💬｜general-chat

[lilac rune]

Strictly speaking, sd-webui cluster

[true mauve]

so I can depoly it on something like an EC2 instance?

[lilac rune]

You can set the address of each node through the configuration file, and receive the task after startup, which will be assigned to different nodes for execution according to the situation.

[true mauve]

Interesting.....

https://stablehorde.net/

you might want to check out this

it uses a similar architecture

[lilac rune]

You can come to the test server through discord in readme and experience it.

I wrote a code generator to generate a large amount of code to build a rapidly scalable sd-webui sdk to develop this robot with this sdk and cluster queue.

[vivid dome]

Ah, I was confused when I loaded VLAD bc in the command line it lists it as backend.ORIGINAL

[coral owl]

What a shithole discord is

[cobalt crag]

How to make a model generate ONLY 1 person?

It's impossible :(

#🏞｜general-with-images

[spiral gale]

makes me curious how @radiant meadow dreamer is verified if you cant add it

[spiral gale]

fr

wierdos who own it

@delicate oxide hey man

[eternal pumice]

Hi, I'm new here! =)

[true mauve]

cause if they let us add the bots people will probally spam generations and overload their servers and its in beta

[spiral gale]

Yeah but to get a bot verified it has to be in 75 servers

and considering you cant add it makes me think they inorganically grew it

also a company that stole the discord

check your dms bc ill probs get banned

[spiral basin]

Hello?

I'm currently facing issues downloading stable diffusion on my laptop

Can anyone be kind enough to help out?

[buoyant terrace]

hi, i used std mac (v 1.5.1) , ModuleNotFoundError: No module named 'insightface' and AttributeError: 'GeneralConditioner' object has no attribute 'process_tokens, who can help me, all my LORA models are no longer displayed and extension ROOP does not work

[proud charm]

What's your issue?

[true mauve]

First what are your laptop's specifications?

[spiral basin]

I can't seem to send pictures here but I can try send it privately if you don't mind.

[spiral gale]

@tall gorge

[delicate oxide]

Ello

[hushed tangle]

Can I use SDXL on RTX 2070 8GB?

[spiral gale]

i dmed you

[buoyant terrace]

hi, i used std mac (v 1.5.1) , ModuleNotFoundError: No module named 'insightface' and AttributeError: 'GeneralConditioner' object has no attribute 'process_tokens, who can help me, all my LORA models are no longer displayed and extension ROOP does not work

[proper sparrow]

anyone have good 3d motion settings ?

translation x and z

[true mauve]

bro why wont Inpaninting give me nsfw?

[bleak matrix]

Good morning, everyone! How are we all today?

[bleak matrix]

If you need help, I would check out #🤝｜tech-support

[still glacier]

yes

[willow locust]

With patience

[still glacier]

it doesn't take that long (with comfyUI)

[willow locust]

how long? not seen the metrics with SDXL

[still glacier]

1 or 2 it/s at 1024x1024

(and 5 or 6 s/it with auto1111 + loading time for base + loading time for refiner)

[willow locust]

oof

[still glacier]

wait

there

I messed up the unit lol

[willow locust]

i assumed it was s/it so all good

i still sometimes use automatic with sdxl just for convience

[still glacier]

the loading time and ram consumption is getting better with incoming 1.6, but the infering time is still awful

[willow locust]

i did a quick benchmark to see what my card gets and i could not understand why i only got 3.5 s/it, should been like 3.5 it/s, didn't relize it was doing all 10 images at once -.-

[hushed tangle]

A1111 vs comfy

What's better and more comfy

[willow locust]

comfy is better, A1111 is more comfy.

[hidden shale]

is sdxl the best model if i want to make a realistic photo for insta with my face?

[willow locust]

depends how your trying to make it honestly from what ive seen. Ive had the best experiences with SDXL, Lora and Reactor

but you can make extremely good pictures with SD as well

[hushed tangle]

Can you do amazing inpainting with SDXL just as with inpainting SD models?

[willow locust]

afaik you can, but ive not tried it.

SDXL is still very young, you have a lot less models that it works with

[lunar steppe]

hi, my model isn't working it stopped on image generation part

[quick umbra]

Hello

[tulip thunder]

it is possible man, c'mon

[cursive fractal]

Hey i am new to discord, dont understand very much how to use it to find help with SD automatic 1111, can anyone direct me to where I could find someone I could ask questions about using SD?

[jolly zinc]

Ok, I have a question. I looked for a Technical Support channel, but didn't see one.

I got SD up and running on my system yesterday, using python 3.11.x I kept getting errors that this version of python is not supported, and so on.

So I installed Python 3.10.6 this morning. Same errors, so I uninstall the 3.11 version. Needless to say, that caused all sorts of issues.

I finally got to an error I can't find an answer for. I'm now getting the error: OSError: [WinError 126] The specified module could not be found. Error loading [PATH].

It can't seem to load torch_python.dll or one of its dependencies. I've checked the path it's using and torch_python.dll is there.

I've had to run with --skip-torch-cuda-test, because if I don't, it gives me the error: RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

Any ideas? I'm running an RTX 3070 GPU, and it worked fine before I uninstalled python 3.11

[warm junco]

There is #🤝｜tech-support
Remove --skip-torch-cuda-test and add --xformers --autolaunch
Then delete the venv folder and rerun the webui-user.bat

[sour needle]

Can any of you guys tell me the best stable diffusion website?

I don't have a good pc..

[weary heron]

With comfy, yes

[hushed tangle]

How does it differ lol

Isn't it just a different frontend

With fancy graphs

[jolly zinc]

When you say "best", do you mean the best website to download Stable Diffusion to your computer, or do you mean the best website where you can create images online?

[jolly zinc]

Thank you, that worked. I figured worse case, I could delete everything and start over. I'm glad it was just that on directory.

I already had the other two arguments in the batch file. What does --api do? I saw it in someone else's batch file, and I'm wondering it is allows connection to the api somehow.

[devout pawn]

i have SD automatic1111 installed, can i just download sdxl and use the model like that or do i have to do something different? cause they dont seem to work

[warm junco]

it runs the api so some extensions and third party programms like adobe plugins can use it

[jolly zinc]

Ok, that's cool. I'll have to do some research then.

[ivory iris]

Is there any way to generate a picture of a model biting their lip? I can't use a LoRA to achieve this because I'm already using another one and it messes the face up.

@trim magnet

[bleak matrix]

I recommend checking out #📝｜prompting-help to info on that!

[vivid dome]

Is Vlad being weird for everyone? I’ve noticed issues a lot on my end. Gets stuck in “Starting”

[near pulsar]

hello.
While using bot room to experiment I am getting the message that my prompt is LONG PROMPT and will be truncated.
But the same time I can see other peoples using much bigger prompts than mine in same bot room.
How I can fix that ?

[hushed quarry]

all of their prompts are truncated too. Everyones is. You just only see the warning for your own but no one has special access to a longer version, the model can only accept a token certain length on it

[near pulsar]

ahh its clear now. Thank you @hushed quarry

what counts as "tokens" is it any symbol? even space or its only words ?

[hushed quarry]

tokens are most closely associated with small words but there are a lot of different types. So a ! would be one token but a whole word of letters could also count as one and for long words it may consistent of multiple tokens seperating it apart. They essentially are a way for the model to break up your words into manageable chunks and then convert those into numers that represent that word (since these models only work internally with numbers)

The bot is running SDXL which uses a CLIP tokenizer that accepts up to 75 tokens (and adds one at the front and end to signal the beg/end of the input)

more visual example of tokenization (from gpt3 tokenizer demo)

[fervent thunder]

the indvisible splitted into 3

[near pulsar]

thank you again @hushed quarry 🤝 perfect explanation,

I am sorry for many questions today from me .. 🙂
Is there some doc available about how to use these image generation bot's ?
for an example I want to set seed.. but it automatically adding seed:blablabla on its own...
and I want to set style i want like style:pixelart
I have checked resources channel and other channels but cant find info about bots specifically.

upd: aha! found how to set style before generation. But still cand figure out how to set seed.

[vestal dew]

Mine has done that for awhile. It loads and loads some things, but eventually goes past it.

I get these kinds of errors: AttributeError: 'Options' object has no attribute 'queue_button_placement'

I had to reinstall from scratch multiple times just to finally get one working. It doesn't seem to be running correctly. For example,. on txt2img screen the refiners drop down box is missing

It sometimes generates, but I have to use the image browser to find an image previously generated, send it to txt2img or wherever. If I just type a prompt without doing that, there are errors

once I send an old image to txt2img, img2img, I can then delete the old prompt, type one from scratch - that works

on the installs that "don't work", I can type a prompot to generate right away, but I can't do things like inpaint. All the menu options are doubled

and there are errors when I generate

vlad asked me to write the workflow step by step to get these errors. I don't have time for all of that.

It's a nightmare, because it happened just when I have something I need to do with SD, images made for a purpose, but I can't make them because old install and new install each have problems

[wispy jasper]

Anyone have an animatediff comfyui workflow they can share with me? especially img2img? 🙂

[vivid dome]

I’m debating going back to Easy Diffusion because of this but I can’t use LoRAs with Easy Diffusion which sucks

Are there any alternative UIs to Vlad that aren’t experiencing these issues but are also easy to use? I’m not super tech savvy and having to use the command line and pip is a pain in the ass

[vestal dew]

A1111

[vivid dome]

Just getting Vlad installed was a pain because I had to go through hoops to get PyTorch working

[vestal dew]

it's a UI, not a model

[vivid dome]

Sorry, that’s what I meant lol

and A1111 doesn’t require me to go through the command line if I need to make changes, right?

[vestal dew]

vlad is a fork of a1111, or started as one\

[vivid dome]

I should just be able to change my VRAM settings from within A1111?

[vestal dew]

A1111 is similar to vlad/sdnext

[vivid dome]

gotcha

[vestal dew]

change vram settings?

[vivid dome]

Yeah, I’ve been worrying that I’m using too much vram with my current settings

[vestal dew]

then do --medvram

I don't think you can change that with a setting in SDNext either

[vivid dome]

I have an RTX GeForce 3070, 8GB VRAM with an AMD Ryzen CPU with 16 GB of standard RAM

but I do know that A1111 at least has a webui-user.bat file instead of just a webui.bat file, right?

[vestal dew]

yes

[vivid dome]

Are these decent specs for doing AI work? I only relatively recently got my RAM doubled to 16 and my GPU switched over

[trim magnet]

thats the minimun to run SDXL

[vestal dew]

yes, but with SDXL not upscaling

[vivid dome]

Oh alright

[trim magnet]

for 1.5 checkpoints u good with those specs

[vivid dome]

so SDXL should run fine as long as I’m not like nitpicky with upscaling

[vestal dew]

right

If Sensei Seagal says it, then it is true

[vivid dome]

alright. As for like models, Lora’s, and Lycoris files, am I good to just copy those files from VLAD and drop them into the corresponding files in A1111’s folders?

[vestal dew]

yes

the dir structure is the same

[vivid dome]

That’s good to hear

What is the difference between Lycoris and lora by the way?

To me they seem like they do the same thing: augmenting models to make certain things possible

or like generating specific characters, ig

There’s a lot of cool stuff that I’ve seen tbh.

[vestal dew]

they do the same thing

[vivid dome]

Does it have to do with what each UI is able to process specifically? I’m just unsure why the distinction exists.

[vestal dew]

lycoris is a more refined lora

it is generally smaller, and it effects style a little less

for example, some people find a LORA character file from an anime character will make an anime looking image even with a realistic model and photo prompt. However, a lycoris from an anime character will make the character look in a photo style

[vivid dome]

I'm trying to go through installing A1111, but I'm getting this popup when I start:

posted in #🏞｜general-with-images

[vestal dew]

Not sure about that.

I didnt get such a popup, and A1111 and SDNext both run

[vivid dome]

I have python installed since I use PyCharm along side doing SD generating

I just am unsure if I need to change my version or just stick with 3.10 for my Pycharm projects

they're not related to my SD stuff

[vestal dew]

try to continue. See what happens

[scenic shadow]

Hello,
I would need some clarification about Stable. Is it possible to have a history of the requested prompts?
Can multiple simulations be run simultaneously?

Thank you.

[carmine lodge]

Hi!
What's the Best way to use SD on mobile? Is there an Android app? Use colab and an app? Use dreamstudio on mobile browser?

[cobalt crag]

Why 512x512 is the default size of every checkpoint or model?

When I try to gererate any image with 762x762

the a.i. always being so stupid

[willow locust]

because 1.5 is based on 512x512

SDXL is based on 1024x1024

both can however be upscaled for extremely good results

SD 2.0 is based on 768x768

so no far from "all" is based on 512x512

[cobalt crag]

Can SDXL generate anime artstyle?

[willow locust]

yes

[cobalt crag]

2D anime artstyle?

[willow locust]

yes

[cobalt crag]

I am bad at english so does prompting too

[willow locust]

https://civitai.com/models/117259/anime-art-diffusion-xl

my favorite for anime art so far, plenty of examples there

[deft sentinel]

hlo

anyone here

[grim valley]

somebody help guide for help?

"RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check"

RTX 4090

[signal mulch]

can i delete a picture that i uploaded?

[trim magnet]

skill issue i generate ultra wide images on 1.5 checkpoints with no problem

[cobalt crag]

Fix it then

[trim magnet]

i cant u just have to learn how to prompt properly

[cobalt crag]

with my low knowledge of english?

[trim magnet]

ill show u an example of ultra wide on 1.5 checkpoint

#🏞｜general-with-images

see its possible even with 1.5 u just gotta know to use proper prompts

[trim magnet]

if u cant write them just go to civitai and copy the prompts from popular images posted there

[wicked gust]

Hi, is anyone here an expert on Control Units? Need some hours of consulting to help us fine configure QR codes with logos

[deft sentinel]

Is there any site where I can free prompts for SD deforum

Something good

[pale latch]

Free, good, or easy. Pick two options.

Free and good means you get civitai.com to collect prompts. Pages where you'll need to manually curate and scrape them.

Free and easy means something like a prompt generator. Easy but they'll be crap for your purpose.

Good and easy won't be cheap. It means hiring someone.

[drowsy fable]

I tried to install control units and when I relaunched the app it stopped working, was hoping I could find an answer to that somewhere (moving the directory solved the loading problem) :/

ops, I meant controlnet, not control units, my bad 😄

[jovial quail]

Hi

[potent jungle]

Does ComfyUI have an OpenPose editor like Auto1111 where I can adjust it overlaid on the orginal image for accuracy? Really the one thing I've ran into where I can't get the same functionality out of Comfy.

[trail lion]

So just use auto to do it and get your processed image, do you need to be a purist?

At the end of the day, they are just tools

[normal verge]

Does anyone know how to use ai to color in drawings?

[tardy thicket]

has anyone here been using "roop unleashed?" I had a question and I'm willing to literally pay someone for some help with installing it locally

[violet arrow]

Hey everyone, I've just introduced some new features to my Deforum Helper tool – an editor and visualization tool designed for Stable Defusion Deforum animations. It's a solution for effortlessly controlling animations within Deforum. These additions include handy frame markers, audio channels, supplementary settings channels, enhanced user interface options, and much more. Give it a try and kindly share your feedback on how I can further enhance your experience. Let's enjoy the process and come together to create some amazing AI art!

https://deforum-helper.vercel.app/

[thorn patio]

when you install visual studio make sure you click modify after it installs and select the python and desktop devlopment c++, I could'nt figure out what was wrong with my roop and it was because I forgot to install those

[sour needle]

Online images

[tardy thicket]

do you have any experience with installing it via terminal on mac os?

[thorn patio]

Nope I don't

[tardy thicket]

thank you!

[sour needle]

Do you guys run ai art locally

How good does a pc need to be for local ai

[bleak matrix]

Good morning, everyone! How are we all doing?

[warm junco]

It should have an dedicated GPU with at least 4gb of vram

8-16gb of ram

[true mauve]

you can always use colab

[earnest solar]

NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
Time taken: 2.67sTorch active/reserved: 2287/2376 MiB, Sys VRAM: 4014/4096 MiB (98.0%)

How to solve this

[fervent thunder]

is it possible to download this on pc?

[warm junco]

What's your GPU?

[cosmic badger]

Why when I try to generate and image it say hang tight, you're still dreaming! 🙂 That's for More than 30 minutes.

[placid void]

im having the same issue

thought it was my connection but its not i guess

[cosmic badger]

Oh ok!

[flint fog]

Same issue: ERROR: @flint fog already has a pending request. its more than 40 inutes now

[earnest solar]

1650ti 4 gb

[crimson beacon]

Anyone have benchmarks for sdxl with 4060ti 16gb?

[proud radish]

Good day, i have "already has a pending request" more than 2 hours, how fix it?

[wild steppe]

#🗣｜artisan-support-feedback message

[jolly zinc]

Well, I started at Stability AI: https://stability.ai/ but I found it is actually easier to download and run it on my own system. Lots of YouTube videos to show you how to do that.

[wild steppe]

#1080946152318443610 has links to SD for both local & online

[quick bobcat]

Had a shower thought. Model/LORA hierarchal genealogy embedded and updated in models. Would be super helpful to see where subsequent models came from and what models a LORA was trained on. We're looking at many years of Stability-provided weights, hopefully, and many many more community-created and merged models.

Even just a spot for a nested dict with guidelines for trainers. The civitai mess is just the tip of the iceberg in terms of where models will be in a couple years.

[strange sedge]

Hi does anyone have stable diffusion installed, if so can i have some help, can message me privately

[nocturne elm]

any prompt tip to generate pitchblack background , where I can just see the object I asked? Like removing the background .

[warm junco]

Okay then you need to edit your webui-user.bat and at the line COMMANDLINE_ARGS=
you add: --xformers --medvram --no-half

[earnest solar]

Ok but when i clip web ui it start script where i put that

[earnest solar]

.

[warm junco]

You right click the webui-user.bat and then click edit

[spice plank]

What should I be using to get a good 1920x1080 picture with hires fix? I am currently using 960x540 with hires fix set to 2x upscale but it still looks like a blurry mess.

[trim magnet]

depends on the model,some were trained with wide resolutions and some were not u gotta try with different ones

[spice plank]

Any popular recommendations? I have most of the popular ones.

Anime,realistic, doesnt matter. Im just trying to get some decent desktop wallpapers

Hm, I think I might jave figured it out. I was using a steampunk lora and I think it might have been trying to add too much detail in not enough pixels so it looked blurry. Turned off the lora and my images look much better.

[trim magnet]

yea some loras do that when you add too much weight try to change weight of the lora to something like 0.4

[trim magnet]

absolute reality for realism and darksushi for anime style,those are really good

[spice plank]

If you are using the bot on this discord to generate, yes. If you are generating it locally, no.

[fervent radish]

Hey, I'm new to SD. I've seen some videos online where they interpolate multiple images and it looks really cool. Does any one here have any clue on those types of videos?

[potent stratus]

what is the difference between a fine-tuned checkpoint and a LORA?

[potent jungle]

A Lora is a specific add on to a checkpoint. You’re basically adding a concept on top of the model for it to understand.

[rigid jackal]

is anyone else getting a problem where their generations just dont show up?

[vestal dew]

in what UI?

I had a problem in SDNext a few days ago. The only fix was to reinstall from scratch. After I reinstalled a few times, it worked. It worked for a few days, but now it's back to not working.

[rigid jackal]

in the server here

[vestal dew]

ah

maybe it's generating but you miss it

it generated for me. Bot 1\

[radiant pebble]

hey new to stable diffusion, does anyone how those ai instagram models work? care to explain in DM?

i just had a few questions on how they keep relativly the same face while changing locations and outfits.

[sour needle]

Just type their name

[uncut pewter]

just type their name

[glass lily]

In which channel can I ask about why sdxl isn't starting up? I basically have a whole bible of code on me I can paste as the error message...which channel is appropriate for that?

[warm junco]

There is #🤝｜tech-support

[cobalt crag]

Which formula of Stable Diffusion allow me to peacefully generate 768x768 size image?

SD 2.0?

[humble onyx]

what is the best prompt to get best results?

like hd or smth

[odd arrow]

how do you i prompt to change color hair for having random between brune and blond for instance ?

[loud atlas]

#1100170312106127410 message

guys wtf is this prompt

what was he going for

[fallow pine]

Base+lora+refiner supposed to give better results than base+refiner or base+lora?

[bright sky]

Hi, I'm new to stable diffusion. Is 4gb gpu (have 2 gpus with 4 gb, but I'm assuming its not possible to use both?) enough to use stable diffisions models locally (assuming 512 x 512 px images) with reasonable time to generate images (generation of single image in like 5 minutes)? Would graphics quality be lower because of weak gpu (compared to e.g. midjourney)?

[frigid sun]

What is a common way to edit real images with stable diffusion?

[pastel lynx]

Hey guys! I'm trying for the first time Deforum extension in Stable Diffusion following a youtube tutorial. When i press the generate button, i got this error message :
"Error: 'Invalid control character at: line 2 column 632 (char 633)'. Please, check your prompts with a JSON validator. Full error message is in your terminal/ cli.
Time taken: 0.0 sec."

Anyone knows how to solve that by chance ? 😅

[ivory grove]

🇺🇦

[placid gate]

I hope you don't me posting this but: Calling all generative AI artists, creatives, and experts in London!

CogX is hosting their annual festival on AI and emerging tech from September 12th-14th. Women in AI, an organization providing a supportive space for women in AI, is an affiliate partner. More on CogX: https://www.cogxfestival.com/

We are organizing a panel entitled "Generative Art: Exploring Creative Potential at the Intersection of Art and Technology" and want to hear from you!

Share your experiences creating generative art leveraging AI for human-machine collaboration. Topics could include:
Creative processes behind generative art
Ethical implications of AI art
Accessibility, aesthetics, and audience reception
Cross-disciplinary perspectives
If you are experimenting in spaces like generative text, music, visual art etc., we would love for you to lend your voice. Please let us know if you are interested in joining the panel or have suggestions for speakers.

We also aim to host an executive function for women in AI - details TBD (12th-14th September in London).

This is a great opportunity to connect with creatives and technologists to shape a thoughtful, timely discussion on the futures of generative art!

Ping me if you would like to join us September 12th-14th. We welcome your ideas on the panel and look forward to hearing from you!

[hollow flame]

When you guys use ultimate SD upscale, do you keep your loras included in your prompt? Or do you only leave words related to lighting, detail, etc?

[tribal oriole]

Is there any way I can use stable Foundation in private mode?

[willow locust]

the upscaler loses context if the loras are not included

[hollow flame]

Thanks!

[warm junco]

You can use both at the same time to generate 2 images 1 with each gpu (two SD installs needed)

An image 512x512 20 steps will take 1:30 minute

[solid reef]

hi , can i generate an image from an other image with sd?

[obsidian geode]

Yes, its called img2img

[cobalt crag]

I want a model that can generate a cool flat 2D anime image

that look like something I sent in #🏞｜general-with-images

[obsidian geode]

Is it possible to use vae-mse-84k with sdxl?

[burnt granite]

Can someone explain what's the difference between normal diffusion (without face restore) against face restore. Like, if the face restore is also doing diffusion with some different algorithm or something. Or is it even doing diffusion at all or not.

I wanted something similar but for bodies

Um 🤔

I meant the way, whichever or whatever way face restore works in restoring/diffusing the face portion. Which is why I first asked what does it even do

To replicate whatever face restoration (the codeformer one) is doing to the body portion.

The way it restores it adds in a lot of realism that way.

I'm aware

I asked here because I'm not deep diving into the code and wondered if anyone who already has more knowledge into it might have known

Something

[late hinge]

I keep getting API request failed errors. I'm a n00b in programming though. Anyone with some python knowledge here who's willing to look at a piece of my code?

[solid reef]

i have a question , anyone know how i can write a description to have a skull with a bandana over his face , i trying but not working ( sorry bad english)

[hasty badge]

https://shangchenzhou.com/projects/CodeFormer/#abstract

[polar totem]

I have an AMD RX 6700 XT 12GB version, and as soon as I try to create images larger than 512 x 512, it says that I don't have enough VRAM. Does anyone have experience with this and can help me? Thanks in advance

[trail lion]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#face-restoration

[trail lion]

What start-up arguments are you using? What version is your UI?

[brazen quarry]

Whatre the current requirements for running sdxl?

Have there been any improvements on sd 1.5?

Ive been away since about May-ish

[fervent thunder]

can someone tell me how do i attach an image?so i can alter it in the #1100170312106127410

[vivid dome]

Noticing my LoRAs that I've installed aren't applying to my images that I make, any help?

I'm getting the following message in the cmd:

activating extra network lora: TypeError
Traceback (most recent call last):
  File "C:\stable-diffusion-webui\modules\extra_networks.py", line 114, in activate
    extra_network.activate(p, [])
  File "C:\stable-diffusion-webui\extensions-builtin\Lora\extra_networks_lora.py", line 12, in activate
    if additional != "None" and additional in networks.available_networks and not any(x for x in params_list if x.items[0] == additional):
TypeError: unhashable type: 'list'

[cobalt crag]

How to not have a skill issue in stable diffusion usage?

someone keep calling me a skill issue

I tried my fucking best to get what I want from SD

But the A.I. is so fucking stupid

I don't even know what prompts I could anymore to generate such a thing

[trim magnet]

go to civitai then go to images and copy the prompts from the best images u can find there

[cobalt crag]

What if their prompts only work for inidivitual models then?

[trim magnet]

then go to the page of the model you downloaded,scroll down and when u find an image u like copy all the prompts

after that u can make minor changes so it generates the char u want

[polar totem]

---medvram and Version 1.5.1 on Automatic 1111

[trail lion]

since you're amd, you cant use xformers, so try setting sub-quadratic under settings/optimizations. I dont use medvram btw, but I have a 16g card, so you may actually need it

also it wouldnt hurt to grab the newest version of auto

[warm junco]

Hey for that GPU you need:
--medvram --no-half-vae --opt-sub-quad-attention --opt-split-attention-v1

[signal mulch]

i need support, i can't generate imagine in bot room

[true mauve]

anyone checked out this model-->https://civitai.com/models/122606/dynavision-xl-all-in-one-stylized-3d-sfw-and-nsfw-output-no-refiner-needed ?

[spiral epoch]

hey there! I am new to stable diffusion (omg another newbie with dumb questions ;)) but actually I have a super simple question...

we did some SD testings and generated images. We are facing continuously issues (and this is what I see also in the images generated in here) with hands/people, currently it seems to be impossible to generate persons without doing some extra photoshop work...

=> anyone in here who can confirm/decline? Is it possible to use stable diffusion to generate brand images and assets without praying every time the images do not contain too many glitches?

[true mauve]

You need face gan for the faces, if you want realistic people use realistic vision or some other model for hands just make sure you have a very good and prompt and if it still turns out bad use inpainting

[true mauve]

what kind of brand images you want? logo?text?designs

[fervent thunder]

does anyone has any idea of how can i re-make an image i found on the internet? The dude made it with AI but i don't have any knowledge or access of the prompt but i want it in better quality

[undone garden]

Swapped my 4070ti for a 3090 and it's sooo much better

Even the speed, I don't know why it was but I wasn't able to use it fully, it was always sitting at 60w 55C while processing

As opposed to 330w 85C for the 3090

[distant flint]

Looks like with controlnets you can regenerate an image with a different style or details keeping the same pose/composition...etc, but is there a way to do the opposite?
Keep the same style, subject, but try to generate a different perspective or pose? Something like a style transfer I guess, but more rigorous?

[rigid furnace]

hey! not sure which channel would be best for inpainting questions but I'll try here and worst case someone redirects me?
So I'm trying to remove humans from photos via inpainting but the problem is that regardless of the prompt, a human-shaped mask is so heavily biased to contain a human that 90% of the time I just get another human inpainted into the photo

[bleak drift]

What a good site to generate a depth map

One that doesn't have a content filter.

[trail lion]

cant you just use controlnet to do it?

[spare fable]

Look into IPAdapter and/or Revision with SDXL.

[bleak drift]

I don't know exactly how to use it.

[trail lion]

so rather than asking for a website, it seems like you could be figuring that out

[distant flint]

I'll take a look thank you 🙂

[hasty bane]

hello, does anyone know any other community that is about creating LoRa?

[shy coyote]

Hello everyone. I hope you all already submitted your comments to the Copyright Office inquiring about ai and copyright law so big tech and big rights holders don't end up dominating these new tools and kill off free and open source alternatives?
In case you haven't https://www.copyright.gov/newsnet/2023/1017.html?loclr=twcop 🙂

[burnt granite]

What can I do to fix tone shifts in stable diffusion? Sometimes rarely, some images will just have a difference in tone when done in a series. Like if I batch processed 10 images one or two will come with a mismatched tone.

[dusky elk]

Hey folks, what's the status now for automatic1111 licensing. From the legal perspective, is it ok to use the code on a corporate machines?

[wispy cloud]

is this site stablediffusionweb.com/ also a part of stable diffusion ?

[cobalt crag]

Is it possible to turn selfie of irl photo of someone into anime artstyle with Stable Diffusion?

[turbid whale]

Absolutely. Img2img is your friend.

[cobalt crag]

I have never do that before

[cobalt crag]

Do you simple workflow for beginner?

I'm using ComfyUI

[turbid whale]

LoadImage node > VAE Encode > whatever sampler node you use

I'm on mobile rn. I don't have an img or JSON to drop at the moment.

[jolly zinc]

I've been watching YouTube videos. I've seen several on training a LORA.

Ah, that I haven't seen anything on.

[pine fiber]

Use kohya trainer

Look up dream booth

[pallid dune]

guys how do i change a parameter at each step, like add 0,1 to guidance scale at each step

if this doesnt exist, someone should make it into

cuz that would be crazy like you get more random stuff at 0,8 guidance scale but it gets more clear when the value goes up

[ocean notch]

Why do generations look so much better here than when I run locally? I can never match the level of quality. I'm I the only one?

[shrewd stag]

Hey everyone! Do any of you have any good ai art recommendations for images that look like nature landscapes from otherworldly, alien planets/universes? I want to start generating images in that genre but first I would like to know what's out there and who is already doing it.

For reference, I'm a big fan of @nvnot on twitter

[hollow flame]

Does anyone know if there’s a way to see the entire image while ultimate SD upscale is running? Going tile by tile makes it hard to tell if the generation is worth continuing, as I’m always experimenting with different methods and upscalers

[eternal pendant]

If I am using a web ui for SDXL how do I get the same style effects that this discord used? What are the added word to make the prompt style? Does any one know?

[true mauve]

Hardware,settings,finetuning

[true mauve]

-->https://www.reddit.com/r/StableDiffusion/comments/15afvnb/sdxl_various_styles_keywords/

[eternal pendant]

Helpful but it looks like it’s missing some style like sticker

[true mauve]

if you want natural or abstract other worlds images the keywords I recommend are Exotic, Otherworldly, Surreal, Eerie, Unearthly, Extraterrestrial, Alien, Celestial, Bizarre, Ethereal.

[true mauve]

These are the styles that clipdrops uses but if you want more of a sticker style then I recommend checking out-->https://www.reddit.com/r/StableDiffusion/comments/153caf7/stickers_in_sdxl/

[eternal pendant]

Thank you!

[true mauve]

An entire checkpoint is hard to make, my exprience suggests you need about 20k images an a100 and alot of fine tunning to make a model you are just better off making a lora for some thing a rev animated model,200 images are not enough you need about 2000 or 3000 for semi decent results

[polar gulch]

Can anyone please share SD model converted for mediapipe image_generation.apk

[fallow pine]

the civit.ai site puts most example for the a1111 or comfy ui mostly? and does comfy ui and a1111 come with different results for a same seed?

[true mauve]

almost same Its like minecraft worlds

[fallow pine]

if some lora models are unavailable in civit.ai does it mean they were privately made and not shared?

[lyric estuary]

Got question, I've trained my own lora models.

I assumed closeup faces with great detail would work on most cases

It worked fantastically on closeups but not on longer camera distance, face deformed and sometimes artifacts appeared.

So is this mean lora need to actually train on both long distance and closeups faces?

Can anyone probably show what their training dataset look like? And probably the settings too? Thank you

[shrewd stag]

Thanks!

[true mauve]

If you only want to make a Lora for close ups the Model starts to prioritize close-ups and sacrifices long shots, mixing long shots and close-ups is not recommended as it confuses the model I recommend you make 2 Loras one for closeups and the other one for long shots and put activation keywords for the Loras so they don't interfere with each other

[lyric estuary]

Ok thanks. So Lora is very specialized/hyperfocused training model.

So that got me thinking how do the ai know what we're training it for

Like say I want to train hands specifically or some sort otherworldly alien shit 🤔

[true mauve]

hands and aliens? I guess there are plenty on civitai maybe some checkpoints too

[lyric estuary]

I know about the premade models in civit and huggingface

Just wondering how things works under the hood so I can utilize and tweak SD to my needs.

My guess is the tags/description I suppose

[true mauve]

Well I guess its always good to tinker

[bleak matrix]

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

[true mauve]

evening for me

[lyric estuary]

Gm

[bleak matrix]

👋

What's everyone up to today?

[digital cave]

Hey, in compatibility it says "Use old prompt editing timelines" to use the thing that made the swap between [red:blue:0.75] at 75% of generation, what's the new prompt to do this ? Is there a patch log somewhere ?

[true mauve]

just spent 6 hours fixing a subprocess bug 🥱

[bleak matrix]

If you need help with prompting, I suggest heading over to #📝｜prompting-help

[bleak matrix]

Get it figured out?

[true mauve]

Kind of

[bleak matrix]

Well, that's a start!

[elder sail]

I am currently playing around with SDXL 1.0, coming from 1.5. It seems like all my pictures are soft / way to much bokeh in SDXL. Is there any way to avoid it?

[lyric estuary]

Revert back to 1.5 lol

[elder sail]

:o/

[fallow pine]

do you guys prefer trained or merged checkpoints mostly?

[bleak matrix]

Learn more about what you are prompting and use negatives. I suggest testing out the bot and learning more about styles/categories! You can always share your prompts and get feedback in #📝｜prompting-help to see where you are at, as well as #✨｜sdxl

You can check out my wip sdxl prompting guide here: https://atypicalconsortium.carrd.co/ as well. That might help you

[covert inlet]

Hey guys is it possible to do a whole 10min animation about precise characters exclusively with auto1111?

[lyric estuary]

no

[thin beacon]

Automatuc1111 just got updated to 1.6 for me and now roop extension is giving me errors ☹️

[covert inlet]

How does people do the animations then?

[lyric estuary]

roop was abandoned, there're forks of it. Works fine on my side 🤷‍♂️

Google around

[lyric estuary]

txt to videos services

there's definitely extension for stable diffusion to make videos but I don't have need for it, so I don't dig that deep and I would guess it would produce horrible result

runwayml is one

runwayml can produce videos out of one image but yeah the result aren't that good. Probably forget about txt to videos for awhile.

Those that have consistent, flicker free result usually produced in conjunction with ebsynth and shit tons of other editing softwares like AE

[true mauve]

Make sure ur resolution is 1024x1024 because anything else will break sdxl

[true mauve]

Zeroscope and runway are acutally half good

[lyric estuary]

did test for em but the results are.... trippy lol

can't prompt text when doing img to videos, so can't direct the AI on what we want, the AI just guess on what to do

[true mauve]

well in runway you cant but zeroscope with a bit of tinkering...

[lyric estuary]

zeroscope flickers

runwayml trippy

so yeah which one is your priority

[true mauve]

Yes and cant forget the shitstock watermark xD

[true mauve]

I l rather use stable diffuison wrapper or animate my self

[pastel lynx]

Hey guys, i tried to install TemporalKit with HuggingFace, but it corrupted my whole Stable Diffusion.. Do you know where i can find a clean version that work of the tqdm.py file ? From \venv\Lib\site-packages\huggingface_hub\utils ?

[languid helm]

what is the least resource intensive model of stable diffusion in hugging face ?

[versed zodiac]

Hi @wintry stream sent you a DM

[true mauve]

chatsalad 19m

[languid helm]

thank you

ah i have gtx1650 mobile gpu, ryzen 5 5600h processor, 16gb ram

is there anything that works with this hardware

pretty well ?

[chilly quest]

Hi, I just update my stable diffusion today and suddenly it become really slow to generate image. Before this just take about 3min to create 4 image batch count. Now I have to wait 30min.

Anybody know how to fix it?

[warm junco]

It will work with any 1.5 based model. Just make sure to use --medvram --xformers

[worn shale]

Me too, it worked really slow, but it definitely use GPU, I can see that wit nvtop, I was wondering if it now upscale automatic, but I can't get my head around if it do, but at the same time I now could use 1024x1024 without running out of vram.

Any how, it started going crazy batch count sodenly was at max and the same for number of pictures, I interrupted it, and the it would only generate abstract shit, I switched between models and I kind of came to it senses, but it really struggles with useful generations in text2image, I first was in doubt if it had taken a bad habbit on after 255 of batch and number of pictures, but reading in to it, it should not impact anything, I guess I have to go to my output folder and set some seeds just to get i back on track,

[fallow pine]

SD upscalers are wayy better than $200 gigapixel scalers.. do you agree?

with gigapixel i used to lose fine details but sd upscalers can do loseless upscaling

that's a bless

[chilly quest]

I havent use gigapixel. But I must agree SD upscaller is awesome. Works like charm. I use SwinIR and I love it

[fallow pine]

Lol they have only 4-5 upscaling model but even like 2-3 of them are usable in usual

whereas u can try out dozens of scalers on sd

[worn shale]

Sure upscale works nice, but not only did it start running slow on SDXL it also completely missed my intention from my prompt, making pictures look like a acid trip. I switched to SD and did a dreampshap reandering, and then it started to produce useful pictures, but what took 4 sec. Before now sodenly took minutes, even at 512x512 and that was when I started to look in to upscale

[fallow pine]

true.. SwinIR and RealisticRescaler are pretty close and they deliver awesome results

[worn shale]

It also added random style in the 255 wrong option reandering, until I switched back and forth to not use random style.

And then I did SD dreampshap, and after over 1 min. 512x512 it generated a useful pic. Normally that takes seconds

[fallow pine]

How do I regenerate an image that made on comfyui and won't show custom nodes when dropped into the interface? 🤔

the image is in jpeg format

[fallow pine]

Who's MADMATIX in civit.ai? is he on here?

[sand holly]

does anybody know what opt-channelslast does?

doesnt seem to have any effect on performance for me

[sudden ruin]

If you mean you're missing nodes and they show red, get the ComfyUI Manager and try to install missing custom nodes from the menu

[languid helm]

cool tnq

[late dawn]

Alright, who was the genius at Automatic11111 Inc. that left hires fix ON by default AND made removed the quick option to disable/enable it?

I don't even have enough VRAM for it

[olive dagger]

when i'm using a different XL checkpoint, does it also need refiner?

if so, do i just use the SDXL refiner?

[fallow pine]

No.. nothing appears from the jpeg image. No nodes, no workflows, nothing..

[sand holly]

it shouldnt be on by default and the quicksetting is right next to tiling and face restoration

[late dawn]

Have you updated it?

[sand holly]

i have the git pull command everytime i double click the user start.bat

[fallow pine]

can somebody help me with the comfy ui? i'm quite new on it and trying to figure out things

it's running super slow for me

[late dawn]

alright I updated (again), seemed like a bug or sth

[sand holly]

hmm

[sudden ruin]

Then there is no meta data in the image

[olive dagger]

hey guys, any news for controlnet XL for automatic1111? 👀

also, do you need to use refiner for custom checkpoints?

[fallow pine]

bro the image comes with metadata on civit

i could still generate on a1111 but it comes with different image

[sudden ruin]

#🤝｜tech-support and did you restart comfyUI

[fallow pine]

yes

i could generate faster in the beginning but now somehow it's delaying much

[karmic brook]

https://discord.gg/czzw5JZn?event=1145771898421121125

starting soon!

[proud shell]

Hi 🙂
anyone knows a good place that offers api support and custom models & loras ?

[zealous cradle]

If I get one more notification I swear

[sand holly]

i just heard that if you change xformers in the optimisations tab to sdp its like 50% faster

went from 37s for a 1024x1024 @30 samples to only 18s

[plucky shoal]

isnt it only for 3xxx and 4xxx models?

also biggest performance boost in Comfui, is AITemplate, as well only for 3xxx and 4xxx, but in my case 50% gain, and 4xxx users probably has even higher.

[wintry stream]

Hey Friends!

Please let me know if you continue to get any pings for the stage event.
We have a ticket with Discord escalated to try and sort out what's going on with the notifications being sent in a loop

[timid wolf]

How to add share my art work?

So can be a part of the team?

I am a Stable Diffusion user!

[torpid crest]

lmao still getting pinged

[wintry stream]

Hi Mystic!

Feel free to post in #🌠｜show-and-tell, #🏞｜general-with-images, and the various topic forums such as #1070829540772937728, #1070829655638151208, #1070829832151240734, #1070829899918610632, and #1070829982252798043!

[hexed parcel]

I have heard whispers on the wind of an event rumored to be taking place here

[torpid crest]

are you here for the grease fight

[hexed parcel]

Yeah that's probably the one

[wintry stream]

When you get the ping, does it show up as a totally new notification as well? It doesn't just replace the old one right?

[torpid crest]

So it's a notification and it's pointing to a message that does not exist, when I click it the notification disappears, I will restart discord now, perhaps its a bug in the app

[wintry stream]

Very helpful - thanks!

Yeah from what I can tell, it's sending a notification for the "Stage Starting" event which doesn't technically link back to any message we can delete. Otherwise I'd happy get rid of the message itself in case it's stuck

[tender nebula]

PING AAAAAAAAAAAAAA

[wintry stream]

But we don't have a way to clear out any notification cache on our end so I've asked Discord to step in & see if they can destroy it on the backend

[torpid crest]

It is linking to a message where the hostess pinged @ everyone

[wintry stream]

In the meantime I've disabled all pings for the stage/events in that channel. Unfortunately, it seems like the notification is stuck somewhere & not that it's being resent

[wintry stream]

Like this right?

[torpid crest]

yes

[tender nebula]

yeah

[wintry stream]

Yeah unfortunately that's automatically formatted & doesn't actually come from a message

It's part of Discord's automatic notification system for events that are linked to a stage

[tender nebula]

kick amli_art

[wintry stream]

Amli gets credit for "posting" it since their user was the one who started the stage

[torpid crest]

yeah looks like it's too automated

[wintry stream]

I've already disabled all their ping perms. The issue seems to be the initial notification continues to send, not that new pings are being sent out in addition.

That's of course mostly just my guess from the access I have

[torpid crest]

[wintry stream]

So I imagine somewhere in Discord's servers the check for "Is this ping already sent?" isn't working properly and it keeps reissuing the same notification over and over

I'll hopefully have an update soon! We have a ticket with Discord as well as a contact I've escalated to on top of that

[wicked hollow]

Shouldn't center stage be offline now? It's still live.

Unless you're keeping it open for troubleshooting purposes.

[wintry stream]

Yeah I just had to switch into it quickly to do some additional checks

[wicked hollow]

Gotcha

[uncut lantern]

Is stable chat https://chat.stability.ai/chat working for anyone ? I can’t use it

[sand holly]

does anybody know a good tutorial for following along to train a model?

[soft marsh]

The pingsss

[torpid crest]

I'm not getting pinged after restarting discord

Yeah the good old turn it off and back on again

Don't underestmate it

[wicked hollow]

I'd probably have to force stop on my phone for that to work there. Easy enough to restart the program on the PC though.

[elfin pivot]

still getting pings after restart

[wicked hollow]

Restarted both myself, now to wait.

[lethal tinsel]

Why are people asking Gigabot the same question multiple times in a single day?

[fervent thunder]

Is this the official sd server

[hexed parcel]

Need that akira meme of the guy saying "leave me alone" and images of the ping pops up in his head

[tender nebula]

this one a fake

the massive green verification tick next to the name of the server is an emoji

[soft marsh]

The pingssssss

[wicked hollow]

What pings? I'm not receiving any pings?

[lethal tinsel]

If you mute those channels, as long as you don't get pinged, the notifications shouldn't show up.

[hazy frigate]

Hello everyone. at some point i came up to a really comprehensive guide about prompting on DreamStudio on stability.ai website. And now I can't find it. I think it prompted me while I was in DreamStudio app. Perhaps someone knows what am I speaking about. Thank you very mch if you share that link

[tender nebula]

AGAIN

[forest trout]

They keep happening!

[lethal tinsel]

What keeps happening?

[wicked hollow]

Hmm, I haven't received one in the last 15 minutes.

[forest trout]

https://tenor.com/view/make-it-stop-noisy-loud-annoyed-irritated-gif-3822323

[tender nebula]

I’m reporting this server for harassment

[wicked hollow]

It's been quiet since I restarted.

[lethal tinsel]

?

[lethal tinsel]

What is going on?

[forest trout]

I keep getting pinged for an event that started an hour ago.

[wintry stream]

The issue is on Discord's end so you're welcome to create a ticket to let them know! We already have an escalation in place requesting them to take a look

[flat dust]

🧐

[wintry stream]

Maybe we should try and domesticate the ping since apparently trying to control it doesn't work 😂

[wicked hollow]

Make it work for US

[wintry stream]

Does anyone know what Discord pings like for snacks?

[forest trout]

[tender nebula]

I’m jus playing 😂

[lethal tinsel]

?

[wicked hollow]

Humans. I volunteer to be first sacrifice.

[lethal tinsel]

Do you know how to mute channels?

Or are you just going to complain?

[tender nebula]

you should just everyone ping the whole server 100 times so the ones that are happening now seem insignificant

[lethal tinsel]

🤦

[wintry stream]

Modern problems require modern solutions 🤔

So what you're saying is, make sure to @everyone on an announcement when we finally get the pings to stop?

[wicked hollow]

[lethal tinsel]

The at-everyone ping should only be used for announcements.

[tender nebula]

yeah but do it 100 times over

[wicked hollow]

This is the way

[wintry stream]

"If you dislike these pings, please direct any complaints to Kane!"

[tender nebula]

come onnn i’m joking

[tender nebula]

blame amli art she’s the person responsible for this 😏

[void ravine]

OK to be clear, there is NOT an InvokeAI thing happening?

[wicked hollow]

Lol. It happened, and Discord had a seizure and keeps spamming the original notification.

Oh noooooo. I just got another notification

[wintry stream]

Time for us to make a "I was here for the 1 million pings event" pins

[round sphinx]

Yo hope everyone is having a great day, I had a question about training sdxl loras, If i use the sdxl base model to train a lora can i use that lora with other models trained on sdxl?

[wicked hollow]

At least we know that it's not a client-side issue and is definitely on the Discord side.

[lethal tinsel]

I would really like to know where the bot is being pinged, so that I can mute the channel.

[wintry stream]

Unfortunately the ping itself isn't linked to a specific message. It's linked to an event that's been deleted.

You could try muting #1005545148211527800 and see if that helps!
If it does please let me know so I can pass on the info

[wicked hollow]

This is how you do it: https://i.imgur.com/hSWe7uD.png

[lethal tinsel]

Okay, see if you can get the bot to push a notification. Technically, it didn't ping, but it is distracting me.

[wicked hollow]

He's been trying to stop the pings all this while, I think it's safe to say that creating more pings is not on the to do list as of yet, heh. You'll just have to wait and see if it pings you again.

[lethal tinsel]

There not pings, if they were there would be a ping sound, and it would show up in your discord-inbox.

[tender nebula]

that’s exactly what happens to me

[lethal tinsel]

However, it's been distracting me for the last hour.

[tender nebula]

go on do not disturb

[lethal tinsel]

I muted the channel. So hopefully it doesn't keep distracting me.

[wicked hollow]

It's also exactly what happens to me. They're notifications, and so we get pinged. Your notification settings may be different than ours but notification = ping.

[lethal tinsel]

No, there not the same thing. Do you have something putting all messages in your discord-inbox?

Anyway, someone's being dense asking the bot the same question multiple times.

It was amusing, only the first time.

[wintry stream]

The notification itself isn't coming from a bot on the server, it's coming from Discord's backend since it appears to be stuck. We have a ticket open with them right now asking for their assistance in resolving the problem. Hopefully it will be delt with soon!

Just adding clarification

[tender nebula]

Going to wake up tomorrow to 99+ on the server icon

[lethal tinsel]

? Isn't the backend of discord server-side?

[wintry stream]

Correct - the notification is coming from our server, but we don't actually have a message or notification being sent on our side. It seems the initial notification for the Stage event is stuck & somewhere in Discord's server it keeps re-sending that same announcement even though we've deleted all the posts & events surrounding it

[wicked hollow]

I was talking terminology. And I get tons of crap in my inbox all the time, I pay it no mind. I'm not sure on the exact relationship between discord notifications within settings and one's inbox, especially considering it is sectioned off into three sections.

[wintry stream]

So basically the pipe is clogged somewhere & that's resulting in the original notification being sent over and over with no way for us to stop it on our side (at least at the moment)

But Discord is pretty great so I'm sure they'll get it sorted soon

[wicked hollow]

Don't quote me on this, but I feel like Discord has been going downhill ever since they were bought out.

But... that's what happens the further you transition into a fully realized corporate presence.

[tender nebula]

nitro super reactions make up for everything though 😁😁😁

[wicked hollow]

Sure, yeah. Big difference. Nothing like switching to mobile and accidentally pressing the wrong button every time you want to react to something, lol.

[forest trout]

"You have used 3 out of your 5 free super reactions."

[untold ocean]

bought by who? 🤨

[wicked hollow]

Oh wait you're right! I could've sworn they were bought by Microsoft but I didn't realize the deal fell through. It does look like Sony invested a fair amount into the company though instead, among others.

When I originally heard the news I was lamenting the fact that Discord would likely receive the Skype treatment. Still, it does seem like Discord has lost a lot of what made it special.

[wintry stream]

Anyone still getting pings in the last 10/15 minutes?

[wicked hollow]

You know what? I haven't gotten any recently.

[flat dust]

I eh got one 25 minutes ago

[forest trout]

I guess if I get another I'll just post here.

I got one about 20 minutes ago or so too.

[forest trout]

I think you might be good. Haven't gotten a ping since.

[tender nebula]

all good now

[misty pollen]

I have trouble understanding why controlnet and high res fix is changes the whole image halfway, is this expected?

[chilly pasture]

anyone else having somy elevator muisc when opening discord phone app today? i think its from this discord channel D:

[warm junco]

Its only the Highres fix. The denois value defines how much the image will differ (how much it get fixed)

[raven sail]

Help, is there any way to see the images that I asked the bot for? I lost my cell phone along with some images that I downloaded from the AI ​​a month ago and I already tried with the search engine putting my name and it does not give me any results

[misty pollen]

I have another inquiry after the update to A1111 1.6 images look blurry now, is there anything that needs to be adjusted somehow?

[spice pasture]

can somebody give me some guidance installing this stable diffusion? i followed this video exactly https://www.youtube.com/watch?v=3cvP7yJotUM&t=4s

but when i get to the point where i doubleclick on webui-user.bat it gives this text-
Traceback (most recent call last):
File "C:\Users\Adam\stable-diffusion-webui\launch.py", line 48, in <module>
main()
File "C:\Users\Adam\stable-diffusion-webui\launch.py", line 39, in main
prepare_environment()
File "C:\Users\Adam\stable-diffusion-webui\modules\launch_utils.py", line 356, 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

[tight quarry]

So, I am not generally a complainer, but stable Diffusion and SDXL seem to have a couple misunderstandings. When I prompted on several systems for " a faery prince playing a lute ...", guess what I get. I get princesses playing guitars. Why can't SD deal with sex.
see a

... and a lute is not a guitar.

[spice pasture]

ye a lute is not a guitar

so i am installing on AMD not Nvidea... i guess that's a problem?

getting this error- 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 . . .

and now i confused

[warm junco]

Maybe post an example image with your settings you used in #📝｜prompting-help

[warm junco]

You used the wrong installation method.
Check this for AMD. (webui directml)
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs

[spice pasture]

thanks a lot i will do that now

[spice pasture]

which settings will people need to see to help me??

[misty pollen]

I will go there but before as a follow up i had a custom trained hypernetwork on my test prompt, and removing that brough images back to normal. Something changed regarding that area maybe? Need to dig more.

[warm junco]

no settings needed, you have to follow the install instructions for any problems there is #🤝｜tech-support

[spice pasture]

posted in tech support

[pastel lynx]

Do we know if ControlNet will support SDXL eventually ?

[bleak matrix]

You can go test it out yourself in #1100170312106127410

[mighty knot]

how does sdxl compare to 1.5? for the longest time people have been using 1.5 over 2 or so, is sdxl the same case or does it have the potential to be better?

[jagged shuttle]

where should i ask my questions?

1. I want to find a prompt generator with image examples. I tell him what to do and he helps me to write the prompt. What do you advice me?

2. https://twitter.com/malicknoir/media i want to do something like this piece of art. What should i write ? 🙂

[true mauve]

sdxl for the most case is better than 1.5 if we are not considering nsfw

[true mauve]

prompt generators with images? I dont think there are many prompts generators that can look at an image and go yea that prompt say maybe try minigpt-4 as it can see images or if you can pay use openai's vision model

as for the style

[jagged shuttle]

well i want to do photos like the twitter i showed you just up. But I don't know what to write to have this style of photos. How should i do?

[true mauve]

for that if you want exact same train a lora

on those photos

but the style seems to be pretty similar to this model

[jagged shuttle]

i have no idea what a lora is

[true mauve]

basically a model on top of a model which tells it which kind of images it should make, for eg you can train a lora on your face and get realistic images of your face from sd which would be immposible with just default sd

you can find premade loras for free at-->https://civitai.com/

[jagged shuttle]

i see!

i feel really limited with what i want to create when i try something, it's never what i expect. I wish i could spend some time wiuth someone good at using it to improve my skills and not waste so much time writing my prompts for bad results.

btw i use : https://playgroundai.com/create
It's pretty great for the moment, but i'm not the best at using it.

[hollow ledge]

hey! I was stable diffusion enjoyer a few months ago. Now I see you have a bot! What settings does it use? When I was creating my images previously localy with stable diffusion they weren't any close to results im getting with bot. Could somebody guide me so I could achieve results localy?

[night mist]

Hey I need to ask. What are the clothes called. It's kind of like a kimono but it leaves the top open and connects the arms lower while you wear a sleeveless skintight black top

[jagged shuttle]

The clothing item you're describing is likely a haori. Here are some key points:

It resembles a kimono but is shorter in length.
Haoris typically have an open front and wide sleeves.
They are often worn over a sleeveless or tight-fitting black top, which is common in modern fashion.
This style is a blend of traditional Japanese attire with contemporary elements.

[night mist]

it also leaves the shoulders uncovered

[jagged shuttle]

The clothing item you're referring to, which is like a kimono but leaves the shoulders uncovered and connects the arms lower, might be a yukata. Here are the details:

A yukata is a traditional Japanese garment.
It is similar to a kimono but is typically made of lighter cotton fabric.
Yukatas often have a V-shaped neckline that leaves the shoulders exposed.
They are commonly worn during summer festivals and other informal occasions.

[bleak matrix]

If you want to learn more about prompting, etc, and getting more results, I suggest using the bot+styles. You can also check out my WIP prompting guide here: https://docs.google.com/document/d/1BxdWqfBJ3QPggHnBCBx3QIkUpHnBWShd_s3zEt2dTLM/edit#heading=h.po8dv7lcq48s for SDXL if you're wanting a more advanced guide on prompting. You can also ask any questions about prompts in #📝｜prompting-help which can help you improve your skills.

[bleak matrix]

Besides what I wrote above, I recommend checking out the #✨｜sdxl channel, esp if you are using ComfyUI, as you can check out the various ways that other people are using SDXL, with their settings, etc.

#🐝｜swarm-ui is also a great place to start

[hollow ledge]

thanks

[night mist]

The kimono/yukata itself doesn't cover the chest area at all but instead it is covered by that sleeveless black skintight top, leaving the shoulders exposed. I know I've seen it somewhere but for the life of me I can't find it now and I want to use it in stable diffusion

[hollow ledge]

so Im assuming that sdxl is some kindof new version that makes everything better right?

[bleak matrix]

#📣｜announcements message

[mighty knot]

yeah obviously

[gleaming garnet]

so, im trying to makea video of AI created avatar talking. Already have the png and voice, but i need an free alternative to D-ID, where i can ipload my own avatar.. any suggestions?

[quartz pumice]

Hi everyone, I am new here and not sure where to ask this question
I am getting this error

ValueError: ['latent noise', 'latent noise'] is not in list

also I can see the following in AUTOMATIC1111 UI under imgtoimg

everything is showing double

latent noise,latent noise
original,original
fill,fill

like this

shared image in #🏞｜general-with-images

[pseudo jetty]

does anyone know of a google collab that allows you to train Checkpoints? i know there are ones for Lora's and Datasets?

[main harbor]

I need opinions on sdxl 🙂 for a noob is it a good idea to work with it or should I stick with sd1.5

[dawn gulch]

@regal raven (going to cc @wintry stream because you're on - sorry!) - after speaking with a member of the Invoke community today, I have been made aware of a crypto scam being run to try to sell some Invoke AI coin. Might be b/c of the stable stage yesterday, but hoping you can help us make sure nobody falls prey to it. Thx
https://www.reddit.com/r/StableDiffusion/comments/167kagr/warning_invoke_aisd_community_being_targeted_by/

[wintry stream]

Thank you for letting us know!

I'll make sure everyone on the team is aware ❤️

[fossil osprey]

does everything that works in 1111 also work in comfyUI?

[distant flint]

I haven't seen a single workflow that works on A1111 but not comfyUI

[chrome marten]

Can someone help me

#🏞｜general-with-images

[radiant perch]

Lets see who wins the showdown

Wow two in pantheon wow

So grateful

[devout totem]

um @bleak matrix i have a question

[near pulsar]

for Image2prompt task I recommend you to try this one:
https://replicate.com/pharmapsychotic/clip-interrogator

[devout totem]

hi so um i have a question

[near pulsar]

where I can see whicj exact model used by these bots?
does it using base model only or there is refiner as well? and where to find all the model parameters which used by these bots?
thank you.

[bleak matrix]

Yes?

[covert needle]

hello

does anyone know where the hell pytorch installs onto ur pc

cus ive had to reinstall it like 3 times and it keeps using up space on my c drive but i cant figure out how to delete it

[bleak matrix]

I would go to #🤝｜tech-support and let them know what UI you are trying to install, as well as where you are trying to install it, and what you have done so far

[vocal vector]

So the #1051259766976040960 channel says to see the pinned post for details, but there is no pinned post for that channel…

[mighty knot]

Just curious since I'm looking for alternatives but which is the cheapest service right now for running sdxl or training it for GPU renting and such instance specs does it need the most?

[warm junco]

Open up a cmd and type
pip cache purge

That will delete it

[wanton hemlock]

where can I find a SDXL controlnet tile Preprocessor safetensor?

Is it not out yet ?

I only have the SD15

[bleak matrix]

Good morrow, everyone

How are we this beautiful morning?

[gaunt finch]

heyo got a dataset

trying to figure out how to do a style lycoris

[proper sparrow]

anyone knows how to keep control net background more stable

?

[distant gull]

I want to write a comment to the regulations.gov Artificial Intelligence Study

[fervent thunder]

how can I add stable bot on my server 🙂

[vapid oyster]

Has anyone tried Ideogram.ai? Have they released the model?

[neat cypress]

just published how to do SDXL LoRA training for free with both Text Encoder and UNET with Kohya GUI : https://youtu.be/JF2P7BIUpIU?feature=shared

[hushed star]

👋🏾 hello ppl, Can anyone give me a push towards a good video enhancer model for A1111?

[fervent thunder]

hi

[summer hawk]

I hope this message finds you well. As a professional Shopify store designer and dropshipper, I'm excited about the opportunity to collaborate and enhance your online store's sales and traffic. Leveraging my extensive experience in Shopify, I can craft a visually appealing, fully functional, and user-friendly online store aligned with your business goals. Would you be interested in discussing this further? I'd love to gain insights into your specific needs and provide you with a complimentary quote.

Additionally, I bring a wealth of experience in sales closing and social media management to the table. With a track record of success in these areas, I'm confident in my ability to elevate your sales figures and bolster your online presence. Let's connect soon to explore how I can contribute to your goals and achievements.

Best regards,

[vapid kraken]

https://colab.research.google.com/github/sagiodev/stablediffusion_webui/blob/master/StableDiffusionUI_ngrok_sagiodev.ipynb#scrollTo=gJBtadxGOK7b

I currently have a big problem. This google colab link just downloads AUTOMATIC1111 for me and updates it to the newest version. Now v1.6.0 just released but I really need v1.5.1

How do I set this back to v1.5.1

$10 if somebody can fix this

[small sand]

imagine an alternate universe where stable diffusion doesn't release and everyone keeps using disco and dalle

[fervent thunder]

im new to stable diffusion who can help me

[covert needle]

i feel like i REALLY shouldnt be trusting what a random guy on discord is telling me to put into my cmd window

lmao

[craggy helm]

I’ve been getting deeper and deeper into sd Python api.. my new project.
Autonomous bot that scrapes the latest news headlines and generates interesting art inspired by them daily, posts them to Twitter and IG

https://x.com/shmeeple_junk?s=21&t=k4r89zK0VnL5xeKgj7vfCA

[sand holly]

is it just me or is Automatic1111 taking up more ram when loading in a model lately?

because my whole system freezes even when only loading 4gb models after it filled my ram completely

didnt do that before a few days ago

[crimson willow]

why when I generate some images in the bot, it suddenly appears as a BIN file to download?

instead of just the image appearing, this is quite annoying

[echo umbra]

Can someone tell me why i'm getting this when I am using stable diffusion in google colab? I had absolutely no issues, but this is an issue all of a sudden

guess I can't paste screenshots in here

Traceback (most recent call last):
File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/webui.py", line 13, in <module>
initialize.imports()
File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/initialize.py", line 21, in imports
import gradio # noqa: F401
File "/usr/local/lib/python3.10/dist-packages/gradio/init.py", line 3, in <module>
import gradio.components as components
File "/usr/local/lib/python3.10/dist-packages/gradio/components.py", line 56, in <module>
from gradio.blocks import Block, BlockContext
File "/usr/local/lib/python3.10/dist-packages/gradio/blocks.py", line 28, in <module>
from gradio import (
ImportError: cannot import name 'wasm_utils' from partially initialized module 'gradio' (most likely due to a circular import) (/usr/local/lib/python3.10/dist-packages/gradio/init.py)

[weary gust]

Hey i'm getting an error while merging models to create an inpainting model

https://cdn.discordapp.com/attachments/1030337643428458528/1147771697412456468/image.png

[near wagon]

why are you getting an error?

[weary gust]

This is what i'm trying to find so I can fix it

[near wagon]

lol

so anyway, you should look at your console window

it'll give more details

if you take a screenshot or paste the error info I could help lead you in the right direction. no promises on fixing it, but it would probably lead us to the problem at least

[weary gust]

somehow I just re-selected the sd v1.5 inpainting.ckpt and it worked lmao

[near wagon]

well there you go

in the future just look at the console window if you want to know where the error is based

[weary gust]

https://media.discordapp.net/attachments/1145845681106468864/1147779614912028754/image.png

that was the error message btw

for img2img should I use the default model or the inpainting model ?

[near wagon]

so if you open extras.py and go to line 135 it'll show you where that error came from. then you can look at the preceding lines to see why it came about. not that you really need to do that now. but just saying, in general it can be useful

[weary gust]

oh okay ty

[weary gust]

and about that what should I use ? @near wagon

[wind magnet]

@bleak matrix Is there way to appeal ban?

[wind magnet]

@sudden marlin Do you have time to answer?

[halcyon galleon]

quels sont les reglages de base pour qu'avec Easy Diffusion v3.0.4 je puisse avoir en gros le même resultat que le test sur le Discord ?

what are the basic settings so that with Easy Diffusion v3.0.4 I can have roughly the same result as the test on the Discord?

[warm junco]

Yea thats true. But you can do your own research. Pip cache purge clears the cached python files giving you 5-7gb back as the were only used for the SD install

If you install SD again they will be downloaded again too

Here is more information on pip:
https://linuxhint.com/pip-clear-cache/

[bleak matrix]

Yup

There's an appeal form.

[wind magnet]

[bleak matrix]

https://forms.gle/FsEPmCzZWjh4suu2A

[coral lark]

What are the best ways to prompt for text in images?

In SDXL

[obtuse marten]

I am out of the loop, Is there an open source self hosted image to video?

[covert needle]

Chen my beloved

[floral umbra]

Would you guys happen to know the best free program that can upscale images "standalone"? As i love topaz gigapixel A.I, but ain't got money atm for refreshing the license. So i need the best free alternative. It's for upscaling textures for a game.
There's upscayl which is great, but wonder if there's better ones.

[sour needle]

[crimson willow]

why when I generate some images in the bot, it suddenly appears as a BIN file to download?

[dense flint]

hello.. how can I get started with my own models? I'd probably start from an existing model and feed it my own images.. how does that work?

[near wagon]

well, get notepad++ or something

[weary gust]

don't worry I figured it out, but that had nothing to do with notepad lol

[neat cypress]

https://twitter.com/GozukaraFurkan/status/1698364539639611852

[stone fox]

Hay guys. I have been using clip drop api and everything works awesome. Except, sketch-to-image. I keep getting internal server error with statusCode 500. help please

[cobalt crag]

What does prompt that have () do?

or {}

[stone fox]

?

[cobalt crag]

or something:1, 1

idk

[hollow ledge]

is stable code released?

and btw does anybody have link to google colab with current version of stable code?

[cobalt crag]

How good at English is required to be able to prompting everything as I want?

I wanna get good at prompting

but I have no one to help me

I don't have a coach

or a teacher to help me with this kind of bullshit

[lyric estuary]

https://tenor.com/view/how-much-money-we-talkin-about-money-how-much-asking-computing-gif-13233042

oh shit no gif

no fun

anyway this is good :

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

never knew () and [] functions before I read that guide, found it on a civit sample prompt

bottom line is you don't need to be grammatically correct 100%, but you'll definitely need correct spelling on each word you prompted. Just use spellchecker. EZ lol

[graceful spade]

hi

stablestudio and sd by same developers?

[somber smelt]

I have installed stable diffusion and all the requirements.
My goals is to create pictures which have different background dresses etc. with the face of a specific person in a realistic style.
Where do I start?
Do I use img2img or should I get a prompt off of the person's face I mentioned?

[lyric estuary]

Why don't you try both approach and see which one fits your need better?

The output are quite random and you'll need to batch em then pick the best ones anyway.

[fossil osprey]

hallo, this might be a weird question, but i have a small team making some AI stuff and im looking for something like a online corkboard where we can post both the pictures and right all parameters to it, for a faster workflow. do you guys have any suggestions? pls dm with the suggestion thank you 🙏

[lyric estuary]

There's no surefire way to generate good results. Been playing with SD for at least couple of months now and I'm still learning as we go.

[lyric estuary]

miro perhaps?

[scenic bluff]

HI anyone knows why some image are uncensored ?

[neat cypress]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/13038

[primal current]

Dear friends, do you know pages, places or people who publish contests, calls, competitions, prizes related to AI art? I got few, I would greatly appreciate if you can give me info, whoever wants I can share the info I have, regards.

[near wagon]

It's a program that edits python, ya dingus. Notepad++. Try typing it into google

[near wagon]

Hello, PeePeePeePooPooPoo, what kind of contests do you know about?

[primal current]

joyn.xyz

[sage tendon]

I want to create a man posing for a photoshoot using specific items (shown in linked tweet) such as a specific shirt, pant, jacket, belt, hat etc. How to do it? https://twitter.com/nalingupta01/status/1698510833456763242?s=20

[fervent thunder]

The idea behind AI to generate images. Is it a combination of algorithms that detect shapes, colors, depth, position, etc? Like google search for images of dogs, it looks for images containing dogs.

I myself cannot draw a dog, much less photorreal. What did they do to make it generate a dog that isn't like a kid drawing a dog with no knowledge of shapes, size, propotions, shadows, light, ...?

[near wagon]

https://stable-diffusion-art.com/how-stable-diffusion-work/ this explains it pretty well

[fervent thunder]

with AI is it possible to specify images from already existing ones? Like. I want the same scene, but change time of day and weather, keep everything else intact. Or change the camera angle. Or draw from a sketch with the silhouette of what you want.

I imagine the thousands of labour hours that would save in cinema, anime industry and TV

[crystal plover]

Morning guys, I am now finetuning a realistic style model, my dataset has 500 pictures(nikon8n7x), regularization has 2500 pictures(realistic). Is these amount is enough?

[near wagon]

Yes

[placid sage]

I've seen clips of people using some interesting workflow for stable diffusion that uses a flow chart of some kind. I think it has something to do with SDXL or something?

I've been lazy and just using auto1111 webui, simply doing prompts and negative prompts. but I want to start diving into whatever that is. Can someone point me in the right direction?

[pastel lynx]

Hey! Anybody ever had the no module named tqdm.auto issue when starting the webui ?

[fervent thunder]

damn

[west prism]

Good morning Sirs

any new Cirnos

oh wait wrong touhou discord

[fervent thunder]

so after I use all my free trials of SDXL I have to pay?

[west prism]

only if u want to

[lyric estuary]

ComfyUI

[slender kraken]

Good Morning Team,

Am i missing a setting somewhere? I am running SD on a 1070, i know it isnt ideal, but it works

My issue is with XL though, it always error's out with a memory error

Under normal circumstances, i would just chalk it up to, GFX card is sh1t, to bad to sad .. but why is it SD cant generate a 512 image, but comfy can generate a 1080 image on the same model with no error?

i have --xformers --no-half-vae in my gui.bat

[warm junco]

You need to add --medvram-sdxl

[slender kraken]

ok awesome! thank you ❤️

does order matter when loading arguments?

[warm junco]

Nope

[slender kraken]

cool, gonna give it a shot now

[warm junco]

Should work for you if your on the 1.6.0 (latest) webui version

[slender kraken]

You are a rockstar!

That worked perfectly, thank you

[cobalt crag]

Is it possible to generate image with ONLY few Steps?

10 Steps or 5 Steps

[still glacier]

try it, result will probably suck but it's possible. Results will vary depending of what sampler you're using.

[still glacier]

Also don't try to generate 512x512 with SDXL, it's not made for such low resolution.

[slender kraken]

with the --medvram-sdxl,

will i be able to go larger without running into mem issues? (just left home, so cant test)

[eternal oyster]

Hi, I tried to download AUTOMATIC111 For Stable Diffusion I installed Python 3.10.11 but it gives me an error when I create the image https://pastebin.com/fS0c5bP4

[still glacier]

larger than without it for sure.

[still glacier]

Well, what's your gpu and did you Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion as suggested by the error ?

[eternal oyster]

GTX 1660 Super, I've already tried those settings, same error

[eternal oyster]

I tried it a while ago and it works

What is module xformers "no module 'xformers'. Processing without..."? @still glacier

[cobalt crag]

so the more steps I add

the more it will be better?

100 steps?

200 steps?

or let's just say 30 steps

[eternal oyster]

So you can help me please, I can't find a solution, I've tried a bit of everything

[warm junco]

That means your not using xformers. To change that edit the webui-user.bat
And at the COMMANDLINE_ARGS= line you add:
--xformers --medvram --no-half --precision full

[still glacier]

1600 series have to use the --no-half flag

yeah what he just said

[eternal oyster]

It works now thankssss

[hybrid harbor]

How do I generate images?

After /dream?

[near wagon]

these questions 😂

[hybrid harbor]

:/

[near wagon]

tbh I haven't ever used the bot. I think there are guidelines listed somewhere though

in one of the channels

[hybrid harbor]

So I do I generate images?

how*

[tepid condor]

I know this doesn't fit in this server but is anyone familiar with KoboldAI? Does anyone know what GPU Layers are and how much I can maximal use?

[still glacier]

There are many many clients/ways to do so. There are the bots channel here for that or you can use some online services such as dreamstudio or do it yourself locally.

[stable pawn]

I installed the new InvokeAI and ran it, but it is slow for it have to load the SDXL models each time I generate a image, that is 20 seconds of loading before it can generate an image that take 9 to 10 second to finish, if I do several images in a row there is no loading, it just when I press the "Invoke" button it start to load, 20 seconds may not sound much, but 3 images will be over a minute and it is soon up in 30 to 60 min of time waiting.

Is this normal or something that may be bad in my instalation?

[empty umbra]

Heya all, does there is a database of most trained words / character in SDV / SDXL ?

I tried to make the 27 club but looks like some musician aren’t going well in SD (vanilla 1.0)

[cobalt crag]

How to use Lora in ComfyUI?

Does Lora and my Checkpoint need to be the same version?

for generate any image?

[grave talon]

After the latest update I can't find the "tiling" button anymore, where did it go?

[slender kraken]

I have a question about a lora, i trained my own model, and it def shows, but want to get my prompt ready, lets say my lora is called m3gan, and its trained on khoya with a class of woman, what would be the better prompt?

Instagram photo of m3gan, lora:m3gan:0.85
Instagram photo of lora:m3gan:0.85
Instagram photo of woman, lora:m3gan:0.85

[slender kraken]

not sure about all the questions pertaining Comfy, but yes, lora and checkpoint has to be the same version

1.5 lora only works on 1.5 models, etc

[cobalt crag]

What kind of stable diffusion checkpoint that is for all-purpose stuff?

Anime, Realistic, Scenary, Artstyle Versatilities

everything at one?

[lyric estuary]

1.5 pruned ema only

Or the base sdxl with its refiner if you got the horsepower

[cobalt crag]

How to "weight" in comfyui?

and what are Separate UNet/Text Encoder weights?

[warm junco]

You can try Dreamshaper v8

[cobalt crag]

Dreamshaper?

hmm oke

since I deleted Automatic1111 now

It's so hard to use, idk why

Imma try that

[warm junco]

Comfy ui is harder than auto imo ^^

[cobalt crag]

Idk, but I love Comfy more than Auto

It's like you gib gab that connect this connect that and boom

[warm junco]

Yea thats pretty cool as you can edit everything and make your own personal workflow

Also sdxl is still faster in comfy

[fervent thunder]

sdxl 1.0 pro only

I tried the other versions, the images are a lot different

[cobalt crag]

I would prefer SD 2.0.

cuz it allow me to generate in HD latent image as 768x768

an upscale model for anime artstyle is not exist :(

I tried find it Civitai and nothing found

[fervent thunder]

what does weight 0 do?

negative means "I don't want" ?

[cobalt crag]

How to "weight" in Comfy?

I only see something like
"Strength_Model"
"Strength_Clip"

[hasty hornet]

Might work on UniPC sampler

[cobalt crag]

UniPC?

What is that

[hasty hornet]

for everything else ~20 steps is kinda baseline (give or take 5)

[hasty hornet]

sampler

[cobalt crag]

hmm ok

Btw

Anyone have Aspect Ratio for me?

so I can do stuff in Latent Image blah blah idk

[candid kindle]

hey guys! how yall doing?

i'm new to this thing... where's the best place to get modules?

[cobalt crag]

Civitai all the way

[candid kindle]

and i saw stuffs about using stable togheter with photoshop, how can i get this?

[candid kindle]

ty 😄

[bleak matrix]

Good morning! Happy Labor Day! How is everyone?

[candid kindle]

morning! fine. what about u?

[bleak matrix]

I think I'm doing dandy! Just working on some rigging and animation in Blender!

What are you up to?

[knotty turtle]

Can Lora models trained on SD 2.1 be used with SDXL?

[candid kindle]

just started using... i tried in the beta a while ago but i quit the AI world (cuz lack of time), now i'm back and its a lot more complex lol

[bleak matrix]

Yeah, the AI world moves fast for sure! Well, you can def check out the bot if you haven't already!

It's amazing how far the technology has come in such a short time.

[candid kindle]

yep, works like a charm, but now i need to work with come inpaint or img2img stuffs and i'm rly lost lol

for example, in dall-e, you can upload a PNG and it will read the transparent pixels and inpaint on it, so you don't have to brush it all... stable have something like this?

[cobalt crag]

nope

[cobalt crag]

That is called img2img

and we do have that

Both ComfyUI and Automatic1111 from WebUI are available for that

[candid kindle]

yeah i installed this automatic1111

and a couple of modules rn

so if i upload a image without background to img2img, will create a background? (if i set up correctly of course)

[pale latch]

sd15 mixes can still suceed with higher resolutions. the base attention is just lower

for upscaling, especially for anime and using something like controlnet tile with ultimate upscaler, you're off to the races

[cobalt crag]

hmmm

[hard marsh]

i need help downloading control_v11p_sd15_inpaint

the inpaint model

[pale latch]

https://huggingface.co/lllyasviel/ControlNet-v1-1/resolve/main/control_v11p_sd15_inpaint.pth ?

[hard marsh]

wtf

how did u get that file

im on files and versions and its not there

[pale latch]

https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main

[hard marsh]

wow ty

im blind or smth

[pale latch]

my eyes can only see the obvious as soon as i ask somoene to help me out too

[fervent thunder]

is it possible to transfer creations from one acc to another?

[pale latch]

account?

[fervent thunder]

yes

[pale latch]

you want to transfer discord bot created pics to a new discord account? no. not possible. just save them. The server logs aren't your file storage

[lavish lake]

I'm sorry to say that it's not possible, the bot deletes images after some time, if you want to save them you have to DM or right-click save them

[fervent thunder]

what's the difference of using a sdxl on nightcafe and installing invoke ai?

[distant flint]

Anyone knows why DPMPP_3m_sde has overall a nice convergence on the preview, but when it outputs the image it's got a ton of colored noise?

(same with dpmpp_2m_sde, just slightly less worse)

most of that noise gets picked up by the refiner no problem, but it means the result is absolutely unusable without refiner

[earnest solar]

where should i add

[bleak matrix]

There's a lot of info in #1080946152318443610

[bleak matrix]

We've also done several stages on how to use SDXL, which you can actually see recordings of here: https://www.youtube.com/@stablefoundationai/videos You can also use #🐝｜swarm-ui as well, too

[warm junco]

In the webui-user.bat at the line COMMANDLINE_ARGS=

[cobalt crag]

How to get UniPC btw?

[warm junco]

You can select it at the sampler dropdown

[cobalt crag]

But I use ComfyUI eh?

oh nvm

I see it lmao

What are these samplers' difference btw?

I have no cloe

Dpm Euler

and many things

[cobalt crag]

Help

my model is broke after I used UniPC

and when I go back to use DPM++

the image generated in pixilated rainbow crystals illustration :(

[fervent thunder]

Samplers are just different math equations

DPM based ones, and euler are used often

[cobalt crag]

Idk what to do

Oh wait

OH!

I accidentally used VAE from SDXL

[hasty hornet]

https://stable-diffusion-art.com/samplers/

[cobalt crag]

cool indeed

[cobalt crag]

controlnet?

what is that?

[candid kindle]

thank you so much 😄

[bleak matrix]

NP!

[fervent thunder]

is it possible to generate the same image again?

[spice plank]

anyone got any lesser know site with loras? I got civit, pixai, liblibai.

[pale latch]

its' a pretty big deal and makes, imo, 1.5 far more powerful than 2.1

there are a few controlnet models for 2x but not as many as what's available for 1.5

[wary belfry]

depending where you created it you can put it in png info in Auto1111 oder drop it into ComfyUI. Then you have the same exact setup. Then when you re using xformers you always get slightly variation, even if everything is the same

[lofty girder]

are there any plans to merge the directml fork of webui to main?

[fervent thunder]

I wonder if AI can save thousands of hours spent generating frame by frame animations and other things in which all you want is to change lighting, weather, position, erase elements, etc.

[wary belfry]

a lot of work, but corridor digital already did it 😉 https://www.youtube.com/watch?v=GVT3WUa-48Y

episode 2 is even better

[dire mulch]

sleep broshi

hu

[trim yacht]

Hey guys, question, A macbook M2 max with 96gb unified ram or a beefy PC with a RTX 4090 24gb ? I know people say get the PC and remote in via other methods but just trying to future proof as both options are not cheap!

[warm junco]

a pc with a rtx4090 will be more future proof and will perform a lot better in AI tasks than a macbook at any time

[rough ridge]

can somebody help me, ive been trying to create workable concept art for 3d modeling for makeshift ork melee weapons for a few weeks now and i cant seem to get it right

i need a mentor or guru

[unborn perch]

Hey guys, I got a question. When I generate I click generate image even after using the same prompt it generates very different images from each other. Is there a way to tell stable difusion to generate images that look more like eg. generated image #2 ?

[placid sage]

Is there any way to use the same models folder for auto1111 and comfyui? Please let me know the trick.

[honest light]

can someone tell me how tu use the style command

[bleak drift]

I find it hilarious that some sites won’t generate a depth map for a photo if it’s “explicit”.

And the other site I was using discontinued, saying to use Stable Diffusion as it’s much better.

Like bruh. The only reason I used the site is because I don’t have a computer built beefy enough for stable diffusion.

[broken cave]

on windows there are a lot of ways. you can specify a models path using the comfyui configuration file. you can create a linked directory.

[solar condor]

hey whats up guys !

im new to this and wondering if i should download thecomfy ui

instead of automatic1111

[fervent thunder]

If your new I would suggest automatic

Everything is streamlined , comfyui required you to know what tools to use as you have to manually put each step in place

[solar condor]

oh okay, and why do people use comfy ui :)?

[lyric estuary]

Flexibility.

Just add nodes gradually as you go or vice versa.

You'll understand this if you experienced with node based softwares before.

Image2image workflow makes more sense in ComfyUI.

Load 2 models (or more) in a single pass, mix and match them for final touch or fine tuning, etc.

[solar condor]

damnnn maybe i should learn that haha.. i already use nodes a little bit in davinci resolve

thanks btw for your answer, the only thing that sucks is that im on mac

and i feel everything is more complicated

😢

haha

[lyric estuary]

Well, if you just started, yeah A1111. It's a rabbit hole and there're a lot to learn.

prompting, controlnet (depth, openpose, scribble, etc), lora, inpaint, etc etc

should understand the basic first and I started with A1111 first tho. Comfy still have a lot of problems on my side, naming I can't get control net to run.

People make fantastic results with A1111.

[unkempt hatch]

if you used Davinci, then you're smart enough. The nodes give way way way more power, so if you're serious, go withcomfy

Someone told me you can fine-tune BLIP2, I can't find any info abouut it (or any fine-tunes). Any leads?

[obsidian terrace]

Hey y'all I'm trying to move from auto1111 to comfy. Trying to figure out how/whether the "variation seed" and "seed resize from" options(that are found in the extra section in auto) are implemented in Comfy. What nodes should I be using for that stuff?

[lyric estuary]

ksampler node?

[obsidian terrace]

I'm not sure which part of it allows me to set a second seed for a generation

[warm junco]

Comfy has a file you need to rename and there you can add the path to the models

[keen marten]

Is there a ppt for introduction to stable diffusion. I'm giving a talk in my company's townhall about stable diffusion and how we can use it in our company.

[umbral mantle]

I want to use stable diffusion control net and everything it has without installing it locally. Is there any online / cloud service ???

[wise aspen]

i have question, if i using controlnet for SDXL, and render will be 1024/768 - what should be preprocessor resolution in this case?

[languid helm]

yo my final year project was rejected by my professor 💀
it was about using a text2emotion model with stable diffusion to automatically create a comic from story book ( without any prompts, the prompts are obtained from the story book ) and controlnets ofc

she said there is no novelty :')