💬｜general-chat

[long talon]

Any linux (ubuntu) room that might be able to troubleshoot a troublesome ROCm Automatic1111 install that connection errors out on?

[native sundial]

if using A1111 you can use AND to dynamically control the prompt: "mountain AND Tunnel, anime style AND realistic, vibrant colors AND pastel colors" for instance

or this: https://github.com/adieyal/sd-dynamic-prompts

[pseudo coyote]

hey guys

Is there a model which generates greetings cards for social media?

[autumn cargo]

this is interesting, thank you! but i wonder if this is exactly what i need because this seems to widen the prompt to multiple alternative words, whereas what i'd want is just for the single word to not be artificially narrowed down by the model. like, i dont want a mountain or a tunnel, i want a mountain every time, i just dont want all of them to look like the exact same mountain. perhaps there is a way to do that with dynamic prompting but it would probably involve writing a script to fetch a long list of words semantically related to each prompt word and cycle through them to sort of artificially inject variation... and im worried that this could diverge quite a lot from the original prompt

[native sundial]

believe there are comfyUI workflows on civitAI that uses ClipVision and Ipadapter to modify the prompts slightly each new generation

never tested as Ip-adapter keep OoM-ing on my system

[autumn cargo]

thats an interesting idea as well!

[autumn cargo]

ok im reading more into dynamic prompts and i think what i said here is already implememted, in the "magic prompting" section

i'll have to try that out

thank you for the suggestions!

[rotund coyote]

bruh how do i cancel a bot request

been like 10 mins still isnt done

[rich dock]

not heard of it, will look it up

[rotund coyote]

its been 3 hours and my image is still generating

how do i cancel it

this is annoying asl

[frosty oriole]

Whats better than 4090 for sd?

[trim magnet]

nvidia A100

[frosty oriole]

it has price of 10x 4090, its as strong?

[trim magnet]

yes specially for training

if u only generate images a 4090 or a 3090 is good enough

[cedar hornet]

hey had to reset my pc anyone got the link for the ai so i can get it again?

[shell hazel]

4090 is the highest you can buy commercially 5000 serries was supposed to come out soon but got delayed til 2025

[white light]

Hey

I'm wondering about background change for

products and food

I see stable diffusion being %99.9 used for humans

[shell hazel]

yes look #🏞｜general-with-images I made food in 1sec

[rotund coyote]

yeah fuck this man I'm js getting ignored

[white light]

No not food

changing background

[shell hazel]

close the the window lmao

[white light]

it's a typical discord thing

[shell hazel]

you can change the background of any image with masking

[rotund coyote]

Yeah its been like 5 hrs still nothing

[white light]

bot request where

this server?

if yes

try leaving the server and return

Also

to get the admins attention

just tag every single one of them

that'll increase the probability of them replying

[rotund coyote]

Oh alr. I just didn't want to be annoying

[noble shell]

hey is it fine to train on outpainted images? I have a lot of images of a subject, but they’re almost never square. What’s better - to train on non-square images or outpaint them? 🤔

[rotund coyote]

@karmic brook Stablebot has been making my image for like 6 hours, how do i cancel it?

[hasty hornet]

That sounds like a ban speedrun strat

[rotund coyote]

nah but dude its saying my image has been pending for 7 hours now

[trim magnet]

just try with another bot theres a lot of bot channels in case one gets stuck

[rotund coyote]

which one?

[trim magnet]

https://discord.com/channels/1002292111942635562/1103708504142925824

https://discord.com/channels/1002292111942635562/1101178553900478464

from 1 to 10 to choose from

[rotund coyote]

same thing man

keeps saying "image still priccessing" with that passive-aggressive ":)"

and theres nothing i can do

gonna use an alt, waiting for someone to help is useless

[trim magnet]

try to leave server then rejoin

[fervent thunder]

I'm you.

[fervent thunder]

and its doing it again

bruh

fix this bruh

next account

[woven oar]

that account would be me

[slender ledge]

Is there a website with good prompts? for my SD local.

[twilit pendant]

https://civitai.com/images/3879801

(image slightly NSFW)

is anyone able to recreate this image in stable diffusion? i literally can't i get something completely different. I use the copy generation data, but it doesn't work. I also downloaded the lora lora:adapted_model_converted:0.7. and the embeddings (negative_hand-neg, EasyNegativeV2, NegfeetV2)

Used the exact resolution of the image in civitai, as well as 512x768. Nothing works, Please help

[distant tangle]

What the best models that its match or almost close with midjourney 6

[torn quiver]

Just use the same prompt or seed

[cedar salmon]

3 new Nvidia cards revealed Jan 8th, 4070 Super. 4070 Ti Super, 4080 Super, i wonder

[spice wyvern]

Hi guys. Any LoRa gurus online?

I'm StRuGglINg

4070ti and it's on 123 minutes... Can't figure what I'm doing wrong . Kohya, installed and running but. . Dang.

[shell hazel]

I have a 4070ti and I can do a batch of 700 images in 30mins Something is wrong.

[hasty hornet]

he's talking about training I think

[shell hazel]

ohhhh maybe needs

[median arch]

what's the point of using hires fix?

it's much better to generate many images and pick one for upscaling instead of upscale everything with hires fix :/

[shell hazel]

maybe img2img batches

[odd geyser]

hey all, just wondering what are the settings for the bot ? like how many steps and CFG scale ?

[white light]

Well it worked when I did it before

It was wholesome too

Like

A family gathering of sorts

[astral pivot]

can someone tell me how much these two are in the 5e-4 type of format?
0.0001
0.0003

[marsh vigil]

0.0001 = 1e-4
0.0003 = 3e-4

[sick cloak]

Hey there! I'm on a 2017 iMac - ( Processor: 3.8 GHz Quad-Core Intel Core i5 580. Graphics: Radeon Pro 580 8 GB. Memory: 40 GB 2400 MHz DDR4. Ventura 13.5.2 ) and would like to try Stable Diffusion. Is it even possible with these humble specs? If not, what would be the minimum requirement - a specific external GPU..? Thanks for any thoughts!

[native sundial]

https://www.pcguide.com/apps/how-to/how-to-run-and-install-stable-diffusion-on-mac/

[torn quiver]

4 hours late

But happy new year 🥳

[vagrant kernel]

anyone knows a good model to remake my selfie and look like its animated/drawed?

[torn quiver]

like pencil on paper?

[vagrant kernel]

cant upload images here, would send a concept

[torn quiver]

#🏞｜general-with-images

[warm junco]

Its possible but would be slow.

[torn quiver]

💀

[sick cloak]

Thanks! 🙂

[sick cloak]

Thanks - defo worth a try!

[native sundial]

Happy new year to those from the Netherlands and neighbors

[brave jetty]

Happy new year 2024 to all! 🎉🎉🎉

[odd oxide]

yo, does anyone want to try some thing kind of crazy that I just made? It's like an interactive experience

(not selling anything lol - solo dev)

[cloud elbow]

hi

[rich bolt]

How do you predict if you will be able to render a file in advance of starting the render? Or was this some other error? I rendered a 35 second video. Then I went to sleep and left my computer rendering a 45 second video and woke up to a 'can't allocate 15gb of ram' error... after it processed the entire thing, but couldn't save it. ComfyUI video vhs saver node.

[wooden tusk]

does anyone know some sort of plug in for chat gpt for discord

[shell hazel]

You can't predict only trial and error like you just did. My max is 700 batches for LCM and only 300 batches for non LCM. I have a 4070ti and anyone with that or better should get the same.

Best to start at 150 then move to 300 and so on and see what you get capped at.

35 sec video at 60fps is 2,100 images. So that would be why it failed, no one can do that many

[rich bolt]

35sec at 15fps I think, that worked. I have a 4090

Thanks

[unborn matrix]

I add these arguments with the previously mentioned ones or just these?

[tame fern]

Happy new year!

[bronze steeple]

Happy new year 2024!,I'm an AI painting illustrator form Chengdu,where is known by panda,wicked spicy cuisine and also FISU!!!Actually I like the world's thousands of ethnic group's culture,and I have made these diagrams to promote mutual learning among civilizations. I hope that a community with a shared future for mankind will become a reality. I am using exquisite AI skills to express the soul of the world's every ethnic groups, and interpreting the song of the four directions of the earth with the civilization passed down from generation to generation. Hope you guys having a nice day!

[bronze steeple]

😀

[sweet ore]

Happy new year!

[tulip forge]

hi guys can anyone help me with stable diffusion i've been trying to do the controlnet qrcode thing but controlnet doesnt seem to affect the output image at all

[eternal oyster]

Hello, i wanted to ask if Stable Diffusion Can check Images For Detect NSFW?

[eternal oyster]

My favourite Subnet Mask yeeeeeee

Anyway Happy New Year

[sick charm]

Hey anyone here know who I give a code fix to?

I can't find the file on Github to fix it, but I fixed it..

sdxl_pipeline.py has errors in it.. so I got rid of them.. anyone want it? 😛

[warm junco]

Just these

[hasty hornet]

there are some extentions out there

https://github.com/jovijovi/stable-diffusion-webui-nsfw-filter

[sleek vapor]

Hey, a question

I'm tryna train SDV2 and one of the steps saying each instance of picture should have a unique identifier

Like this: https://i.imgur.com/d2lD3rz.jpeg

Tho my question is, for example I want to train a load of icons, but they are different icons and don't represent the same thing or meaning, but they all have the similar style.

Does this mean I should use same identifier for all of them while naming it or each should have their own identifier as name based on their meaning?

[eternal oyster]

i need to send a image to stable diffusion and get a response if is nsfw

this is a filter not a checker

[hasty hornet]

there are plenty of different libraries / plugins / extentions...

https://github.com/s0md3v/ifnude

[stark hornet]

is there work already on the next version of the base model ? like now that there's midjourney 6, will there be a big update here soon, too ?

[buoyant moss]

I've run like 200 thousand prompts thru Google Gemini API, it can tell you if a prompt is sexual or racist

[bold crow]

Does anybody know how to add DwPose into stable diffusion?

I've been trying and it's not working. I've gone to extension, install from url, and then pasted in the correct url and it comes up with a lot of text once installed and doesn't seem to work.

[hardy dove]

It is possible to sell an image made with SD XL?

[eternal oyster]

How can i execute this commandline "--no-half"?

[shell hazel]

right click the .bat > Edit > add the line to the end

[eternal oyster]

Thanks

[warm junco]

What's your GPU ?

[eternal oyster]

1660 Super

[warm junco]

Make sure your webui is updated

[warm junco]

Okay then also add --medvram --xformers

[eternal oyster]

ok

thanks

[warm junco]

That helps to improve the performance

Np

[inland egret]

Hello lads! I just wanted to ask something, do you happen to know if it is possible to create posters for an event on sdxl? I could use a hand to make a quality one, or even some tips would be fine. Thank you!

[lean sequoia]

Hey all. Happy New Year. I'm building a new system to run Stable Diffusion. I plan to use an ASUS WRX80E-SAGE SE AMD Motherboard with 2 x ASUS ROG Strix GeForce RTX 4090 GPUs. Running on Windows 11. Basically, I'd like to know if I'll be able to take advantage of the 2 GPUs in Stable Diffusion and will I see an increase in performance for generating workflows using animateDiff? Or am I just throwing my money away? Any suggestions appreciated. Thanks. PS... I'm very new to this community so if this is the wrong place to ask, please let me know.

[marsh vigil]

i'm also curious if two GPUs improve matters. i'm also curious how you power something like that 😅

it does seem like dual GPUs currently aren't worth it. not because the work doesn't suit them, but because software support just isn't there. you might be better off building two smaller machines and distributing work to them over the network

this is solely based on me reading up on the state of things in the last five minutes (!)

[warm junco]

Hey, that would be a sick setup.
You have to know that you can't combine two GPUs for one workflow (image)
But each GPU can run their own Webui instance so you can generate simultaneously

Stableswarm ui is also good for using multiple GPUs I heard

[lean sequoia]

Hey CS1o. Thanks so much for the reply. Using 2 GPUs for the same workflow was exactly what I was curios about. I'm currently running a single RTX2080 and things are sloooww!!! Also I get "out of memory" when trying to run workflows with animateDiff and a decent number of output frames. You saved me from a big disappointment.

[lean sequoia]

This is a good idea -> building two smaller machines and distributing work to them over the network. Going to look into that.

[warm junco]

make sure you use --xformers OR --opt-sdp-attention and you also should add --medvram-sdxl for better performance in animatedDiff and for sdxl in general

[lean sequoia]

OK. I did not know that. This is great information. Thank you.

[warm junco]

i would say first try --medvram-sdxl --xformers
and if you get any errors with animated dif, then use --opt-sdp-attention instead of xformers

you have to add these args into the webui-user.bat at the line COMMANDLINE_ARGS=
then save and restart

and no problem, for any other tech question feel free to ask in #🤝｜tech-support

[lean sequoia]

I figured that is where they go but thanks for clarifying. lol I was debating between this chat and tech-support. Thanks again!!!!

[warm junco]

no problem, if you havent used the commands before you will be shocked how fast it is now xD

i mean normal generation time and sdxl generation time

animated diff can be a slow experience still, but should be faster too

[warm junco]

dw-openpose is already included in the latest controlnet version, no need to install anything other than controlnet, and of course downloading the openpose controlnet model

[marble steppe]

will SDXL run on a Linux machine with 3 AMD cards?

[tawny surge]

Wondering where the best place for me to run SD is; I have an Apple M1 Max laptop with 64GB shared memory and a Windows gaming PC with a 3080 with 10GB VRAM, what are the tradeoffs of running SD on one vs the other? Would the M1 be better for training with all that memory and the 3080 be better for generating with the faster GPU?

[marsh vigil]

is --medvram-sdxl necessary on a 4090? i'm not sure if you were mentioning that in regards to the RTX2080. i'd think with a 4090 (~24gb), you'd want to use as much memory as humanly possible 🙂

[rancid sierra]

Hello there, i am teck savy guy and i am really interested in the stable difusion project, i downloaded python 3.10.6 and the SD git, and had many set backs, now after a whole day of trial an error i managed to get it to start working but it's taking almost an hour, can someone help me ?

am using a 1650ti 4go,
i had an issue with the batch saying i had both cpu and gpu selected (cuda) thing so i added this line of code:
self.logvar = self.logvar.to(self.device)
than i got an other error and the ui suggested adding --disable-nan-check
that what my batch file look like now:
--xformers --autolaunch --medvram --no-half --disable-nan-check

[warm junco]

Hey, nope not necessary for more than 11gb vram

[warm junco]

Hey, remove --disable-nan-check
And make sure the model you load is 1.98gb or larger

[rancid sierra]

it just finished after 45min it's just a black square

[warm junco]

Come to #🤝｜tech-support
Restart the webui-user.bat and make screenshot of the cmd and post it there

[rancid sierra]

ty will do

[rotund coral]

Happy New Year everyone!

[neon bridge]

hey hi ❤️ happy new year for all of you!! hope u had a nice night and wish u a lovely 2024!!

sorry for bother u all, can anyone help me to how install stable diffusion to mi AMD computer??

[warm junco]

hey, your 3080 would be the much better option, faster and better supported, for training idk, for mac stuff you should ask viking in #🤝｜tech-support

[warm junco]

hey, sure i can give you my Install Guide in #🤝｜tech-support

[neon bridge]

thanks a lot, sorry i mistake the channel

[neon bridge]

i saw a tutorial on the official AMD youtube but i dont understand a few things

[warm junco]

dont follow that xD, onnx and olive isnt well supported rn

[neon bridge]

ohh is this a bad vid??
https://www.youtube.com/watch?v=kw0WT5sDBIY

how should i install stable diffusion on my pc if i have AMD graphics

[warm junco]

give me a min, i post my guide in #🤝｜tech-support

[neon bridge]

thanks ❤️

[warm junco]

Here you go: #🤝｜tech-support message

[neon bridge]

thanks a lot!! look a bit hard hehe i will give it a try!

[warm junco]

its nearly the same like the AMD one of YT, but will work better

[misty zealot]

hey i was wondering how can i prompt stable diffusion to give the same style of image with little variation

[pale latch]

image prompt adapters

[misty zealot]

what does that exactly mean?

[pale latch]

https://stable-diffusion-art.com/image-prompt/

[misty zealot]

thanks

[hardy dove]

If im using revAnimated as checkpoint should I use any refiner?

How can I polish the eyes so that they are well done?

[snow carbon]

was the emad ama posted anywhere?

[odd oxide]

Hey, right now I'm creating a stable diffusion based text to video app. If anyone thinks that they have quite a bit of experience in stable diffusion image prompting, shoot me a DM or @ me here!! Looking to work with some people

[lean sequoia]

Oh my. You were right!!! So much faster with the command switches. Thanks 👍

[static idol]

anyone know how extra spaces or new lines in prompts are handled?

i've got a lot of { | } randomization in my prompts and it isn't easily human-readable without a lot of spaces

[umbral crater]

Hey y'all, new to this discord, would this be a good place to ask a few questions about the diffusers library?

[lean sequoia]

When we discuss Stable Diffusion in this channel, are we expected to discuss only Stable Diffusion running the Automatic1111 front end? Or are discussions about other front end UI such as ComfyUI acceptable? Thanks. I find myself bouncing back and forth between the two.

[shell hazel]

its of a mix of everything, discord bot, automatic1111, and comfyUI. Even other lesser known like StableSwarm

[pale latch]

fooocus

[golden knoll]

is there a discord bot that can use the inpainting tool that midjourney uses for theirs? If not, how does discord call that tool, because I can't find it in their documentation.

[fervent thunder]

I really wish I could add this bot to my server

[slow lake]

Hey everyone, can you guys tell me what's the best value to stop NSFW pictures coming from SD Models?
Like what's the best adjustment value?

[inland egret]

guys I was looking at a photo on civitai and I see "30 nodes" in workflow, what does it refer to?

[sudden ruin]

ComfyUI probably

[sudden ruin]

putting it into the negative prompt as well as a few other keywords

[sudden ruin]

Probably not gonna happen, since its running on their GPUs and they cant let you run that for free. Also why do you need it on a private server

[inland egret]

guys for you using a refiner during the generation is important?

[plush knot]

hi

[sudden ruin]

Please dont advertise your own messages in other channels

[sudden ruin]

Depends on the model, your steps, samplers and so on

Most XL finetunes dont need a refiner

[inland egret]

ok thank you!

[bronze steeple]

No,I'm not advertising somthing,I just wanna show my genuine and pure love to every ethnic cultures from all over the world.The world is only wonderful when it comes to diversity.

I love this Earth,I cherish every parts of the world's culture.

My purpose for making this intricate pictures isn't "advertising" something!Please give me more excuse,and I'll do more better!

[sudden ruin]

Sending your images in one channel is enough

Also you literally said "subscribe me", what is that if not advertising

Anyways my point stands, please refrain from doing so in the future

[bronze steeple]

Oh,that's my error

[reef fiber]

After install Reactor Stable Diffusion is not starting anymore. Somabody can help ?

It says in prompt

File "D:\Deep_AIGeneration\stable-diffusion-webui\extensions\sd-webui-reactor\scripts\reactor_faceswap.py", line 414, in ui
with ui_components.InputAccordion(False, label=f"{app_title}") as enable:
AttributeError: module 'modules.ui_components' has no attribute 'InputAccordion'

[sudden ruin]

#🤝｜tech-support Also did you update your webUI

[reef fiber]

yes, i did. I also disabled roop.

it says version: v1.4.0

[eternal oyster]

Can i train a stable diffusion model with custom image?

[reef fiber]

my fault, it was not update, now its v1.7.0

[eternal oyster]

And how

[lean sequoia]

Got it. Thanks.

[bleak matrix]

Good morning, everyone! How are you today?

[sudden ruin]

works now?

[reef fiber]

Yes , it does. Latest version seems to have the attribute. Thank you.

[zealous hamlet]

[Help] 🥹besides PIKA,and PlaiDay.. are there other AI video generators on discord ?

[shell hazel]

good video takes way more than a bot could ever do

[reef fiber]

is there any successor for DeepFace ?

[zealous hamlet]

check out spaces on huggingface

[fervent thunder]

hey ive got a problem

look in #🏞｜general-with-images

[clever swallow]

Hallo, does anyone here know much about the watermarking? Does it auto mark it like an artists

[fervent thunder]

why the format automatically turns into Video? how to do picture format?

[warm junco]

With format:image

[buoyant moss]

I wonder if there are any plans for SDXL 1.1

For example trained using v-prediction and 0-SNR

Of course, I wouldn't expect anything from Stability for free

And they should def focus on money making first

[inland egret]

Hi guys, I just wanted to ask you something. Why does it give me an error on xl every time I try to add 1.5 embeddings? PS: I'm using kaggle

[reef fiber]

I get continously this error message: UnboundLocalError: local variable 'h' referenced before assignment

Is this a known issue ?

File "D:\Deep_AIGeneration\stable-diffusion-webui\repositories\k-diffusion\k_diffusion\sampling.py", line 651, in sample_dpmpp_2m_sde
h_last = h
UnboundLocalError: local variable 'h' referenced before assignment

changed the sampling method, which fixed it

[torpid ridge]

hi all, i'm from Martinique i'm searching for animated studios can anyone help ?

i have all the character designs already done and the storyline too

i want to put it in anime i already have the manga set up

[unreal path]

Is there a way to default to image format instead of video without including it in every request

[maiden mortar]

So do these windows every go away

No many how many times I go to start here it shows me the same thing over and if I back out this room the rules are on every screen what am I doing wrong?

[sudden ruin]

Cause they are made for 1.5 and not XL

[sudden ruin]

Nope, not that I know of

[sudden ruin]

Accept the rules at the bottom, might need to scroll down

[sour needle]

Stable diffusion in colab is working again!

[fervent thunder]

anyone that can help e

[maiden mortar]

It worked maybe it was on a delay

[pale latch]

the one that never goes away for me are the nitro popup ads

i use discord in a browser window just ebcause of how much spam the desktop app delivers

its only a matter of time before discord has to show their investors they can serioulsy monetize the platform. that means ads in chatrooms.

enshitification is on it's way. the community leaders ought to be planning for a migration somewhere else because discord WILL be enshitified. it's their only possible destiny

[limpid timber]

irc bot when

[native sundial]

Vencord with FakeNitro to keep at least Nitro popups at bay while getting some nitro benefits (like max res streaming and highest quality audio) 🙂

doesn't give all nitro stuff but the most used functions it works with

[minor smelt]

hello all, not sure if this is the right channel .. but wondering if someone can point me to how to store models on a network share and reference them from Auto1111? I tried creating a shortcut, but that didn't seem to work. I also tried the --ckpt-dir but that didn't seem to work either

[limpid timber]

fun and games until discord starts putting some magic word metadata into messages and banning people without itt-

[native sundial]

google on how to make symlinks .. and have fun with all the issues that comes with those

[minor smelt]

on Windows? I know how to use symlinks, but have only ever used them on Linux

[native sundial]

same principle then 🙂

[native sundial]

how would they add metadata to text?
any methods is a federal lawsuit for breaching every single privacy law within minutes upon discovery

[wind bolt]

https://learn.microsoft.com/en-us/windows-server/administration/windows-commands/mklink might need to run from an elevated command prompt

[minor smelt]

thank you. I think I got it working!

really appreciate the assist

[limpid timber]

not really metadata, sorry for bad wording

[limpid timber]

basically just some random magic string sent along with the message in hopes of catching unupdated fake discord agents

that´d probably also catch unupdated true discord agents idk

i am sick and my brain isn´t braining 😦

[minor smelt]

any discussion on current GPUs price/value for Stable Diffusion?

[warm junco]

cheap and much vram = 3060 12gb

[mellow meteor]

[smoky igloo]

hey guys, just wondering when i do generate images using models from hugging face the resulting images are not high quality and pale, but here in bots channels and other places I see people produce a high-quality images from sd
can you please tell me why?

---

i used in hugging face the sdxl and dsxl turbo , and the results were came out with a grey background and with near low-quality results (missing foot or face)

[sand holly]

Does anybody know a webui which is good for SDXL Turbo?

Like where you enter a prompt and it automatically generates a new image

because having it keep generating images seems highly inefficient to me

[buoyant moss]

SSHFS-Win

Mount the network share, and access it as you usually will

[hardy sierra]

Hi, does anyone know how I can use more power from my graphics card to generate in comfyui. it seems to only use 20% to 30% or less and I have a 4060ti 16gb.

[sand holly]

have you set the powerplan to "maximum performance" in your nvidia control panel?

alternatively you could use something like msi afterburner or nvidia ansel to get more out of it

i have the same gpu btw

[hardy sierra]

can i send u a dm?

[sand holly]

sure

[trim magnet]

thats just normal even my 1060 only goes up to 50% its because its not using the 3D capabilities just the cuda and memory

[cinder heron]

does anyone know how to instill Stable Diffusion Version 2

[shadow bear]

Hey where can i get a desktop version of this? ive seen lots of diffrent videos that have random websites .. is there an offical link?

[sand holly]

depends on what you wanna do

[sand holly]

look on Huggingface

[cinder heron]

Ok

[topaz parcel]

Do you have any recommendations for managing downloaded models? I am starting to run out of space. How do you guys manage your models? I don't want to keep deleting them and downloading them every time. So is there a 'best practice/process' to store them so they are always available to SD?

[shell hazel]

you buy more hard drives LMAO

my model folder is 250gb so far

[warm junco]

If you use auto1111 you can add --ckpt-dir "D:/Path/to/Models/"
In your webui-user.bat and move the models to the other drive.
You can also use the Link Shell Extension to create a Symlink in the models folder and your outsourced models.

[warm junco]

SD 2.1 is just a model you can download and use in any local webui

[warm junco]

For an install guide check the pinned tabs in #🤝｜tech-support

[dire hound]

Hello, I'm seeking clarification on a specific point that I haven't found a definitive answer to: Are the rate limits applied collectively to the entire account, or are they individually applied to each API key associated with that account? I am thinking of a limit of 150 requests for every 10 seconds.

[short basin]

Hi guys and girls, I'm new here, I'm going to buy new GPU and my choice now is between RX6800 XT and RTX 4060TI 16GB, as daily linux user I'm more likely buy AMD card but I want to know what experience do you have with these cards to run SD?

[high wigeon]

Hello there! I'm curious about the cool image-making magic the bot uses – which stable diffusion model is it? Have there been any other models before this one? I stumbled upon a dataset called DiffusionDB with prompts and images that apparently came from this Discord server. I'd love to know which model or models were behind creating those images! Thanks😊

[sudden ruin]

As far as I know the bot currently runs a version of SDXL, but probably not the official release one, dont quote me on that tho
Other models before are SD 1.4, SD 1.5, SD 2.0, SD 2.1 and the Turbo Versions

[high wigeon]

Thank you! I noticed that some of the images in the database were generated in August 2022, so they might have been created using an older model, possibly version 1.5 or similar. Do you believe these prompts are still applicable for newer diffusion models? Thanks!

[sudden ruin]

Yeah prompting has changed but you can still run older models locally for example

[topaz parcel]

Perfect - I'll try that, and store the models on an external drive. And see if that works.

[topaz parcel]

So do you keep all your models on external drive/s ?

[high wigeon]

Is there any information available about the specific model that the bot was using in August 2022?🙏

[fervent thunder]

when will we have AI movies

[languid plover]

what does it mean if write specific word in bracket while you are writing prompt ?

[high wigeon]

I saw a short course called ChatGPT Prompt Engineering for Developers by Deeplearning.ai and they suggest a tactic of using delimiters to clearly indicate distinct parts of the input
Delimiters can be anything like: ''', """, < >, <tag> </tag>, :

for example:
text = f"""
You should express what you want a model to do by \
providing instructions that are as clear and \
specific as you can possibly make them. \
This will guide the model towards the desired output, \
and reduce the chances of receiving irrelevant \
or incorrect responses. Don't confuse writing a \
clear prompt with writing a short prompt. \
In many cases, longer prompts provide more clarity \
and context for the model, which can lead to \
more detailed and relevant outputs.
"""
prompt = f"""
Summarize the text delimited by triple backticks \
into a single sentence.
'''{text}'''
"""

https://www.deeplearning.ai/short-courses/chatgpt-prompt-engineering-for-developers/

[languid plover]

thanks bro,

[sharp siren]

Hey guys,
How can I create my own AI influencer? Looking for tips and guidance!

[earnest depot]

Hello, i have 2 questions about batch in img2img 1: why every image is overwritten and only the last is saved? 2: how to have same filename output as original input? Is a manual solution even to this?

[warm junco]

2022 ? that can only be SD 1.3 or SD 1.4
The very old models can be found here:
https://huggingface.co/CompVis

[warm junco]

depends on the webui your using. in auto1111 round brackets are used to enhance the words strenght

[languid plover]

I am generating image
In this server

[warm junco]

ok, then i dont think it will do anything

[languid plover]

What about
Leonardo ai
Playground

[warm junco]

idk, you would have to ask the team of them what they implemented

i think they support brackets

[rich ravine]

guys can i use stable-diffusion api in my site free?

[broken bloom]

Certainly not

[rich ravine]

what about with money?

[high wigeon]

Thank you!
Do you believe these prompts would be applicable for research on newer versions as well?

[broken bloom]

Yes, you can i have sdxl api

almost with 200000+
Image generation

[warm junco]

which prompts?

[rich ravine]

how much is that?

[broken bloom]

Different provider cost different amount

I have an api I can give you for free I don't use it for most of the time

@rich ravine are you a developer?

[rich ravine]

yes

[rich ravine]

can you give me for free?

pls DM me

[broken bloom]

You can check it here my api don't have any security you can get the API code in 'files' https://huggingface.co/spaces/FumesAI/FAST-STABLE-DIFFUSION-XL-DEMO_SDXL

If you code in other languages except javascript I can give you python code also

[rich ravine]

thanks bro ❤️

[broken bloom]

Welcome

[cerulean rivet]

Does anyone know when we will be able to play with Nvidia perfusion to train AI models?

[limber saddle]

anybody know how to stop my prompts from being made into videos by the bots? I'm only interesting in images.

[limber saddle]

use /format image

[fervent thunder]

Any articles on inpainting at full resolution programmatically? I've made an implementation of my own inpainting masks separately but the results are too inconsistent with the rest of the image. Any hints in the right direction much appreciated.

[dawn tendon]

good morning guys, does anyone know how to insert a specific image in a specific point through inpaint (even in a different way)?

[inner lark]

Please tell me, is it possible to view the version of the selected node directly from comfyUi?

[sudden ruin]

I dont think so

[grave quest]

Does anyone know a good comfy ui workflow for regional prompting similar to what we have with latent couple in the webui?

[karmic radish]

Hello @iron pawn -- did you ever get to the bottom of this? I am hitting the same error when calling the image-image endpoint -- the same code previousley ran fine

[near crag]

and how to redraw pictures

[zinc sun]

Hi, how can i tell stable diffusion to show a men standing and not only the chest ?

to zoom out

[native sundial]

prompt shoes 😄 usually works

[zinc sun]

But now, half of the images are crop to show only the shoes to the hips

[native sundial]

what is your prompt, also #📝｜prompting-help

[honest radish]

hi guys, just downloaded a model but cant run it.What is the problem?

[native sundial]

what model...and what errors are shown in your command prompt

[honest radish]

juggernaut model , shared.opts.onchange("sd_model_checkpoint", wrap_queued_call(lambda: sd_models.reload_model_weights()), call=False)
File "D:\webui\modules\sd_models.py", line 764, in reload_model_weights
sd_model = reuse_model_from_already_loaded(sd_model, checkpoint_info, timer)
File "D:\webui\modules\sd_models.py", line 713, in reuse_model_from_already_loaded
send_model_to_cpu(sd_model)
File "D:\webui\modules\sd_models.py", line 573, in send_model_to_cpu
if m.lowvram:
AttributeError: 'NoneType' object has no attribute 'lowvram'

[native sundial]

https://civitai.com/models/46422/juggernaut this one?

[honest radish]

https://civitai.com/models/133005/juggernaut-xl

this

[native sundial]

What are your computer specs?

[honest radish]

intel r core i5 8400 cpu 2.80 ghz 2808 mhz , 4 core, 4 logical core , nvidia g force gtx 1060 6gb

[native sundial]

Do other SDXL models work fine?

believe you are at the absolute minimum and need to use some launch parameters and might be limited to use SD1.5 models due to lack of VRAM

[honest radish]

didint try another model yet. What can i do about launch parameters

[native sundial]

ask in #🤝｜tech-support as i am not certain which would be needed

[warm junco]

--medvram --xformers --no-half-vae

[honest radish]

how can i change my launch parameters

[warm junco]

right click and edit the webui-user.bat
at the line COMMANDLINE_ARGS=
you add the arguments

[shut frost]

Is there a tool that would help me improve my prompts? maybe a guide book, or even beter a prompt generator ai

[native sundial]

Put the most important thing of the image at the front of your prompt and each following thing after it, is usually the rule of thumb

[native sundial]

https://stable-diffusion-art.com/prompt-guide/

[primal arch]

https://twitter.com/almmaasoglu/status/1742634325957390461
dang this is crazy. do we have anything similar for sd?

[native sundial]

NCNN upscaler

as used by: https://www.upscayl.org/

MJ is just running it at their million dollar cloudservers

pretty sure tons of the innovations by MJ are just existing comfyUI nodes running on enterprise level cloud servers etc

gatekeeping anyone to actually see the processes

[primal arch]

thanks!

yeah do we know if mj is using any open source models in any kind related to sd models?

[sonic compass]

can i upload picture and make it to a movie?

[native sundial]

try https://pika.art/

completely free and takes maybe 5 minutes for 3 seconds, which you can extend with 4

[sonic compass]

ty 😄

[dim garden]

Okay sorry if this is a dumb question or has been asked 100 times but I am at a loss here. I have been watching videos on using Loras and have followed said videos to the letter but where they show the Loras showing Up in the LORA tab mine is still blank

So is this common or is something in mine broken? I place them in the right folder but still blank

Sorry see this is in wrong spot

[pale latch]

You need to refresh the lora list after you copy some into the folder, for automatic1111

comfy i think you need to refresh the page

[obtuse vault]

i've used chainner for upscaling so this might sound dumb
when using hires fix
do i have to use it when generating every image or can i apply it after generation

[finite cloak]

no #✍🏼｜rules-and-tos rule 4

[bright ridge]

Hi

can we use lora for img2img generation?

#💬｜general-chat

#🤝｜tech-support

[honest radish]

------------------------------ --------- 2.0/2.6 GB 11.7 MB/s eta 0:00:56ERROR: Could not install packages due to an OSError: [Errno 28] No space left on device

 ------------------------------ --------- 2.0/2.6 GB 11.7 MB/s eta 0:00:56

Traceback (most recent call last):
File "C:\Users\erenu\OneDrive\Masaüstü\Yeni klasör (4)\webui\launch.py", line 48, in <module>
main()
File "C:\Users\erenu\OneDrive\Masaüstü\Yeni klasör (4)\webui\launch.py", line 39, in main
prepare_environment()
File "C:\Users\erenu\OneDrive\Masaüstü\Yeni klasör (4)\webui\modules\launch_utils.py", line 378, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
File "C:\Users\erenu\OneDrive\Masaüstü\Yeni klasör (4)\webui\modules\launch_utils.py", line 116, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "C:\Users\erenu\OneDrive\Masaüstü\Yeni klasör (4)\system\python\python.exe" -m pip install torch==2.0.1 torchvision==0.15.2 --extra-index-url https://download.pytorch.org/whl/cu118
Error code: 1
Press any key to continue . . .

i am getting this error during download.What can i do

[tidal salmon]

hi guys, can someone tell me, are there some differents in quality between using discord bot, and using sd locally?

[warm junco]

hey, yes, the discord bot can only use basic prompts, no extensions, no upscaling, loras or community made models.
But if you compare only the base SDXL model localy with the bot here boath on 1024x1024 you dont have a quality difference

[bright ridge]

@warm junco any idea?

[tidal salmon]

gotcha, thanks

[warm junco]

yes we can

[bright ridge]

may know how, any article?

[warm junco]

do you use a local webui?

[bright ridge]

yes

[warm junco]

then its just like using the lora in txt2img

prompt for something, set the denois to 0.5 or 0.7 if yo u want massive changes. then add the lora with the trigger words and generate

[bright ridge]

thanks let me check

[bright ridge]

@warm junco kindly check dm

[junior dirge]

has anyone else encountered this issue?

[ripe cradle]

Hi, a friend has asked me if I can help create potential baby images from two parent images. I was wondering if anyone could offer any pointers on how to do this?

[honest radish]

trying to download a model but it says there is no space left.Cant i use my D disk for models ?

[warm junco]

yes you can, by adding a command in the webui-user.bat

come to #🤝｜tech-support if you want to know how

[timber raft]

should I get a 4070ti or second hand 3090?

[trim magnet]

3090

[marsh vigil]

what's the price difference?

[flat flare]

any idea how come on last step of image generation the image becomes screwed up?

xD

like some filter that goes over and makes everything layered with random color and overly sharp

[marsh vigil]

@flat flare #🏞｜general-with-images message

[flat flare]

ty

[marsh vigil]

(that might not be the right answer!)

[flat flare]

i tried in tech support

xD

[iron orchid]

model diffs are just insane

[iron orchid]

isnt the 4070 ti super rumored to be 16gb vram?

[void wolf]

Hi, i'm a beginner in SD, i try to create some picture with my head but i can't do it.

Where can i find a good tutorial. (maybe I don't have the good version of SD)

[iron orchid]

im also learning

but what i did was first search up youtube/text guides on all the basic functions of the UI you are using

etc a1111

and then start looking up other's works with full workflow and begin immitating

then start playing around with settings to understand it more

[void wolf]

I watch youtube tutorial but that doesn't work on my computer. I create a cpkt model with collab google. but when i want to use it on SD, it doesn't want to change the checkpoint

[radiant acorn]

Hi there,
Is the limit "This API is rate-limited to 150 requests every 10 seconds." connected to 1 account or 1 API key on the account? In other words, As I can generate 10 API keys on my account, can I have 1500 requests via 10 api keys from 1 account?"

[ornate flame]

PLEASE someone tell me that SD3 is coming out tomorrow

[topaz parcel]

On my Lora tab, I only see Loras that are compatible with the current main model I have selected. E.g., If I have an SD1.5 model selected, I will not see any of the XL Loras I have installed. Maybe it is that?

[pale latch]

that SD3 is coming out tomorrow

[ornate flame]

Thank you, now I can be content

[pale latch]

ok! just to be clear though, nothing is releasing tommorrow

[ornate flame]

No, too late for that. If it doesn't release tomorrow, I will sue you

[warm junco]

whats your version of SD? (webui)

[void wolf]

SD 1.7.0

[pale latch]

the newest version of automatic1111's webui is what you mean. Stable diffusion is a seperate code base and the model weights themselves. . yeah it can get tricky and tedious. Open Source nameing conventions! Every package has a brand!

[void wolf]

When i go on the automatic1111 page, it's install the 1.7.0 version. I can't find another version

[warm junco]

yea thats the latest webui version, what model are you using?

[void wolf]

I used model found on civitai EpicRealism to creat my model with my own photo

[warm junco]

ah so you trained a model or a lora?

[void wolf]

I want to make some picture with my face and i don't know if i need a model or a lora (I'm a beginner). I try to made one with dreambooth

[warm junco]

ah okay, yea then you would need to merge it with an other model

[void wolf]

Maybe. Is it another solution to make picture with my face?

[pale latch]

you can use an extension called "reactor" to do quick and dirty face swaps

[native sundial]

or this: https://www.youtube.com/watch?v=1mEggRgRgfg

old but might be a bit easier for beginners

[pale latch]

uses old versions of software too

[native sundial]

as long as the methods still work 🙂

[boreal basalt]

Can auto1111 or any UI run on amd GPUs, I found a git rep which is supposed to work on amd GPUs, but when running it in CMD I get a message that no nvidia drivers where detected and by the time it takes to make an image it seems like it is running on my CPU

[native sundial]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs

[boreal basalt]

Yea that's the guide that I followed but as I said I get the message in CMD and the amount of time it takes is a lot, also I can only generate in 512x512 if I try higher like 1024x1024 it just doesn't start, my GPU is RX6600

[warm junco]

you should follow my install guide for amd here: (with performance settings)
#🤝｜tech-support message

if you have any questions feel free to ask in #🤝｜tech-support

[boreal basalt]

Thanks I'll try it with your guide, also sorry for asking in general chat I didn't notice there was a tech support channel

[warm junco]

no problem 🙂

[void wolf]

I try this but I don't understand how to make different picture. It makes only one

I try to do this too.

https://youtu.be/ksG_ZjEKuLA?si=VoqJf6UZn1TF2AuH

but it doesn't work on my computer

[boreal basalt]

Hey thanks for the help it's working now

[warm junco]

Perfect np 🙂 be aware that if you encounter an out if memory error, to restart the webui and then try with a lower resolution again.

[hardy dove]

Anyone know if is possible to sell art made by stable diffusion or there is any problem?

about copyrights etc

[potent spire]

@hardy dove you can sell products but a) depending on what you generated you risk a copyright infringement and b) you dont have copyright over your generated stuff unless heavily edited

im playing with vector logos (Adobe Firefly in Illustrator) for my future business that includes game development amongst all and if i take a logo as it was generated then everyone could technically use it

which is a bad idea

[mellow tendon]

Hello, im trying to generate subtitles in english from a video in russian, is there any AI's that pop into mind for this task? ideally it would be nice to have it locally since the videos im trying to translate are lessons ranging from 1 to 2 hours long each and most translators ive seen have file size limitations, thanks in advance!

[warped jewel]

When generating clarification images is there a difference between using "photo of woman" vs girl or female?

[frozen kraken]

test

[unique glacier]

can anyone help me find an AnimatedDIFF discord?

[native sundial]

???

[shell hazel]

its the same thing, the only time it would change is if you added an adjective like "old"

[shell hazel]

#🎥｜animation

[shell hazel]

if you live in USA the law recently passed that all AI has no copyright protections and you can do whatever you want

[native sundial]

sauce?

[sudden ruin]

Yeah provide a source with that

[shell hazel]

You can't claim AI as your own with copyright law

and no one can copyright strike you for AI

otherwise places like fiverr.com wouldn't have an AI section to sell

https://www.fiverr.com/categories/video-animation/video-art/ai-video-art?source=category_tree

[native sundial]

fiverr and similar sites don't care about copyright infringements

they just pay any fines and go on

[shell hazel]

the most recent thing is Dec 6th 2023 was the deadline for congress to submit what they thought about AI copyright. That is it. lol

https://www.copyright.gov/ai/

[native sundial]

it is government so any results will most likely be announced next month or even somewhere half this year

any comments on the matter cannot be submitted for them to use in their decision starting Dec 6th 20203

[shell hazel]

the most you could get in trouble for is stealing IP but that's it anything else like an artist claiming that hair was ripped from their image or some bullshit like that won't hold up in court today, or next year, maybe 3 years from now...maybe.

[native sundial]

Last 90 Days (10,370) lots of papers for them to read 😄

pretty sure they will just follow china and allow Ai generative works be copyrighted .. no matter the trained data

[solid iron]

File "C:\Users\kerim\Downloads\sd.webui\webui\modules\launch_utils.py", line 384, in prepare_environment
raise RuntimeError(
RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
Press any key to continue . . .

what is problem ? Cant download

[warm junco]

What's your GPU?

[solid iron]

nvidia ge force 1050ti

i was able to download it when i got win10 i changed to win11

now i cant

[warm junco]

Make sure your GPU driver is updated.
Then follow my install guide here:
#🤝｜tech-support message

[solid iron]

its updated

[warm junco]

Good. Let me know if the guide works for you, for any problems feel free to ask in #🤝｜tech-support

[marsh vigil]

does anyone know if there's a custom node available that can control denoising strength via a mask/greyscale image in comfyui? i'd like each pixel in the mask image to be treated as a denoising strength for the corresponding pixel in the image being denoised (where mid-grey means a denoising strength of 0.5 and so on)

seems like a very obvious thing to implement, but i've not seen a node available anywhere

[bleak matrix]

Good morning, everyone! 🌞 How are we all this morning?

[sullen cypress]

quite good friend

my computer shows not good enough to run stable diffusion....need change a better computer.

[bleak matrix]

I'm glad to hear you are good!

[bleak matrix]

If you can't run SD, you can always play with the bot! There's also https://beta.dreamstudio.ai/generate which you can check out, as well as https://clipdrop.co/

Both are web-based--pretty fun to create while you are on the go

[stiff valve]

curious if anyone else is doing dedicated AI rigs. I'm debating between a Jetson AGX Orin, a Zimablade with a GPU, or a LP Sigma with a gpu. I'm trying to go as small as possible, and still get teh same or better performance than my main PC

[bleak matrix]

I personally am--can't vouch for anyone else--but having a PC dedicated to AI, I think, is something a lot of people may eventually want

[sullen cypress]

thank you friend

[sudden ruin]

Got one at the office with a 4090 which I mostly use for AI stuff, the iMac is for work

I switch between those 2

[bleak matrix]

No problem!

[bleak matrix]

Good morning, Boto! How are you? (Also, 110% for having things to switch to.)

[sudden ruin]

Good hbu

Also got the logitech mouse and keyboard that lets you switch PCs

Really useful

[stiff valve]

Got one what? I listed 3 machines

[sudden ruin]

I got a built one, not one from your list

[stiff valve]

Ah, gotcha

[bleak matrix]

Pretty good! Just enjoying my usual morning routine.

I have so many devices I end up switching to, but at this point, it's a toss up between my tablet, and my phone.

[sudden ruin]

I mean same for me, iPad, Phone, PC at home, PCs in the office and a few other things people wont know here

[frank rampart]

ive been wondering, and trying with various levels of success

but could i give Stable Diffusion A1111 a sketch and ask it to finish the drawing for me? like what if i gave it something like this and asked it to ink and colour it in comic fashion, with a sick model or lora base:

(shiz i cnt attatch images)

[bleak matrix]

Don't forget the Steamdeck, hahaha! Can run Blender on that, too. I'm glad I have something to switch to when I'm doing something with SD. It's kind of a pain when you're modeling, and you just go back to #🐝｜swarm-ui but then you have to wait to render something

[bleak matrix]

You can go to #🏞｜general-with-images for pics!

[stiff valve]

i'm mostly frustrated and looking into options because I cant do everything i want with one system

even as beast as this is

[fervent thunder]

hey would anyone b able to help me with training?

[stiff valve]

like marathon training? i would start by trying to keep a brisk pace and walk 5k, then up to 10k when that's too easy

😄

[fervent thunder]

training a model?

[stiff valve]

if she's hot enough, she won't need much training

[fervent thunder]

ur jokes suck lmao

[stiff valve]

dad jokes FTW

lol

[vagrant raptor]

Hi all! I just found this discord and I'm struggling hard if anyone can help with AMD Radeon install of Automatic1111. I had it installed and running. It wasn't perfect, but fool that I am, trying to make it better, I ran git pull and everything broke. I tried multiple reinstalls the same way I installed the first time, but I can't get A1111 running and Comfy and Invoke are either incredibly slow or not working either.

I should take this to tech-support. Oops.

[frank rampart]

thnx, i did that

[bleak matrix]

Check out #🤝｜tech-support

[bleak matrix]

You can find resources on training in #1080946152318443610

[stiff valve]

while not as "pretty" as webui, I'm kind of digging the style of comfyUI

[solid iron]

Downloaded a model but faced with this error. What should i do ?

AssertionError: We do not support vanilla attention in 1.13.1+cu117 anymore, as it is too expensive. Please install xformers via e.g. 'pip install xformers==0.0.16'

Failed to create model quickly; will retry using slow method.

[upbeat oxide]

anybody have some good prompts to make one of those Disney poster you see everywhere?:)

[stiff valve]

any disney poster lora's? i'd do that first and reference any examples prompts from inside the lora

[upbeat oxide]

sry im new to this, what is a lora?

[sudden ruin]

Install xformers, also #🤝｜tech-support

[solid iron]

where is the link

[sudden ruin]

Which version of the webui do you run

[solid iron]

version v1.7.0

[sudden ruin]

whats your launch argument like in the .bat

[solid iron]

run

[sudden ruin]

okay

[golden knoll]

is there a way to generate ai photographs on a private channelk

[sudden ruin]

Nope

[golden knoll]

that sucks

[sudden ruin]

What do you need a private channel for?

[long talon]

Anyone running Ubuntu?

[upbeat oxide]

why does my bot keep making videoes insted of pictures? can someone help plz

[sturdy smelt]

how generate ? say command pls

[warm junco]

Dream

[sturdy smelt]

th

[warm junco]

Format:image

[bright nest]

been awhile since ive been here, anything new since SVD and sdxl turbo?

[upbeat oxide]

thnaks bro!

[hushed sparrow]

quick question: is 1XL or 1.6 cheaper/better value for money?
just not sure how to compare them apples to apples

[echo vapor]

Thank you so much for ComfyUI!
Win11, Chrome,ComfyUI updated every day.
Issue: I select multiple nodes, holed Shift, click, and drag to move selection and all selected nodes jump over off-screen stact on top of one another, and there is no undo. What am I doing wrong? thanks

[mortal geode]

will dall-e 3 ever be open source and downloadable?

in 2-5 years?

[stiff valve]

hmm....new idea, instead of a dedicated pc, maybe i'll just get an A4000 to install in this system with the 7900XTX, best of both worlds, then i can use the AMD for my SD stuff, and the NV for my LLM stuff

[gloomy bison]

openai used to be open source before microsoft bought a bunch of the company

[pale latch]

they were clsoing up before microsoft got in to it. AI safety reasons

ethical deployement or whatever

[bright nest]

there was a node i had and lost in a reinstall that i can no longer find, it would save +just the prompt as a txt file next to the image. anyone know what it was?

the node connected right to the text box/string/clip text encode

[undone kraken]

Help please. For no reason at all, my generations are suddenly getting deep-fried. It's worse with some samplers (dmp++ sde) and worse at higher cfgs (even 6 or 7). I didn't change anything

[warm junco]

if you want to try out LLMs, go checkout GPT4all, works with cpu and also amd gpu

[warm junco]

do you have an example? you can post in #🤝｜tech-support

[undone kraken]

Yes I do, I'll do that, thanks

[primal arch]

what is the tool again to train your own images and produce more based on the given image?

[stiff valve]

tahnks! i'm currently using oobabooga, and sillytavern

[strong hornet]

guys

i remember

there was like

free stable diffusion

fully functional without need to pay

im lost cannot find it

[stiff valve]

SD is free, most here recommend starting with A1111 and some youtube vids

[stiff valve]

doesnt look like GPT4All works with ST, but its insanely fast in comparison, although its running a 7B model and my ooba install is running a 13B, so that makes more sense

[topaz burrow]

How to prompt using an image, can someone help me out please

[stiff valve]

easiest method is use GPT 4 and tell it to describe the image in prompt format so that you can duplicate it.

[dusk verge]

stable diffusion

[hard orbit]

new to stable diffusion, over the past 5 hours ive got it installed finally. how long does it typically take to render 1 image?

i put this on a machine with a 1660super, is that too slow?

[shell hazel]

a lot of variables change the render time, block size, CFG, Steps, Refiner..etc so if you're asking just basic nothing changed it takes me 2sec to render an image on a 4070ti

[hard orbit]

well it took 6 mins to render a solid black image. Im doing somethign wrong LOL

[shell hazel]

512x512? 1280x1280?

[hard orbit]

everything is default so i think 512x512

[shell hazel]

yeah 6mins way too long lol xD

1660 super is really old though so maybe

[hard orbit]

well yea and it didnt render an image either

i have a machine with a 3090 but didnt want to run it on there might have to grab a new GPU for this machine

this is the data at the bottom, perhaps it reveals an error ive made?

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 800296630, Size: 512x512, Model hash: 31e35c80fc, Model: sd_xl_base_1.0, Version: v1.7.0

Time taken: 2 min. 41.3 sec.

[shell hazel]

probably just the hardware

nvidia 4000 series is where AI is shining atm because it has AI built into the card

[hard orbit]

well this ancient GPU isnt getting it done for sure

oh it isnt doing the black image thing now added -precision full --no-half

[warm junco]

You need to also add --xformers --medvram

For better performance.

Your 3090 would take seconds btw xD

[hard orbit]

its the gaming rig though... about to buy a 4070 for this one...trying to determine if pwoer supply will hold

[warm junco]

Okay dont got for 8gb GPUs if you can

[hard orbit]

i dont understand GPUs how is a 4070 cheaper than a 3090

[warm junco]

3090 was the highend GPU two years ago. It has double the vram and its ultra rare to find one that doesn't get used for mining. Thats why its so expensive

[hard orbit]

the other has to be a 3070, it was like 2k 2 years ago

it wouldnt still be 1700 today i wouldnt think

[warm junco]

Nvidia prices are a joke in general xD

But 3090 is just rare. Like a 1080ti was rare after rtx 20 launched

[hard orbit]

so what does one buy today?

[trim magnet]

a used 3090 is good price

if u dont mind used

[hard orbit]

i prefer new, i like warranties

[warm junco]

4070 is a good card. But also a bit pricy

[trim magnet]

yea the 4070 is great or if u have more money u could also go for a 4080 16gb

[hard orbit]

i dont think i have enough power supply to run it actually. by the time i grab decent power supply and the gpu might be wise to grab a cheap cyber power prebuilt machine

[trim magnet]

whats your current psu?

[hard orbit]

i think its a 650

it is

[trim magnet]

4070 its a 200watt card and can run in a 600 watt psu so its more than enough

[hard orbit]

im seeing like 580ish on amazon that doesnt seem too bad

[trim magnet]

the TI version gives u like 12% more speed on sd so if u can by that version u should go for it

[hard orbit]

ti is a decent jump in price like 300 bucks it looks like. but maybe a 4060 ti is better than 4070 non ti

[trim magnet]

no the 4060 ti is the worst version because it only has a 128bit bus so when mem gets full it slows down

[hard orbit]

noted.. so 4gb more but still worse than a 4070

[trim magnet]

yea lower bandwidth so extra memory doesnt help unless u are not filling the memory

[hard orbit]

is there anything from the AMD side that is a better option?

[trim magnet]

not really ,to run sd on amd at decent speeds you have to use linux because amd hasnt added support for their ai stuff on windows

even a 7900 XTX 24GB gives u the same speed as a 4070 ti on windows

[hard orbit]

got it. well looks like this asus 4070 12gb white edition is about as good as it gets for 589

looks pretty cool too

kinda wanna snag a white case for it though

[trim magnet]

yea all white build looks sick,specially with the white rgb fans

[hard orbit]

yea .. down the rabbit hole I go lol

[trim magnet]

u could also wait til end of this month since nvidia is about to release their 40 super cards maybe the other gpus will drop in price

[hard orbit]

already ordered, i don't have a ton of patience... if this one was semi usable then maybe but this is totally unusable

[trim magnet]

how long does it take to generate a 512x512 img on your 1660 card?

[steep star]

@bleak matrix

[hard orbit]

ill run another one...

it took at least a minute to just start

[trim magnet]

you probably need to add this arguments --xformers --no-half --medvram-sdxl --medvram

are u running a 1.5 sd checkpoint or SDXL?

[hard orbit]

sdxl...its still running ill share time when complete

this is the arguments im running --precision full --no-half --disable-nan-check --medvram --xformers

[trim magnet]

oh thats why,sdxl wont work on that card,it will probably take like 15mins

technically it will work but it will take a lot of time

[hard orbit]

so i can down grade to likw 1.5 and get decent results

[trim magnet]

yea u can use a 1.5 checkpoint and it will take like 1 to 2mins to render an img at 512x512 and if u use high res fix it will take like 5mins

[hard orbit]

so a 512 is kinda useless though, I dont get that, is it to run that until you get the result you want and then upscale it/

RuntimeError: [enforce fail at ..\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 6553600 bytes.
Time taken: 7 min. 24.6 sec.

it failed at 7 mins

[trim magnet]

512x512 is the default resolution of 1.5 checkpoints u either gen a lot of 512 imgs until u get one that u like then reuse the seed and upscale it with high res fix at 2x or 2.5x

a few 1,5 checkpoints can go higher than 512 resolution but depending on your card it will take longer the higher the resolution

[hard orbit]

ok makes sense. well it took me 5 hours to get xl installed.. might tackle 1.5 laster toniight and give it a go. I assume i have to satrt from scratch

[trim magnet]

if u already installed automatic1111 u just need to download a 1.5 checkpoint then place it on models folder and also a vae and place it on vae folder

[hard orbit]

oh i just drop the Dl file into the models folder and it shows up in the GUI?

[trim magnet]

yea just click the blue button to refresh model list and it will show there

[hard orbit]

whats the vae i dont recall doing that the first time

[trim magnet]

its the model that decodes the image if u have a wrong vae it will either show a black img or a washed img

[hard orbit]

my folder structure i think is wonky also... i have D:ai/stable-difuision-webui/models and in that there are 12 files codeformer, deepbooru, and others

vae and vaee approx also is in tehre

[trim magnet]

yea vae goes in vae folders then u have to go to settings so vae menu appears on top so u can select it

[hard orbit]

launching again, well see if it does anything different ...same arguments apply for 1.5?

[trim magnet]

go to settings tab > user interface > quick settings and add this to the box sd_vae and then restart the webui and it will show up at top

after u add that it should look like this sd_model_checkpoint, sd_vae

[hard orbit]

so far so good, that went as expected... reloading now

same arguments still in place

[lofty basalt]

hi

imagine getting hacked, lol

[hard orbit]

things are happening 1.5 was in the drop down clicked it its processing now...so in therory i can run this when the 4070 arrives in a few days i can slap that in and switch back to xl

and what ive learned using 1.5 will be valid on xl

[trim magnet]

yea the only difference is that for XL u need to use 1024x1024 resolution

[hard orbit]

cool. no idea what it is doing now though, i assumed it was going to count to 100 and be done but its up to 178 now lol

[trim magnet]

yea it takes a while to load unless u have a lot of ram

[hard orbit]

i have 64gb of ram

[trim magnet]

then it probably taking a while because its first time

[hard orbit]

the good machine came with 64 upgraded it to 128 and switched that 64 tyo this machine

1660 with 64 gb ram is a bit comical

got an error: File "D:\Al\stable-diffusion-webui\venv\lib\site-packages\torch\nn\modules\module.py", line 1143, in convert
return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking)
NotImplementedError: Cannot copy out of meta tensor; no data!

[warm junco]

Would be a good Minecraft Server xD

[trim magnet]

come to #🤝｜tech-support and post screenshot of your cmd there cuz u cant post pics here

[hard orbit]

true

[hard orbit]

done

[median arch]

Hello. I want to create my own model that will react to various words which will generate different faces, poses and objects that are based on my own images I have.
Where should I start? (I don't want lora training)

[polar bane]

Hello guys,
New here so I'm asking : this SDXL Bot, can I just ask him to do iamges for me just like that ?

Also (not about the bot), does having a better gpu means better renders ? Is it possible that my bad renders are due to my small gpu (GTX 1060) ?

[warm junco]

Hey, yes you can use the bot in the bot channels for free

Also a GPU with more vram would be faster but you can also get good quality with a 1060

Upscaling helps for that for more help you can ask in #🤝｜tech-support
Like how to get a better performance for it or how to use upscaling

[polar bane]

Ok ok

Thanks

And can I "spam" the bots and try minor changes in the prompt ?

[warm junco]

It queues each persons prompt so if others are using it too you'll have to wait. But you can write multiple prompts, and the bot will generate each of them when it hits the queue

[polar bane]

Ok, so that's why I can take a little time

I'm still impressed you can do that on a discord server

[warm junco]

Yea its awesome 🙂

[fervent thunder]

How to upscale only inpainted regions in webui?

[golden kiln]

tenho uma conta com mais de 2700R$ em jogos pra PC qm quiser, chama

[median arch]

guy I literally cannot find any tutorial how to train a new model based on another model

everything is lora based

[hasty hornet]

this?
https://huggingface.co/docs/transformers/training

[limpid rock]

Hello everyone!
Kinda new here, i was hoping maby anyone could help me with an issue (its most likely the easies fix ever) 🙂

[hasty hornet]

just go straight to question, skip foreplay

[limpid rock]

oh alright!
When i copy generation data from an image and the paste it all into stable diffusion, it cant create the image beacause it cant find the upscaler

4x-AnimeSharp in this case

any idéa how i can get that specific upscaler?

[hasty hornet]

https://huggingface.co/spaces/0x90e/ESRGAN-MANGA/blob/main/models/4x-AnimeSharp.pth
This?

[limpid rock]

Thanks man! appreciate it!
Any idea what folder i put it in?

[hasty hornet]

probably /models/ESRGAN/

[limpid rock]

Thanks man, il try it ouy

out*

[median arch]

I found this https://github.com/victorchall/EveryDream2trainer

Should be fine but I don't know if trainers are designed for various images.
What if I want to fine tune a model for a person and few objects?
Everyone is doing models for specific concept but I want a lot of variety

or what if I want a model that will generate different faces?

[floral umbra]

You don't need to. you can use lower, but recommended to use pixel count of at least equivelant of 1024x1024, or 1,048,576 pixels

[frosty hamlet]

guys may i get some help with stable diffusion on linux?

i installed rocm and stable diffusion already running

but feels like something is not working properly

[vocal badger]

why does it says multiple pictures of nsfw are made with stable diffusion?

does it that work tho

[frosty hamlet]

also because i have ubuntu all choppy

[native sundial]

#🤝｜tech-support and share your terminal outputs

[vocal badger]

i just bought the stable diffusion

where do i get my role memebr?

[native sundial]

if you have your discord linked it should happen in 1-2 hours

[vocal badger]

okay how do i ask a reembolse

[native sundial]

[e-mail] if the support page on the site doesn't work

[vocal badger]

wheres the support page link, sorry for asking too much but this site is getting me confused

[native sundial]

? in the bottom right

https://beta.dreamstudio.ai/support

[warped jewel]

For dreambooth, how do you train continuously? I'm following the zero to hero guide using his settings. 300 steps per epoch, and I'm using 16 training images. Clicking the Train button generates 1 sample. Most recommend 30k steps to train.. If my math is right, then I have to manually click train 100 times... I feel like I'm missing something here

[chrome mirage]

hello, i have a question. why is my ui white? ive seen people use a blue one

[primal arch]

depends on your system theme if you didnt play around with your settings
if you have light mode on your system it will use light and vice versa for dark mode

[chrome mirage]

can you see the chat general with images?

id appreciate it

[limpid rock]

any tips, tricks or prompts to make faces look better?

[limpid rock]

Also can anyone tell me howi use an embedding?

Sorry for nwebie questions

[sinful sparrow]

you mean something like https://civitai.com/models/116230?modelVersionId=125849 ? if you're using automatic1111 for the ui it goes under wherever ur stable diffusion ui is\embeddings

[limpid rock]

was that a reponse to my question? sorry 😛

[sinful sparrow]

oh yea that was for the second question

idk about first one prob someone else knows more

[limpid rock]

Ok ty!!

[limpid rock]

So basicly just download it and but it in folder?

[sinful sparrow]

yea

for the automatic1111 there should be a folder and then ull see the actual embedding under the tab Textual inversion cant past image in this channel should be fairly easy to find

next to Generation

[limpid rock]

Thanks dude, i will try that.
Basicly i copy generation data and use the same model, loras as as ones i find but i never understood what to do with embeddings, idk if that has anything to do iwht it but it never turns out as good as the ones i copy 😦

[sinful sparrow]

ohh sometimes the data might say what lora +embedding and if its found on civtai the site i mentioned above u can just grab it from there

u could try the same exact seed once u get the embedding to double check everything is correct

[limpid rock]

Thanks man!
and how di i know if i have the automatic1111?

[sinful sparrow]

its a ui im not sure what u actually have, the automatic1111 is one interface https://github.com/AUTOMATIC1111/stable-diffusion-webui from here

there are many others you can use, its like how bing ai there is a ui for its model so u can type a prompt and ask it to make images, stable diffusion is same but theres more than one way

[fervent thunder]

Where do I generate AI porn images if I can't run it locally on my machine?

[covert dirge]

ei do you have AI tools that i can use to fix distorted face and hands?

[lean sequoia]

When using Automatic 1111, I know you have to include the Lora you want to use in the prompt. ie: lora:xl_more_art-full_v1:0.5 along with any trigger words. In ComfyUI I use a "Load Lora" node, so do I still need to include the Lora in the brackets along with the trigger words?

[forest trout]

I mean the node still needs to know which Lora you are picking.

[lean sequoia]

Understood. I definitely select the Lora in the Lora Loader. Just wanted to know if I need to also include it in the prompt.

Like in Automatic1111

[valid jay]

Is anyone on the staff willing to be interviewed for my school project? Its three questions

[forest trout]

Pretty sure its implied but it never hurts to double check.

because you can also add the strength inside the lora node.

[graceful gyro]

which is the oficial web page from this IA?

[orchid birch]

thanks for the help before guys

i have a new issue...would love your thoughts

I just downloaded sd on my computer, and is able to open the 127 ip address. but it keeps saying this "expecting value line 1 column 1 (char 0)

[full lark]

This one https://stability.ai/

[full lark]

Try adding this to the webui.bat "--no-gradio-queue"

[broken bloom]

I lost all hope from stable diffusion after using midjourney v6

[sinful sparrow]

midjourney is very good from a consumer perspective, SD still has way more control for the power users

[orchid birch]

to the cmd? I tried typing in the --no-gradio-queue

didn't work ) :

do you know why it's saying "expecting value line 1 column 1 (char 0)?

[full lark]

Nono I mean to the commandline arguments, I send you a picture on #🏞｜general-with-images channel

you have to open your webui.bat, if you dont have the commandline arguments just add this to the 8 or 9 line "set COMMANDLINE_ARGS=--no-gradio-queue "

[orchid birch]

thank you, only issue is I can't type anything in the cmd? so I can't add --no-gradio-queue

wait do you mean by editing the file? mind blown

just fixed it!!! thank you so much!! why did it work though?? that was magical @full lark

[tulip cairn]

Helloooooo

[iron orchid]

global harmonius makes outpainting borders seamless but it changes the original image : (

[snow umbra]

Hi, has the option to attach an image to dream prompts been removed?

[dusty jewel]

guys, what do you think is currently the best face swap extension / workflow for auto1111 ? Im using reactor, after roop is no longer and have some success. But its still not fully there yet

[fervent thunder]

im using reactor too

mapping the face is easy but not sure how to map emotions

its not possible with reactor, but was looking for some other extension which does this .. doesnt change the face, but just the emotions / expression

[potent spire]

Id use Photoshop for that

Thats what i do at least

[warm junco]

hey, reactor is very good, but i played around with IP-Adapter and its very good too for face swapping, you can even convert a face into an other style.

[warm junco]

is possilbe with IP-Adapter in ControlNet

tried it rn works pretty good

[fervent thunder]

can you share 1 example pls ?

[warm junco]

sure give me a sec

ill post it in #🏞｜general-with-images

[pliant ermine]

Are p40 cards useful?

[void chasm]

Hey folks, I wanted to ask if anybody is creating good icon images for items in fantasy/sci-fi games.
I have almost no experience with diffusion models and with the tooling around them so I'm not sure what is the good way to start to be able to do this.
Can you point me to the best way to do this? If there is already reasonable service doing this that could work too, but I would also love to learn to use Stable Diffusion for this.

[jolly spire]

hey does someone have settings for img2img + controlnet for a good upscale similar to magnific.ai?

im trying to add some details to a human face

[agile kindle]

hi

[kindred flame]

when trying to generate a pic with a1111 I get a very bad quality... what can I do to fix that?

[fervent thunder]

Try using a higher resolution or downsizing the image before processing it through the AI.
Or maybe you should just give up on your dreams and go work at Wendy's.

[peak tendon]

Friends, if I wanted to make ai images where I appear, is training a model of me the only way? Isn't there any ai utility where I can upload my pic so it uses it?

Pay me and I make a logo for your business

[fervent thunder]

Pay me and i will do it

[peak tendon]

Then no macaroni

[fervent thunder]

can you run SD locally?}

stable diffusion ..

[peak tendon]

I'm not asking for free logos, I'm asking if anything exists, I think my question makes more sense

[fervent thunder]

You need a very good gpu to generate photos for free locally on your pc

[warm junco]

There are some paid services that do this.
But localy you would need to train an model or lora of you. or easier using Reactor or IP-Adapter to faceswap (both extensions for local webui)

[fervent thunder]

just to experiment id use huggingface spaces, which use SD
https://huggingface.co/spaces/fffiloni/sdxl-dpo

[peak tendon]

What is Reactor or IP adapter?

[warm junco]

Extension for face swapping

[fervent thunder]

like deepfake for imgs

[grave hollow]

easy, how to create a image? whta prompt should I use?

[iron depot]

I'm trying to learn the terminology and usage of AI, SD, and its front-ends. I just read https://beta.dreamstudio.ai/prompt-guide and they say that the prompt weighting range is -1 to 1 but then immediately use 2.0 and-5.0 in the very first example. Why is this?

[elder anvil]

/help

[hasty hornet]

I don't see 2.0 or -5.0 there

[iron depot]

it's in the first picture under Advanced Topic: Prompt Weighting [New in DreamStudio!]

of the buildings

[hasty hornet]

you mean .5? that's same as 0.5

[iron depot]

moving to #🏞｜general-with-images

[proven acorn]

are we allowed to ask technical questions here?

[warm junco]

you should ask in #🤝｜tech-support 🙂

[tulip cairn]

No clue how to use the bot so I will just watch other people do it ✨ ✨

[iron fable]

How can I combine two images 1 my friend and 1 girlfriend AI generated picture

[tulip cairn]

Yo some dude is dming people to make them join their server

[cedar vortex]

Whats a good website for response chaining with LLMs?

[fervent thunder]

https://flowiseai.com/

[sudden ruin]

You can open a #1010934719455707218 ticket and send the screenshot in there, we will take care of them

[cedar vortex]

I'll have a look at that, thank you.

[marsh vigil]

nothing revolutionary, but might be of interest to people here: https://blog.io7m.com/2024/01/07/consistent-environments-stable-diffusion.xhtml

[mental sand]

hi guys, i'm a total noob to stable diffusion, just installed the program, installed a model, followed instructions online, did everything in order, but the generation commes out a grey box. anyone wanna help a beginner out, just spend like 10 minutes in vc help me get started? it would be a blessing, thanks in advance!

[warm junco]

Hey, what's your GPU? And what's inside your webui-user.bat?

[mental sand]

gtx 1070

@warm junco what do you mean what's inside? would you like a screenshot?

[warm junco]

Sure, right click and edit it. But paste the screenshot in #🤝｜tech-support

[mental sand]

@warm junco @echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=

call webui.bat

[warm junco]

Okay, at the line COMMANDLINE_ARGS=
add the following: --xformers --medvram-sdxl --no-half-vae
Then save the file.
Then check the file size of the model you tried to use. It needs to be larger than 1.98gb

[mental sand]

@warm junco i add it after, not under? so as COMMANDLINE_ARGS=--xformers --medvram-sdxl --no-half-vae

[warm junco]

Yes

But that is just for performance and compatibility.
The file size of the model still needs to be checked

[mental sand]

@warm junco actually the model started working

[warm junco]

Then everything is good

And your ready to play around ^^

[mental sand]

yeeboi

ima watch some tutorials and start having fun

[warm junco]

For any questions feel free to ask in #📝｜prompting-help or #🤝｜tech-support

[mental sand]

thanks

[weak tusk]

So now they hooked a brain up to a computer chip, but is that really Ai? Or regular i?!!: https://workflowpedia.com/brainoware-the-dawn-of-biohybrid-computing-and-its-implications-for-ai

[gritty ravine]

Hey all! Newbie question here! How do I get the dropdown menu for styles? How do I add new styles from a spreadsheet? Where do I put the file into a1111 folder?

[fading crypt]

Hey, what are the prices for SDXL and how many images per day can they do?

[edgy glen]

Hi everyone, I’m new to SD and previously only played with image generation using DALL-E via ChatGPT 4. I just installed SD today and started to play with it. One thing I’ve been unable to accomplish is any sort of text generation. I tell it to create stylish typeface for a string of text and it always comes out garbled. Is there a specific model that’s best for text generation? I also tried to guide it with ControlNet but haven’t succeeded yet. Anyone trying to do simple text graphics as opposed to characters and scenes that can point me in the right direction?

[onyx kiln]

I would try using a control net in addition to img-to-img generation on some image that already has the text you want in it. Then you might be able to prompt for the style you want

[edgy glen]

ah ok, thanks. Yeah, the image I fed to controlnet was really rough and drawn in MS Paint, but hopefully it'll be good enough. I didn't realize I should use img-to-img. I'll give that a shot!

[onyx kiln]

Good luck! I remember there was some text diffusion model (a model that generates images containing text), but I cant seem to find it anymore

[edgy glen]

ah ok, I'll have to keep looking. Is it hard to train models? I think I'll want to do a lot of simple stylized text generation with SD if it's capable.

DALL-E is so much easier in some ways, but it always misspells stuff and also doesn't quite listen to me most of the time.

[trim magnet]

its loras for generating text in SDXL models there are two https://civitai.com/models/176555/harrlogos-xl-finally-custom-text-generation-in-sd and this one https://civitai.com/models/221240/texta-generate-text-with-sdxl they only work on SDXL models, for 1.5 models you have to use controlnet either softedge,canny,depth or qr controlnet

[edgy glen]

thank you for your tips, I'll have to look into that to more fully understand some of the terms, as I've just started with SD today. But I assume I can just download the model on civitai for those and refresh, and also look for the controlnet models.

[trim magnet]

yea any SDXL model will work with those loras like JuggernautXL is one of the popular ones then just add that lora type the trigger words to generate your text

[grave quest]

Does anyone know any good retail stores in the us for graphics cards? I don’t live in the us and was looking to compare prices to decide from which one to import from

[edgy glen]

the only one I'm aware of these days is Best Buy

[trim magnet]

amazon/newegg for new cards and ebay for used ones

[edgy glen]

oh yes, if online stores is also what you meant, newegg is good

[grave quest]

Oh I see, I tried Amazon but all in there says refurbished

And most reviews on refurbished are 1 star bar product ones

[trim magnet]

click on show only 5 star items

this is prob the cheapest one there its a Yeston https://www.amazon.com/YESTON-GeForce-Express-Computer-Graphics/dp/B0CM9B19GZ

[grave quest]

Yeston is supposed to be good; I read about them and was thinking of importing from Brazil one of their cards, thanks for the link

[iron orchid]

Inpainting in txt to image vs img 2 img? What's the difference?

[grave quest]

You can’t do impainting in text to image

[iron orchid]

Outpainting sry*

Like edit OG image to give it a blank canvas

Then put it into txt + inpainting controlnet

[edgy glen]

I've loaded texta and tried to do a quick test, but I get a NaN-related error. I disabled the NaN check just to see what the resulting image would look like, and it's all black, which makes sense to me. I've had this same problem with the broom003 model as well. What's interesting is that if I then try v1-5-pruned, it will generate an image, and then if I select broom003, it will also generate the same image (I kept the same seed). So I guess this is just a quirk in automation1111 where it preserves a previously-working model in case the newly-selected model produces errors?

Do you know how to handle the main problem I'm having, which is the NaN error?

[trim magnet]

whats the name of the model u are using

[edgy glen]

I'm using texta and broom003, both give me errors whether I generate a "cat in a hat" or "display "hello"" prompt

[trim magnet]

i dont know broom0003 is it a safetensors file? where did u get it?

[edgy glen]

yeah it's a safetensors file. It's just a random one I downloaded from civitai: https://civitai.com/models/243271/broom-lora

[trim magnet]

thats not a model thats a lora

im talkin about what model you are using the one that shows at top left of the webui

[edgy glen]

oh I see, I must have copied to the wrong folder and gotten my terminology mixed up. the top left shows broom003.safetensors

so I gotta get an actual model and put it into the models folder. I thought the lora was a model

the one that works in the droplist is v1-5-pruned-emaonly.safetensors

[trim magnet]

yea you are using a 1.5 model and the harrologs and texta loras are for SDXL models not 1.5 models

[edgy glen]

I'll read up on how to do that, thanks

found this, seems like it'll get me on track: https://ambcrypto.com/blog/how-to-use-lora-in-stable-diffusion-a-step-by-step-tutorial/

[fading crypt]

Subscription says $13 a month of 1200/24h image generations?

[edgy glen]

well I'm getting closer now, thanks! I asked for "Hello World" and it printed "HELO WOLD", but on the final step, the image turns all black. at least it's showing some promise 🙂

[trim magnet]

its probably black because you need an sdxl model try to download this,then put it on models folder,hit the blue refresh button then select it from the dropdown menu https://civitai.com/models/133005/juggernaut-xl

[edgy glen]

thank you, I'll try that now

[fervent thunder]

Whats the best way to swap the whole head instead of just face via reactor

So to transfer hair color and style too if possible

[edgy glen]

that's so much better, thanks! The results remain on the screen, and it's closer now -- "HELL WORLD" lol

[warm junco]

Using IP-Adapter maybe

[jade wadi]

I ain’t smart enough to use stable diffusion, is there a group for nsfw Microsoft image creator? I can make boat loads of them

[fervent thunder]

Hugging face spaces for some free nsfw but long wait. For unlimited and quick there are paid options only

[jade wadi]

Yea ok, I make them all the time and post in some telegrams I’m in just never see other peoples prompts and have to make up my own, would love to just find people who share good ones that sorta bypass the safety features

[gloomy elbow]

Hello! I'm a new Professionnal Membership subscriber (but with another email, I can prouve ownership)

How can I register this Discord account for access to the Professionnal Channel?

@lavish lake @bleak matrix @sudden ruin

[fervent thunder]

Is there a quick way to determine if two faces are of the same person?

I have lots of images and i want to automatically swap the face of a particular person in all of them

This will require a bit of coding. First i need to detect all the faces of that person, them swapping is ez. However idk any ez to use model that does face some sort of face id

[sudden ruin]

We have no insight on that sorry, I think its a different discord and youll get an invite once its up and running

[gloomy elbow]

Ok. So for now we just paid. There is no console nor I have received a ToS or something, a licence or anything similar. Do you have any ETA to communicate ?

[sudden ruin]

I have asked the higher ups for some more information on this and will get back to you as soon as I got an answer

[dusk verge]

yoo whats new

[stuck wyvern]

hi

[sudden ruin]

https://hothardware.com/news/nvidia-unveils-rtx-5880-graphics-card-with-14080-cuda-cores

[gloomy elbow]

Thank you!

[inland egret]

guys, quick help. I should make an AI influencer and create a checkpoint with her face. But I don't know what subject to take as inspiration. Do you have any ideas? A thousand thanks

[marsh vigil]

john candy

[unborn briar]

Hi friends, I was wondering if I was in the right place to ask for help. I'm having trouble training a LORA. I'm using Kohya_ss, with an RTX 2060 Super. I apologize if I'm in the wrong place.

[wintry stream]

Hi mbCrypto!

The channels you are speaking to are not yet public, but coming soon!
Please keep an eye on your emails and you will be getting an invite to join there once it's ready 🎉

Looking forward to having you on board 👍

[jade moth]

Hey guys, I wondering if it's just me, or Hires.fix takes way more time then generating the actual image on A1111 ?

[warm junco]

Thats normal, what's your GPU? And do you use xformers ?

[jade moth]

I'm using Rundiffusion, an online platform that is pretty powerful. They probably have xformer installed...

[warm junco]

Okay then make sure to not set Hires Steps to 0

10-15 is mostly enough

[jade moth]

I usually set it around 0,51 and 0,80 depending on what I want. I love the hi res denoise effect and how it changes the image. Even with low Denoise strenght it takes ages...

[warm junco]

I dont mean denois

Can you adjust hires steps?

[jade moth]

Sorry I missread, I'm tired.

Yeah I usually keep the same amount of steps than the generation.

So "0". Lol. Changing that now.

[warm junco]

Yes thats why it takes so long xD

[jade moth]

Thanks it is slightly faster, although I feel I need a lot of hires steps to get the skin textures I want, I think I'm doommed to wait 1 minut per gen...

Thank you for the guidance bro

It seems I cannot post images here.

[sudden ruin]

#🏞｜general-with-images

[eager lantern]

anyone can guided me on how to train a base model with my data set

[vivid palm]

So I have a dreambooth model I've trained on a person. Let's say I want to try on a specific puffy jacket I find online. Is there an extension or img2img workflow that could help me somehow plug in that jacket and generate a reasonably good photo of my model wearing it?

[fervent thunder]

Is Texas in South America or north 🥹

Ok 🥹

[pliant ermine]

are tesla cards like the p40 reasonable for sd?

[fair hedge]

hello, im new here

i mostly use midjourney but I am searching for ways to do image to image generation in ways that the output looks exaclty like the input. for example, graphics designs that i want to put on ai fashion pieces. Are there any solutions with stable for this?

[warm crown]

Just starting off with Stable Diffusion - automatic1111 but I am getting errors every time I try to run the GUI.
Is there a support forum somewhere where I can post my error message and try to get some supprot?

Thanks!

[valid sandal]

In dire need of assistance. I'm working on a project that has a deadline tomorrow and returning errors that I cannot figure out the solution for. NotImplementedError: No operator found for memory_efficient_attention_forward with inputs: query : shape=(1, 16384, 1, 512) (torch.float16) key : shape=(1, 16384, 1, 512) (torch.float16) value : shape=(1, 16384, 1, 512) (torch.float16) attn_bias : <class 'NoneType'> p : 0.0 cutlassF is not supported because: xFormers wasn't build with CUDA support Operator wasn't built - see python -m xformers.info for more info flshattF is not supported because: xFormers wasn't build with CUDA support max(query.shape[-1] != value.shape[-1]) > 128 Operator wasn't built - see python -m xformers.info for more info tritonflashattF is not supported because: xFormers wasn't build with CUDA support max(query.shape[-1] != value.shape[-1]) > 128 requires A100 GPU Only work on pre-MLIR triton for now smallkF is not supported because: xFormers wasn't build with CUDA support dtype=torch.float16 (supported: {torch.float32}) max(query.shape[-1] != value.shape[-1]) > 32 Operator wasn't built - see python -m xformers.info for more info unsupported embed per head: 512

Up until aa few minutes ago SD A1111 v1.7 was working fine. I'm on Linux Zorin 17 based on Ubuntu 22.04 with Nvidia Geforce RTX 4070Ti 12GB, Pytorch2.1.2+cu121

python -m xformers.info
Unable to find python bindings at /usr/local/dcgm/bindings/python3. No data will be captured.
xFormers 0.0.23.post1
memory_efficient_attention.cutlassF: available
memory_efficient_attention.cutlassB: available
memory_efficient_attention.decoderF: available
memory_efficient_attention.flshattF@v2.3.6: available
memory_efficient_attention.flshattB@v2.3.6: available
memory_efficient_attention.smallkF: available
memory_efficient_attention.smallkB: available
memory_efficient_attention.tritonflashattF: unavailable
memory_efficient_attention.tritonflashattB: unavailable
memory_efficient_attention.triton_splitKF: available
indexing.scaled_index_addF: available
indexing.scaled_index_addB: available
indexing.index_select: available
swiglu.dual_gemm_silu: available
swiglu.gemm_fused_operand_sum: available
swiglu.fused.p.cpp: available
is_triton_available: True
pytorch.version: 2.1.2+cu121
pytorch.cuda: available
gpu.compute_capability: 8.9
gpu.name: NVIDIA GeForce RTX 4070 Ti
dcgm_profiler: unavailable
build.info: available
build.cuda_version: 1201
build.python_version: 3.10.13
build.torch_version: 2.1.2+cu121
build.env.TORCH_CUDA_ARCH_LIST: 5.0+PTX 6.0 6.1 7.0 7.5 8.0+PTX 9.0
build.env.XFORMERS_BUILD_TYPE: Release
build.env.XFORMERS_ENABLE_DEBUG_ASSERTIONS: None
build.env.NVCC_FLAGS: None
build.env.XFORMERS_PACKAGE_FROM: wheel-v0.0.23.post1
build.nvcc_version: 12.1.66
source.privacy: open source

[gloomy bison]

how do i do videos in #1100170312106127410

nvm im dumb

just click video after making image

if there is another way tell me

[slim zinc]

Will this new Ryzen 8000 series "Ai" of any use for SD users? The video names hugging face partnership. Source: https://youtu.be/LlTpLD0whIo?si=eLrX9Kg6BBvMtL0v

[broken smelt]

#🤝｜tech-support

[tame basin]

Hey everyone!

[elder moth]

Hi guys anyone her knows some kind of TTS program thats actually sound real and good?

[hasty hornet]

11labs

I think there's decent alternative dougdoug was using on streams, but I don't remember what it is and if that's even an alternative or he's using 11labs still, eh, not sure

[high vault]

Hello, does anyone knows the best and most efficient software to generate images on Mac M1?

[shell hazel]

There's not many software that can, I can only think of Photoshop's generative AI fill. Everything else is ran through Code in a command prompt in windows.

[shell hazel]

Took them long enough

But my guess is Nvidia will come out with some that blows AI generation away with even better performance and render speed

[high vault]

dayum, ok thanks!

[trim magnet]

here read this https://github.com/viking1304/a1111-setup/discussions/2

[high vault]

Sweet, thanks!, I've been using Fooocus for a few days and is really cool, but sadly is SDXL only, I wanted to try many checkpoints and loras that are SD 1.5 so A1111 is definitely the next one to try and also compare the performance of my little M1 base chip hahah

[lyric egret]

Which Cuda version works best with A1111?

[pure aspen]

hello ,I’m looking for a comfyui workflow to have 300pp and not 72pp? if anyone has an idea?

[wind ingot]

@pure aspen I'm guessing you mean ppi? That's mostly irrelevant , its just a guidance for print size, you can have 300ppi and 72ppi image with same pixels

[pure aspen]

@wind ingotjust upscale? in comfyui