was reading the chat even before and you also sent some messages like that) btw if I remember correctly, you were also acting sulky even in the first message I replied
What I meant is that if you just want to create pony and release it for free to the community, but you're impaired by possible costs, there are solutions, like kickstarter, etc
i'm often disappointed and that can come across sulky. i accept that.
English not being your first language isn't really much of an excuse for anything when you write paragraphs as easily as you do.
let's stop there, I see there were a lot of misunderstandings and different points of view, now you even sent a link that I won't open 😄
dpmpp2_sde_x is ancestral?
dont sulk
hmmm, you sent a link, probably to a guy making a weird face
I dont believe it works with that one
it doesn't, I tried
I didn't sulk, I just said that I won't open it
it only works with the samplers nobody really used much like eular (non ancestral) Heun, etc
Which seems a really weird limitation
I have some experience with Heun. It can produce some interesting (realistically messy) stuff at low cfg scale
at least with 1.5
Back to SD3, its fine if they wanted to make a license for commercial use, it's the way they did it, so I hope they earn money to cover the costs and earn more than that
There are many solutions, they all have tradeoff, so far XL tradeoffs seems better.
gotta go. I have 2 episodes of star trek discovery s5 to watch... wish me luck... lol. Not sure I'm not gonna fall asleep
just downloaed the wrong sd3 model for comfyui, bummer.
fair enough. I'm focusing on 1.5 myself. so.... :p
But I tried pony XL, and failed at getting what I was looking for. More on that later (maybe)
SD 1.5 it's good for scenes and backgrounds
What were you looking for? maybe I can help you, I've been using it for some months
that would be helpful for solo gamedev people to protoype there game asets.
yeah, I think some already used it for that
I mean they can all produce interesting as a rule, but overall the ancestral samplers are just much more heavily used. You just have to scan the civit AI models very few recommend using non ancestral
I know, and if i ever get into gamedev i might want to aswell.
For SDXL and Pony models I have seen a lot that recommend using DPM++ SDE Karras
Or just the DPM++ Karras
Reading reddit and this release appears to be a disaster. What a shame
Yeah those don't work with SD3
@wheat hazel FYI, I'm trying diffusers branch for sd3 now. Not yet release by it seems to be downloading the model for the new sd3 pipeline
Thyri went to watch 2 episodes of Star Trek
The results I'm seeing from SD3 look pathetic
No, it wasn´t and isn´t, no finetunes needed here, have been working almost exclusively with XL´s base model since release.
This was supposed to be the last model we would need? Doesn't look like it....
So it seems like Stability's enterprise license requires the usage guidelines be implemented. This means services have to censor if they use sd3.
Honestly, good. After that x event where brigades of trolls flooded the service with nudes of anyone they could, mostly celebrities but also just of people replying in the thead, more needs to be done. Not government censorship. Just private parties regulating like they always would. We don't live in a hedonistic society
how lol
btw you need a license for commercial use, even if you only want to sell the output images
the 6000 image limit is a shining beacon of hope now. they should still do more
OK so that's interesting. If you set scheduler to exponetial it produces horrific outputs as base, but if you throw the steps up to 90+ it actually produces some nice outputs
lol sd3 is DOA, its absolutely horrible at anatomy if you want anything with a human in it just use sdxl
it's an architecture. this is the base model that was trained at great expense. Finetunes get made to base models.
if this was a refine of sdxl like cosxl is, you would have a point.
it isn't
Da'sulkiest O'sulk Asulker
dead or alive :3
Dead on Arrival for some
Those are good fighting games
I'm really curious what kind of prompts and parameters the people use that get good outputs. I'm trying to gen a "not absolutely horribly and deformed" looking basic Anime image (sfw of course) - and it's just not anything like the 2B images we were shown.
im currently trying all sorts of parameters to see what works best
that (sfw of course) remark makes me think the other way
love that you needed to explicitely clarify that anime was going to be sfw. We know what most users are doing.
that could all change thought going forward now that theres toon crafter workflows. People might actually start making anime with a message and purpose with SD now
Do you guys really think peope only make anime for nsfw or something?
Well, if I can get anime-style output quality anywhere close to SDXL
they have, not as good as fine-tunes yet working
Just making sure because people are apparently trying to get nsfw.
nah but why the remark? haha
It does decently on backgrounds, but messes up a lot on people. And seems totally unable to draw children.
Just weird you guys both jumped on that part of the comment, guess the rest of it has been plowed ground already though
Information should not be censored
have you seen civit? it's basically assumed at this point that people using models for anime are well... y'know.
That's just the meta right now. things will change. it doesn't speak to anyone's individual use
oh yea totally aware its basically a porn site now
oh yeah anyone trying to create legitimate expressions and art content are leaving civit . even deviantart is less degenerate these days
Leaving it for what?
I just didn't want to hear the usual "haha y u tryin to make booba?" remarks. Guess it backfired in the most streisand way possible. 😉
other content sharing communities, not model hosting communities
Yes, but even the messages will be censored. Don't say damn and expect not to get 4 asterisks. In the image.
🙂
I think that's because they don't want the model to be used to draw children
Which pushes me to OpenAI instead, I guess.
it can do children with proper latent contexts. try norman rockwell
Doesn't work. I need it to look like anime.
I see
SD3 doesn't seem to do accretion disks either. ...I realise that's a niche interest, but XL succeeded. :p
álgum BR aqui ??
I think for some who just want to make images of a happy family with children it's bad that SD3 can't draw them
Ah, well. The children in this story are, in general, ... not happy.
Adventures are someone else having a terrible time far away. In this case beyond the fourth wall.
But as it stands... DALL-E doesn't mind, and GPT-4o looks promising for whenever they unlock that. So I'm not too bothered that SD3 can't.
Good news. Seconds ago on diffusers main
Add Stable Diffusion 3 (#8483)
Is SD3 SD2 all over again? I wonder if people will bother to train on it. Someone on reddit posted a prompt of "person walking towards the camera". All four pics had the person with their back turned
amd windows users if you need help: https://youtu.be/VOg8XFR47p4?si=MnR7q-BVhl0eJDYZ
could probably divide it into two simple parts - "walking", "facing viewer"
Yikes
idk, biggest problem I think is non-commercial use only, can't do much with it...
Not yet. It's a start, but it needs to be finetuned badly.
Better than SD1.2
Successfully installed diffusers-0.29.0
Bingo!
I'm waiting AAM make with SD3
I need it
AAM?
I'm looking forward to good upscaling.
Have you seen the licensing on this thing? Wonder what fine-tuners will take up on the job.
Anime Mix XL
What does that mean exactly? What's the difference? Did they include training scripts already?
The guy who did PonyXL was here earlier inquiring staff about Enterprise license... They basically told him to go where sun doesn't shine.
The 'pro' license limits you to 6,000 pictures per month... it's easy to exceed that just with throwaways and regularisation images
This has nothing to do with training. It is all about being able to create one's own pipelines with a simply python program using diffusers
you misunderstand. that's if a service is serving up 6000 images
Sad, so now i'll wait for Hunyan-Dit
And my discord bot has done about 2x more than that just handing pictures to whoever asks
Have been dabbling around with Stable Cascade for a couple of weeks and along with the Invictus Redmond fine-tune checkpoint it´s actually quite pleasant (already without) and way more what I would have SD3 thought to be (so far). The training set isn´t that large yet the overall quality and artistic output are very good, superior to what SD 1 .5 and SDXL can do, especially sharpness, shape accuracy and very much so dynamic colour range-wise including nuances in bright and dark lighting environments enabling Old Master colour contrast plays.
if you're charging for use of that bot, you may need to get a different license from stability. contact them. the $20 a month isn't intended for you
oh wait. 6000/m i misundertsood. it applies to anyone making money with stable diffusion.
After what happened to the ponyxl guy earlier though, I'm not going to bother.
"If you are above any of these thresholds, please contact us to discuss an Enterprise License."
I first thought the limit was just for services not any commercial use of it
hey guys does anyone know any really good virtual try on? right now i am using IDM VTON but the prints are getting f'ed up. I also want to transfer the backside of the shirt, is it even possible currently?
so if a game developer is doing over 6000 images a month , since they intend to publish the game commercially, would have to have an enterprise license
The website pairs that (the 6000 limit) with "Ideal for developers with less than 1M active users"
Which is just hilarious.
I don't think so... I think those 6000 are for serving images commercially... Gamedev resources would probably fall under yearly revenue caps.
if you aren't using it commercially, i don't think the 6000 limit applies to you. your bot seems to fall under free license. i could be wrong but non commercial use is non commercial
It's non-commercial at the moment.
pretty sure stable horde will have SD3 on it
The thing is, I was planning to add a paid tier for access to some features powered by GPT-4o... once that API unlocks.
since you're over 6000 images a month then, going commercial would require you to negotiate an enterprise license with stability's sales people
I can imagine how that'll go.
"Hi, yes, I'd like an enterprise license. My revenue is about $10."
if your revenue from charging people for access is only $10, don't charge access. just eat the cost
It might be ten times that. Who knows.
you'll have ot do some business modelling. figure out how many users a month will pay you and predict revenue from that for negotiating with their agents
While the base model is pretty bad anatomy wise, its really good with everything else and textures as well. Id say this is as bad as 2.0, however in that situation people didnt bother because 1.5 was still so good. So nobody put extensive effort into finetuning. We will see if the effort is put into SD3. Hopefully so.
It's less 'charging' and more 'passing on costs', so..
you can't just go into these licensing arrangements with "mights" . They charge you based on how succesful other generation services are
then you've got no problems then
For now I'm implementing a usage limit.
Sucks that the API model is most definately uncensored and generates people just fine.
They just censored the hell out of the Local model.
Oh, the enterprise license form also requires you to own a company. Welp. :p
if it's non commercial, you don't need the usage limit as i understand it
Look, the thing is that almost nothing counts as non-commercial
It's an extremely restrictive term
you can do business as yourself. if you intend to collect revenue from discord, i recommend setting up a business to colelct that revenue
financial literacy is a great idea. oftne local libraries will offer free courses. there are good ones to pay for too. having a business is easy and it's smart to run revenue throuhg one
prompting mostly honestly, though through IPAdapter+ and input images as well 🙂
Hi, I'm trying to figure out this whole commercial licensing. 1) Where can I read more about the 6000 images/month you are talking about? 2) Is it about using SD3 locally or via the API?
i have bad news kinda, sd3 does not run on my pc even though i have 8vram (amd). it keeps crashing my pc.
This is good
why so?
you should be able to run it if you dont use the t5 text encoder
the computer tells you to wait for finetunes
bummer?
how do i even run pixart because im trying comfyui but it keeps getting errors.
pixart isnt a good recommendation honestly, sdxl finetunes are far superior
Does SD3 run in AMD?
Yes but for now only with comfyui
Wait so they only released a limited version then or the version was only made for comfyui?
have you managed to find prompts that increase text coherence ?
ComfyUI is made by SD Staff, so there's no surprise it got 0-day support.
hey guys, i tried to install stable diffusion but i'm getting over and over the same mistake, RPC failed, fetch-pack. fatal Early EOF, invalid index pack output
Ok odd they wouldn't make it for most of different formats?
It kinda is at this day is age? Thats like relasing a game on only certian versions of ps5s but then again it's free.
error: RPC failed; curl 92 HTTP/2 stream 5 was not closed cleanly: CANCEL (err 8)
error: 5177 bytes of body are still expected
fetch-pack: unexpected disconnect while reading sideband packet
fatal: early EOF
fatal: fetch-pack: invalid index-pack output
Format is same for everyone, it's just other platforms will have to implement it on their own like they did with all the other diffusion architectures.
Please come to #🤝|tech-support
SD3 will come to Auto1111, but we have to wait
sure love
What about forge?
anyone know how I can easily toggle on and off sections of a workflow? (I want to deactivate upscaling until I have found a generation I like and then activate upscaling again)
Forge is deprecated for normal Users
right click on group, bypass group nodes?
what is the point of the multi prompt workflow
Hey yall. Is Comfy UI safe to download? Someone posted on Reddit last week that one of the executables contained malware
It is safe as far as we know
I use it
oh wait...my system.,...its going...
wait..don't...
ah! seriously its ok
yea
The original Comfy UI files aren't malware, the comfyui_llmvision node by applebotz had malware in it
What do we need to download to use it?
To use SD3 ?
You would need to use Comfyui or wait until Automatic1111 gets an update.
Ah ok
you can also use stableswarm, which has a more easier to use ui but is pretty much the same as comfy
SD3 can change a life!!!!
Thats correct I should mention it more often
I'm working on a science-fantasy novel, and I've been thinking about either making illustrations or a comic book. So I suppose generally speaking, I'm not really looking for anything particularly fancy. But my usual "benchmark" prompt is for a scene where two hunters from an antique/mythological/pre-medieval era find the corpse of a giant bear lying in the snow, in the middle of a grove. Nearby, there's a strange woman hidden in a bush, and there's a track of blood and chaos between them.
Yes 🤗
I HAVE A CAT!!! Finally!!!
your kitten me
SD3 finally works with diffusers!
i think this solves the resolution issue https://github.com/GavChap/ComfyUI-SD3LatentSelectRes
comfy is really good
thanks
so, I spent 10 hours with this and I don’t want to use it anymore until finetunes appear
this absolutely does not obey the prompt
it can be made to make good images, mostly not related to people, but it doesn't care at all about your prompt
It got the "Hello World" text perfect
True. Tried one full prompt that worked good with SDXL finetuned model and it happened to mistakenly generate "underwear" instead of full body suit.
comfy has one massive issue: its execution model
wat dat means
yes
I wish I didn't get floating point exception so often. Damn my old cpu i9-13900k
comfy uses a backward execution model. Basically, it starts by identifying the output nodes, and for each output node, it traverses the graph backward: it gets the nodes connected to the output nodes, and so on, until it can execute the workflow.
"anatomy has been improved in SD3" 
i turn nodes off, but yeah i hear you. its similar to unreal engine nodes. i like it its easy to debug, but this isnt that
https://i.imgur.com/yDXA82i.png seems like 1gp is the limit
i'm starting to wonder if sd3 will turn out like if's release where after like 9 months the stable diffusion discord server will close, stability ai will file for chapter 11 bankruptcy, and we'll never see a commercial release
All girl are asian by standard. "Girl" - asian girls, "Instagram model" - latin and tanned girls, "beautiful woman" - partly european girls.
yea, but it's worse than unreal engine blueprints. UE blueprints have a forward execution model.
Comfy being a backward execution model you can't do basic stuff like: lazy evaluation, loops, etc.
basically, comfy is not even turing complete, in fact
and you can’t regulate this in any way with a negative prompt or anything else, just use the right trigger words
do you think anatomy got improved and reviewed by people who, unlike stellar blade developers, know what a woman looks like?
nope. Never at all 🤣. What's the problem with stellar blade ? (I didn't play it yet)
basically, there were people who got angry that the main protagonist of the game was a hot woman.

oh yes, and if the ai doesn’t like something, it will replace your girl with an asian one and it won’t care what you write
Well I have a counter example but it is grotesque. and there is no kill-it-with-fire thread here I can post it tool
only one?
So far...
HA!!! It did what looks like nude and I didn't even ask for that.
SD3 is like 9GB of data to make it work. And you can obtain similar or better results with SD1.5, which is only 2GB, and some tricks
that's my feeling right now. Maybe I'm wrong
Ok. Well this was a big waste of much of my time today. I get better looking people at 15+ fps 1024x1024 with my rt video pipeline. sd3 does give good clothing detail and outdoor scenery but ehhh.
you guys are comparing base models to fine tunes
you won't be able to get such realistic photos with SD 1.5 and SDXL
SD3 base. Not just any base.
I don't care. I'm comparing models from June 2024.
Perhaps with the base that MIGHT be true
and any finetunes look less realistic than SD3
realistic... you mean... mangled?
details, skin, lighting and everything everything everything
SD3 indeed "looks" more realistic. But that's all.
where can I download that SD3 model? I obviously don't have the same
it looks like a retouched photo
it's base 1.5
that's what I'm talking about
Did you got normal one or the one with integrated clips ? And does that make a "big" difference ?
Do you want to do this with diffusers or the other heavy weight ui?
I downloaded all the repo
you got a link to a realistic SD3 image?
Did sdxl generate as much deformed body horror as I am seeing at 28 steps?
Guys whover isn't getting good results with your prompts including myself I expected a bit more 😦 , I suggest to take a look at this post by stabiliy https://stability.ai/stable-assistant-gallery?utm_campaign=Stable Assistant&utm_content=186946249&utm_medium=social&utm_source=twitter&hss_channel=tw-1281048162602369024
nope. SD3 Actual state is far worse than sdxl for anything other that realisticness
think with your head and you will understand what kind of shit this is
I am only doing realistic right now.
all images are either a close portrait or hidden arms and legs
By realistic, I only mean the look in terms of color, lighting etc...
Because in terms of anatomy, faces, peoples etc... it's worse
time to move into pixart alpha
I heard about it. What does it add to sdxl ?
I'll try some 'pro' prompts.
It's just a diffusion transformer model similar to SD3.
try if you want. I did it from a prompt that was perfect in sdxl and it performed so bad with SD3
I was trying to cut sd3 some slack and assume it was my prompt skills or the lack thereof. 🙂
I been trying these and they are working so far, then modify with something you want, another thing I saw in the paper research is that the "t5xxl_ text encoder" ,If you don't you use It you will get a better performance obviously sacrificing some quality with the texts in the image but is not a game changer to be honest, I'm currtently using "sd3_medium_incl_clips.safetensors" so If interested try It yourselfs
That's hard to make with only prompts, you need to inpaint a lot with a high resolution image and maybe some loras to achieve the science-fantasy style that you want
How do we try the alternate encoders?
As you know we have 3 encoders files right? one of those has an 8fp and 16fp versions the "t5xxl_text" , well If you don't use those which their weight is 4 and 9 gb you won't se a game changer downgrade in image quality as far as I have seen, so you just use the "sd3_medium_incl_clips.safetensors" which has the two little text encoders files baked and thats It
also you could use a depth or canny controlnet for the giant bear, then inpaint the rest and details
I would share the images I'm getting but I don't have the option to send images

last chance to "save" the company
it's god awful and worse than even XL or Cascade
Ah yes... safety!
I'm using the 10gb model and all the 3 text enconders, the upscaling example workflow needed them, first I removed the TripleCLIPLoader node and just used the 10GB model, after that, I used the 3 clip text encoders to regen the image, same seed, same prompts and I liked the one that didn't had the TripleCLIPLoader node, the pupils and some details looked better
doh! SD3 doesn't know who Maetel is! completely unusable! :p
everything is much worse than you think
I remember people saying the same thing when SDXL came out "SD 1.5 with all the finetunes deliver way better results" now we have arguably better SDXL finetunes... I expect the same thing will happen in a few months with SD3
So I think for me it's better to just not use the TripleCLIPLoader and instead just use the sd3_medium_incl_clips_t5xxlfp8.safetensors (10.9 GB) model
SDXL didn't have a completely lobotomized base model. The issue isn't the finetunes the issue is the model lacking knowledge and needing half an essay to produce a good image
public model doesn't want to obey at all
Hahahaha that's true
Also I forgot to mention the tripleCLIPLoader takes AGES to load
0.9 wasn't great, 1.0 was usable ish but not for nsfw
NSFW isn't even the issue it's the fact it
- struggles with BASIC anatomy.
- doesn't know certain concepts at all.
- takes comical amounts of VRAM to run in comparison to the quality
Took me 101 secs to get an upscaled image with the default settings, but the base model was already loaded
without the TripleTEXTLoader, took me 108 secs to load for the first time the model
I tried again, after the model is loaded and without the TripleCLIPLoader, now it only took me 80 secs
So after testing a bit, I think the results are better without the TripleCLIPLoader and only the biggest 10.9 GB model
that workflow was made for the ones that didn't download the 10.9 GB model
I'm gonna let the SD3 model sleep for a long time, until we start seeing the modified versions, in the meantime before I leave I recommend you to try the Mobius XL model, It is for me the most capable XL model right now produces images you havent seen in other models
Which is why no one is gonna put in the effort, yeah
the upscaling workflow
this one? https://civitai.com/models/490622/mobius
everyone did either XL or 1.5 
what workflow(? I downloaded all from the huggingface
The upscaling one
Yep as you see in the images below do not make justice to the model neither the download count If you want I can send you some of the ones I made
Yes send me some, please
How can the model, after so many years, not understand basic human anatomy? Very strange....
It does
Then why am I seeing horrific chernobyl mutations on reddit?
Guys, how to use stable-diffusion-3-medium for free?
The AI is not the problem
After all, people write all computer programs
I am sure a version of SD 3 exists that can draw a human with perfection
Older technology can do that
what are you talking about bro? how to use table-diffusion-3-medium for free?
oh sorry!
The sdxl base was better than the old sd1.5 base. But in general I never found a way to get non-blurry backgrounds with the original sdxl.
I would have assume that sd3 would have learn from past mistake and had to make "less" excuses for poor body shape and other attrocities.
"Man, SD3 Released! Can't wait to see what's changed when I wake up tommorow!"
wakes up, checks Reddit/Forums:

Man, I realize the need for a basic censorship filter and all, but this base model is completely r-word in terms of anatomy, human physiology, and I cannot generate any images with humans in them being proper except like 1 out 15 times... my prompts are simple; just why? Who exactly thought that removing all knowledge of human bodies from the training dataset would have no consequences?
Has the flaming of it started? I need to join the party. 🙂
I have seen some write ups in the Stable Diffusion Reddit on how to start with SD 3
ARE YOU ALL HERE TALKING LIKE AI BOTS. ARE YOU REALLY LIVING PEOPLE?
no
i llm3
Na, I'm Actually a Kansen
Hello, it seems you are wondering if we are really living people. Of course we are!
ENOUGH I AM ALREADY CONFUSED BECAUSE OF YOU. EXPLAIN CORRECTLY HOW TO USE stable-diffusion-3-medium
Go on Reddit, the subreddit's already on fire
NOTHING IS CLEAR IN GOOGLE SOME GITHUB AND THE SITES ARE NOT UNDERSTANDABLE. I WANT IT TO BE FREE AND I COULD JUST ENTER A REQUEST AND THAT'S ALL AND NOT SUFFER WITH THIS CRAP
We know your weather sucks but don't take it out on us!
Have you tried the Reddit?
Sorry guys, I just really would like to know how to use this) So I’m a good Russian
pay money for mj
Their weather is actually great at this time of the year, was at Russia last year.
Wow! This look fun. sdxl didn't have this much hate. I know because I was one of the few that initially complained a lot about the blurry backgrounds for sdxl.
Do you have friends from Russia?ъ
Yes, I lived there for a while.
oh my bad
Да but not much tbh
Is it true? And in which city?
The 1.5 model was uncensored. The world hasn't ended. No one cares. There are far worse issues in the real world than an uncensored model.
I never actually had any issues with SDXL. all my use cases turned out fine. SD3 however, this is a flaming garbage pile. wonder if the butchered human generation would be good for like fallout 1 Centaurs?
О ты знаешь Русский?
Pushkin
hiiii
Hi everyone... I am wondering if there's anyone who would be willing to help me figure out an issue I am having opening stable diffusion
great to jump in
sankt petersburg?
We should probably censor the entire internet while we are at it. Maybe have a few big corporations decide what sites we can and can't visit.
Surely that will go well.
I dont know how to tell you this.........
Yes. I have a good friend who lives there. I stayed with him for a while. And in return he stayed with me in Florida!
cough.."good friend"
Do you know Russian? Can you write some words in Russian?
Ya znaio necleshwkom, no ya na mago picat no roskai keyboard aaaa-(lmfao)
Gay?
oh so you know Russian well
just a bit, i make very obvious mistakes))
Ты понимаешь меня?
Привет!
da
о
good ai bro to speak in russian
pizda?
Have you been in Russia?
Yes, lived there for 2 years
Saint Petersburg
You like it? And why did you decide to leave?
Damn Twitch, guess you gonna put that install on the backburner
I used to work there, but then formed a business in the US, and suddenly war happened so couldn't use local banks lol
Saint Petersburg: ❌
Leningrad: ✅
Shit, you're cool! Which side are you on, Russia or Ukraine?
Ohhh
Nice!
What state are you in now?
My ol' Pappy used to tell me, "Be on the side that has Mcdonalds after the fightin is done"
Sorry, cannot give out such information))
lmao
😢
Vkusno i tochka
from russia
Вкусно и точка!
The McDonald's restaurant chain in Russia has been renamed "Tasty and full stop" McDonald's in Russia has decided on a new name. Now the chain of fast food restaurants will be called “Tasty and full stop”
Just talked to someone. And they just dont like AI especially AI generated images, You know, i jus dont gi it. Its still so strange to me that people feel that way.
when McDonald's left Russia
Most people do not. We are in the minority at the moment. And many people REALLY don't like it.
yes, russia simply scammed people by taking away their business
And after the end of the war, won’t you come to Russia again?
In what way did you cheat? McDonald's themselves left Russia, they released such a chain only by changing the name
if banks open again, actually i might!
Hurray, I'm waiting for you!
I liked Russia a lot; really quiet but had very cool people there.
And we love you! It’s true that on TV they always like to say that America is bad and that it will soon attack Russia! But these are propagandists, ordinary peaceful people love you!
@fervent thunderI feel like its the same old, if you are of this skin color, etc. then you are not allowed to do this, etc. Youd hope humans have changed for the better, but it seems this blind hate is still strong but just in different forms.
Its a bit unusual of a comparision but its similar to me as i tihnk it comes from the same place.
Yeah, I don't trust news from either. Always gotta go there and see for yourself))
You have to realize AI is seen by many as the death of The Artist. In many ways it will be most likely, but new AI artists will most liktely prosper. Just hard to see that right now.
Don’t your news reports say the same thing, only in the opposite direction, that Russia will attack America?
Most of the business was taken away by racketeering. And then they introduced laws that gave even more opportunities to take away business.
Luckily, I don't watch news anymore :D But probably in some political magazines of course 100%
Who are you for in this war?
By the way, what do you mean by lmao and fr? I don’t understand what kind of abbreviation this is
I'm backing the War Machine business in the USA. They are currently, undefeated?
S D 3
I always saw comments on tiktok and youtube Americans constantly write FR And Lmao what is this
I always saw art as a thing you do because you love it not because you want to make money from it so that factor doesnt effect me.
Who can you be for in war, such a waste of human lives and materials.
this is not a war but a special military operation
"laughing my *ss off" "for real"
"War never changes"
аfr - for real? at what plan? I’m translating the context here through a translator like: is this true?
This is not our own war, even Putin himself is already saying this
I don't think I said "fr" at all lol; probably someone else
oh wait, just read this
ahh how terrible Google translate is, I translated your text, it gave me some kind of incoherent thing
Напиши на русском
Ахахахах
It is just an expression of "laughing", you can just read as-is, no need to translate it
By the way, who do you vote for in the elections? In America, do you generally respect Biden or not? I heard because of him you have a lot of migrants and even the Americans themselves don’t like him.
Okay, what does "fr" mean?
Im going to look for a translate plugin just so i can read russian on the go. Hopefully such a plugin exist
fr = for realz. i tihnk
google translate
B... bro
Hurray come on!
fr stands for "for real," which is basically slang for "I agree"
@soft ironI mean with just one click. its too much clikcing to go to google translate
ahh thanks!
I thought it meant France, your version works better
Oh, are you Russian? Your nickname is simply Vk (Social network VKontakte from Russia)
he doesn't write anything useful or dangerous, don't worry
"for real?" can also mean "really?" but it's more likely "I agree"
Hahaha no it's my dog's initials
@soft ironlol
Oh Ahahahaha
Do you know streamer Dizzy Dizaster?
Twitch
my girlfriend can't be replaced
Is your girlfriend Russian?
Azerbaijan
Oh it's close to us! And how does she really know Russian well?
Ah Azerbajijan, where they "poke badgers with spoons" according to some british guy
I DONT trust americans as an american myself. Especially their news or government.
Is it true? What if it’s no secret what your news says about Russia? Give examples, it’s just that absolutely all the channels on TV hate America and threaten to destroy you under the pretext that you will attack us
Never trust people. They are the only ones who can screw you over.
@fervent thunderoh i meant to include Don't. But forgot to type it lol.
you can't paint your emotions with ai
bro&
?
I was a photojournalist when Photoshop was released. It was the Devil I tell ya! I actually liked PS, but not digital cameras 🤣
almost everyone speaks russian
I'm talking about her work, she is an artist
bROO
@teal pagoda
Is it true? What if it’s no secret what your news says about Russia? Give examples, it’s just that absolutely all the channels on TV hate America and threaten to destroy you under the pretext that you will attack us
🙃

Fine 🙂
Putin?
Putin is a reasonable guy isnt he.
Do you like Putin?
As long as America has all the technologies in the world, nothing threatens America
@fervent thunderAmerican policitians are just as bad if not infinitely worst.
yup
What country are you from?
I didn’t understand from the translation, do you like Putin or not?
We have lost a lot of our credibility. The truth is some of it was undeserved.
Well, at least tell me, from a European country?
America has invaded, infiltrated, spied on, overthrown, nuked, bombed, killed more people and more countries than any other country, for the worst of reason and lies. So the US cannot speak or fear monger about any other country or poeple. Constantly fear mongering about China and Russia when America is the one with the worst track record on anything. Its beyond a joke for the US to try to lecture anyone or anytihng on anything. I say this as an american. But i will end it there with 'politics',
How do you feel about everything that is happening in Russia and Ukraine? And do you think America and Russia are friends?
I think it's a proxy war between the USA and Russia
nothing new to see
Do you support these US invasions? And do you support the same invasion from Russia?
Well, tell me why he attacked Ukraine?
Have you heard anything about Navalny?
Well, okay, in a nutshell?
damn, general chat going on and on about politics
Да бро! Русские здесь!
POLITICS
I think everyone should be banned
We got yo ticks in here!
NO
get u some!
ok buddyideologist
Its different because Ukraine is largely a part of Russia, and Russians live in Ukrain, and Ukrain were killing Russians since at least 2014. So Putin invading Ukraine is different from the US invading foreign countries for money, oil, etc. So Putin is protecting his people in Ukrain, while the US is stealing resource and killing people for money and greed. So its very differnt comparing Russia and the US
I have a lot of fun chatting with you from Russia! I like how you love Russia, you Americans are cool!
Im not saying its okay for Russia, but its a better reason to 'invade; then the US.
regardless of whatever this or that in terms of culture or politics or whatever I am a very simple man, i interact and I see people, and I only judge on their individual actions, too hard to put the weight of their country as the whole of their identity and personality
Luckily we can say that here.
I think everyday people have more in common than we have differences, it's our governments that are the problem, they're usually made up of people who have no clue what it's like to live as a normal person
Like, if someone is a jerk, they are a jerk, nothing too complex
the country needs to be banned too
You are a worthy person!
I guess sd3 has rigor mortis setting in. Perhaps they'll do a post motum.
People always had a fuckton of things in common, but get too enamored with the narrative of conflict and differences, and are like "My people is special, I am special, this that, us vs then"
funny that it became 10x worse after current social media
Yeah I fully believe social media has made everything so much worse
this is human nature and there are researches about this
Who here will bow down to the AI overloards when they become capable enough. Although i like AI, i wont.
But it doesn't mean that it is the case absolutely always and for every single one, it is definitely very possible to stay away from that mode of thinking and just treat people on a case by case basis, and judge their worth on their own actions and consequences, not that of their people or their group or whatever
You guys are all from America! Do you really believe about Russian propaganda? It’s crazy, I didn’t think that people from America generally love Russians, so much so as to support Putin! You are cool!
you are a big positivist 

Putin is far more intelligent than any US president we have had in the past 20 years or so. He rarely ever needs to read from a teleprompter - at least he seems to be more intelligent.
What about Donald Trump?*
but he is just as much a fraud as other politicians
Show me the pictures of leveled cities.
I dont trust Putin either but i respect him more. He walks like he is a badass lol - its a joke.@soft iron
God there is so much hate towards anybody saying anything that's not 100% positive about SD3 in the #🆕|sd3 chat
Don't talk about politics :)))
I love discussing politics with completely different people! Damn you are from America, this is tough, this is a different world, every second child from Russia would like to come to you in the USA
Sure buddy secret american labs in ukraine, because they don't have 50 better developed countries do do that in
Including their own
kek
What are those images gen'd with?
Well, let's discuss this!
Ah so it's compared to 8B?
yes
complete lack of following the prompt
From comparisons I've seen
Americans, is it true that you have Spider-Man flying in New York?
SD3 2B follows prompt WAY better than SDXL or anything similar
Bruh how's that related
I'm just from Russia:)
can sd3 run on a 6GB vram thats the question
The magic sauce is the MMDiT block's huge amount of attentions, densely well made blocks,
and the wonderful alignment within T5
Волшебный соус*
?
Yes, tho it's a narrow margin
I got it to use a little bit over 4GB VRAM with lowvram fp8 on ComfyUI
ye, Волшебный соус
I'm just hoping for controlnet because controlnet on SDXL is... uh..
Соус для еды?
Mistoline and Openpose are good, others indeed succ
not for food, for "the secret thing" or like "the key change"
or the "why it works" and the "why it's better"
iv been downloading a SDXL model for the past 3 days just to test if it even works on my 6gb vram gpu. I have 1mb internet speed.
А, мы просто в России не говорим с таким контекстом это слово
I haven't really found anything that works as well as on 1.5, but to be fair I haven't really looked in the past few weeks
I see,
English is also not my native language, so I get your struggle my friend
I have 02:53)
It is 20:53 in here
А какой твой родной бро?\
My native is Portuguese
Португалия это тоже очень классная страна
Это Европа?
can sd3 do "a cow sitting on a frog", by default?
Ah, I am from Brazil, we speak "Brazillian Portuguese", I am from South America
Жесть! Так вы всё равно из Америки! Очень круто! Я бы хотел побывать в Америке, но из за Визы это очень сложно (
Likewise, I would like to visit Europe but the process of getting a visa is complicated
А Россию хотел бы посетить?
Definitely, I have it in my plans to visit Russia in the future
Урааа! Возможно мы даже и увидимся
Indeed, when the day happens I will send you message about my arrival
Хорошо! Будем разговаривать друг с другом через google translate Ахахахах
hahaha true!
no, it's impossible for it
meta was amazing at it. but i tried sd 1.5 and none worked.
someone was able to get a cow sitting on a frog in sd 1.5 but i think they did a bunch of tricks
hi guys is it possible to create image on discord privately instead of public channel? like u can do with midjourney
Yes, you can run Stable Diffusion on your own computer
Oh the SD3 new version too? where I can download it?
Check out #📣|announcements
Thank you Midare
Did sd3 end up being a very good model to work on or a disappointment?
without finetune it's still useless
😭
gm gm
请教各位前辈,谁能告诉我如何不部署使用stable diffusion软件?
I would like to ask the seniors, who can tell me how to not deploy and use the stable diffusion software?
Yes, and Putin is a saint.
Do you love Putin?
Of course. He is the world's hero.
I agree
Of course, you do. You know what would happen if you didn't.
hey
What would have happened? Explain? Putin is actually the best
)))
have fun getting drafted into the "special military operation"
so you can be used as cannonfodder
in the meat grinder
Yes, calm down) Toxic, I'm kidding
Just read some of your newer laws about what you can say and not say.
So am i.
😁
We know we know))
Luckily Russia hasn't banned porn.
And if it's not a secret, how do you know this? Are you from America?
@fervent thunderSome people were supposedly able to do a cow sitting on a frog with sd3
Yes, but we hear the news from Russia.
They actually show the actual news shows from Russia here.
It's just that some Americans wrote in the chat before that Putin did everything right and the news in America is terrible! Do you have News in America about how Russia hates you and wants to destroy you? It's just that they say on TV that the Americans are enemies
We see it all. Russia isn't as bad as China. Everytime CNN starts reporting on something from China the local CNN channel in China gets blocked. When the story is over it comes back on.
This discussion is the definition of off-topic
The US is constantly saying evil things about Russia, but especially about China.
Yes. I miss the sd3 hate. 🙂
And which side are you on, Ukraine or Russia? And in general, how do you feel about people who support Putin? And would you like to be friends with Russians as an American?
I would have no problem with being a friend of someone from any country in the world. But we've seen what actually happened in Ukraine and the degree of destruction. We have heard Russian threatening using nukes. And so much more.
When you hear the words from Putin's lips it isn't propoganda
pootin
on the ritz
is wide
Well, you have to understand that not all people from Russia are subject to this propaganda) For example, I am absolutely adequate and I am for peace
Obviously that'd be better
Would you like to visit Russia? How will this whole war end?
The same apply to people from China. Many can access the interent through firewalls. The question to ask is why people in these kinds of countries need to use proxies.
And is it also true that you have this: we in Russia just had a popular meme recently like: a typical party for Americans (And they start saying typical phrases from movies like: I came to a party with Stacy's girlfriend and this nerd didn't finish his broccoli and he wasn't allowed to go to Stacy's party) Is it really the same for you? Or do your films create stereotypes about it?
I've visited China 5 times. With Russia I'm not sure if I'd be able to go home if I said the wrong things.
yes, you will be killed
What on earth are you talking about. broccoli? Stereotypes about broccoli????
Don't worry, you can just not say anything ahahaha
Not sure if it would be that bad.
Not just about broccoli, but about behavior in general, is it true? I mean, how it's presented like this: BABY STACY'S BIRTHDAY IS TODAY AND SHE DIDN'T INVITE A NERD THERE BECAUSE HE EATS BROCCOLI, BUT her favorite James will be there. I mean, do you really have that?
I'm not sure what you are talking about. We have all kinds of movies. Some are comedies which will joke about all kinds of things. Politicians, celebrities, culture anywhere including America, ...
As in various American films and TV series))
Yeah, the problem with visiting countries that have extremely tense relationships with your home country is that if something goes wrong, nobody is going to help you, and you don't necessarily need to have done anything wrong, there might be a misunderstanding, or you lose your passport etc...
It's unlikely but it is something to consider
Yep, and as tensions get worse the danger increases.
China has a huge tourism business. I've always felt safe there becuase if a local were to harm a tourist they get in serious trouble. That doesn't mean they might not try to sell you a fake watch. 🙂
Yes and being unable to use an embassy is a very strong consideration when traveling internationally
Right
Why don't you guys get all worked up! If you are in Russia, write, we will meet) I will show you all the beauty of Russia! Just don't say anything too much or you'll end up in jail. 😁😁
yes, they will simply kill you in Russia
We don't get all the worked up becuase we know a troll when we see one. 🙂
no you'll appologize to the sate and confess your love for father, then you'll fall out of a building
Come on, what kind of troll am I) I'm just kidding.
you're wrong, this is what happens in Chechnya
My brother went to Russia on a tour group from a cruise ship St. Petersburg I believe. He said he does remember seeing "guys" carefully monitorying the group from the outside.
https://tenor.com/view/muajaja-the-simpsons-gif-7251407 but it says chechnya instead
Perhaps that is what I meant. You are doing such a great job about joking about things I have no clue how insane you are or are NOT. 🙂
Oh cool, I also live in St. Petersburg, and it's not surprising that they are being watched, I would also follow) it's interesting, an AMERICAN came to Russia is a whole event!
Suprising. In China you have to visit some small village for you to get strange looks. In a major city like ShangHai it is all high class folks.
yesterday was a bit messy
The verdict on Trump, Biden's son, or sd3? 🙂
Well, imagine that for Russia, when an American comes to it, it's a whole event for Russia! And we are very glad to see you here) It just seems so unusual for us, I want to ask questions and see the reaction of Americans
Americans wouldn't give a crap if a Russian came to america. unless it was Putin himself.
Well, you compared it)) We just have this mentality, for us Americans it's just like other people from another planet) That's where the interest comes from
sd3 only for comfy right now?
Why are you silent?
i see
Ты бабка гренки?
SD3 is out, but you're talking about Russia?
Some places are like that. Probably because they are more isolated. America is heaviy mixed (having all sorts of people living there) so it doesnt interest them.
And imagine a Kazakh and a Kazakh will come to America, but you will all be shocked there!
today is russia day
Today is SD3 day, and this is a SD server
lol
no in russia it's a holiday russia day
that's why SD3 came out today
Well, not today, but yesterday)) Russia Day was a holiday, I didn't go to college because of it 😁
SD3 has died.
In Russia, SD3 downloads and installs you.
thats what Microsoft does
A Russian that paints his toenails!?
Why does the Discord finger painting emoji look like toes 😐
I never noticed that and now I can't ever unsee it
Either way, finger, toes, ...
No way, it's just that this smiley face in tiktok is popular now
Have you watched Russian films? Or have you ever played Russian games?
SD3 allegeldy, it;s more important than those clowns.
do i donwload the 10 gig or the 6 gig?
10
We have enough great movies I don't need to watch some film I don't know the language of. My last movie was Dune and it was a 10 out of 10.
But I did watch the Korean film that won the Oscar for best picture named Parasite.
russian movies have their own vibe
if u onyl watch americna movie sits very limited
for instance in hungarian movie sbobbs ar enothign who cares
And play the game The Last Of us? I really like it! Or have you watched Interstellar? Insanely cool movie
nudity is more casual in european films
There are exceptions. Great movies can come from anywhere.
and americna movies since the matrix past 20 years have been just clorified comic book editions
Yep, I saw interstellar long ago
the silver screen now is a comic book
Non American movies seem more realistic, in some cases.
stop making movies for chinese (no offense to the chinese in here your cool)
It just has to be a good movie. And I have to understand it.
Do you know the game Atomic Heart? From Russian developers
Ever see Enemy at the Gate? Very different perspective from the typical US view of war.
@knotty fog @low moon @dawn mulch Guys?
I am not a game player. My 4090 is for AI
DO YOU HAVE 4090? TIN IS VERY EXPENSIVE!!!
Not for Americans. Every citizen is given a 4090 in the US.
Are you kidding? Like absolutely everyone just gets RTX4090? What nonsense is this, then why don't they give it in Russia?
Because communism is a failed system.
Bro, can you please pass the RTX 4090 for me? I'm dreaming about her, please!!! Since you are all being given for free, can you please send it to me? I beg
Yes, I am kidding. I retired after 45 years as a top end software engineer so I can afford a 4090(as long as my wife approves)
man people seem to be dogging on sd3 a lot on reddit
Probably can't be shipped to Russia.
Yep, big time. sdxl had it easy when it came out.
anyone here into historical ai art?
i imagine this failure will cause sai to go bankrupt
Only if there are no marble nipples. 🙂
Its because the models have more limitations than sd 2.0 when it came out :S
Why? If you have imposed sanctions on us, this does not mean that you cannot send a parcel to Russia) It can be done! Will you send it? 😁
Maybe Emad knew what he was doing by leaving.
Of course the new model were necesary and community finetunes will "fix it"....but the non comercial license doesn´t improve it, there are people who were going to train sd3 finetunes that now are not going to do it
i can't wait to see news posts saying "Stability AI, makers of Stable DIffusion are filing for chapter 7 bankruptcy"
That's not the point in saying people will fine tune it. How many fine tunes are there of sd2. It was such crap no one bothered. If SD3 was a good starting base then perhaps so but it is quite bad as a base.
Russia 🇷🇺 America 🇺🇸 Belarus 🇧🇾 Ukraine 🇺🇦 Kazakhstan 🇰🇿 And the rest of the country, we are all people with you! And we are all very friendly with you! Let's live in peace we love you all 💗
Emad had the most intelligent idea of removing nudity from the dataset (midjourney and dalle didn´t do it) and remove copyright (midjourney and dalle didn´t do it) and that affected the model ability to figure out how humans look
Meanwhile there is the news out of Cuba. Probably nothing but a lot of eyes watching it now.
Am I speaking the truth? After all, where will we go without love and friendship? That's right, nowhere!
And Cuba is good too!
sd2 didn´t have finetunes because it was hard to run and sd 1.5 finetunes were more than decent to compete
hopefully they'll make an sd3.1 addressing these issues but it might not be likely
I doubt SAI will release another model bro
press x to doubt
We are all good and kind people, regardless of the country and nations, we are all wonderful people! Let's live in peace friendship and love ❤️😁🥰
They took so many bad decisions, idk how many $ they have left
American Booys 🇺🇸
They should have trained 2 models, one that the community can run and other one that only a100 can run on the API to compete with mj/dalle3
Can't wait to see news headlines saying "Stablity AI, makers of Stable Diffusion are filing for chapter 11 bankruptcy"
Let's talk
sure
Guys, let's chat) as a Russian, I am very interested in learning how you live in America))
You are so unusual!
Guys? 🙃
Well, you don't want to chat, okay, bye bye 😢
Alguien me puede ayudar con el error código 1 ?
Qué dice el error bro?
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't checkout commit cf1d67a6fd5ea1aa600c4df58e5b47da45f6bdbf for Stable Diffusion.
Command: "git" -C "C:\Users\USER\Music\isaias\stable difucion imagenes ia\stable-diffusion-webui-directml\repositories\stable-diffusion-stability-ai" checkout cf1d67a6fd5ea1aa600c4df58e5b47da45f6bdbf
Error code: 1 eso bro
ojala me pudieras ayudar llevo desde ayer intentando descargar esto pero siempre me da un error diferente ya ando bien estresado
@full lark
Oof la verdad que no vi nunca ese error, estás usando comfyui o alguna otra repo? @olive phoenix
no se que es eso bro soy nuevo una disculpa
puedo hacer que salga otro error y no ese pero me pone que ocupo actualizar driver de invidia y yo tengo grafica de amd
y el otro es este RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check @full lark
de ese ya busque videos y en el diffusion-webui-directml no me da ese error y en el stable-diffusion-webui-forge si me da ese error
I heard of atomic heart
what restrictions or limitation are known so far for sd3?
so far users have confirmed that it can make a cow sit on a frog. sd 1.5 cant do that i think.
But can SDXL?
dont know, i never used it
But can I sit on a frog?
But can the frog solve an aquatic equation
it might be able to if a cow is sitting on it
What was the issue with SD2 was it that it did not contain NSFW materials by default or was it impossible to include NSFW at all or was it some other issue?
from what i know, SD2 failed almost as bad as google image did. to this day, google (gemini) still claims it cant generate humans (because their fail was so bad.)
We live free. Much food.
Also not in fear of being thrown out of windows.
Thats a lie. Many Americans dont feel that way. We are not all one entity who think feel, and see things the same. The 'much food' though is mostly true.
For non-human generations IMO I give 9/10, it's really good for a base model, -1 for complex stuff being basketball court lines not aligned/incorrect
@hallow nacelleDon't sd 1.5 do that perfectly fine - courts and other alignments.
thats the thing that bothers some ppl seem to think if it can do sd1.5 quality images then its a "good enough base model". i'm like dude, who trains a model for months just s it can look like something they released 2 generations prior
@rigid harborfor some reason i cant make out what you are saying lol. but i guess you are saying that sd3 might just need some training and new models to do it?
Same, and yea needs training for non-human stuff and it's good, I think SD3 ultra is on-par with current MJ?
SD3 large*
Oh i didnt know it has different versions.
Guys If you saw Purz on twitter SD3 works great as an image to image tool I recommend to use It that way adds a lot of detail
is there like a model that i can use locally like stable diffusion but actually for lower end computers?
how long till they release SD3 Large?
If people could finetune sd3 large, it would be better than mjv6, in specialized styles
i tried running pixart sigma locally but it was just way to hard to setup and it coudent even create an image it was just too slow on my end.
um - you sort of, still, need an nvidia GPU
bummer i have amd, what am i suppose to use then? i might be stuck in sd 1.5 for now?
you can try setting up comfyUI or auto1111 but isn't AMD your CPU?
Yea some finetune is needed, not sure how they'll deal with the anatomical issues, it's gonna be rough
intel is my cpu, amd is my gpu, you might be gettign me confused with someone else, also i did try comfy.
If the finetune is big enough and uses better captioning that sd3 used ( cogvlm ) i dont think it would be that hard to fix
Good
Too bad descentralized training isn´t a thing :C that would make it WAY easier for finetunes
S.D 1.5 with lcm is the minimal available for local I think, If you can't run that then you have to upgrade your pc and If you are pacient enough and can save a good amount you could buy in the future an entry rtx series 5000 by later 2025 that would give you access to any AI program
i feel like there is barley any love/support for amd in the ai spcae but i could be wrong.
i think you're going to want to upgrade that machine at some point, but you can still try to set up auto1111 or comfy and use SD 3 with that. https://youtu.be/Di1KqPXxx2Y?si=CyYMHhaZCzVhNy4N he just put out a tutorial on setting up SD 3 with comfy
i know i should ugrade but ima highschooler who has no idea how to make money, and i really dont want to work with food/fast food.
It's not that, you just had the bad luck to have AMD because that company just focused for gaming for along time and Nvidia was always about a diverse group of consumers,like designers,artists,also gaming but not the only focus
also amd is sometimes cheaper than nvidia.
AMD has always been the cheaper option but to get those prices they have to sacrifice functions , I don't know how old you are but next time when you're about to buy something that could be expensive make sure to investigate,I'm telling you because before I bought my GPU I was about to buy an AMD card , but there's a lot of youtube videos that warns that AMD is for people that just wants a good entry for gaming and not too demanding design programs
Now you know that,just be ready next time and get all the cons and pros before buying something from pc or anything actually👍
noted, hopefully i learn from my mistake but atleast i can game.
by the way If you can use s.d 1.5 there's to much to do with that model yet
i can use sd 1.5 thu directml and comfyui, im testing hyper and lighting version of sdxl but ther enot as fast as something like sd 1.5 lcm.
Yeah If you want to test some XL models,There's a website called Playground AI do you know It?
last time I checked they let you generate 100 or 1000 images a day for free
ive heard about it but ive never used it. in some cases i use this free site called dezgo.
i might have to check it out then if thats the case.
https://playground.com/ I just entered again It's 50 images a day
which is still a lot If you think about It
Do DoRA's go in the LoRA folder?
yeah i guess.
Id like to know that too. I downloaded one a while ago and threw it in the Lora but never checked ever since.
yes use like lora
SD3 kicks Michael Jackson's ass.
Is there SD3 Lora Trainer yet
Is it just me or does stable diffusion make the hands worse than ever?
There's also imgsys where you vote between 2 images to help rank them: https://imgsys.org/
🙂
/dream,
trying to test the new sd3medium with the code from huggingface and nothing works, it fails on loading:
Exception: data did not match any variant of untagged enum PyPreTokenizerTypeWrapper at line 960 column 3
any idea?
What's your GPU?
Stable diffusion works with nearly every AMD GPU
can we no longer prompt in the bot channels?
Hi all, I need help creating a ComfyUI workflow to turn images of empty rooms into furnished interiors without using inpainting. Looking for tips or examples on maintaining perspective and room size. Thanks!
what are you loading it into? i've got it running in Comfy with no issues
I need to download the SD3 10gb Clip
SD3 medium has amazing potential but with how SAI is running things i am getting SD2 vibes all over again
go read the conversation in #🏞|general-with-images that @loud solar and i just had
I am not denying the endless possibilities of SD3, and the model itself is very amazing, i just have a problem with how SAI is running things
they could be bankrupt and gone by now, you know. let them run their company so they can survive
i have been in the community longer then most of you, i had been with SAI since its beginning and i very much love SAI and their contribution to Open source AI, but with that being said they could've atleast trained the model on a better dataset
the dataset is just fine. you work for them?
I dont work for SAI but i have beta tested for them before, and i can say with full confidence that if they used the same dataset they used for 1.4 or 1.5 or even XL, medium would've been astronomically better
okay, then i'm going to guess fully 80% of the peopel on this discord have been using stable since 1.4 was released and a LOT of people, myself included have also beta tested
they didn't get rid of anything that I can see
are you using their demo site or are you using something like comfy where you can adjust thinks like the model sampling, the scheduler, etc?
cause i'm getting perfectly good, realistic, bodies
can it be sampling issues? on ComfyUI
then start tweaking settings. if you are using the workflows that come with the weights, the ModelSamplingSD3 node value is critical
because many people dont know the combination of Euler + SGM Uniform dont worked on SD3
and so are the samplers and schedulers. use the wrong ones together, you get garbage.
or Euler Ancestral aka Euler A
No that just plain ignorance, i used two prompts for anatomy for SD3, one using a painting style and the other photographic, and painting style is much better then the photographic style in anatomy, it means the dataset is heavily leaning towards painting rather then photography
what sampler and scheduler you used btw
so me sitting here for hours, testing minute changes, running through my thousands of 1.5 and SDXL prompts, and seeing results I expect is ignorance? i'm done talking to you.
use uni_pc for the sampler to start and sgm_uniform for the scheduler. and set your ModelSamplingSD3 node (in comfy) to 1
and go from there
maybe we should make a XY plot on ModelSampling and each sampling + scheduler combination
SD 3 does not use U-net - everythign is totally different. you have to learn it from the ground up
SD3XL when
that's sort of what I started workign on but i'm a long way from publishign anything. i intend to do a grid of gens for each combo - but you're welcome to do that before me if you want
By default, DPM++ Normal worked just fine
the denoisers will work well
but messed up if using Euler Normal
Tell me your settings for Comfyui and let me verify your claim, if you are correct about your statements of SD3 then i am sorry
euler isn't going to work.
yeah, and also SDE
go look in general. i've posted screen shots and posted my settings.
any combination with Euler will fucked up
or better yet
I am trying DPM++ 2M Karras right now
i just DM'd you my workflow
@desert daggeri just read you wasnt going to talk to him ever again.
i'm a softy
i realy just want everyone to get a handle on how to use SD 3
it can do a cow sitting on a frog so its okay.
OK DPM++ 2M Karras dont worked fine
sd 1.5 had trouble with that i tihnk
but is the frog okay?
yeah. stick with the denosing routines
Havent checked if its okay or not
ddim_uniform seems to work best
give me a second to launch Comfyui and i will get back to later
take your time. play around with it. you want a copy of my prompt worksheet?
make sure you do an update all first AND that you've downloaded all the checkpoints and clip files first too
SD3 High when
Howdy guys im wondering if anyone can help/ give a breif example of tagging for a background lora or location? I can't seem to get it to work
Tried tagging like you would see from triggers on civit and no tags but the only way to describe it is I can either have a picture from the dataset rendered or a person in front of that checkpoints understanding of a location
hi,is uni_pc+ddim_uniform best?
What's best atm is -> SDXL
Whats the top 1.5 models for photoreal?
RealisticVis, AnalogMadness, EpicPhotogasm, ICBINP, EpicRealism
Honorable mention to the original Deliberate, but its odd to prompt for and a little to Anime inspired on certain gens, but it can get great results also

Not shabby
Deliberate v2, ICBINP, PicReal and Photon v1
Although I am biased to ICBINP 😉
Are you its creator ? I love ICBINP tbh its great 🙂
Thats me 🙂 Glad ya like it!
Deliberate V2 I never really liked way to Anime for my tastes, and V3 was terrible
EpicRealism and AbsoluteReality both decent as well
1.5 lcm icbinp best lcm model
Thanks 🙂
hello gaymers
happy pride gaymers
if you arent a part of the LGBTQ+ community
dont you dare call yourself a gamer
do you want to die?
no
LGBT is illegal in my country
I would love one of those. And a comfyui workflow if you've got one; the provided workflow doesn't seem to be the best.
Bert#2929 is nothing... it's Bort#2020 you gotta watch out for
your name bort too?
is SD3 available to download now?
gives me 403 uppon filling form
where do I download it exactly? iI do not see any links @velvet obsidian
ah I git clone it?
form must be completed without errors otherwise it won't give you access
its failing git cloaning saying my creds are not good enough?
curse my luck , I knew some day not being rich would catch up to me
okay I got it
can anyone help me with stable diffusion?
its producing weird images
it also seems to ignore my loras am traumatized
/help
Hi everyone.
I need your help 💁 I saw announcement - new SD3 released Please let me know how to check this model ?
Can I test it on Foocus or Automatic 1111?
i got too much power
Only Comfy UI and some others
Hi, Thank you. Can you guide me how to test this model with comfyUI?
And Please let me know why I can't use this model on Fooocus.
You can watch on YT it might be easier
because everything except comfy is shit
Its not ready yet for other UI
No Comfy is bad only good for well known people
if you don't "well known people" pay money for mijorney
?
why support terrorist groups?
turning people into castrated people with propaganda is not a good idea
not everyone, there are other people and that's normal
I don't need it, I'm very close to psychology
@upper panther can I DM you? I'll be happy to send you both
all people have always been people and will remain so
you're just old and that's why you have thoughts that it was better before
I´ll delete before i get banned XD


