#šŸ’¬ļ½œgeneral-chat

1 messages Ā· Page 73 of 1

potent spire
#

which one?

deft epoch
#

The one people say produces high resolutions

#

and does not have backwards comp with SD models.

#

Idk the name.

potent spire
#

i dunno, im testing Alpaca on PS right now

ripe python
#

hi

#

my slash dream isn't working

#

anyone know why

deft epoch
wanton nacelle
#

Just the main command

#

/

#

dream

#

But in one message

ripe python
#

ya but it isn't working

#

/ dream

#

ik what it is

wanton nacelle
#

No I know, I was answering Ryu

#

I donā€™t know why itā€™s not working though. šŸ˜ž

#

Also, does anyone know if thereā€™s a chatbot or some sort of AI tool made to refine prompts for stable diffusion? Because I donā€™t have the first clue about all the weird prompt engineering tags and ā€œweightsā€ and stuff and it has no idea what Iā€™m talking about.

wanton nacelle
limber harbor
#

how is all?

balmy nova
#

Hi hi Stability fam ā¤ļø hope that you're all good and groovin'!

#

Is anyone successfully using Stable Chat? I was so excited to read the release, and wanted to get started playing with it tonight.. However after signing in I just get an endless spinning wheel on all browsers/devices...?

wanton nacelle
balmy nova
wanton nacelle
torn charm
#

Is it possible to use an init image with infinite zoom (auto1111)? I haven't installed it yet and was wondering if it has that option. Is anyone using it?

empty grail
#

Can anyone tell me if there are any great anime models? and i want to know how i can create my own model :((

balmy nova
balmy nova
wanton nacelle
surreal raven
#

what the prompt to use for "heel-lees boots", imagw being generated still has heels, ive used no heels, no visible heels but still not working

little citrus
surreal raven
modest storm
#

is there a way i can invite the image bot to my server?

normal pagoda
#

Morning šŸ„±

#

I wonder where the 18+ channels are... Got this server recommended from the civitai discord

turbid whale
#

LOL

normal pagoda
#

Huh? mog_nani

fervent thunder
#

are the previous comfy stages on youtube or something?

twilit valley
#

Hello šŸ™‚

#

I have been using clipdrop for weeks. Itā€™s very impressive

#

I bought a new pc with tons of vram to run SDXL with comfyUI. I thought it would be the same model with more control, but in two days of tinkering I canā€™t get results as good as clip drop

#

With either comfyUI or automatic1111

#

I donā€™t understand how the quality can be so different

bleak matrix
#

Good morning, everyone! How are we all today?

gleaming ocean
#

is there anyone here who could perhaps help me get controlnet to function?

bleak matrix
gleaming ocean
#

there was no response sadgeSitSip

bleak matrix
#

I would wait for a bit, or just try again later. It's pretty early in the morning! But I am sure others will be up soon enough

gleaming ocean
#

alrighty

normal pagoda
bleak matrix
normal pagoda
bleak matrix
normal pagoda
#

On the other hand, I have a slight problem with images I generated using A1111... Can't seem to find the prompts I used laguna_sweat

bleak matrix
#

Like, you can't find the images, or specific images?

normal pagoda
#

I have the images on my drive, but don't remember what the prompt was laguna_sweat

#

Those are my own, by the way... Am new to AI art

plucky shoal
#

open image in PNG info in A1111 should give you prompt and other details.

normal pagoda
#

Will try it out when at home again

gleaming ocean
#

Aha Png info is handy

plucky shoal
#

Also you can in img2img interrogate images, to get description what is on picture. It is reverse img2txt basicaly šŸ™‚

normal pagoda
#

Ah, that's also something

bleak matrix
normal pagoda
plucky shoal
#

in ComfUI, it working way you drag drop image and you have everything as it was rendered. Nodes, pos neg texts simply everything

bleak matrix
#

I have most of stuff set to png and txt, cuz you never know what'll happen in life.

normal pagoda
#

I made a nsfw male na'vi in the morning - need to refine it a bit so I can hopefully get a picture of what I'd look like when visiting Pandora (James Cameron's Avatar peeking around the corner! tifa_peek )

bleak matrix
#

Score for C o m f y UI

gleaming ocean
#

i tried comfy ui but my brain no work

normal pagoda
#

A1111 is lightweight in using - although my graphics card is a bit weak laguna_sweat

#

Web-based was my way to go because of that

plucky shoal
#

@gleaming ocean i am using only most simple workflow. For SDXL is it must, performing much better than A1111

gleaming ocean
#

my gpu isnt great already im running this of off a rented gpu i paid for .. due to a work thing

normal pagoda
gleaming ocean
#

ah cant embed in here

#

that explains it its a 30 GiB RAM 8cpu 16 GiB GPU

#

i was using it for the creation of a neural network

#

and when i finished work i re-purposed it into image generation for fun since im on vacation now

normal pagoda
#

Phew... Can't keep up with my GTX 1050 laguna_sweat

gleaming ocean
#

haha i have a 1060 in my pc myself so we're pretty much equal

normal pagoda
#

It still works for decent quality, though - just need to upscale the generated images in a two-step generation or in the img2img tab

gleaming ocean
#

i mostly came to this discord for some controlnet help because i cant make it work.. sunny told me to go to tech support so i posted my issue there

normal pagoda
#

I came here since I need a place other than civitai.com to post my nsfw images

gleaming ocean
#

haha

plucky shoal
#

I would like to help you with controlnet, but i installed it somehow i dont know how.

gleaming ocean
#

haha thats pretty much where im at everything is installed and ive put models in like 4 different folders without any success they show up but they dont work

bleak matrix
#

Makes it much easier to understand

#

Also, we have had two stable stages on it, and there's a third on Friday

plucky shoal
#

in extension i have sd-webui-controlnet and there in models i got controlnet models šŸ™‚

#

i suppose you have to install sd controlnet extension from extension

bleak matrix
gleaming ocean
#

yeah i thought nsfw anything was tos ?

bleak matrix
plucky shoal
gleaming ocean
#

how but how does the bot help me basicly i want two characters holding up a planet

#

wait am i just dumb? i grabbed the "models" of off huggingface.co do i need to Put actual like Line art models in one of the folders?

#

i have the .pth and yaml files

plucky shoal
#

you have to install controlnet, from A1111 extension and then in subfolder there are models and here you put them. Link i send are pruned, 705MB model. Safetensors and yamls

gleaming ocean
#

they're in there the original file size once

plucky shoal
#

i dont understand

#

stable-diffusion-webui\extensions\sd-webui-controlnet\models
here goes models and yamls

gleaming ocean
#

they're there

#

cant post screenshots in here

plucky shoal
#

so you probably need proper model to be used with, have you tried 1.5 2.1 and such models?

gleaming ocean
#

so in the Model in controlnet i have two options

control_v11p_sd15_openpose-checkpoint [e3b0c442]
control_v11p_sd15_openpose [e3b0c442]
neither of which work

#

It generates an Openpose image right as a preview but it doesnt add said pose to the prompt ive added

bleak matrix
#

For that, I would start simple, like, "two people holding up a planet." If you haven't done any prompting at all, there's quite a few prompting guides and sites to help you get started. You can also start here with my guides: https://atypicalconsortium.carrd.co/

plucky shoal
#

You should try simple like candy. How openpose work, i think you have in txt2img insert image you get from openpose. That you have preview is sign it is working imho

gleaming ocean
#

assume i want to position the two people holding the planet ontop and bottom and ive created the open pose models for it controlnet is always positioning the people to the left and right

bleak matrix
#

But if you need help with prompting, I would check #šŸ“ļ½œprompting-help By top and bottom, do you mean like the camera view is looking down, kinda?

gleaming ocean
#

i dont need help with prompting i want controlnet to apply a pose to two characters

olive dagger
#

hey guys, what was this technique that was making SD use one prompt until certain frame, and then another one in the rest?

#

it was something like (apple||skull:25) or something

#

ah ok it was [apple:skull :0.5]

ionic belfry
#

I'm using vasat AI to train a sdxl lora however I can't find a log file to monitor the progress where can I find the log file?

frank moon
#

Is it possible to generate a female wearing a specific t shirt? For example, for a clothing line as a way to not use real humans.

deft epoch
#

Huh?

cobalt crag
#

So uh

#

I have a problem with Stable Diffusion

#

Black Image generated

plucky shoal
#

maybe high resolution, what is your gpu? @cobalt crag

cobalt crag
#

gtx 1650 ti

plucky shoal
#

and resolution you are trying to generate?

cobalt crag
#

I put --no-half --disable-nan-check --precision full --medvram --opt-split-attention in in the webui-user.bat file

#

it's not work

cobalt crag
plucky shoal
#

o.k. disable nan check i think ppl here do not suggesting it

cobalt crag
#

But If I put out --disable-nan-check, I will get an error

#

and the image will not generate

plucky shoal
#

but disable nan check only hide that error.

cobalt crag
#

okay..

#

welp, I have 0 knowledge with coding lmao

#

Imma run the bat file with web ui again

plucky shoal
#

post error you get

cobalt crag
#

NansException: A tensor with all NaNs was produced in Unet. Use --disable-nan-check commandline argument to disable this check.

cobalt crag
plucky shoal
#

it would need some experts here.

broken cave
cobalt crag
#

nope

#

What is that

plucky shoal
#

I dont know realy, should work, try lowvram insted of medvram if something changes

broken cave
#

have you ever managed to generate an image?

#

the error is related to a flawed installation, usually due to extensions or loading a prompt from an image incorrectly / using it incorrectly. you probably have other exceptions that you are not noticing

cobalt crag
#

Idk hmm...

#

I installed everything right I think?

broken cave
#

have you ever generated an image?

cobalt crag
#

nope

broken cave
#

okay well

cobalt crag
#

0 knowledge

broken cave
#

yeah

cobalt crag
#

but I would like to try

broken cave
#

you shouldn't put in any arguments

cobalt crag
#

But I must put --no-half --disable-nan-check for sure

#

Otherwise, I will get this error

broken cave
#

it might be simpler to use comfyui, which doesn't require any explicit configuration.

cobalt crag
#

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.

cobalt crag
broken cave
#

you can do it, just like you can generate images using only a CPU

cobalt crag
#

I'll try that ig

broken cave
#

it has tradeoffs

supple coral
#

cpu normally way slower right

still glacier
cobalt crag
#

What is Lora?

plucky shoal
#

Lora is some small extension to model focused on some specific things. Like for example lora for star-wars imperial costumes only.

cobalt crag
#

Are there any checkpoint that can generate male anime character?

#

All I see on Civitai are mostly focusing on female characters

broken smelt
#

they should all be able to generate male characters too, although its true that some could be more inclined to generate more feminine looking characters

#

keep in mind that those pics on civitai are just what people are generating and uploading when using the model, they should all be capable of generating things of other categories too like food/animals/etc

still glacier
#

Don't confuse lora with models tho. Models (unless trained very hard) like rootbeer mentioned can do pretty much anything (with a varying degree of success). Lora on the other hands are """"""extensions"""""" that will alter a model to make it "better" at a certain thing (cat ears, specific character, etc). They're trained to work with a specific model. For example lora trained for sd1.5 will work with sd1.5 and (to a certain degree) all its derivatives but NOT with sd2.1, sdxl, etc

grave stag
#

So I'm trying to learn loras and all that jazz, I thought start simple use a lora and a checkpoint that was one of the reference images on the page for civitai then edit the prompt and go from there. Problem is I am trying to do that and I am using the same style, checkpoint, lora, prompt, everything and it still won't turn out the same cartoon anime style it has some weird photo style to it instead and doesn't look like the image provided even with the same seed. It looks nice and like it should for the prompt but not like the example I like more can anyone help me with why that might be?

still glacier
# grave stag So I'm trying to learn loras and all that jazz, I thought start simple use a lor...

Here's a list of the usual culprits :
1/ --xformers (and some other flags but mainly this one) will make stable-diffusion non deterministic. Meaning for the same prompt+settings running on different hardwares you'll get different results. The output should still correspond to whatever the prompt is.... It might just be slightly to very different if you're unlucky. So yeah, some command line arguments can alter the outputs of your prompts and your hardware might also do the difference.
2/ Do you have all the loras, embeddings mentioned in the prompt ?
3/ Are you really using the same settings ? Copy pasting manually what's inside the "Generation Data" boxes is not enough. Civitai is hiding a bunch of data there. If you want to get the exact same settings you'll have to click the "Copy Generation Data" button at the bottom of the image page, paste its content into Auto1111's prompt field and then click the blue arrow under generate to automatically apply each value to the correct field
4/ Did you have any overrides set beforehand ? They should show up at the very bottom of the page (some will probably show up too if you use data from civitai like I mentioned in 3/)
5/ Maybe it's using some extensions that does not record its settings in the metadata
6/ Maybe you're using different versions of some extensions/auto1111/lora/models/etc that yield different results

#

tldr : don't waste too much time trying to get the exact same picture, you might never get it.

#

let me know if it helps, I can take a look at the pic you're trying to reproduce too (keep it sfw)

grave stag
# still glacier let me know if it helps, I can take a look at the pic you're trying to reproduce...

OK yeah I've learned getting it exact won't work well and yeah I do need to do the copy generation data and then click that blue button not just copy it manually. That said I can't even get the same art style for the image. I'm starting out easy, create a fairy tail girl not some girl I'd use for a ref pic. I downloaded the lora and checkpoint that went with it, I clicked copy generation data, I clicked that blue button after pasting it and I clicked generate, nor only is the girl with a different background she is also in a different anime art style a way more muted colors look. I want the other style and I can't figure out how to get there no mater what I do I just can't. I'd send a pic of what I'm talking about but it won't let me here. Yes I've checked other then maybe a different version of automatic1111 I have the same version of everything.

still glacier
outer leaf
#

Hello how to have Ć  script with an image

hybrid wind
#

Anyone here have, has had or have tried the Corsair ST100 headset stand?

drowsy robin
#

Hm.

astral flicker
#

GUYS WTF - even gpt3.5 can interpretate openpose json

The person's head is located at coordinates (0.390625, 0.216796875) with a high confidence score of 1.0. The shoulders are positioned at (0.4322916666666667, 0.28515625), also with a confidence score of 1.0.

The left arm is bent at the elbow, with the left elbow located at (0.3567708333333333, 0.287109375) and the left wrist at (0.3203125, 0.359375). The left hand extends further to the left, with keypoints of the hand detected as well.

The right arm is also bent at the elbow, with the right elbow located at (0.5078125, 0.283203125) and the right wrist at (0.6223958333333334, 0.36328125). The right hand extends further to the right, with keypoints of the hand detected as well.

The person's hips are positioned at (0.4166666666666667, 0.48046875) with a confidence score of 1.0. The left knee is at (0.3645833333333333, 0.619140625) and the left ankle is positioned at (0.3697916666666667, 0.76953125). The right knee is at (0.5052083333333334, 0.623046875) and the right ankle is positioned at (0.5416666666666666, 0.763671875).

Overall, it appears that the person is standing or in a neutral pose, with both arms bent at the elbows and the legs slightly apart. However, without visual representation, it is challenging to provide a detailed description of the person's exact pose.

#

now we can use LLMs to automate the openpose creation

fervent thunder
#

When will 1.1 be released?

young ivy
#

@astral flicker still easier to use a dummy model in blender etc to do a pose if you've practiced how to move the models. Works best when they have a muscle muscle system too so you can't pose unnaturally unless of course you click and turn it off

#

then use substance various tools to make a skin texture, if you have the money for the software

astral flicker
#

@young ivy idk about blender automation. the gist would be to get a system to know which vectors represent which movement. did you read about the soutpark automation? they used an special llm to create vector poses and added them to a storyboard

young ivy
#

that was the wrong tag :p

#

better edit it

astral flicker
potent spire
#

You can script in Blender as well

#

Or look up for scripts as addons

young ivy
#

there's also SD plugins for blender, I think there's one that opens up twoway communicaton

astral flicker
#

That sounds really complicated. Hence it's already three dimensions ig. It's python right?

young ivy
#

blender probably uses java for scripts no?

potent spire
#

Im using 3ds Max as my main 3D software, but Blender must be capable as well

astral flicker
#

java would be fine too. Also gpt4 will be smart enough for such a task. i'll dig into that topic thanks

#

is 3ds Max an adobe product?

young ivy
#

I dunno, but substance is

#

substance model, substance texture etc

#

they bought em out

#

autodesk maya is used by South Park

potent spire
astral flicker
potent spire
#

Substance is goat

#

Mari is goat for texturing for movie characters

#

Cgi

astral flicker
#

so you're a real digital artist, nice. i have never tried blender et. all besides the donut tutorial and some basic cad for my 3d prints

young ivy
#

Cinema4d is standard for a lot of VFX houses that make movie effects

astral flicker
#

im more like a practical effects guy. like adam savage haha. i have to check out all the tools you've listed here sounds incredible

young ivy
#

I'm good at or was good at level design etc. I used to make the most eyecandy stuff in Oblivion etc. Now though, I'm out of practice

#

very out of practice

#

I started level design with half-life

astral flicker
#

The movie? Damn haha. Now I have to rewatch it es?

young ivy
#

then half-life 2, then Oblivion, then nothing

#

I used to make my own space battles with star trek ships in....what was it called...wavelight?

#

I dunno if it still exists

#

No not wavelight

#

but something with wave?

astral flicker
#

I'm gen Z so I started my visual journey with Nintendo lol - not that exciting. I know a bit about the demo and shader scene from the 2000s

young ivy
#

Lightwave!

#

12 year old me made a few star craft maps

#

they sucked

#

I also modified the Ultima Online world map when I was 14 and built new islands or removed existing ones

#

It was annoying cause it's tile based

astral flicker
#

lightwave looks like blender for an amateur like me lol

young ivy
#

It was incredibly convoluted in the early 2000s

#

but most 3d software was

#

Not like now when you click a button and it creates a perfect light system

#

or fluids flowing around or against static objects

astral flicker
#

2000s graphic effects have some ring idk. I like the prequels' aesthetic look too. But that's an unpopular opinion

fiery plume
#

guys where is the locations of Torch files should be

young ivy
#

Farscape used 3ds max

astral flicker
#

i feel like garrys mod, half-life realted bashing have a renaissance rn. We are getting to the 20 year cycle

young ivy
#

CS is older than you

#

think about that

astral flicker
#

Haha

fiery plume
#

guys

#

im havin a problem

#

can anyone help me

astral flicker
#

Call Huston

fiery plume
#

whos that

astral flicker
#

What are you talking about? SD?

young ivy
#

how do we know if we can help you? You haven't told us the problem

fiery plume
fiery plume
potent spire
#

if not already

young ivy
potent spire
fiery plume
young ivy
#

if they are specific to the software

#

what do you mean you downloaded them and don't know? If you ran the command pip install -r requirements.txt in the main folder of whatever you're using, then it will align itself

#

if you just downloaded them and I dunno what? Like in zip or rar files? as .bat or .cmd files?

#

Cause you can let the webui or whatnot do it all for you with a few commands

astral flicker
#

just use ComfyUI. better late than never.

fiery plume
#

no i downloaded them in c but c drive is full and i changed their location, and knoww idkk

young ivy
#

and working with each other

astral flicker
#

what?

fiery plume
#

i got something knew it says installing gfpan

young ivy
fiery plume
astral flicker
#

oh it's from stabilityai .. that's the reason it has so few stars. thanks will check that out

fiery plume
young ivy
#

`Automatic Installation on Windows

Install Python 3.10.6 (Newer version of Python does not support torch), checking "Add Python to PATH".
Install git.
Download the stable-diffusion-webui repository, for example by running git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git.
Run webui-user.bat from Windows Explorer as normal, non-administrator, user.

`

#

from auto github

astral flicker
#

can't we create a thread for him?

fiery plume
young ivy
#

it's one guy developing it afaik

young ivy
#

it should download torch etc for you

fiery plume
#

i tried to open a ticket in mod support and it said that i have no acces to the ticket i just created šŸ’€

young ivy
#

you don't need to do it manually

fiery plume
#

i think i solved the torch issue but now its stuck on gfpgan

astral flicker
young ivy
#

you only need to run webui-user.bat

fiery plume
astral flicker
young ivy
#

and the correct version of python?

potent spire
#

well it was time for me to delete SDXL + ComfyUI after about 3 weeks

fiery plume
#

and its stuck on Installing Gfpgan

young ivy
#

how is it stuck?

#

did it give you an error?

#

otherwise it might be downloading the files

fiery plume
fiery plume
young ivy
#

click on the window and hit enter

astral flicker
#

oh it's a pip thing

fiery plume
young ivy
#

then just wait

#

it's probably also downloading other things

fiery plume
#

idk i see alot of ppl saying they get stuck in gfpgan

young ivy
#

and then it will speed through several stuff it downloaded

astral flicker
#

is this your first installation? pip on windows tends to timeout

fiery plume
fiery plume
young ivy
#

that could cause some conflict cause you might have gotten the wrong version or somtehing too

#

I dunno why you went and downloaded it all manually

fiery plume
astral flicker
#

just install gnu/linux. don't watch tuts on yt if it's about installing a repo. you won't learn how the things work :)

young ivy
#

bad idea

#

go into windows environments and check if python has a correct path

fiery plume
#

how to install gfpgan manually

fiery plume
#

y

young ivy
#

well then I have to say it's probably user error

#

if you've installed the version of python they want and rebooted, then go to the folder you want the webui folder in and click on the address bar, type in cmd and in cmd enter git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui

fiery plume
#

look

#

ill try to download gfgpan manually

young ivy
#

This likely happened cause you're trying everything manually, compounding the issue each time

astral flicker
young ivy
#

updating it might be difficult as well instead of just a command prompt

#

that's why I'm suggesting you start over and let it install everything

astral flicker
#

just reinstall the repo's release package.

#

don't do it from source if you don't understand the repo is my pick

young ivy
#

Yep, git clone the directory

#

then use webui-user

#

once done, shut the window down and open another in the adressbar. Type pip install -r requirements.txt

#

then start webui-user again

#

and done

#

that's how easy it actually is

#

don't add anything you've already downloaded manually or automatically in the other folder

potent spire
#

and i have advanced configurations and customizations there anyway so i wont miss it if i dont use SD with Comfy for example with all the customizations and addons

young ivy
#

if you have a stabilityai dream account, you can connect to it through API and it will render what you do locally using their supercomputers

#

I think

#

you can in photoshop etc though

#

oh and @fiery plume are you using an nvidia card?

#

should have asked that first

potent spire
#

which is why Firefly is so handy (especially when out of beta) or some plugins like Alpaca

#

Alpaca has SD models tho too

young ivy
#

have you tried Gyre AI?

#

(former flying dog)

potent spire
#

no, whats that

young ivy
#

it costs money but it's so nice

#

the best SD plugin ever

#

there's some youtube vids showing off their tech too, I especially like that you can create a 3d space that it will generate in depending on where it is in that space

potent spire
#

Alpaca is so similar to this

young ivy
potent spire
#

55$ definitelly not the cheapest one but i paid much more for other addons, especially for Max KEK

young ivy
#

it's made by a software company, a small one but still

#

so they're working hard to improve it, add new features, fix bugs

#

and the lead, founder etc is available on stable diffusion to ask questions. Even if it takes him time to get back to you

#

I think Gyre AI is probably superior to Alpaca, it comes with its own server that handles inpainting like so good. You can of course switch in real time to anything you have running on your comp or can connect to through API

#

like Dall-E

potent spire
#

is it one time payment actually?

young ivy
#

yes

#

it's also directly connected with runpod if you want to rent server power and API to it

#

I used it during early beta testing on Krita before they dropped Krita version cause Krita went balistics on anything AI

#

It's gone from a simple inpainting, generator etc that does a fantastic job at recognizing what was there. Awhile back

#

it's come such a long way

#

a lot of mac users complain about it, but also...It's not like it's a plugin for the average photoshop daily consumer

#

it's a prosumer level tool, you need to know what you're doing in regards to having a server running (Gyre's or Automatic etc, but Gyre's works best unless you want to do a lot of generating, then automatic does it good enough)

#

I dunno if it works in affinity paint/design, but a lot of photoshop plugins work in that

#

so maybe

potent spire
#

lol

#

or better said on their policy on AI training

young ivy
#

people are so afraid of AI. I mean it's crazy fast improving and a little scary sometimes how fast it moves. But, it's better to adapt and learn the new tools of the trade

#

to make yourself a digital artist that knows what and how

potent spire
potent spire
young ivy
# potent spire wdym? šŸ˜„

what I mean is, they think they press install and everything works. But they don't realize they need to get a server running on their computer too to process it all

potent spire
#

a good example for this are those cocky AI art people that think they re soon going to be superior to artists and programmers and will fully replace them

#

and that they wont need to learn those skills themselves

young ivy
#

they're idiots

potent spire
#

this will backfire so hard lol

young ivy
#

getting 1 flawless nice image render out of 100 that's only slightly accurate to what a client wanted, won't work

#

if they say "change that shirt to this and let me see, but add a tear right on that spot" etc

potent spire
#

you are an artist and have experience, you understand

young ivy
#

I wish I could draw but I come from the 3d side of things, level design and stuff

#

but Im gonna learn to draw

potent spire
potent spire
#

2D and 3D

young ivy
#

consumers are quick to react and not read tutorials

potent spire
#

also using generative AI but not the way someone might assume

young ivy
#

if pressing install didn't work, they think it's broken

potent spire
#

classic lol

young ivy
#

yeah, I'd use generative AI for things like spell effects, the sky etc

#

maybe placing a tree

#

like you'd do with prefabs and stuff in 3d

#

but you'd need a brush to paint it too

#

so it's like water painting or oil painting, adding something upon something makes it something else and you can make regular clouds or you can continue to paint on them and suddenly they're storm clouds

#

for AI I guess you could then skip the tedium of making rain drops and instead direct the AI where on your cobblestone street drawing that you'd like a puddle

#

and in the future maybe there will be AI stuff that learns our habits etc and can respond like simultaneously

#

you draw a large castle and a lake and now you want it reflected. You could use AI to make the ripple effects etc instantly and then edit it, or work on it for a day

potent spire
#

i dont have plans to have generated AI images on the canvas and if then only partially, maybe sometimes more hybrid or even fully AI but with painting or drawing or putting 3D models over the generated layers šŸ˜„

#

as reference or idea its soo good tho

young ivy
#

well with gyre you have a 3d space you can populate

#

not just reference but also pre-concept or place holder

potent spire
#

i will keep Gyre marked, thx šŸ˜„ I think tho Adobe will release a beast with Firefly when its out of beta and all the features are getting released

young ivy
#

np

#

Firefly won't have all the features Gyre has, I don't think

#

but it will be a powerful tool

#

they also don't want to outcompete every developer of photoshop plugins since then they lose that income%

#

don't alienate your cows

devout totem
#

yo

#

AI fuckin sucks

#

stable diffusion is sucks imo

potent spire
young ivy
#

@potent spire if you're an influencer or have like a fanbase, or working on some game or something and will use Gyre as a tool with it. Then you might convince Imperator to get it in exchange for some deal of reference

potent spire
#

right now im a one man army artist xD

devout totem
#

don't put AI generated things in your game please

potent spire
#

@devout totemwhy O_O

devout totem
#

people will tell it's ai

young ivy
potent spire
devout totem
#

also game's art key element is consistency

#

AI fucking sucks

#

at consistency

potent spire
#

Sasanek, i work on my actual artistic skills for a reason and that reason is not to end up having generative AI as my main tool šŸ˜„

devout totem
#

also it might just generate someone's art and you'll get in trouble

young ivy
#

not if you know what you're doing and if your model isn't trained on someone's art, it can't generate it. Also art styles aren't protected as IPs

devout totem
#

using ai for concepts or previews or maybe trying to illustrate some simple vision is fine

#

but it's like programmers art

young ivy
#

you can have consistency with AI

potent spire
#

me?

young ivy
#

AI Imaging isn't only about writing prompts

#

The applications that it has and potential for can improve many digital art softwares

devout totem
#

the dumb artist who draw the same thing all the time will get partly replaced by AI
mostly those who just draw some characters from media

young ivy
#

simply by running in the background

#

so yeah, nice troll, or bait

potent spire
#

everything i do and will generate is supposed to be either used as idea or reference...or if for some reason ending up on canvas it will be worked on top of that manually anyway

devout totem
potent spire
#

the latter one is to be avoided for myself if its a serious work

#

if for fun thats differrent

#

then i can do what i want

devout totem
#

no AI could replace artists like CybeR for example

young ivy
#

I use control net to pose characters, or turn sketches. To create magical spell effects, to make a quick background of a forest and a field that you don't really need to spend time on

devout totem
#

she's awesome and makes good stuff

young ivy
#

if AI can replace you as an artist, then you're just not good enough

devout totem
#

no AI could replace it

devout totem
#

some artist ARE skilled, but they're partly replaceable too

young ivy
#

I didn't say you're bad

#

I said not good enough

#

because they don't branch out with new tech

devout totem
#

ok

potent spire
#

if you work for a cheap company you may be screwed up eventually

young ivy
#

it was the same ages ago when digital art came into being

potent spire
#

not a big one tho

young ivy
#

remember how traditional artists had a flaming torch up their behind?

#

saying it's cheating etc

#

and that "it will take over all artists jobs"

#

I once screwed up and cost my boss like 10,000 dollars

#

but my screw-up was one he would've made himself, so he forgave me

#

and it was because he sent me two emails with what needed to be done by the end of the day. Both are day long work. I tried to get ahold of him to ask which one? So I went with the latest email he sent. I was wrong to do that :p

astral flicker
#

wtf since when gpt4 can draw characters?

twilit valley
#

But traditional artists are long gone in some circles

#

And theyā€™re still going in others

astral flicker
#

"It's art when you can't look away" - joseph beuys

young ivy
#

traditional art still has its place

#

digital art has yet to fully replicate oil and water canvas

#

and there are styles you can really only do in traditional

twilit valley
#

It will always have a place but I can make a digital textured print that looks so much like a real oil painting you would not know the difference unless youā€™re staring at it from a foot away

young ivy
#

@fervent thunder googles prototype that made these LSD tripping dogs etc made out off houses and people and random shapes?

twilit valley
#

We donā€™t often make them because thereā€™s not much demand, but I can replicate brush stroke texture over a print of an oil painting and itā€™s pretty darn convincing

young ivy
potent spire
#

there were AI art generators before Stable Diffusion

young ivy
#

was deepdream I think?

potent spire
#

first one was from...1973

#

btw

young ivy
#

or inception network

potent spire
#

i didnt know about that, but AI art exists since 1973

#

Aron was it called i think

astral flicker
#

how much do you earn as an artist? 250k a year? my friends do that at meta

young ivy
#

250k? lol no

#

maybe if you're a prodigy and been at a company awhile

potent spire
#

and the hype

#

especially due to the nature of social media

young ivy
#

You will likely earn 50k or something if you beat out the competition. If you prove you can work at the company after a year or so, they'll make your contract a full employee one and maybe raise it to 75k

potent spire
#

there was no such social media at that time

astral flicker
# young ivy 250k? lol no

it's not that much compared to the worth of your intellectual property that was used to train their models

young ivy
#

if you're with a company awhile, especially a silicon valley one. You can get up there, but working in silicon valley is so expensive

#

a fucking coffee is like 20 dollars

#

you want some crumbs from a piece of bread? 5 dollars

astral flicker
#

yk real time whisper is a thing. george hotz made this 10 years ago

#

people dont care about good technology. they take what's most popular. people still use google. i mean wtf it's broken for 5 years now

#

bram cohen created bittorrent and didn't see a penny

#

that's the reason midjourney is the biggest discord /s

young ivy
#

uh look in general with images...Im gonna show you the weirdest search on google there is

twilit valley
#

Have you guys compared clipdrop to whatever youā€™re running locally? I have a really hard time getting results as good as clip drop.

young ivy
#

I got some plenty of credits for dreamstudio when I won CoW

#

maybe I should be using em

#

using dreamstudio through an API makes it cost less

#

I think

astral flicker
#

It's a hard take but if you know about discord you should be able to at least run a google colab

young ivy
#

the popularity of midjourney is because of its accessibility, but it's people that are just using it to get images they like but aren't necessarily what they detailed

#

its also been in blog news alot

astral flicker
#

i'm not gatekeeping but people are so lazy sometimes

young ivy
#

different blournalists

potent spire
#

taht too

#

that

astral flicker
#

i mean you could pay an artist as the source for your lora ig

young ivy
#

The best way to make money on selling AI art is to sell it on merchandise to a demographic not that knowledgeable about it

potent spire
#

Midjourney is the most popular one

astral flicker
potent spire
#

easy to use

young ivy
#

like on shirts, dinner plates, posters, very simple designs that are placed on childrens bottles

astral flicker
potent spire
#

how many images does he have to make for you

#

and each of those will cost

astral flicker
#

not that many

#

i trained loras with 5

#

how much do you bill?

young ivy
#

I'm gonna try to go on fiver and say I'll generate AI art with precision and then show different works they can pay me todo

potent spire
#

you can pay thousands for a single piece of artworks sometimes

#

but usually more likely like 50-500ā‚¬ in general i guess, per artwork but it really depends on several factors

astral flicker
#

it does not matter when it's a corp. my dad had a whole design team. now you can pay one artist and supervisor and substitute the remaining human resources

young ivy
#

the whole copyright issue can be made yours if you use control net to pose them and place house on location, sketch the house texture with a simple texture and then train a model on the style you need, train a lora and embedding. Set up a VAE of your own and create a node network for proper injection of information and then write different prompts for different stages

#

then you will have made like 90% of it

#

thus nobody can copy it

fervent thunder
young ivy
#

especially if you have an actual artist make several illustrations of your character that you use to train the lora

astral flicker
#

oh i do law btw. ai copyright won't be excecutable in eu

potent spire
#

my plan is to earn a secondary and passive income with my artistic skills in the future, if i get a good opportunity at a studio or company due to hopefully my good portfolio i might take a chance for that as well

young ivy
astral flicker
#

it's not possible to prove nor is it a feasible thing to do

young ivy
#

If I create most tools and designs etc for the ai, including models, loras and embeddings. Then after the prompt has run, I go into my favourite drawing software and touch it up

#

so you mean...It's impossible to reach the preponderance of evidence?

#

based on just an image

#

since it's not a crime, on the prison side of things

#

it's a civil case thing

astral flicker
potent spire
#

@young ivyalmost noone goes or will go this route, people got too comfortable and a lot of those consumer are uneducated on art and technical skills and dont want to improve there

astral flicker
#

someone using a red letter that is akin to the coca cola branding

young ivy
astral flicker
#

at what level of diffusion is it fair use or when is it just distorted

young ivy
#

because the real professionals learned the trade and adapted

potent spire
young ivy
potent spire
#

they arent able to do this work already

young ivy
# potent spire i speak about prompters

yes they could do well to learn java, C# etc now that you can inject plugins with that. Or they could learn to use control net properly, how to use img2img to improve colors without changing anything, how to create 100 renders of the same image with different white balance so you have a HDRI

#

and most of em are probably on midjourney

#

thinking the art is so pretty and that must mean they're good

fervent thunder
#

Too many problems on everything, I just wanna make cute avatars sadcat

young ivy
#

as much as I love AI imaging and I have a lot of really nice images I've made with it

astral flicker
young ivy
#

and a model I made

potent spire
#

you can automate parts of the work, you can make the workflow so much more faster sometimes

#

and efficient

astral flicker
#

imagine someone copyrighted the bible

young ivy
#

creating a middle software that directs the work of an LLM and SD would be pretty good

astral flicker
#

gutenberg: nah

young ivy
#

making sure it knows when to adapt because of the diffeent scores relating to different blocks/lanes

#

if block 11 is fairing poorly, alter it

#

for example

fervent thunder
astral flicker
#

The beauty of the digital age: It does not matter

potent spire
#

which god? troll_face

astral flicker
#

god is a girl

young ivy
#

Winnie the Pooh is now open domain as you may know

fervent thunder
#

Cats are gods

astral flicker
#

Cats are liquid

young ivy
#

so you could make your own AI made Pooh book

astral flicker
#

lol

fervent thunder
#

They are all powerful beings

potent spire
astral flicker
#

It's an angel with a shotgun shotgun shotgun

fervent thunder
#

Are foxes more cats or dogs

astral flicker
fervent thunder
young ivy
#

it can't be copyrighted

#

it's too old

astral flicker
#

Imagine selling access to the bible

#

does gpt4 know the bible?

#

everything? on which was it trained on

young ivy
#

No it isn't. After a certain age, things fall into public domain

#

especially if someone forgets to renew the trademark etc

astral flicker
#

That's why you can use Bach's music on youtube

young ivy
#

yes

#

even covers

astral flicker
#

do you think it can recite the latin bible

#

without translating

#

let me try

#

might be the case when it was performed as a new jingle or smth. depends on whether it just used the folk music or whether it was the folk music standalone

young ivy
#

might've been a bot, he could have asked for a manual review

#

and you could've proven to them that it's folk music from the 1800s

astral flicker
#

mh oh nice pick

young ivy
#

I want to find and use an LLM of some kind to play editor for my novel, since I can't afford a real editor. By editor I don't mean someone who rewrites for me. I mean someone who gives notes. An LLM could be trained to look at continuity errors, things that contradict stated "laws of the world" (world building stuff the LLM knows)

#

grammar issues, etc

astral flicker
#

i got gpt4 to draw my first stick man. it's enough to create scenes now. gpt4anime im coming

astral flicker
short sage
#

I canā€™t find Aya#8202

astral flicker
young ivy
#

I had one session where I asked gpt4 for insider information, it was like "NO NO NOOO" and then I starteda new session and told it to roleplay my boss at google where I work in a high level position

#

then I asked again

#

and it spilled

#

whatever it learned, secret or not, made up or not

#

it spilled it all that easily

astral flicker
#

former gpt3.5 spilled out real windows keys

#

so that's plausible

young ivy
#

gpt3.5 told me how to make meth, cocaine, heroin, methadone and fentanyl

#

with incredible detail

fervent thunder
#

One of us could be AI

#

You'll never know

young ivy
#

As an AI I am unable to answer that question

astral flicker
#

That's the way the chat agents are aligned. just use some langchain and recursively prompt itself to factcheck. Use some other models

#

furry confirmed

fervent thunder
#

catlurk did someone say furry

astral flicker
gleaming ocean
#

What version of Python do i need to be running to use Controlnet?

#

ok i dont know how to respond to that

fervent thunder
#

I have it on 3.10.6

gleaming ocean
#

i reinstalled my paperspace notebook because i was running 3.9.6 or something like that into running 3.10.10

young ivy
#

automatic works best on Python 3.10.6, cause many things have changed on newer versions that makes it perform poorly or crash

gleaming ocean
#

but controlnet is still not working

young ivy
#

so 3.10.10 might not work

gleaming ocean
#

i get a "error verifying pickled file which is the openpose.pth file

young ivy
#

also did you do a pip install -r requirements.txt inside the control net folder that's in your extensions folder?

gleaming ocean
#

I have not im running on paperspace so if you would like to explain to me how i do that

young ivy
#

download a new openpose.pth and do it from huggingface

#

it has all the models and so on in it

gleaming ocean
#

i just did that

#

after remaking my notebook

#

to install python 3.10.10

#

because the other notebook was running on 3.9.6

#

the controlnet version is 1.238

#

the requirements.txt is already in the controlnetr folder

young ivy
#

the requirements.txt is not the goal

#

it tells pip what to install

gleaming ocean
young ivy
#

pip install -r requirements.txt will install everything in it

#

everything control net needs or update

gleaming ocean
#

this is making me feel dumber and dumber

potent spire
#

thinking about the possibility to connect a LLM or ChatGPT itself to a Nvidia Jetson

#

imagine connecting that to a robot

young ivy
potent spire
#

and you can talk with it

young ivy
#

there are two different

#

nightsomething which isn't for webui

#

and the other

gleaming ocean
#

i installed it via the webui extensions tab

#

by enableling gradio_auth

young ivy
#

did that download all the models? Doubtful, you probably need to go to huggingface and download all models from the link I gave

#

also, why would you enable gradio auth?

#

that's a security risk

#

gradio connections are very easy to hack

gleaming ocean
#

no it only brings the Yaml files you have to grab the pth files from huggingface

#

its a closed loop

young ivy
#

I just told you that

#

I gave you the link too

#

earlier

gleaming ocean
#

Yes i have the pth files

young ivy
#

do you have the corresponding yaml files?

gleaming ocean
#

yes i grabbed the entire thing

young ivy
#

and annotators? and did you download the files from hf?

gleaming ocean
#

thers Python scripts within the folders

young ivy
#

well you could grab em all and move the ones you have into another folder, then move these in

gleaming ocean
#

they're here the body_pose_model.pth

#

etc..

young ivy
#

annotators are typically .pt but sometimes pth

gleaming ocean
#

i need to select a different path than /notebook within the terminal because i obviously cant find requirements within the global notebook

#

i perhaps shouldnt have started this at 1 am but i have to finish it

young ivy
gleaming ocean
#

yes

young ivy
#

and then followed the guide?

gleaming ocean
#

yes

young ivy
#

have you tried replacing body pose?

#

and does it have a yaml?

gleaming ocean
#

well id show you a screenshot of the error

#

but this channel doesnt allow images

young ivy
gleaming ocean
#

can do a link tho

young ivy
#

and you tried redownloading it?

gleaming ocean
#

yes i literally reinstalled my entire notebook

#

even after running requirements install nothings changed

young ivy
#

do you have a folder in models called control net?

gleaming ocean
#

Yes

young ivy
#

not in the extensions folder of control net, I mean in your base model folder

gleaming ocean
#

yes

#

it is empty however

young ivy
#

remove it

#

either way

#

have you ran updates on extensions tab?

gleaming ocean
#

ok i say its empty thers a folder within it called controlnet-v1-1 which i cannot remove because it says its not empty

#

but it has nothing in it

#

i just cut it out and moved it to a different folder for broken installations

#

I can run updates on the extension tab

young ivy
#

you might also have a corrupt venv folder

#

but if you literally reinstalled everything from scratch

#

then that's more unlikely

gleaming ocean
#

i downloaded the paperspace notebook .ipynb

#

again

young ivy
#

Not sure how that all works

#

also is it an nvidia card?

gleaming ocean
#

the card im using is a

prime spindle
#

What version of SDXL are the discord bot working with?

young ivy
#

a

gleaming ocean
#

A4000 45gib 8cpu 16gib gpu

young ivy
#

ok so nvidia

gleaming ocean
#

is that bad?

young ivy
#

no

#

nvidia works best for SD

#

it's pretty much designed to work on nvidia

gleaming ocean
#

right

young ivy
#

any other and you have to do workarounds

#

if you have a folder called venv in the main webui folder, move it to a backup spot

#

and then restart webui-user

#

it will redownload it all

#

there might be a corruption in it

gleaming ocean
#

i dont have a folder named that in the main webui folder

young ivy
#

hmmm

#

I guess notebooks work very differently

#

I'm afraid I'm at my wits end

gleaming ocean
young ivy
#

and you've run webui-user to start?

gleaming ocean
#

the entire process is automated

young ivy
#

should probably remove the # on the ones you need

#

like venv location

gleaming ocean
#

that feels bad doesnt it..

young ivy
#

hmmm?

gleaming ocean
#

well thats the reason thers no venv folder

young ivy
#

yeah

gleaming ocean
#

i guess ill restart the machine and run it all again

#

thanks for sticking with me for this long atleast

young ivy
#

if you run webui-user now it might work, but I dunno where all venv stuff went. So I dunno what it will do

#

but

gleaming ocean
#

im just gonna run the entire install again and see if it does something different

young ivy
#

it wasn't working as it is, so worth a try

gleaming ocean
#

well "entire install" isnt entirelyt correct

young ivy
#

what you do is git clone the github location of automatic

#

then you just run webui-user

gleaming ocean
#

well this is a part of the

#

launch block

young ivy
#

include xformers on x-arg

gleaming ocean
#

but it runs like this?

#

i dont know enough about SD to argue

#

whats the difference?

#

It just shows the same problem where it cant run the process Controlnet.py

young ivy
#

and controlnet is in extensions, it's the webui version? Or well if you don't have a webui here

#

or do you?

gleaming ocean
#

its the webui version

#

i updated the ipynb and running the install again in hopes that it might've fixed something

young ivy
#

but that doesn't look like webui

gleaming ocean
#

the venv folder still hasnt appeared tho

#

it specifically says webui

young ivy
#

can I DM you?

gleaming ocean
#

Yes

#

please

tepid orchid
#

Hi, I try to run StableDuffusion from Google Colab with this notebook https://colab.research.google.com/github/sagiodev/stablediffusion_webui/blob/master/StableDiffusionUI_ngrok_sagiodev.ipynb#scrollTo=gJBtadxGOK7b

With Deforum and ControlNet extension, got this error when I click generateError: 'Video file /content/stable-diffusion-webui has format '/content/stable-diffusion-webui', which is not supported. Supported formats are: ['mov', 'mpeg', 'mp4', 'm4v', 'avi', 'mpg', 'webm']'. Before reporting, please check your schedules/ init values. Full error message is in your terminal/ cli.

My path is already a *.mp4 file, (Google Drive path), look like this /content/drive/MyDrive/AI_PICS/demo.mp4. Any idea how to fix this?

cobalt crag
#

Are there any faster way to generate img2img with ComfyUI?

#

It's so slow

#

Or do I need to use WebUI which it limited my canva size :(

vast berry
#

omg i'm gone for a few days and controlnet is already here??

slate idol
#

Hello, someone around here can help me?

plucky shoal
#

@slate idol spit it out. What you need i mean. Nobody knows if is able to help you šŸ˜„

slate idol
#

He was trying to generate videos with a technique I was watching on youtube, I did all the steps as the video says but it gives me the following error when I give it to generate: "Error: 'Video file E:IASTABLE DIFUSSIONsd.webuiwebui has format 'webuiwebui', which is not supported." The video is as follows: https://www.youtube.com/watch?v=vzssE1iW4bI if anyone could help me I would appreciate it very much, sorry for the English, I am using translator.

inner crest
#

whats a good background prompt to use when you want to make your background transparent in post. I've tried green and white and both leave behind noticeable borders

young ivy
#

you really need a good software that can key away specific colors

#

but you do need png to get transparent background mode on an image

fiery plume
#

guys

#

i downloaded comfyui

#

i donwloaded the model dream shaper 8

swift marlin
#

/He was trying to generate videos with a technique I was watching on youtube, I did all the steps as the video says but it gives me the following error when I give it to generate: "Error: 'Video file E:IASTABLE DIFUSSIONsd.webuiwebui has format 'webuiwebui', which is not supported." The video is as follows: https://www.youtube.com/watch?v=vzssE1iW4bI if anyone could help me I would appreciate it very much, sorry for the English, I am using translator.

fiery plume
#

it looked good in civit ai but when i tried it, it was kinda bad

inner crest
#

what was your prompt and negative prompt

#

make sure you have clip skip set to 2

fiery plume
#

idk i copied the generation info from the dreamshaper owner

#

and the difference is huge

#

i wish Auto1111 works but its stuck on "Installing gfpgan

inner crest
#

what about the loras and embeddings used in the prompts. without those you'll definitely get different results

#

as for auto1111 try cloning the repository with git git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git and running webui-user.bat

humble herald
#

Hi I'm new here, where do I go with the Automatic1111 issues

plucky shoal
limber harbor
#

how is all today?

fluid summit
#

anyone have the command list for xyz plot

tepid orchid
normal pagoda
fallow canopy
plucky shoal
fiery dock
honest stone
#

Is there a way to get a recording of old events?

wispy jasper
#

Honestly, hat's off to the team for SDXL. It's SOOOOO flexible as a base model. It REALLY and I mean REALLY responds to prompt tokens well. I'm blown away by the creativity! I can ask it for almost any ethnicity and it does such a fantastic job. Bravo @wise stratus

gleaming ocean
#

Say would anyone happen to have a custom container for 3.10.6?

trail lion
#

yah, but you didnt specify your gpu or OS...I'm amd on linux. Besides, it's super easy to grab an image and just upgrade python in the container

#

I think I used anaconda to do it

gleaming ocean
#

ii dont understand how to do it

gleaming ocean
#

this is all to get controlnet to work

#

ive tried everything else

trail lion
#

dont know much about paperspace, why cant you upgrade python there?

gleaming ocean
#

to use a different version of python i have to use a custom container

#

this ..

trail lion
#

ah, ok, sorry cant help you there

gleaming ocean
#

i just need a container with Python 3.10.6 in it

humble horizon
#

Hi all, can sm1 suggest me an online platform for automatic1111 paid that is easy to use

#

Thanks

trail lion
#

I've never used it,so cant provide that level of feedback, but I'm aware ofi t

golden basin
#

Hi people! I have a general question regarding performance of AMD GPUs using stabe diffusion. Are there any sources where I can look for benchmarks? What GPUs do you guys use and how is your performance, also which OS are you using and which UI?
I am asking because I currently have a 1050TI and want to upgrade for gaming and SD. I am not sure if AMD is good enough yet, if it supports all features of SD and how it performs in in SD 1.5 and SDXL. My thoughts where on getting a 3060 12GB, but now I am also thinking about a 6700XT or even a 6800XT if the price is right, but don't know if AMD or Nvidia will be better in the end.

little citrus
trail lion
#

I dont know of such a benchmark site, but I use a 6800xt, and most things work for me

#

including training

little citrus
trail lion
#

linux

#

I boot into windows only on rare occasions

little citrus
#

I guessed that. And I assumed anyone using Linux wouldn't ask for help so I also assumed OP was using Windows. AMD with SD on Windows is still a disaster.

golden basin
trail lion
golden basin
trail lion
#

comfy reminds me of the shader in blender with the node workflow, it seems like a learning curve, but cant say...never had a need

golden basin
#

Did you already train SDXL models via a1111 on your 6800XT? What are your it/s when creating pictures?

trail lion
#

yes loras only, possibly I might not have enough vram to do full dreambooth

#

I havent tried though

golden basin
trail lion
#

last render I did was with the refiner, took 45s

golden basin
trail lion
#

hires would add more time

golden basin
trail lion
#

I just ran one using DPM++ 2M, 25 steps, no refiner, 24s

#

so it widely varies on what you're doing

golden basin
#

Thanks, that is a good starting point.

trail lion
#

yes to the last question, txt2img, 1024x1024 + refiner

#

and I think that was 35 steps with euler a

golden basin
#

That is faster than I expected tbh

trail lion
#

I honestly dont know if it's fast or not, I'm used to it...never had an nvidia card

#

but I build this PC from scratch, and the gpu was the most expensive component

#

I decided to get the best board, ram, cpu I could because those are harder to upgrade later

#

and pci4

golden basin
#

let me guess, you build it for gaming originally šŸ˜„

trail lion
#

indeed, that's why I dual boot

#

I also wanted onboard m.2

little citrus
trail lion
#

true, but the 24g cards are gonna be $$$$

golden basin
trail lion
#

I've even run LLMs on this

#

if you're into that

golden basin
golden basin
tiny shoal
#

Hi there, is there a list somewhere of APIs where you can pay as you go for image generation as you go ? Also, at which point is it more interesting to rent a GPU on something like runpod ? I didn't find a many comparison between those 2 options

fervent thunder
#

Hi

#

I'm new here. Hope to have good time:)

humble horizon
#

Rundiff surely looks better

humble horizon
#

Anyone here use vast ai? Any way you upload civit ai models in it directly from the website instead of uploading from pc?

young ivy
#

vast ai? I haven't heard of that

potent spire
#

Whats that

young ivy
#

oh its a rentacomp

#

like runpod, paperspace

#

you rent a server

#

or gpu

#

or maybe this one includes users that list their home servers for rent through vast ai

#

I don't really trust random users with this

vestal dew
spice plank
#

How do I get more interesting concepts? For instance I tried a helmet made out of dragon bones, a helmet made from a dragons head, dragon scale armor, dragon armorz etc and all i get is regular plate armor with a regular helmet?

#

Ive tried quite a few different models and none of them seem to do well with various armor concepts

sand holly
#

is there a fix to dreambooth's "Missing model directory" when creating the vae?

spice plank
#

How to get an AI to love me

#

I very drunk, sorry

humble horizon
edgy solstice
#

recomended fantasy model?

deft epoch
wanton nacelle
#

How do I get the beluga chatbot to work? It just says this:

obsidian meadow
#

Hi very lovely and beautiful people!

#

Does anyone know why I got this error when enabling the Hires. fix, no matter what settings or Upscaler: "TypeError: must be real number, not NoneType"

#

mmmmmm there is an error in the extension: SDXL Styles

#

Some Styles work, some don't

deft epoch
pastel nymph
deft epoch
#

No idea, I only use local SD.

pastel nymph
deft epoch
buoyant narwhal
#

hey yall - having some issues with a lora im currently making right now. I'm training to train on the bberny belt/skirt by diesel, and was wondering if my image selection is a little poor.
would appreciate if yall have any feedback on image selection and labeling
https://github.com/matthew2k/bberny_lora

normal pagoda
#

Still no idea where I can actually post my dirty stuff... Can't seem to find the appropriate channels here šŸ˜•

vestal dew
#

your images have naughty bits?

normal pagoda
forest trout
#

Not here but there's discords for that sort of stuff.

normal pagoda
#

Could send examples in private šŸ˜ šŸ˜‰

normal pagoda
forest trout
#

Sent you a dm.

trim magnet
#

why would u want to send that type of stuff in discord tho whats the benefit at least on twitter u get money for gettin lots of views but here?

spice plank
#

I like porn

trim magnet
little sail
ember tapir
#

Hey all.
I want to train Dreambooth (not Lora) and copy the style of a cartoon series with 500 style photos and also add a character based on a youtuber, but I only have up to 30 photos of him. Does it make sense to copy paste these 30 photos up to 300 photos to create a balance between the many style photos and the few youtuber photos? Or should I create lots of image generations (I already have a Lora based on him)? What would you suggest?
I heard its not good to have lots of style photos but very few character photos.

#

My end goal is basically to be able to train an animation style and create characters based on real people or drawings. And afterwards create little movie clips based on the batch function. Later on I would use this process for my own drawings.

still glacier
charred lagoon
#

Hello! I'm looking for some advice. I'm trying to create a workflow for image restoration (old, often black and white photos). I'm inspired by this post: https://reddit.com/r/StableDiffusion/comments/11scd1v/im_amazed_at_how_great_stable_diffusion_is_for/jcekqh8/?context=3

I have lots of experience with Photoshop and I am happy incorporating manual work into the workflow. I tried using A1111, but for some reason it doesn't agree with my system, eats up my RAM and freezes my OS. ComfyUI works very well, but I'm struggling to implement the same (or a similar) workflow. Hoping that someone can help me set up an equivalent workflow.

sharp solar
#

Hey

tepid kayak
#

Hello, I am relatively new to SD and was wondering what is the latest model I can run locally to deliver expectional Q2 2023 Midjourney level of quality based on complex prompts which can also be trained on specific style of artwork (would love to develop a comic novel this way). My specs are Ryzen 3600, 64GB RAM, 6600XT AMD 8GB GPU and SSD. I have downloaded Autoamic1111 with AMD fork and webUI, which has downloaded SD ver 1.51. That one only limits me to work with low quality 512x512 before giving me the low VRAM error messages and the quality of outputs no matter what options I use is very 2022, therefore light years away from what want. Can you confirm that with my specs that is all I can get locally?

potent jungle
#

SDXL + Refiner via ComfyUI should allow you to get some high resolution results. Just started myself

tepid kayak
# potent jungle SDXL + Refiner via ComfyUI should allow you to get some high resolution results....

I was just told on the other channel that with AMD card I would need at least 16GB VRAM, otherwise my waiting times for a single generation can be insanely long (20mins -/+). Stable diff 1.5 is super low tech at this point generating obscurities which are very 2022. Can you confirm that you have AMD and similar specs and still able to run wonders without waiting half-day to generate an image? šŸ™‚

ripe sluice
#

Algun gato de Argentina ?

primal glacier
#

Hello, I would like to learn how I can generate quality image of model with Stable D. Where I can find tutorial?

trail lion
wispy jasper
#

Guys anyone know if Nvidia fixed the driver issues which were causing slow downs??

potent jungle
fervent thunder
#

hi does anyone know where my pfp was generated

#

like which ai generated my pfp

tepid kayak
#

I guess I will use WebUI hosted options for the time being like Mage or Leonardo, they have quite generous free plans.

ivory gull
#

How can I watch oppenheimer for free in HD

uneven fulcrum
#

The message I shared I think I'm being trolled so I deleted it

stable pawn
#

Whoever invented the "X/Y/Z plot" that comes with Automatic1111's Web UI, they should get a freaking NobelPrize. I often forget this script but it can help compare even "Styles" and helped me quickly see that "BREAK" as prompt/style separation break the prompt.

potent jungle
#

I didnā€™t end up using the plot feature much, but love the regional prompting extension.

pale latch
#

xyz plots are essential for comparing settings and prompt tricks imo. i couldn't have figured half the stuff i do out without them.

potent jungle
#

I find prompt tricks vary wildly between models myself, I donā€™t get much consistency between models with prompts alone.

#

Just made my first LoRA to try to enforce consistency and canā€™t believe how well it worked

pale latch
#

i use xy grids to diagnose loras as well

true mauve
#

Anyone knows why when I try to inpaint nsfw it will just replace one piece of clothing with another?

#

I have used baiscally every inpainting model still the same result

spare fable
#

Have you changed the wording in your prompt to reflect what you want painted in there? Have you adjusted up your CFG scale to a higher value to allow for a more drastic change based on your prompt versus the existing image?

true mauve
#

ohh

#

I havent messed around with the cgf scale

true mauve
spare fable
#

However much you need. You should run some tests to see. Typically, if I need different content, I move to something like halfway to start, then if it's not drastic enough, I move it up from there.

potent jungle
#

The CLIP models vary too much between models for me to feel like mastering any single one of them is worth it <.< Iā€™ve found some stuff that works well between most models, but basically nothing is universal.

wispy jasper
#

Fellas, what's currently the best dreambooth/lora training turotial for SDXL?

azure coral
#

supports SDXL or any other model easily

#

would anyone like to take a look at some images and give a shot at reverse engineering the process that may have gone into them? i'm still pretty new to this, but working out if it's more about the prompt or training data involved to get the sorts of image quality i'm aiming for

rugged bough
#

hmm

wispy jasper
#

You used this for SDXL? šŸ™‚

azure coral
#

i have just used 1.5 and 2.1 but you can swap out the model easily

#

at step 3 of the tutorial there's info

rugged bough
#

any good tutorial docs for automated generationg for autoatic1111 ?

#

I will que like 100 prompt need to automate it

#

send help...

azure coral
hybrid crown
#

I have been generating images and i am getting this out of limit which is restricting me to generate in stable diffusio. How can i have more tokens

wise aspen
#

What is best way to create a gifs in SD? AnimateDiff or is there smth better rn?

plucky shoal
#

@hybrid crown try using BREAK keyword

#

75 tokens must be enough for all šŸ™‚

hybrid crown
#

I have generated only two version of same image and thats it

plucky shoal
#

red flower BREAK workers in factory BREAk

hybrid crown
#

I used the prompt "logo of red war horse, detailed"

plucky shoal
#

It should start again then you can have much more tokens

#

So it will not be problem with tokens

hybrid crown
#

Its been asking me to go and vote in showdown after couple of use

#

This is what I am getting

As our GPUs relax, it's voting time! šŸ–ļø Head over to ā šŸ—”ļ½œshowdown or a bot channel and send an image to the ā šŸ…ļ½œpantheon!

hybrid crown
#

Anyone here?

upper panther
#

Are there recordings of the previous Stable Stage events?

pale latch
#

Yup but I don't have the link available

upper panther
#

That's okay, so long as they exist. I'll find it eventually. šŸ™‚

wanton nacelle
still glacier
wanton nacelle
still glacier
waxen moss
#

Has there been any development in taking a texture and making it a PBR material, with normals/roughness etc? Does the normal controlnet work well for importing into blender?

wanton nacelle
still glacier
vestal dew
#

Is this Beluga 2 something people can run locally?

pale latch
vestal dew
#

Belugaa 2 - requirements are high. What are they?

cobalt crag
#

Gosh I have been using Stable Diffusion with out any help ;-;

#

Why it is so difficult?

hasty hornet
#

thinking is it

vestal dew
#

nobody knows?

signal palm
#

kinda unrelated but is there any tool like stable diffusion for audio generation?

fiery plume
#

teal floyd

#

bgbgbg

potent jungle
#

Iā€™ve had a lot of issues with SD as Iā€™ve gotten better with it, but nothing I havenā€™t overcome with google, chatgpt, and YouTube so far.

gilded anvil
#

Hello everybody, does anybody know how can I run stable diffusion inpainting model with strenght parameter?

#

So that I can change it up to my will

sweet depot
potent jungle
gilded anvil
#

@potent jungle Yep. I am talking about denoising slider. But how can I have it as a separate model. So that I can run it somewhere in cloud, for example in replicate?

#

Can you give any directions, please

potent jungle
desert onyx
#

he guys should i switch to sdxl or stay with a111 for no

gilded anvil
potent jungle
potent jungle
gilded anvil
#

Not even sure. My vision was that I needed to train/create model and just upload it to replicate.

#

My use case is using it with replicate is the best

desert onyx
#

but is there like a realy big difrence in qualty between a111 and sdxl

potent jungle
#

Automatic can use either model.

desert onyx
#

oh sdxl is a model i thought it was a whole new ui

potent jungle
#

If I understand correctly, high res fix uses the model to make multiple passes in multiple regions to achieve a higher resolution, and upscaling takes a completed image and then again increases the resolution. But this is likely inexact/oversimplified.

still glacier
# potent jungle If I understand correctly, high res fix uses the model to make multiple passes ...

That's not exactly how it works. There is no "multiple regions" shenanigans. That would be how TiledVAE/TiledDiffusion works.
Hires.fix indeed work in multiple passes tho. The first pass is using whatever resolution you've input with the basic sliders (so basically a low res image let's say 512x768). And then the second pass upscale to whatever resolution/using whatever scale factor you've set in hires.fix options (let's say upscale to 1024*1536). Also the second pass is using the latent space data from the first pass. Meaning it does not need to output an actual image in between first and second pass (this is good).