#💬｜general-chat

1 messages · Page 56 of 1

vast ingot
#

40GB

cold siren
#

anyone tried llama-cpp btw? seems to be much smaller but i still have no idea if its even working, it just hangs there and sometimes writes some random words

tall gorge
#

i mean if you drop the framerate and step counts then it could be doable

tall gorge
cold siren
#

don't you have to have like, same # of VRAM for these?

vast ingot
tall gorge
#

💀

wide juniper
#

Is Instruct-pix-2-pix improving at all?

alpine bear
#

some samplers are better than others at certain prompts or concepts

#

takes trial and error. x/y plots are super helpful.

shy lagoon
#

hey is there a prompt that would increase the resolution for the pics ? I have 512x512 pics but even when i use upscalers they seem to not be high resolution

edgy terrace
#

ei! someone can help me plz?

wanton fossil
#

hi , someone cna help me plz.
i want use stable diffusion at FastAPI.

How can I use stable diffusion in FastAPI?

crystal perch
#

may i ask if there is a free recharge of the credits in a month or so?

lapis kindle
#

hello i can't install stable diffusion can someone help me ?

frank badger
#

Does xformers and medvram affect result quality? It seems to me like images is less detailed now and often colors became pale and washed-out

shy lagoon
#

Any idea how I could prompt in dreamstudio so that it creates the style of midjourney ? The art focussed commerically used kinda of style ?

warm junco
warm junco
broken gull
#

anyone got any good tutorials on how to use Lora and what they're same for textual inversions?

modest schooner
#

how to start use the tool for free

mighty gazelle
#

why do TIs work in negative prompt but not LoRas?

shy lagoon
tropic bridge
#

Is there a help channel?

astral goblet
#

chatgpt's knowledge has a cutoff date. Also, it doesn't get "facts" right often. Especially about social media type events

#

It's impressive but it's really just a rube goldberg machine that adlibs out text

karmic brook
#

thanks bunny_thumbsup don't think general chat is the place for such a sensitive charged discussion

#

please take the conversation elsewhere

strong lance
#

You are pathetic for a ranging on mods, just stop the topic and move on

#

if you dont feel like it, leave

karmic brook
#

Emotionally charged/sensitive topics aren't the best place for this. Instead of feeding in to one another, it's best to calmly diffuse. I am glad to ban or timeout those who cross lines and are harmful to one another. Typing that I lose credibility as a human before I can even reply to you is very disrespectful. Please drop it.

strong lance
#

This is not a political discord

#

are a Jewish users lawyer?

karmic brook
#

asking to not further a dangerous/harmful conversation is not out of bounds

#

this is not okay to say to someone either

#

so long as this conversation is dropped things are fine. Please be nice to one another.

warm junco
astral goblet
#

woah when did the server turn into a southpark episode? the only debate about hitler's death is from hero worshippers of him

#

and that's all i'll say about that

#

yeah. seems really fucked. that was pretty egregious

#

right

#

i seent it

karmic brook
#

@uncut junco , you have been timed out for furthering this very dangerous and sensitive topic after being warned. The other involved in the conversation has been banned based off the messages sent here.

Please take this conversation elsewhere. This conversation does not belong in #💬｜general-chat , and just ask you please respect this. This is for AI and stable diffusion and not heavy charged conversation that can become explosive or reactionary.

In the meantime please do some things like eating your favorite food, generate some images, go for a walk, take a breather-- Fighting in a chat is not the way nor will it ever be the way.

As some people cannot reply quickly sometimes, it is harmful to dehumanize someone on that principle and to blatantly disrespect them and call them pathetic as well. I was taking the time to read the messages from both parties before acting and halt before coming to the actions now taken.

Give others the benefit of the doubt and a little time please. I apologize for any harm or misunderstanding.

astral goblet
#

I feel like the health of the community is shifting for the worst. Toxic trolls are reaching critical mass. Mods are playing delicate hands for VERY trolled out topics, such as "hitler isn't suhc a bad guy" . I see it in the subreddit too. The toxic vocal minority is starting to dig their heel in. I don't like where it's going. I think this is a very serious issue that speaks towards toxic seeds being cultivated.

vast ingot
#

Trolls are reaching a critical mass, I can attest to this too. But I think @karmic brook was right to first try to stop the escalation. Action has been taken now, but when a situation like that happens, taking rash or hasted decision isn't a good idea either. and when there are walls of texts to go through, it takes its time to decide for sure, especially when both party stay to push further and further. Time out was necessary to analyze, and act fairly. But yes of course, for sure, this isn't language or rhetoric that can stay around.
I feel like you may be right on some parts, but I don't want to generalize either. Even if I may not be fast enough and not have the immediate judgement on simple situations all the time, I need to keep on trying. So does every one of us.

strong lance
#

I like how AI answered my question to this riot.
When someone offends another person, it can elicit a variety of emotional responses from those who witness it, including anger, sadness, and empathy. When a third party takes it personally and defends the offended person, it may be a sign of empathy or a strong sense of justice. Psychologists may view this as a positive behavior because it shows that the person is willing to stand up for others who are being mistreated or disrespected.

However, it is important to consider the context and motivation behind the third party's actions. If their defense of the other person is excessively aggressive or stems from a personal agenda rather than a desire to help, it may not be an appropriate or effective response. Additionally, some individuals may be more sensitive to perceived offenses and may overreact in situations where others would not feel offended. Overall, psychologists would likely view this situation as complex and multifaceted and would consider the individual personalities and motivations of all parties involved.

shut hedge
proud nova
#

Best VAE for anime style?

fervent thunder
# karmic brook <@988275769723748432> , you have been timed out for furthering this very dangero...

It's more harmful to let someone believe some things, get affected by propaganda or bs media.
Some people NEEDS to be told they are saying dumb things, whether they like it or not.

How else would they know that if everyone sitting there and playing along.
Idiots should have someone saying to them they are idiots or they'll never know.

I agree - that's not right place to discuss it, but also I feel like you're doing more harm than discussion itself did, but...maybe I'm missing something, since it's cleaned now.

#propaganda_needs_to_be_stopped or whatever.

karmic brook
#

It's important to keep the space here a comfortable one for users, so that's the action I took after assessing the situation.

fervent thunder
#

makes sense thinking

astral goblet
#

Very cool approach. I can see in hindsight that reacting in the moment might've been narrow focused

#

Fruits got the juice

karmic brook
#

thanks all HeartPink

#

all good, thank you!

fervent thunder
#

I'm on opposite side of barricades, I'm living in Ru and I've seen whole country go nuts in a finger snap...
Propaganda is really bad...nomatter how stupid it is , it just works if people keep repeating it over years...

#

I'm sorry for my country being responsible for all this crap...

broken smelt
#

these are only positive comments but like fruit said, mabey move this to #🌶｜off-topic

astral goblet
#

It's sometimes scary how suggestible people in general can be. Not so much individuals, but seeds can be sewn so easily

#

i dont mean, in general chat either lol.. words. so confusing

coral orchid
#

is step/s and it/s the same thing?

astral goblet
#

I'm really digging into archives surrounding illuminati diffusion and nothing i can find indicates that it was ever under a restrictive license. Rail M is what I can see mentioned on all remaining sources. The author purged everything.

I think it's really important to call these licensing blunders out and say "Hey! That's not cool!". I mean, the guy does great work but he was in here earlier blaming community members for deleting it. passive aggressively suggesting that anyone merging illuminati is hurting the wide community rather than his personal business deals around secret licenses

pulsar oasis
#

One model among many.

astral goblet
#

i woke up to find it and my immediate thought "who the hell does he think he is?"

#

well, it was certainly ONE model. It's a gooder

pulsar oasis
#

Yeah it was a good one, and it allowed all the merges to get better

astral goblet
#

now its all wrapped up in the same BS that fantasy AI was lighting fire to and leaving on our doorstep

pulsar oasis
#

These models come out so fast its tough for either one of them to act like their model is that important. In a month nobody will be using either one

astral goblet
#

Philosophically, i'd compare it to someone branching the linux kernel, releasing it under the same license, then getting pissed when people migrate the changes to another branch

#

"those changes were licensed!"

pulsar oasis
#

but i do like both of the models. I use them and get good results. Do you believe all the deepfloyd release hype? Is it reall imminent you think? Seems like it is more serious than the last few months of SOONing

astral goblet
#

yup

pulsar oasis
#

I really want to know how DeepFloyd will be released, as in what form. How we will use it

astral goblet
#

i'll believe it more next week when it happens

#

on hugging face and then ported to auto by day 2

pulsar oasis
#

yeah, that is what i was thinking too. Probably have a colab right away, but then someone will make it happen for auto. I really like the idea of being able to pull that model in to auto and use it like the others

#

same for SDXL

astral goblet
#

uhoh but now we have to worry that model authors dont want their wide public release to be refined at all 😮

#

it might become important for any company funding the development of public models to declare their position and philosophy around community refinement very clearly and openly. This illuminati bs where the author just suddenly declared it was never meant to be merged, after it was released with a permissive license, is a giant debacle

coral orchid
#

If they want to hold the IP of something for 100 years I don't see why others can't use it

#

Would it be possible to sue them?

astral goblet
#

derivative works will always be derivative works. no matter how they're made or what process is used to get there

coral orchid
#

okay I get it. yeah usually that will be in the user agreement

vast ingot
#

it still seems to be

#

and yes, I do hear you and the regulation that is coming around it

coral orchid
#

no they won't

vast ingot
#

but I do think we are close to a paradigm change there

coral orchid
#

they will try to gather as much data as possible in the process and then try to monopolize

#

I think with this AI age, those guys with tons of graphic cards are the real players

vast ingot
#

I do think S.AI is quite a lot working on regulation, making sure this just doesn't get purely gutted by politicians, by making sure the necessary breaks are one, tuned for diferent location differently.
At least it's what it seems from all the news they gave us about future version, and my personal hope

astral goblet
# vast ingot but I do think we are close to a paradigm change there

i'm seeing this stuff blow up faster than the dotcom bubble did in the 90s. i like to think i was young then but had a decent perspective over the rollout of tech. By "this stuff" i mean ML. If we consider the paradigm shifts that came out of that dotcom period, we'll probably see more from this

vast ingot
astral goblet
#

oh. itll come in like a wrecking ball no doubt. it won't be a graceful period of change

#

if you've got an "oh shit handle" near by, hold onto it tight

#

you know those little handle grabs in some cars? thats what i call them

coral orchid
#

It's going to be more cyberpunk like

placid mortar
coral orchid
#

But that is not the matrix thing 20 years ago?

astral goblet
#

ever seen the movie "MindStorm" ?

#

Stranger Days another good one

#

and don't get me started on Total Recall

frosty turret
#

ya know I wish there was something we could do about people going into Voice chats and sitting there with mic muted.. that kind of defeat's the purpose.. ya know

broken smelt
#

theres a bunch of other channels if you want to use one for yourself

frosty turret
#

no thats no my peeve, my peeve is they just go into the vc, mute themselves and just sit there. never replying to any one that might want to actually vc, ya know what the channel is for. why do that do that. it's grrr. so... I don't know . just P*'s me off

#

but anyhow...

broken smelt
#

it's a public space so its equivalent to someone sitting down next to you at a really large table at the library. no big deal if they dont want to chat imo. youre free in either case to swap voice channels if their presence makes you uncomfortable

frosty turret
#

no it's more like someone sitting down next to you at a conferance so people can network and then they just sit there and put tape on their mouth while you try to actually talk to them

#

so yeah I disagree

#

VC is suppose to be about interaction and ya know Talking on voice. thats the point i'm trying to make

broken smelt
#

yea i dont disagree. what about the people who want to simply listen in and be included? some dont have mics. i actually know someone who is physically mute irl who does the same thing

frosty turret
#

then to be transparent they should I don't know include that in their name somehow so others know , would be helpful, and it's so easy to do.

#

and going to VC with no mic is sorta like entering a race with no car

#

but I thought I would just mention it, don't wanna debate or argue the point. i've made my point so i'm gonna back out of the conversation. have a great day.

fervent thunder
#

Hola

#

Hola, gracias por recibirme

astral goblet
#

what do they hurt by just listening?

#

i default ot mic muted. it mutes immediately when i'm done usually. otherwise i could be the dummy with a wide open mic when i'm afk. I'd think this is a common habit because nobody wants to be that dummy

#

unmute when i actually have something to say

frosty turret
sand owl
#

Dead chat

astral goblet
#

plus, i don't even see the harm

#

ever heard of something called manufactured drama?

frosty turret
#

yeah it's a term made up to try to get out of argument to discredit those that are trying to make a point and make them look like the one's who are the cause of the issue when they are just point out something. so yeah I may have heard of it.

#

still doesn't make sitting in a vc and not talking or responding to anyone right or even the correct course of action.

#

nice try tho

fervent thunder
frosty turret
#

but anyhow, nothing I say isn't going to change anything i'm just getting this off my chest so it doesn't bug me as much. ya know .. .venting.

#

but again i've made my point said what I had to say so. yeah. just gonna back out.

#

and I know my grammer sucks

#

😄

fervent thunder
#

hey there, was quite busy and havnt checked SD since 2.1 release, is there some models with good reputation released since then?

rapid comet
#

can i get some technical help please? im recieving the following error and cant find the referenced settings in the error.

#

"NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check."

lunar vigil
stone cosmos
#

Wow, so much to know with this stuff. any particular direction one would advise I start learning if I am most interested in character creation type imagery?

lunar vigil
#

where is the ai art

#

im here for ai

alpine bear
alpine bear
fluid zealot
#

hello all, does anyone know what prompts to use so that the result generated image is going to include a specific text, like if i am making a logo and i wanted to have the name with the logo in the result?

stone cosmos
fervent thunder
#

Making text and then generating new image based on that with controlnet would be another option

brazen quarry
#

What's the general speed like these days since the first release?

#

I needed a 1070 TI to generate 20 images in a minute I think

#

I'm looking to generate 1 image per 1-2 minutes on a shitty laptop

#

Is this out of the question still?

brisk bloom
brazen quarry
#

With a 1070 TI this was easy for it

#

erm, for my desktop

#

I was just wondering if there were any performance increases in the past few months 🙂

#

I basically want to just replace "xstarfish", a random background generator, with stable diffusion lol

brisk bloom
#

a bad laptop generally doesnt have a decent gpu, so it will use the cpu. thats really slow. no improvements there

buoyant marten
#

olá pessoal

#

algum brasileiro ai?

iron dome
#

Boa noite! Alguém sabe como resolver esse erro aqui?

buoyant marten
#

RuntimeError: CUDA error: CUBLAS_STATUS_ALLOC_FAILED when calling cublasCreate(handle)
Time taken: 0.02sTorch active/reserved: 2505/2590 MiB, Sys VRAM: 4096/4096 MiB (100.0%)

brazen quarry
#

so it will use the cpu
This kind of makes no sense. Is it because the lack of CUDA or similar support?

latent hawk
#

Which software is currently the best for video stylization with SD? (e.g. to make it look like japanese animation)

ripe cloak
#

Does anyone know how to put a woman leaning on a car in stable diffusion. Having such a hard time. Especially since it keeps messing up the hands lmao

#

Damn I got it finally. Took 100 years 😛

#

over a thousand images made. Finally got what I wanted 😛

open hawk
#

hello chat

#

quick question

#

does StableDiffusion work on Linux?

fervent thunder
open hawk
#

🙏 🙏

fast heart
#

Hi all,I have a question about Image Upscaling.
I wrote a program in python, and use pyinstaller in order to package my program as .exe file
After I run the .exe program, I got the output error:

Traceback (most recent call last):
File "zoom_Img.py", line 7, in <module>
File "PyInstaller\loader\pyimod02_importers.py", line 352, in exec_module
File "stability_sdk\client.py", line 28, in <module>
File "PyInstaller\loader\pyimod02_importers.py", line 352, in exec_module
File "stability_sdk\interfaces\gooseai\generation\generation_pb2.py", line 16, in <module>
ModuleNotFoundError: No module named 'tensors_pb2'
[14548] Failed to execute script 'zoom_Img' due to unhandled exception!

How can I solve it?

Remark:
python version 3.10
pyinstaller version 5.9
run the following pyinstaller command:
pyinstaller --onefile --hidden-import=tensors_pb2 --hidden-import=stability_sdk zoom_Img.py

oblique cipher
#

hi , can anyone show me how to use multiple control-nets as hand most like pattern . I tried with size 512x512 using open pose and depth method but the hand result is always different from sample.

#

3>setting control-net of mehttps://upanh.tv/image/iXRkga

arctic magnet
#

Can someone explain how does the base model matters when training a lora ?

#

why should I pick one or another ?

#

I don't find anything about this

#

Hey

vast ingot
#

Hey there 🙂

#

well that base model question is a good one 🙂

#

to answer you I need to specify what LoRA is, how it acts when you use it

#

LoRA is close to what a model is : it's parts of a models. it's the weights that would have been modified in that model you trained it on if you had trained it using Dreambooth, almost

#

LoRA on their own aren't usable : to use them, you need to apply them on to a model

#

they then make a "new model", adding their differences in weight inside your choosen model

#

so your choise in base model imports a lot : all that is not trained in your LoRA file will come directly from the base model you choose

#

best thing to do usually is

  • either to use the same base model the LoRA was trained on
  • or use a model that is close to the types of render you want (anime, photoreal, ...)
arctic magnet
#

nice, thanks for the input

#

makes perfect senses 🙂

vast ingot
#

a LoRA is never permanent on a checkpoint : it loads in memory while you use it, but the model file doesn't change at all

#

so no, a LoRA can't break your checkpoint for good

#

but a badly trained lora can make worse pictures than not using it, sure

next haven
#

Does anyone know why my program is not updated automatically? It seems that everything is written everywhere.
@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--theme=dark --xformers

git-pull

callwebui.bat

but on startup it says:
fatal: not a git repository (or any of the parent directories): .git

vast ingot
# next haven Does anyone know why my program is not updated automatically? It seems that ever...

you'll need to go into #🤝｜tech-support for better answers on this but I see 4 problems in your case already :
1/ you didn't install using "git clone". instead you installed by downloading a zip folder. This will prevent you to update using "git pull", you'll need to either reinstall correctly, or continue downloading the zip folder to replace your old code.
2/ it's "git pull" and not "git-pull"
3/ it's "call webui.bat" and not "callwebui.bat"
4/ it's "set COMMANDLINE_ARGS=--theme dark --xformers" and not "set COMMANDLINE_ARGS=--theme=dark --xformers"

next haven
#

Thank you.

#

Point 4 was corrected, points 2 and 3 were the same for me. For some reason, this is a copy so wrote. I'll look into it first, thanks.

#

Can you suggest how to do this? "This will prevent you to update using "git pull", you'll need to either reinstall correctly, or continue downloading the zip folder to replace your old code." I don't understand at all.

#

launch.py: error: unrecognized arguments: --xformers and not set COMMANDLINE_ARGS=--theme=dark
Для продолжения нажмите любую клавишу . . .

vast ingot
# next haven Can you suggest how to do this? "This will prevent you to update using "git pull...

#🤝｜tech-support really for those in the future please.
To do this, you need to reinstall. To go into a new directory, open a console and do git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui
This will download the code again, in a way that will be possible to update easily in the future for you.
Then just move your models from your old install to the new install, and run the new webui-user.bat. wait for it to install, and you are good

vast ingot
next haven
#

🙂

#

My mother told me in childhood that I should learn English, not listen to her. Now everything is so difficult to understand, but the main thing I understood is that I need to reinstall the program and reinstall all my 113 models.

jolly flume
#

Hello, can we use sd 2.1 base to generate icons? I tried to fine tune it with AWS services icons, but quality is poor. I used modified DreamBooth script to allow multiple subjects in one training

next haven
#

Where is technical support located, where should I write?

vast ingot
#

click on it 😉

vast ingot
next haven
#

🙂

vast ingot
jolly flume
# vast ingot I have a harder time training on 2.1 than on 1.5 personally but the project you ...

Thanks for the advice. Let me take a look at that channel. It is poor that, with few epochs it doesn't generate the details of the icons properly, and with 150+ epochs, it is darken and it forgets any other subject the base model has.

For context, I have about 30 icons in the dataset, each with 15 of 770x770 px images with the icon being translated around (to avoid duplication). No prior preservation. Using modified DreamBooth script

vast ingot
#

so, 30x15 = 450 total pictures ?

#

each of the 30 concepts has the same problem ?

#

you are describing 2 different epochs, low and very high. Did you look at your Loss curve on the 150+ training, to see what was the sweet stop you should stop on ? did you do more checkpoints in between to test ?

jolly flume
# vast ingot each of the 30 concepts has the same problem ?

Yes around that. Yes all having similar problems. The loss curve are going down, but at epochs 110+ there were spikes going up. Average loss at end of epochs was about 0.0098. I guess it overfits. I didn't do checkpointing though. Good advice to start doing one. Let me take a look at your link

vast ingot
#

take your curve, and try to find the lowest point in loss before the spikes on 110

#

this is where it has the most change to be good

#

in your next test, do multiple checkpoints, every 10 epochs or so, starting on epoch 80, and compare those

jolly flume
#

Thank you! Your github is very useful. I was looking for definition of GAS and batch size, and didn't find good explanation elsewhere.

What would be a good prompt for an icon of AWS Lambda when fine tuning it?

vast ingot
#

"AWS Lambda" I don't understand what that means

jolly flume
vast ingot
#

well, the full captioning strategy of such a dataset isn't very easy...
1/ I think 15 lambdas is a LOT TOO MANY. I would use 3, max.
2/ I would do the same for each type of icon if they are this simple
3/ for each icon, I would caption it "icon $type $variation"
where $type would be "lambda" here
and $variation would be another token describing how this icon varies. here for example, "hollow" or something. no need to specify the orange for example, since this is not a variating part, and will be trained inside the "lambda" token because of it

#

class prompt on this... "image" ? I wouldn't use prior preservation on this at all tbh

jolly flume
#

I see, but if there is only 1 type of Lambda icon, all the 3 training samples will be all the same, just translated around? So far in my 770px, the icon is about 50px in size, so I just move it around for variation. Should my captions be "icon lambda right", "icon lambda left" ?

Below is my result with 12 epochs of ~450 images with 30 concepts, lr 2e-06, GAS 2, Batch size 5

oops, can't attach image?

vast ingot
#

we are in the only channel without images currently

sand owl
#

SD private testing

sturdy mountain
#

hello

vast ingot
#

Hey there !

next haven
#

@vast ingot Thanks, they helped in those support. Only I lost the shortcut that launches the program and forgot what it's called, can you tell me where it is? The one with white letters on a black background 🙂

#

I remembered, this is a file in a folder.

vast ingot
next haven
#

Thank you.

vast ingot
#

sorry i didn't answer I was on something else and didn't see it before you just pinged me

next haven
#

Thanks for your work. without you it would be very difficult for me.

sleek osprey
#

Looks like deepfloyd IF is bigger than SDXL, which makes me question what the point of SDXL is?

#

i guess SDXL is being trained for aesthetics?

torpid rivet
#

Hi guys. I have 2 questions:

  1. can I used SDXL with API?
  2. I can't see the negative prompts example for REST API usage, just like you send text_prompts, how would I send a negative_text_prompt?
shrewd leaf
#

Hi guys, is GTX 1650 sufficient for stable diffusion usage?

edgy terrace
#

hi

#

i have an intel irisxe graphic

#

can i use this with sd?

#

stable diffusion

vast ingot
#

self promo like that, without even a presentation and in the wrong channel, isn't welcome :/ try to explain what you do before promoting it like this, and head over #🌶｜off-topic since this has nothing to do with SD

shy lagoon
#

is there is a reliable way to let sd not show any hands because v1.5,v2.1 and the new one beta just dont manage to make them look good, i tried negative prompts "hand,hands" as well as positive brand "no hands , hands hidden" but they dont work ?

warm junco
past sentinel
#

i cannot get normal finger count myself

dim shard
#

hi i want to use stability ai to create a text to image app i want to how much credit the restApi cost per request

vernal cliff
#

join kahoot

#

260700

pulsar oasis
shrewd leaf
coral orchid
#

I mean for him

tacit loom
#

hi

#

does anyone want t join my server

#

anyone there

coral orchid
#

Guys can you upload your venerated images to online stock websites?

#

Do you actually owe the images?

pulsar oasis
white mountain
astral goblet
# coral orchid Do you actually owe the images?

Copyright is a complex set of law and is often dealt with on a case by case basis. I'm in the camp that AI generated works may sometimes be a new creative work, and other times be derivative of copyrighted material. It really depends on the intentions and process of the author to whether it would qualify or not

#

An iconic photo we all know and love, maybe, is the XP background. Those rolling green hills. If you generated a ton of stock images to replicate the aesthetic of that original background and named them "XP Rolling Hills" then that i don't think would qualify for copyright protection and would be derivative work.

#

Another case, which i think is one with a lot more nuances and courts will probably have to chew on through a couple different circuits until somtehing is figured out. What if you make a prompt that has Stock photograph in the prompt and fired off 1000 of those, curated the best 30% and published. I'm not sure enough intention of work is there. What sort of creative act was done to make those images and how are they transformative and not derivative? Questions!

Courts will figure ones like that out.

#

i think though that documenting our creative processes will be something photographers and artists using AI tools will have to learn to do. It may be our key defense to why we own the work we make. It's all about that creative process and proving that you were indeed the one who had the intention and will to create the work.

vagrant glacier
coral orchid
jagged vigil
#

what are other good sites to see pictures with prompts for inspiration what works good like civit.ai?

sly eagle
#

I wonder why everytime i want to do 1 character it always draw 3 or more even using negative promts

astral goblet
#

If you're cloning another photographers style and selling that image as "style of photographer, by me" it could be infringing because of that intended expression, literally admitting that it's derivative of that photographer's work. While a lot of people say style can't be copyrighted, and sure it can't, it's not that simple. Derivative works exist too.

#

https://en.wikipedia.org/wiki/Campbell_v._Acuff-Rose_Music,_Inc. << here's a sick case law study about derivative vs fair use argument. precedent is really important in this part of law. the case found that a parody is derivative but can also qualify for fair use if the artist provides enough transformative intent from the original song. Also 2 live crew are sick.

#

discord screws up the wiki link and you need that . on the end

#

or not wikipedia fixed the bug

sly eagle
#

I'm teaching a certain style to sd, but for reason, it learned somehow but can't draw bodies or poses

#

what i did wrong?

astral goblet
#

no idea how to train styles here sorry.

sly eagle
#

no prpob

vast ingot
#

I do have a guide on training, not only styles

#

but I don't have time to answer questions right now

#

so I can link the guide and you can ask complementary questions here, i'll answer later if you want

astral goblet
#

perhaps you need a better regularization data set? i'm noticing when i'm training people, the regularization images should have many other similar people. similar only to a degree though.

I ask only because i'm trying to figure out styles myself. will come back to this later as i have to go to the world famous city of nanaimo, where they keep the nuclear wessles.

fervent thunder
#

Guys, I need gpu guidance but is this the right section for that? I mean, as it pertains to stable diffusion.

sly eagle
surreal sable
#

Hi, I'm want to start using SD locally but my internet is kinda bad so I wanted to know the download size, in some places says 20 gb and in others 12 gb required.

warm junco
fervent thunder
#

Guys, will the 4070 12GB be good enough for model training and such?

surreal sable
sly eagle
#

The problems start when i change the size of the canvas

hallow plaza
#

Is there a good AI tool to download images of a subject? Like if I said download all sprites of RYU from street fighter that have at least 128x128 resolution, Is there something that could search the web and automatically download it for me?

ember grove
#

In what channel can i generate the image?

#

I didn't remember

arctic topaz
#

Hey, it's my first time and I couldn't find the info in the Start Here or FAQ about what channel I can just play around in and start creating images.

hallow plaza
ember grove
hallow plaza
small spear
arctic topaz
#

where do you create images then?

hallow plaza
barren laurel
#

is reddit down?

#

nvm ig

#

sometimes it just doesn't load my stuff

atomic mantle
#

I want to make a style LoRA based on my partner's comic. I've put a fair bit of effort into extracting high res crops from each panel. I think I can give it hundreds of examples of backgrounds, and I envision a future in which I could generate backgrounds in the same look.

But before I invest all the time in cleaning up training data, I'd like to know, what is the plausible way in which the style LoRA could be used to create new backgrounds for future comics in the correct perspective and angle? Could I give it a 3D image from sketchup? Could I give it a fairly sloppy sketch? Do you also type a verbal description of the background you want? (e.g. draw a loose sketch of a hallway, and then specify "Hospital hallway, green walls" as well?)

Is this a feasible thing to try to do, basically?

nimble musk
#

Hey everyone, I'm trying to figure something out for a new pc build and it's hurting my brain

#

is anyone able to provide some advice on using dual 3090s. I understand stable diffusion can only use one but I want the other for LLM's

#

or anything else

#

is anyone able to advise if it's a bad idea or not or things I should take into consideration? I'm trying to get my brain over the motherboard PCIE lanes I need right now

#

Thanks!

small spear
small spear
nimble musk
#

I suppose it's making sure i have the platform set up correctly (hardware wise) to enable me to just freely explore

small spear
nimble musk
#

or do the LLM's only boost into the GPU performance when spitting out responses

#

and not while idling

#

any issues with using the E-GPU route?

#

sorry to bombard you 😄 I appreciate you entertaining my idiotic questions

small spear
nimble musk
#

no problem, thanks for replying anyway!

dim sparrow
#

Guy i'm trying to install disable fusion but i'm getting an error, can someone help me out?

atomic mantle
#

He also warns at the end that he couldn't generate the exact same scene from another perspective, which is a bit of an issue for comic making. Is that just inherent to this, or is there another method (maybe even just feeding it the previous render, so as to say "make sure the details of the scene match this")

small spear
#

also, it may depend on the setting/theme of your comic

small spear
#

no, i would do photo, 50mm, cat (the fov thing idk)

#

the software cuts up natural sentences to find the embeddings, sentences do help if you specify 2 cats with different attributes tho

iron dome
#

Pessoal, boa noite

#

Galera alguém sabe onde que acho essa opção para verificar qual é o modelo a usar?

#

1c0ed90d69

slender jay
#

Hi i need some help

small spear
slender jay
#

whatever i prompt i cant get any good results in terms of quality

small spear
slender jay
small spear
broken smelt
#

are you using automatic1111? and its base model?

slender jay
#

for example

#

ou i cant send img here

broken smelt
stable pawn
#

How important are "negative prompts"? I seldom use them and only do so if I see that my prompt generate unwanted result often or all the time, but I ran in to more and more folk that now claim the negative prompt are almost essential for make good images. So what do those negative prompts actually do, what would be a good way to test the strength and usefulness of those prompts?

small spear
#

try monochrome, greyscale, bad quality, lowres, blurry, text, or removing stuff from existing gens

stable pawn
#

I tested "bad quality" and know what AI do not know what thay is for "quality" is unspecific, and if I specify in my prompt "color photo" I seldom need to use "drawning, grayscale, 3d" and so on in a negative prompt.

small spear
#

if you are happy with your gens thats fine, but in my experience (realistic anime AI images), i cant live without them

stable pawn
#

I know AI is trained on tags, but what do the tag "quality" and "bad quality" stand for? Question is for a method to test negative prompts to see how they work.

small spear
stable pawn
#

Can you give me an example of a prompt that always give "bad quality" or "worst quality" in a image so I can confirm that the negative tag "worst quality" give me better quality?

sly eagle
#

Something similar happens to me, it gives what i want in 512 but when I add 1024 is just turn crazy

#

does anyone recommend me a model to train with human poses?

small spear
broken smelt
broken gull
#

Anyone who knows if you can use diffusers in auto1111 and where I can find the information to do it?

tacit loom
#

h anyone there

shy lagoon
astral goblet
#

you might be favoring roko's basilisk if you start training ai to mutilate creations. careful here. 🤡

sly eagle
#

for me embedding worked better

#

omg maybe i made a mistake hahaha, I trained my style in embedding, it draws it well but for some reason can't draw bodies

fleet coral
#

Hi guys, I joined because I had a problem with some files on my pc

#

Umm is it okay if I say it here?

tawdry turtle
#

No go to a tech support discord server

fleet coral
#

Bruh, is part of stablle

stable pawn
#

I test negative prompt but my English skill is lacking so; is "unsharp" a real word or just a photoshop routine?

fleet coral
#

every time I try to run it I get an error for a "dll file"

tawdry turtle
stable pawn
#

@tawdry turtle Great idea, I look for words to test.

tawdry turtle
fleet coral
#

I just need someone who has the stable and can give me the dll files

#

copy paste and fixed

tawdry turtle
#

Just go to their GitHub and get all the dlls from there

#

Which OS are you using

fleet coral
#

window

tawdry turtle
#

Bruh window sucks, get Linux already

fleet coral
fleet coral
tawdry turtle
#

Okay makes sense. Yeah go to their GitHub and get the dlls from there

fleet coral
#

all? I want it to work, no explode agony

tawdry turtle
#

Okay, I DMed you for the files

#

It’s my sleep time soon tho

fleet coral
#

okey

#

thanks .3.

fervent thunder
#

Maybe it does for anime, idk

craggy grove
#

There seems to be very little discussion here

shy lagoon
#

do you think there is going to be a way to have models with predictable output to certain prompts ,besides finding out thorugh trial and error ?

strong hedge
#

How to generate images

tacit loom
#

hi anyone there

storm island
#

hello, can you guys help me with it: Couldn't find Lora with name cuteGirlMix4_v10

tacit loom
#

gn

stone cosmos
#

how the heck does anyone understand all this jargon XD

viscid orbit
#

how do i increase the allocated memory

#

for stable diffusion

#

i downloaded it on my computer

#

and i put the url to run it online

stone cosmos
#

Good answer. I... don't know if I'm up for that level of math knowledge. Documentation can be confusing in this subject. But I shall persevere. Everything moves so fast.

viscid orbit
#

and i downloaded it by following royal skies download tutorial

stone cosmos
#

I'm with you up to 'grid of pixels,' but nothing beyond that, lol. But i will look into these concepts.

arctic magnet
#

hey, I'm not sure to understand how to properly caption images for loRA training

#

I've seen a yt tutorial where the guy say that describing the picture help ai not to focus its training on what's described

#

it's confusing, i tought that i had to actually caption what i'd like to be trained 🤔

#

not what i want to avoid beeing trained

limpid tusk
#

when u guys switch models using the x/y/z plot script, does it keep displaying when switching models? Mine shows the progress images for the first model then stops until it's done, not sure if its just me or not

alpine bear
#

Oi guiz, not sure if this counts as spam or not. The account only has 4 messages in the server, and they're all this link.

vast ingot
#

I cleaned it

#

👍

bleak matrix
#

Good morning, everyone! sunsmile How are we all this beautiful day?

normal jasper
#

my images look very uneven if I use inpainting. is there a way to process a image to combine all the elements in an 'even' way?

stable pawn
robust frost
#

GM all.

bleak matrix
bleak matrix
bleak matrix
robust frost
#

More adventures in AI art!!

bleak matrix
#

Dun dun DUNNNN!!!

robust frost
#

lol

#

Has someone in the community deployed their own instance of SD please?

#

Looking for a bit of architectural help if thats possible

bleak matrix
#

Haha (we all need some dramatic flare in the morning lol)

stable pawn
#

I will not talk to mush about the religion of AI prompt, but I see that many use like "best quality" as a prompt at same time as "worst quality" as a negative prompt, and I just have thoughts what folk want to archive with (cute:3.5).

ebon beacon
#

anyone know of a tool i can just drop in 50 images and have it create a video transitioning between the images?

bleak matrix
#

Yeah, you can certainly use worst and best, just like you can use the words well and good. It's feel like it's a game of word association tag. There's a lot of fun things you can do, and some do work really well across the board.

bleak matrix
#

Though

coral granite
#

hi how to fix a "in queue" i put the prompt and click start

#

nervermind, its work now

#

but i have a another question, how to generate from text to image high quality images, my generated image looks like from dalle mini

bleak matrix
#

There are plenty of examples on the server to look at, as well as community guides, etc

normal jasper
bleak matrix
normal jasper
bleak matrix
#

Well, there's a lot of ways to do things. Personally, I would say to lower the denoising for how much your output image is going to change in img2img, and generate a few very close images to your base image. Use the seed you want, too, ofc. There are many different ways to upscale and combine upscalers to see what happens/get what you. #1003034183716835418 is a channel dedicated to working with them, so you might find some useful information to help with your project

#

I'd also post your settings/images

normal jasper
#

thanks @bleak matrix

real kayak
#

Is it possible to train your own images with stable diffusion?

#

lets say you have 20k nice dslr photos of family and friends

#

is that a huge undertaking?

bleak matrix
bleak matrix
real kayak
#

Thanks will look around, have you ever done it?

#

kind of wondering about what the benefits would be as well

bleak matrix
#

Yes

real kayak
#

I mostly have family and friend photos - thought maybe if I trained it I could generate new art of them in a way.

#

but I have no clue :p

bleak matrix
#

There's a lot of awesome things you can do, so it's def worth learning

#

I recommend you start with learning how to use SD, as this will help you fundamentally understand how to make things with your friends/fam in them

tidal moth
#

can someone please help me with stable diffusion? for some reason it wont let me launch it

bleak matrix
icy grove
#

hey

tacit loom
#

good morning

#

guys

tawdry turtle
#

It’s evening here

#

For the unfortunate people of the world

tacit loom
#

its morning here

tawdry turtle
#

It’s evening here

bleak matrix
#

👋

#

Soon it'll be afternoon for me haha

drowsy dock
#

Ever since I started messing around with SD it's all I can think about.

#

Preoccupied thoughts throughout the work day...while I'm eating...browsing the SD subreddit... any suggestions where to go from here? I currently use Automatic1111 GUI and am learning to navigate that. I guess the next step would be to find a "model" and try other renders from that?

wild steppe
trim ingot
#

hi

past vortex
#

Man, 2.1 kinda sucks

crystal barn
#

will there be an everyone ping when SDXL gets released? Or is there any other way I can get notified because I only really care about working locally

fresh marten
#

is stable diffusion xl open source

stoic hearth
#

I also only care about working locally. SD is fun because it's a local model. I don't want to pay per use or subscribe to some online bot.

charred mesa
#

SDXL is live, but still not available for offline use yet 🥲

#

not that I could probably run it lol

wild steppe
gray wolf
#

The websites are so goddamn slow now because of the SDXL ping lol

rustic isle
#

The SDXL model just vanished for me lol

sonic carbon
#

its completely down now i think

broken pollen
#

I assume that when sdxl gets out of beta Stability will open source the model

sonic carbon
#

L

rustic isle
rustic isle
#

it still can't do cube stacking tho

broken pollen
sonic carbon
#

it really is an improvement

#

take a look at my small comparison

rustic isle
#

wow that is a significant improvement

sonic carbon
#

both are variations on the same base img

#

so they both had a bit of a leg up

cerulean bane
#

model disappeared

wraith bridge
#

hello everyone, cant see new model on dreamstudio

half wigeon
#

my gradio for 1111 is quite slow and take a lot of time to load, is there anyway to fix?

vital iris
#

what did you do to thispersondosnotexist

grim frigate
#

SDXL beta right? So we still have a chance of seeing better 'hands'? 😭

solemn sky
#

where do i download the new model? 😮

copper crystal
fervent thunder
#

I poped up too late to the AMA. 😂

#

I wanted to ask what were to major roadblocks to having a model that can generate HD sprites, but keep the style coherent, similar but not the same as the video models.

wild steppe
#

SDXL should be available on DS again 👍

wild steppe
tawdry summit
fervent thunder
#

Nice ama

fervent thunder
fervent thunder
fervent thunder
tawdry summit
#

Anybody happen to know of someone having applied AI for organic synthesis or other chemistry?

solemn sky
#

yo, where can i download the new model SDXL ?

calm plover
#

came late to the party want to know more about "IMMERSO"

#

only available on dreamstudio website?

wild steppe
solemn sky
#

oooh so it's not an open beta

#

i don't have or work with dreamstudio

#

i'll wait for the release then 🙂

wild steppe
#

model is not yet fully baked, you can give us feedback over at #✨｜sdxl to help ❤️

calm plover
#

its web based sign in with discord to play

raw ore
#

I was hoping we would here some news on deepfloyd

solemn sky
#

how can i leave feedback if i can't use it xD

calm plover
solemn sky
#

i'm on Auto1111

wild steppe
forest trout
#

Just sign in with your discord account.

river furnace
#

Wait so just to be clear, SDXL is going from a paid credit system on DreamStudio, to an open-source model to be used on local web UIs?
(Sorry I know very little about Stability AI)

tight talon
#

hey so im trying to get everything set up and it's giving me this message when i run webui-user.bat
venv "C:\Users\1anda\stable-diffusion-webui\venv\Scripts\Python.exe"
No Python at '"C:\Users\1anda\AppData\Local\Microsoft\WindowsApps\PythonSoftwareFoundation.Python.3.11_qbz5n2kfra8p0\python.exe'
Press any key to continue . . .

but my bat file looks like this
@echo off

set PYTHON="C:\Users\1anda\AppData\Local\Microsoft\WindowsApps\python.exe"
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=

call webui.bat

solemn sky
#

i mean, doesn't matter much does it, i have all the tools i know how to work with on auto1111, like upscalers, img2img and inpaint... the works. plus i can generate as many images i want. what's the benefit of using dreamstudio? even "edit" says "soon" so i can't even edit the images, right?

#

hi happy

#

dunno how dreamstudio is better when it doesn't even have an editor

raw ore
#

from testing it seems like sdxl seems to be very similar to previous models in terms of fidelity and prompt adherence. Deepfloyd looks better though

solemn sky
#

its grayed out and says "soon"

river furnace
#

Man AI is progressing so quickly. It's almost scary dude.

untold ocean
inner prawn
#

SDXL open source model release when? Or are we clamming up like closed AI?

solemn sky
#

xD

inner prawn
#

If it happens we riot 😂

untold ocean
#

The plan is to open source SDXL

inner prawn
#

Good.

river furnace
#

Crazy

hallow karma
arctic sedge
inner prawn
#

Mid journey forces you to use discord to generate images. I want to run locally with the ability to fine tune

river furnace
#

I wanna make a dedicated AI rig. What are some of the most ideal cards to utilize, price to performance? And at that end, can you even multi-gpu with SD?

inner prawn
untold ocean
next fable
#

que es este server

#

what is this server

wild steppe
river furnace
#

I have a 3090 myself 😄 super quick with gens

#

I've been meaning to try it with alpaca, but my PC is goofy and hard-locks when I start the model for some reason lol

arctic sedge
river furnace
#

Same with any 2.1 SD model lol

solemn sky
wild steppe
river furnace
#

Not too spendy either

storm pollen
#

Does anyone have a couple of key bullet points from Emad's talk just now?

winter pike
#

not yet its still in beta

#

in dreamstudio (paid) or free in pickapic.io (free but random, you dont know which model youre using)

#

its a russian roulette between dreamlike-photoreal-2.0, SDXL 2.2 beta, and SDXL 2.2.3 beta

barren mulch
#

what's the best upscaler ?

solemn sky
#

how long is a rope?

#

... depends on where you cut it.
just like upscalers, the best is depending on your needs.

broken smelt
#

i havent ever used anything other than the 4x ultrasharp model after discovering it

solemn sky
#

and that's probably great for the art you make, but nobody knows what ne3zy makes xD

leaden mountain
#

Hey everyone, as sometime somewhat new to SD I've been using automatic1111's implementation and getting cool results. What's the next step I can take to start exploring making more realistic and cohesive images? I see the images in POW and they look so much more cleaned up and dynamic. Any tips on achieving this level of control on my prompts?

solemn sky
#

well you can download ControlNet, that'll be a new universe for you to explore

warm aurora
#

Does someone have the recording of the AMA?

wild steppe
karmic brook
broken smelt
broken smelt
fervent thunder
wild steppe
daring girder
#

when will SDXL be open source?

vapid tangle
#

did u guys check out the new model?

#

maybe I'm bad at prompting but it seems just cartoony with alot of things

winged peak
#

Does anyone know stable diffusions native render resolution?

#

Stable diffusion XL*

untold ocean
#

the model currently on dreamstudio is native 512x512, but that could change with further revisions of the model.

verbal vale
#

is there any recording of today's AMA going around?

wild steppe
fervent thunder
#

Has anyone used easy negative embeding? I wanted to ask, I placed the files in a folder and then in the sb embedings folder is that going to be a problem?

manic wraith
#

Is there any video showing off SDXL? Some kind of "before/after" promo material or blog post?

balmy flume
#

hey, how am i going to create ai pictures on my own pc? can someone help me with this or we just need to use de website?

manic wraith
#

paid is easier
local install is more versatile
it mostly depends on if you have more patience or more money

balmy flume
#

i have patience but i don't have money and website cencors almost all of it

#

thanks for the info

manic wraith
#

actually if I google I get a page whose URL is "voldyold" but just "voldy" is the right one, hosted on Rentry

manic wraith
hushed quarry
balmy flume
fervent thunder
#

I thought Emad said some models would release this week

#

or there'd be some big news hasnt been any has there?

balmy flume
#

@hushed quarry hey, am i supposed to use all models on the step 3 or one of it

verbal vale
#

what's the hold up on getting the AMA recording uploaded to the youtubes? We on dial up?

#

any longer and it will no longer be relevant, at the speed of AI progress

winged peak
cerulean bane
fervent thunder
cerulean bane
#

whats the soem models?

#

actually he probaly meant "some" nevermind

sharp plaza
#

I know sd is mostly gpu intensive

#

but does a better CPU help load times

#

well not the actual image generation

#

just launching the program

dry trellis
sweet jetty
#

Hi guys im new and i want to make pics first of all will a 1660 ti make the job?

#

second where can i dowload the program and plugins or smth

fervent thunder
fervent thunder
maiden bough
fervent thunder
#

Hello! Can anyone recomend a good video about Khoya GUI and training LORAs. I have read the git hub installation instructions, but still a bit confused.

gritty ivy
short crypt
#

Hello everyone! I'm wondering if anyone knows any tutorials/guides on installing stable diffusion from stabilityAI without any webUi, so basically generating the images from the terminal as shown on the github readme. I'm trying to install it on macOS (M1) and i'm having some trouble

flint lodge
prime patrol
#

is there like an art proof name for a background that would contain a placeholder character? I basically want to use ai to make backgrounds for my character designs.

short crypt
flint lodge
short crypt
#

but this is for webui

flint lodge
#

webui is easy

short crypt
#

i specifially want to not use it, since i'll be generating images from a python script

short crypt
flint lodge
#

yes its stinky

#

the other colab is better

#

allowing other models to run on it

#

textual iversions

#

and loras

short crypt
#

but i can't use webui with a python script

#

i need to execute it with code

flint lodge
#

just use the colab itself

#

why do you want to run it on your potato laptop

short crypt
#

I mean I get that the performance will be worse, but regardless I need to run it locally

flint lodge
#

run it with colab or i will throw you into the waifu diffusion paradox

short crypt
spark arch
#

hey

warm gazelle
#

Dreambooth on M1 anyone?

glacial crystal
pine comet
#

hey i need help setting up chilloutmix ai can someone help me thanks

dapper crater
#

SDXL is shockingly bad

#

should be called sd 2.1.5

civic jolt
limpid tusk
#

is there a way to use prompt s/r on the negative prompt?

civic jolt
#

Having a really hard time referencing architectural prompts to ControllNet (openpose). I still get humans as results with architecture in the background..

vast ingot
#

openpose isn't made for buildings at all, it forces a human into the frame

#

mlsd is the best in this category, but canny and depth can help too, as well as HED

civic jolt
#

I´ve tried depth at first and it kind of worked. Out of 300 frames only 70 had human like features (e.g real faces or legs). The problem with depth is that it tends to cut of the arms. His Videos are tracked really well^^

gaunt falcon
#

Is there another website like civit.ai where you can browse models that isn't filled to the brim with NSFW?

terse swift
#

hi

vast ingot
#

Hey there

#

what's happening ? what results do you get ?

#

you can't post pictures in general chat

crude cove
#

Hi everyone,
I'm looking for AI artists for a paid gig, used to working with warp or controlet, vid2vid etc

hybrid sphinx
#

Does anyone know if using movie styles, such as those from Game of Thrones, for generating an avatar could be considered a copyright violation?

zenith thunder
#

need a lawyer for that

tight prism
#

Anyone have a suggestion for a easy and free background removal tool?

zenith thunder
#

ai gen or not

#

you would be on better footing if you paid an artist to make the avatar and used the ai gen as just Ref

#

thats for a lawyer to say

#

they can sue for any reason they want to pretty much

hybrid sphinx
#

It's so complicated ! I don't think they have an answer 😄

zenith thunder
#

if they feel you are diluting their ip sure

#

under current copyright law as a normie non lawyer understands it

#

if you want to have copyright of your content you need the final creation to have significant alteration by a human

#

the more humans you have involved the better

#

when it comes to copying characters or designs from movies there has been cases with companies using photographers and makeup artists

#

for copying a look

#

so its not unheard of

hybrid sphinx
#

No I'm not looking for that, I just wanted to make sure if I'm able to sell it as a Movie package.

zenith thunder
#

is that image just a text prompt?

#

that sounds like you are in a legal grey area

#

that dosnt sound like enough human input to me

#

i wouldnt do it unless you own the original images used

#

for this purpose

magic quiver
#

I have questions abouts image prompt, are they kept in the servers or deleted after?

tight prism
#

They are stored with the pictures generated, you can also recall the last promt with a single click

#

In <PNG Info> you can recall a promt from a picture. and send it directly to txt2img

undone garden
#

Automatic1111 is producing generic tile style patterns every time I try to generate an image, no matter the prompt, no matter the model. What am I missing? Invokeai works fine

jagged vigil
#

Has someone a short summary of all the important aspects for the prompts, which I can feed GPT-4 for prompt creation?

vast ingot
mellow meteor
uneven forge
#

oh noes civitai died again

soft siren
#

hi, I want to ask if i use the photo generated by stable diffusion to train Lora (with similar face but not the same face) and then i use this lora to generate photo with weight 1.0 will i get a consistent face which is the mixture of the photo i used or how

#

for example, I use 10 Jisoo photos and 10 Lisa photos to train Lora, then use this Lora with weight 1.0, will I get a consistent face that is half Jisoo half lisa or will it just sometime look like Jisoo and sometime look like Lisa

vast ingot
#

cowdance Did you know that cows have spiritual powers and serve as messengers between the physical world and the spirit realm. In certain cultures, cow hide is used in ceremonial clothing and accessories because of its believed mystical properties. These beliefs lead to the idea that cows are secretly shamans, possessing wisdom beyond our understanding. cowdance

vast ingot
marble plaza
#

is sdXL free from dataset copyright problem?

jagged vigil
#

@vast ingot Thanks works great!

tidal moth
#

IM FINALLY unmuted..

winter pike
fervent thunder
#

So hellou everyone ! I need your help ://. So you maybe had saw a videos there AI (They say its stability AI) create a photos of anime girls. So I want to know how to make that kind of stuff on stability ai. Now you thinking that im simp. NO. I using renpy game and I need original photos. So maybe can somebody help me tell how to use it ?

tribal brook
#

Question: ¿Can I ask questions about Stable Diffusion WebUI in this discord? If so, ¿In which channel?

wild steppe
wild steppe
fervent thunder
#

@wild steppe Thanks <33

coral pebble
#

Hi guys, I'm totally new here. Trying to use stability.ai with my discord, but won't let me accept the terms of use. The button is grey. Why? 😦

stable pawn
#

Star Wars, worlds longest triology.

modern quartz
#

Hi i have problem with installing 1111 . i did install python but i get these errors what should i do ?
ERROR: Could not find a version that satisfies the requirement torch==1.13.1+cu117 (from versions: 2.0.0, 2.0.0+cu117)
ERROR: No matching distribution found for torch==1.13.1+cu117

celest flame
#

Is their any AI that can take a line picture and colourise it?

small spear
small spear
modern quartz
small spear
# modern quartz yes

im not an expert but looks like something went wrong with your pytorch install

modern quartz
topaz lantern
modern quartz
#

what is torch ?

dry trellis
#

its part of the python framework that handles machine learning stuff

fervent thunder
#

This is a great server already loving it

sly eagle
#

Hi! is there's a way to color an lineart image in SD?

pine comet
#

hi i need help installing stable diffusion

wild steppe
small spear
fervent thunder
#

What is better to use

#

Controlnet or Dreambooth?

gleaming shale
#

very awkward question, how do you pronounce Emad? i'm gonna be at the ML meetup next week and don't want to fumble his name if i ask a question

fervent thunder
#

What is better to use?

dry trellis
fervent thunder
#

best model for creating a space helment based of the marauders game

#

ngl, so far been using it for porn, please help me branch out

potent fox
#

Hi guys. Do you have a success story regarding a locally installed ChatGPT-alternative which could be connected in some way to SD to generate prompts?
I am thinking about things like trained text-models, LoRAs or such stuff, to realize kind of a "virtual friend / partner" finally. I hope that sounds not too weird.

verbal vale
#

I take it the AMA is never getting uploaded?

modern quartz
#

do you have any idea how i have to use
torch-1.13.1+cu117-cp310-cp310-win_amd64.whl ?
i had problem with installing 1111 and when i googled it said i have to install that
but now after i download it it wont open bc its not exe
i dont have any idea what should i do

umbral yacht
#

I want to create some balenciaga meme type images. How could I approach the quality of what is seen in the memes?

burnt lichen
#

Any progress on this? (Incorporating something like UV-coordinates into SD for the purposes of keeping a character consistent across frames/poses/images?)

sinful shale
#

Question! Is there a channel where I can ask about commissions?

fervent thunder
#

anyone know other discords around AI in general

#

like a central server

pine fulcrum
#

hi

#

does anyone here have Auto GBT, im having trouble installing it

wild steppe
wild steppe
fervent thunder
forest trout
#

Anyone wanna watch the Coachella Livestreams? I'm watching them in the Harmonai discord server if anyone wants to join. Gorillaz is performing rn.

copper crystal
#

nothing to do with each other i think

#

unstable hate stability for taking nsfw images out of the training

#

i think they make all the sexy anime models

#

yeah 1.5 took time to grow momentum is is getting there. sdxl this week i hear. maybe

#

well 2.0 itself had wonky results for some cases.. okay a lot of cases. they investigated and realized 0.9 was a bad setting for the filter and it took out WAY too much

#

they set it something more precise for 2.1, like probably 0.99

merry niche
#

Any idea how I get real photorealistic images? For example, my prompt just now was SDXL beta was: Photorealistic dog sitting in an airplane. The output was more of an animated image instead of a photorealistic image. Does it randomly change to animated vs realistic or how does this work?

copper crystal
#

no no nothing standard. okay so SD was trained on billions of images total set. On a huge cluster of GPUS, like 1000s of them. For 2.0 and 2.1 they used a nsfw filter on the set before training it, so it culled any sort of image that had boobies or butts. none of this is standard cause theyr'e writing the rule books as they go.

#

base 1.5 from runway ML, trained on stability's GPU cluster, is able to do porn before it is even refined at all. not bad either.

halcyon briar
#

Ayo sup guys

copper crystal
#

probably not. they'd get investors either way. billionaires don't give a fuck

#

emad been saying kids and pornography together in the dataset presents huge problems. if the community wants to refine that sort of thing into a model later then its not like stability the company can stop them.

zealous maple
#

Hello, am I able to fine-tune the 2.2 XL model on my own data-set?

copper crystal
#

not yet, and if you had a lot of gpu power. that kind of training power should be available in the cloud if you dont own it

zealous maple
#

I mean can I fine-tune the model they offer via API?
otherwise, I think they havne't released that model, right? Is there a plan to do so?

copper crystal
#

there is a plan to release it. sdxl is still in training is why its only through websites

alpine bear
#

@marble cedar you on a laptop 4070ti?

#

Did you do custom cudnn installs

#

Then your performance is far slower than it should be.

#

It should significantly improve your numbers

rich lintel
#

Does 4090 pulls 600w when rendering images?

alpine bear
#

350 - 400 depending on the image size

rich lintel
#

Is it only benchmarks can 100% use 4090s power?

alpine bear
#

No, typically image generation can fully utilize the 4090 - however you need to have the right cudnn files, python version and nvidia drivers to fully utilize the card.

ember flax
#

I have used Dream Booth to train my girl friend and created a checkpoint based on chilloutmix. Shall I use this custom trained checkpoint for my Textual Inversion as my checkpoint? Or I use a clean chilloutmix checkpoint for my Textual Inversion checkpoint?

vast ingot
# ember flax I have used Dream Booth to train my girl friend and created a checkpoint based o...

what do you plan to train in this TI?
If you plan on training again on your GF to improve quality with double training, it can help a little yes, but rarely the best solution to go for.
In any case, to answer you :
the best quality you will have is by training your TI on the model you intend to use it on. It will work on different models, but it will perform best on the model it has been trained on

elfin pivot
#

Hey has anyone tried SadTalker in automatic 1111 yet ?

real zodiac
elfin pivot
#

dang lol

#

hoping it could do larger

#

I've seen other methods for animating AI character heads

real zodiac
#

for something usable you'd have to run the whole video through controlnet video2video or similar

#

and there are a looot of frames so

obsidian geode
#

the problem with all these talking head research demos is they all use the vox dataset which is stuck at 256x256 grid

#

like wake up this isnt 2010

elfin pivot
#

but man its hard to figure out

real zodiac
#

pretty much, the talking anime head thing only outputs 256 too

#

thin spline-whatever it was called as well

elfin pivot
#

someone suggested live2d

obsidian geode
#

Yes they are all based on thin plate spline model

elfin pivot
real zodiac
#

live2d you'd have to cut out parts, make sure there are no holes and such, it's a bit more advanced

#

could be feasible with the segmentation models

elfin pivot
#

I was looking at Thin Plate Spline Motion Model stuff but most of the tutorials need a video

#

I just wanted to audio to head animation

real zodiac
#

they'd work on a phone recorded video tho bigbrain

obsidian geode
#

For the strongest single image variation from face you have to go thin plate + dreambooth

#

however since its 256x you lose the skin

#

if you dont find good face you go further and further in the anime realm and no realism

#

in stable, if the unclip models were stronger, then you would have a free alternative to midjourney image variations

#

for now, you have to go dreambooth

#

I have tried to apply realistic skin patches using control net and a realism model to get back the skin, it is too slow and would need photo restorations skills

#

the post on reddit are fun until you realize the noses arent the same

real zodiac
#

yeah also the small details like mouth movement tend to not upscale very well with controlnet

elfin pivot
#

got any good videos on how do it ?

obsidian geode
#

getting back the mouth can be done in gigapixel upscaler

#

it is the only feature were it is ok to do so

#

all of this can of course be avoided with a large number of samples face of the same subject, then lora to keep the skin

#

my problem is I have a single face

real zodiac
#

pretty long process for a funny meme

obsidian geode
#

you are trying to control net a meme?

real zodiac
#

nah I'm just saying most people just want to do like a funny short video with a talking character, training loras, upscaling skin and such is a bit too long of a process for this

#

unless your use case is more professional

obsidian geode
#

yes I was talking about making faces

#

not animation

real zodiac
#

we could avoid like most of this with a higher res model tho

#

ah I see

obsidian geode
#

for me the faces are just not good enough

#

and having a realistic skin wont do, I want a concept

#

and I want extreme control over the face

steep topaz
#

hello

meager bear
#

can i make images of my friend to prank him, how to do this?

fleet kayak
#

^^

proud locust
#

Can someone tell me if there is something wrong with Dreambooth? I have been trying to train over and over again today. It generates class images that look nothing like the training pictures and then generates samples that look nothing like what I am training. It's just wasting my time. I upgraded my graphics card and upgraded my Automatic1111 which made me end up having to install torch 2, etc.

fervent summit
#

Newbie here, hi y'all. I have a comic hero, flat art, and I'm trying to get Mage to render it looking photo-realistic with inpainting img->img. It keeps adding it's own hero behind my hero so he gets extra cape and arms that look great while he's still flat. Is there a way to get around this? I tried on DreamStudio to 'copy' to a photo and it just kept adding it's own badges and belt buckles. Imaginative one, that one.

tidal moth
#

hey guys im new to stable diffusion and im having trouble installing it and making it work, can someone please help me?

bleak matrix
#

Good morning, everyone! sunsmile How are we all today?

tidal moth
bleak matrix
tidal moth
bleak matrix
#

You can post a screenshot of the error in #🤝｜tech-support , and you can also search the server for the same issue using the words of your error to see how/if it has been solved

tidal moth
#

if thats ok with you

bleak matrix
#

I just woke up, so I am not really awake at this particular second. But someone should be able to help you in #🤝｜tech-support , and if not, you can always check back a little bit later and ask again.

tidal moth
languid vector
#

hey! fairly new to sd - can someone please briefly explain to me the difference between fp16/32 and full/trained/merged models? i do understand that embeddings and loras are "additions" to existing models but the others i do not get so much. pls help 🙂

vital dragon
#

hey guys, maybe someone knows what can be a reason that I don't havev DPM++ SDE Karras?
It's the only sampling method I'm missing, and it's actually the one that I want to use agony

vital dragon
subtle umbra
#

Hey

#

does stable diffusion have that image extender feature

#

can anyone tell me

#

like dalle where it extends an image

#

based on the previos image

alpine bear
#

it is called outpainting

bleak matrix
subtle umbra
steady grail
#

Konichiwa

bleak matrix
steady grail
#

Uh?

#

Yeah, good morning

#

You got a lot of puppies going on there

bleak matrix
#

Hahaha, yeah, currently puppies are the theme!

#

I love animals in general ❤️

languid vector
vast ingot
#

that's about it yes. fp32 is more precise, and can be converted into FP16 for a lighter file usually too.

#

they take less space, and the difference in precision is very minor in terms of picture changes

#

but to finetune the model, it's usually better to run on higher precision (fp32) if your hardware can manage it

languid vector
rotund zephyr
#

hi

copper crystal
#

i've seen this that precision is lost and that makes sense, but whats the real results? are generations a little fuzzier?

vast ingot
#

no, they feel the same quality to me, but you do see small difference. it's hard to qualify the result as "worse" visually though

#

I mean running the same parameters on just the 2 different models

copper crystal
#

I dont think i coudl go to civit and guess which images were 16 and which were 32. guessing 32 might be better for training?

vast ingot
#

the picture is just different in some small details

vast ingot
#

training benefits a lot more from higher precision

copper crystal
#

this makes sense. i might be understanding!

fervent thunder
#

It depends on sampler you're using mostly...

Too high SFG scale could overbake , sometimes even too long negative prompt makes it weird.
(at least negative imbeddings based on my expirience)

copper crystal
#

keep experimenting. i was going to offer tips but they were immediately argued upon. There's a lot of toxic attitudes that want to argue every tip.

Remember this. nobody is an expert in this field. Not even the PHD doctors have it figured out and they're the most knowledgeable in the field!

Your most reliable form of discovery is going to be experimenting. try an xy grid generation with cfg and step counts if you want some sweet experiments that give a great lens on what works best. Z axis is good to use for different samplers

#

someone saying you can't do that is probably just being negative and wrong.

dim shard
#

hi is it nessary to get the engine list first before every request can i hard code a engine id it that a good idea

fervent thunder
#

Matrix is good way to do it, yea.

I was doing matrix for each new model first time to see how different parameters and samplers affect results

#

hello, i found this server bc i have some questions about stable diffusion, may i ask here or there is a channel for questions?

warm junco
#

If its a General question feel free to ask here

fervent thunder
#

tysmmm

glacial oar
#

Hello

warm junco
#

Hi

valid rain
#

Is deepfloyd supposed to be released this week?

silk nebula
#

3D render of a paper dragon, studio style photography

reef knot
#

What is the best way to achieve "similar image"? Like I upload an image of a woman in a modern living room using laptop on a couch with a small yellow dog, and I want the output to be a brand new image but with same elements and sam description. Something like in Midjourney, using "describe" to describe a picture, then using the description as a prompt for a new image. Thanks!

fervent thunder
#

thinking civi dead again

fleet kayak
#

i was gonna try and find

#

loras

fervent thunder
#

yea I wanted to find new models to play with

fleet kayak
#

fr

#

plus i legit was just setting up my colab

fervent thunder
#

yep, same 😄

fleet kayak
alpine bear
#

wot?

#

you're trolling

#

chill

#

kek means lol, its a laugh

#

kekekekek is laughter

#

tf about the frog, pepe the frog?

#

what?

#

You are either a child who is trolling, or terribly misinformed

#

I am genuinely not sure

#

ㅋㅋㅋ (keukeukeu) is the childish Korean equivalent of the English "haha". Since this is often used in StarCraft matches, Blizzard, Starcraft’s developers, decided to reference it in World of Warcraft: when a player of the Horde faction types "lol" using the /say messaging command, members of the opposing faction see it as "kek".[1]

drowsy garnet
#

Anybody knows what happened about civitai?

alpine bear
#

pepe the frog is an online meme. It was a meme before it was ever used by the right, and it is a meme now.

#

@vast ingot

fervent thunder
drowsy garnet
#

Thanks bro

vast ingot
#

yeaaaaaaaaaah politics isn't at all a topic that, if it's starting to create drama and trigger/troll people, it's not welcome in general.
we are an SD server, not a politics server. DMs if you need, but not in #💬｜general-chat . #1008652627204132904 or #🌶｜off-topic if you can manage to keep it civil, PMs or muting each other if you can't

#

good night Happy

bleak matrix
#

Rest well!

fleet kayak
#

civitai is up lets go PU_pepeCatHug

#

i said that too soon Sadge

#

ok its back

#

im just gonna

fervent thunder
#

thinking dead again

young ivy
#

ok since my home channel of anime didn't speak up, is there anyone in here who hasn't ever had a Nitro Subscription? I have a code for 3 months of free Nitro on Discord, all you gotta do is raise your hand

gusty summit
#

How do you generate photos of old historical figures, like Julius Caesar? It just shows him as a stone sculpture/bust for me

fervent thunder
#

just to be able to upload bigger videos lol

young ivy
#

check DM

fervent thunder
young ivy
#

did you see it?

fervent thunder
#

Where’s the bit

#

Bot channel

past harbor
#

One of the last channels

young ivy
fleet kayak
#

oo damn

#

nvm i cant claim

#

gl to whoever

fervent thunder
#

yea, I couldn't too

soft sparrow
#

is there an AI tool to increase the quality of compressed videoes?

soft sparrow
fervent thunder
#

thinking civi back up...but can't login...eeeeeeeeeeeeh

soft sparrow
#

if yall cant log in then change your passwords

#

because then its likely a scam

fervent thunder
#

I'm talking about civi, it has nothing to do with discord

fleet kayak
fervent thunder
#

thinking You can't scam me if I literally have no idea how to login to my own discord account

fleet kayak
#

I NEED LORAS despair

fervent thunder
#

But civi don't let me loging, button to auth with google doesn't work agony

fleet kayak
#

NOO ITS DOWN AGAIN Sadge

fervent thunder
#

thinking we killed it

young ivy
#

but probably used up by now

fleet kayak
#

not trying to diss sd models but legit 75% of them are just the same thing in either a different style or type of woman Sadge

#

either that or furries

fervent thunder
#

yea, but whatever, it's fine

fleet kayak
fervent thunder
#

That's what always happens when community is huge, even with programming

fleet kayak
#

fr

soft sparrow
young ivy
#

uuuh

#

can't tell if serious

soft sparrow
#

ill just bing it edit: couldnt find anything

young ivy
#

the link to nitro was from a game pass ultimate code offer to whoever hasn't had a nitro before. It runs out or ran out on April 26th and I dunno if it's been used up or not at this point

#

you get these promos at random as an ultimate holder

fleet kayak
burnt lichen
#

GUYS! I'VE GOT IT!

#

Latent Space Averaging!

fleet kayak
#

yes

burnt lichen
#

Instead of packing all your images together into a giant image most rigs can't even process, like https://github.com/LonicaMewinsky/sd-webui-keyframer does, you could just run through all the input images, average them together, then reuse that image's latent space for the whole animation. Even better, Latent Space Import/Export. Would it work as well as a LORA? I have no idea. For all I know, it could. Somebody should get on this.

#

Unless it's already a thing. In which case, download where?

mint lily
#

is a1111 sd2?

#

(I'm looking at controlnet's instructions on stuff and it says "use this model.. or that one if you are at sd2")

vast ingot
# mint lily is a1111 sd2?

no.
A1111 is a tool that uses SD.
SD is a technology, there is no SD2 on this exactly
But "SD" also refers to the SD1.5 and SD2.0 or SD2.1 models
A1111 works with all those types of models
I'm not sure what they are referencing as SD2 in your video

#

there are 2 sets of models for controlnet though

#

controlnet is something that comes "on top" of your model

#

if you use a model 1.X, you need controlnet for 1.X

#

and same for 2.X

mint lily
#

actually, I'm at 1.4 in a1111, so I'm just trying to update my model at the moment.

vast ingot
#

ok so first thing

#

1.5 > 1.4, simply

#

same requirements, same dataset, more refined

mint lily
#

but i was reading controlnet's training page since I'm contemplating on some of the issues I'm having with generation (a skeleton with just one muscle.. but the canny edge detection loses the distinguishment, screwing up my images)

vast ingot
#

then there are 2.X models, those have other requirements

vast ingot
#

you are intending on training a new controlnet you mean ?

#

or just a tutorial on controlnet ?

#

(sorry, training is a keyword around here)

mint lily
#

@vast ingot training.. wanted to, possibly, make it more capable at some tasks with a specifically-trained model

dim shard
#

can i generate 2k image with stability api

#

i am trying to upscale a image but i get the same diminsion size

mild orbit
#

Hey guys i was wondering if theres a way to train non square images yet?

small spear
mild orbit
south granite
#

hello is there a LoRA for making massive structures ? I'm trying to make giant pillars of stone but it's not working.

umbral tulip
#

any sprite sheet models out there already? if not, I assume it's difficult to get it to work well enough?

small spear
small spear
small spear
vapid tangle
#

did u guys check out the new model?