#💬｜general-chat

1 messages · Page 67 of 1

frank spruce
#

mage.space

#

basically the amount of computational power required to generate AI images makes it impossible to give away free at scale

#

like they could be mining bitcoin with that shiz, or not paying cloudflare

frank spruce
#

relative to a simple webpage i meant, its still in the pennies or so depending how its implemented but that can add up with a thousand users running 100 prompts every day for a year, at 3 pennies each, that’s over a million dollars

#

but the real issue becomes the 0.1% of user who abuse and make fake account etc with a free service idk, why dont u implement it xD

tulip yoke
#

AUTO1111

wild steppe
spare urchin
#

have you guys been able to get any good web designs ?

frank spruce
#

whats stable diffusion got to do with web design

spare urchin
#

by producing images?

frank spruce
#

whats the extension to autofill lora/hypernet/embeddings correctly, i cant spell them right or remember their full names

warm junco
frank spruce
frank spruce
#

ok weird idea i think if u have a bunch of textual inversions at low values at once it can cause some weird kind of aliasing inside the tensor?

wooden briar
#

What setting to change when you're getting duplicates in a scene, for example multiple cherries when you only want one?

frank spruce
stark mason
#

what is this server about

wooden briar
frank spruce
wooden briar
#

Ok, I'll try that ty

frank spruce
willow needle
#

Any recommended positive/negative words for prompting out hands? Figured it would be a sufficient fix while I learn after-creation image editing.

fervent thunder
#

Hello

frank spruce
charred hawk
#

Hello all. I'm a bit new to SD, but my god, it blows my mind what has become possible in the past year. Nice to be here. Exciting time to be alive.

fervent thunder
#

Any veterans know about this?

#

the whole deterministic xformers thing

#

lot of people saying xformers speeds up generations but results in caching issues or something? Left over denoising and an inability to reproduce the same seed.

willow needle
#

Thank you very much, I'm just learning controlnet so I will certainly give it a shot, I thought I had searched civit for all the badhands negs but I didn't see that one, awesome.

verbal osprey
#

When will SDXL pop off? Sometime this week?

frank spruce
#

just experiment and choose what u like

willow needle
#

You're right about that, I learned real quick to stick with one neg prompt for one thing(hands in this case), lol. Unfortunately the LORAs I'm currently working with must have been trained on bad images or without care for the digits. Prompting the hands out has given the best results, though Badhand4 certainly helped.

fervent thunder
#

What's the difference between a Lora and a stable diffusion model?

#

I thought they all went in the same folder but it seems they don't.

pearl fog
#

A stable diffusion model is kind of like a full game, while a Lora is like DLC.

#

The model trains a bunch of different stuff or a whole style, like anime. Meanwhile, a Lora is for a specific character or a specific artist

fervent thunder
#

This whole time I kinda just ignored the word lora, thinking it was more like a genre or something, and just assumed all these small lora models were just smaller shitty nichely trained models with no flexibility. x)

signal saddle
#

Hello guys 👋

#

I just joined this server, could anyone please explain what’s the first step for creating animation with stable diffusion? Courses? Classes? Resources?

#

I’m very interested but have no clue where to start

boreal cosmos
#

there are thousands of youtube vids on that, I recommend watching some recent ones

radiant bronze
#

My Kohya is creating Loras that don't work

#

Does anyone get this problem

fervent thunder
#

Can someone that's good with cmd help me?

#

Trying to install this thing

#

And the patching step after downloading the extension is just giving me an error

gray vapor
#

I heard Stability.ai is developing stable-lm variants for India, Indonesia and other countries. Where can I find further information on these efforts? Is this the right channel for this question? (I previously posted in the stable-lm channel but no answers)

magic solar
#

This might be a weird question but has anyone here trained a object detection model using generated images or do you think it is possible?

radiant bronze
#

Kohya is completely broken for me, I dont understand why there's no good finetuning tools out there

harsh fox
#

Is there a way to change the race of the generated characters? I seem to get a lot of Asian themed people even if I put Canadian, American, Caucasian, etc. Or do I need to use a different checkpoint?

sudden ruin
unkempt hatch
#

if I set bucket size to 1024 and my dataset includes 512x512 images, what will happen? It'll train the images as is, or upscale them?

fervent thunder
#

Hey, I was just wondering, will the Ryzen 7 Octa Core 5800H be good for programming or should I prefer intel over it?

hasty hornet
#

thinking for programming what

bleak matrix
#

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

thorn pilot
#

Morning Sunny! 🙂

bleak matrix
thorn pilot
bleak matrix
#

I'm awake! Slept terribly but I'm gonna make up for it by drinking this orange juice...and playing Hitman XD

thorn pilot
bleak matrix
#

Yeah; I think all the rain has just been making it harder to breathe, so I keep waking up. Sometimes I wish I was like a cartoon and could just bap myself asleep, ahahahahaha

#

I'll prolly work on more tutorials today

#

How about you?

thorn pilot
# bleak matrix How about you?

It's been raining here too today. I worked, then played a bit of diablo and now just playing around with SD as per usual weekday 😄

bleak matrix
thorn pilot
#

just wish there wasn't the work part 😄

bleak matrix
#

W o r k

#

The thing I constantly do

#

Lmao

#

But also can't seem to give up because endless curiosity

thorn pilot
#

I like my job, I think I just need a break, a week off

bleak matrix
#

Well, take some vacation time! We all need some r&r

thorn pilot
#

I really should, can be a bit hard sometimes when you're the only one who can do the work 😄

bleak matrix
#

R I P the feels

#

This is very difficult, esp if you are self-employed. No rest for you--lmao, 24/7, at times.

#

But sometimes ya just gotta take care of yourself

#

So, what are ya makin in SD?

thorn pilot
#

kind of like my pfp

bleak matrix
#

Ooo! I love the idea

#

Your pfp is so cool

thorn pilot
#

thank you

#

it came from a prompt that was producing some amusing images: grumpy old bones full of __colour__ dust so that had kaleidscope dust 😄

fervent thunder
#

Hey

fervent thunder
primal current
#

the new trailer of Messi's project uses AI

pine fulcrum
#

does increaseing the number of images generated per batch decrease the ammount of time the generator spends on each image?

wintry stream
#

You could lower your steps to decrease the ammount of passes per image though

pine fulcrum
wintry stream
#

It's more of just queueing up more passes automatically so you don't have to keep clicking the button

pine fulcrum
pine fulcrum
wintry stream
#

Nothing to be sorry about, rather you be clear and make sure you get the answer you need 🙂

pine fulcrum
#

🙂

#

thanks

hasty hornet
fervent thunder
#

overall im asking if it will be a good choice>

hasty hornet
#

There are too many aspects of it...

Like on one side for most things you don't need "da best" cpu, on the other hand, who knows, maybe you're gonna code your own minecraft with terrain generation and such, so you'll need better cpu and multithread the shit ouf of it, you know... (even then, not sure if amd \ intel gonna matter)

mortal sphinx
#

How do I download Stable Diffusion

#

It's different since I last used 2-3 months ago

karmic brook
fervent thunder
frank spruce
#

sometimes CPU is a bottleneck

#

especially if it isn't multithreaded

#

ok i got an idea guys

#

we could have an artificial prompt breeding engine

#

so we have a bunch of prompts

#

and we only let the ones that produce the highest ranked images breed

#

so it can automatically refine a prompt to make images that are as pretty as possible, and fine tune all the weights and make something reaely cool

#

all without any human input

sudden violet
deft epoch
#

I bought a 3060

#

so I can proompt things faster

#

1660 is really really bad for generating images.

#

Hope I didn't mess up.

frank spruce
#

does it have lots of VRAM?

deft epoch
#

12GB.

frank spruce
#

nice 🙂 i hope it does good with the new SDXL stuff coming "soon"

deft epoch
#

Altho I don't know if VRAM alone is enough when it comes to proompting

#

cuz my 1660 has 6GB

#

and it takes FOREVER to draw high quality images.

#

It's honest to God one of the most unoptimized gpus for ai art.

frank spruce
#

try my settings, 640x768 10 steps SDE karras, no upscaling

#

no hires fix

#

even tho only 10 steps quality ends up very nice with sde karras

#

i notice between 2m and 2m karras, difference is semitransparent stuff is much nicer on karras, like single hairs

#

if/when you do upscale, use the 4x-UltraSharp upscaler, its the best quality

crisp gyro
#

Quick ?: I want to take a photo of a family on vacation and replace the family with a different one. What's likely to be the best / easiest path? Inpaint? Controlnet? Something new I don't know about since I haven't been keeping up on this stuff in months?

#

I've got the latest AUTOMATIC1111 (AFAIK).

frank spruce
#

if you have a photo of the family you want in the image, i would outpaint around them

warm junco
crisp gyro
#

I was rather trying to create strictly AI-generated people.

deft epoch
#

Man, I tried doing that with the 1660 and the results are pretty bad

frank spruce
#

with only like .25 denoising

deft epoch
#

I honestly do not know how to do upscaling separately in img2img

#

which is why I opt for both at the same time

#

during img gen.

frank spruce
#

i watched that video like 5 times literally i swear by it

frank spruce
deft epoch
#

Also about the resolution

#

when I first started using stable, the people who taught me the basics said to never go beyond 512x512.

frank spruce
#

but its also a lot faster to do it in smaller tiles than to do a big big image all at once

frank spruce
#

thats why i like 640x768 its a compromise

deft epoch
#

Oh, hands and feet are my nightmare when it comes to img gen.

warm junco
deft epoch
#

How do I check the version of the stable I use?

warm junco
deft epoch
#

Hmm

frank spruce
#

im using revanimated right now i love it

warm junco
#

On civitai you see Information about the models and they mostly say on what they are based on

deft epoch
#

What if I find a model I like that the author doesn't specify the res it was trained on?

frank spruce
#

when in doubt, experimentation is your friend

#

its plausible a model will do better in a resolution it wasn't trained on, just a crazy idea

warm junco
frank spruce
#

but yeah just go between 512 and 768 and you be fine

#

unless youre a SDXL tester

#

what res do they use

warm junco
#

1024x1024

frank spruce
#

i just found this "variation seed" thing what is that

deft epoch
#

About seed

#

I was also taught to always use -1.

frank spruce
deft epoch
#

There was once a bizarre case

#

I looked at an image made by another person

#

then I copied their exact prompts, configs, seeds and also used the exact same model

#

but the image generated wasn't the same

#

any idea why?

frank spruce
#

were you missing a lora or other file

restive prawn
frank spruce
deft epoch
#

Oh, and about that

#

what's the difference between extensions and

#

textual inversion?

frank spruce
#

extensions are mods to your auto1111 ui

#

textual inversaions are files u link that look like normal tags but arent

restive prawn
#

essentially you can download something else to change the process, could be anything

#

for example Loras

deft epoch
#

I can't seem to make textual inversions work.

frank spruce
#

but ther also is loras and hypernets that u invoke with angle brackets

deft epoch
#

Extensions are fine tho.

#

That's why I asked for the difference.

frank spruce
#

also check your model and CFG and sampler if u want to reproduce a result

#

and make sure u use some basic negative textual inversions like bad dream and fast negative

#

"blessed vae" is pleasant and realistic, default vae is too much saturation and detail in comparison

twin loom
#

does anyone know much about the qdiffusion thing where you can run it from a google drive remotely?

#

for some reason it's not letting me click generate

rancid quail
#

Greetings! I made a fine tune model on SD 1.5 (it's called RPG v4.0 on Leonardo, Civit, Mage, etc...) and I am in the process of training / finishing 5.0. Curious to know when we can start working with SDXL to fine tune or train?

#

I run 6x A10 (AWS GPU) for the training and pipeline. Interested to move my workflow to the new SDXL if this will be possible.

frank spruce
jade trail
#

how i use stable difusion online?

sudden violet
thick marlin
#

A friend of mine works in the entertainment industry and is looking for a laptop under $2000 that he can experiment with ai image gen and deepfake video for production, he doesn’t want a tower because he wants portability, he’s interested in the MSI GP66, does anyone have any better recommendations? I’m a Mac user so I’ve been out of the loop on powerful laptops PCs for a while.

meager marsh
#

AMD gpus out? The Nvidia offerings are shit. I'm guessing since there is no CUDA that work arounds don't work as well.

soft moss
#

Hey !!! new enthusiast here. just hoping to say hi

restive prawn
#

you can get cards dedicated to data processing that are better for this for less than gaming card prices, however those are only for if you build a pc pretty much

#

Graphics Card Ram Size ‎8 GB is what you are looking at

#

I looked up the laptop

#

it will work, just limited

earnest lichen
#

are we supposed to be able to join the event going on right now?

#

it tells me restricted access after logging in

thick marlin
# restive prawn it will work, just limited

Yeah I agree he's pretty dead set against a tower, I use a PC desktop setup for all computer vision related tasks. You came to the same conclusion I did, 3080Ti with 16GB VRAM, it's a passable value for sure, it's just not a great experience to be on a gaming laptop. I'm surprised at how limited "powerful" PC laptops are still.

restive prawn
#

16 gb general ram

#

gaming is gonna be having you take this

indigo ermine
#

This is semi off-topic, but what are the chances stable diffusion or one of the models from civitai compromised my PC? Antivirus turns up nothing. A hacker stole my Facebook and managed to somehow get around the 2FA so I'm trying to think of any way they might have been able to manage that.

restive prawn
#

here's one easy 16 gb option

#

32 GB

#

perhaps you don't like ebay tho

soft moss
#

your statements about "not being enough" is for what purpose? im trying to use this conversation you are having to learn, but im missing some context. this is to train models? or just to use them?

restive prawn
#

600 bucks 12GB ram, plenty of cards like these on nvidia websites

restive prawn
#

with 8GB you would be able to train loras on the lowest settings

soft moss
#

total or dedicated vram?

restive prawn
#

and generate pictures quite slowly and with limited resolution even with extensions that make it easier

restive prawn
#

regular ram to a small extent, but that's easy to handle

soft moss
#

cause im using a laptop with 4gb of vram and so far i did not felt it was "not enouth", but maybe just because im used to it being slow ?

#

sorry, 4gb of dedicated vram

restive prawn
#

how are you even using stable diffusion?

soft moss
#

🤷‍♂️

#

i swear it sats 4gb of dedicated vram, and its just chugging along at 8.10s/it

restive prawn
soft moss
#

it goes faster if i close some other stuff as well

restive prawn
#

straight up there's plenty of ppl who use 8gb vram and when they click generate Cuda out of memory error

soft moss
#

where can i check the exact video card? i know its an nvidia, its a TUF gaming laptop

restive prawn
#

task manager

#

performance

#

then you should see GPU

soft moss
#

NVIDIA GeForce RTX 3050 Laptop GPU

restive prawn
#

scroll down

#

should say

soft moss
#

picture in general with images

restive prawn
#

I run into issue with mine a lot

soft moss
#

maybe im to new to have run into issues so far ?

restive prawn
soft moss
#

i have to close everything cause now a days everything seems to want to use 3d

restive prawn
#

I have to close most things or my system will crash

soft moss
#

i can still watch youtube, or play videos games while waiting, but it makes the generation take way to long

#

games AKA blue stacks and honkai star rail

restive prawn
#

I will crash if I watch YT def

soft moss
#

maybe general system memory is more important than what suggested?

restive prawn
#

Dedicated ram does greatly affect speed I can bet that though

soft moss
#

for sure. there is no discussion

#

but my wallet preffers that i wait, rather than buying a beefy tower pc.

restive prawn
#

But I guess for memory error situations it's about shared gpu memory

soft moss
#

if i get serious enough with this, and i see an avenue for at least 10 bucks a month, ill considering setting the google thingy so i can have a speed up

restive prawn
#

600 bucks and you get a 12 gb ram card

soft moss
#

in the US

restive prawn
#

A grand for 24 gb

#

Near 2 for 36 gb

soft moss
#

my salary is about 1k usd per month

#

and the imports add markup to all the hardware

restive prawn
#

Lets just say my card is several years old with 8gb ram for a reason

soft moss
#

XD

restive prawn
#

I point it out because it's a better deal, not because it's cheap

pine fulcrum
#

anyone have any advice for getting more then one figure in a image?

#

im really having trouble

soft moss
#

image 2 image ? and work on the parameters

#

its called denoising strenght. read the tooltip if you hover over it

weak sundial
#

Hi, I seem to be having issues with using textual inversions like EasyNegative, BadDream and UnrealisticDream. When I don't use them, the images come out looking perfect but when I add them as negative prompts, the images go horribly wrong. Does anyone know what I am doing wrong?

peak portal
#

maybe your model is too far from those

#

the weights have to be somewhat similar for it to work

#

try on a different model or two

earnest lichen
#

easynegative works best with novelai based models, unless I'm confusing it with easynegativv2

weak sundial
crimson belfry
#

I really enjoy SD over all the rest out right now

#

i wish i had a computer to actually run it on my own system, I think that is the biggest setback for many right now.

thorn pilot
#

how does comfy resolve the order of nodes? I noticed if I have two models, it's loading the second one further down the pipeline and running the conditioning steps, then it goes and loads the first model. I can't find a way to set the order manually

ivory plank
#

where can i ask for a help?

sudden ruin
ivory plank
rose smelt
#

Hi! Is there any solution for 9x16 ratio (TikTok/Reels) SD video generation? Pls share 🙌🏻

sudden ruin
#

I mean the solution is the render it in that resolution, or what do you mean

#

im on 3.10.9 so id guess so? hmmm

#

I remember having some problems with python too from time to time PU_Sadge

warm junco
#

Its due the non latin letter

#

SD works with python 3.10.6 - 3.10.11

soft moss
#

whats the exact error and where do you see it ?

warm junco
#

you would need to rename your username, or you can try to install python and SD in an other destination. Like on an other Drive

deft epoch
#

So, just a quick question, not asking for support

#

once my GPU arrives and I replace the older one, will I have to reinstall python and such?

#

Or just the gpu drivers?

soft moss
#

it does look like that is the issue. can you change the user username ?

soft moss
fervent thunder
#

how does this stability ai work?

soft moss
#

i beleive its not SD or python related, but rather Windows related. its the enviroment that is not able to support the start of the bat file. and wrinting this message, i toufh, have you tried runing it from powershell ?

fervent thunder
#

how can I create pictures

soft moss
#

but the best way to understand it is to try it out. also covered in that guide, but you can google, there are free services that let you generate a few images and then charge you if you want to keep going

pearl fog
#

Then type whatever you want to make

soft moss
#

create a second user ?

#

secondary user and then run it from main user if you dont want to swap, altough i still think there should be a way to migrate your current user to another folder

#

test it, but dont think it will work. sustituttion will, in the end, be the same. the bat trying to run the same command line

#

so, you can try powershell hoping it is UTF-8

mint peak
#

gm

#

Can someone tell me how I can use the SDXL V0.9 engine via API? Whats the name of the engine exactly?

Edit: found it

fringe yew
#

hello

#

@wild steppe im sorry for pinging but i dont underestand what this server is about agony

pearl fog
#

It's about stable diffusion, an image generation software

fringe yew
#

ohhhhhhh

#

and how does this exactly work

#

like

#

what do we have to do idk

pearl fog
#

You can test the bots in the bot channels if you want to make stuff

pearl fog
fringe yew
#

okok ty

#

and after that

#

what do i do lmao

#

😭

surreal perchBOT
pearl fog
#

Wait for it to make some stuff for you

fringe yew
#

okok

#

ty

#

man im actually an interior designer at the age of 16 and idk what to do

willow coyote
#

How to add camera movements to zeroscope-v2-xl? What I currently do:

  1. Generate image
  2. Upscale using swim2sr (https://replicate.com/mv-lab/swin2sr), stopped using gfpgan as it messes up my details
  3. Turn it into a 5-second still high-def MP4 animation (I use which https://replicate.com/anotherjesse/zeroscope-v2-xl doesn't accept images)
  4. Feed it to zeroscape (model 570w) with instructions to add birds, pedestrians and slight fog so it doesn't seem like a still photo
  5. Not gotten this far but: feed it to zeroscope again but this time with the xl model
  6. Use ffmpeg to split the video back ito images, add some neat post-processing scripts, and then reassemble

How would you guys improve this workflow? Any cool tricks?

willow coyote
# fringe yew man im actually an interior designer at the age of 16 and idk what to do

Me too I'm an architect (we basically got taught the same stuff I think?). I mostly repurpose my design skills to make Rails apps with huge income potential. Since we studied this stuff at school, and our curriculum included all the other design genres from graphic to branding, so I'd say you have a huge advantage over 99% of the crowd that is self-taught.

fringe yew
fringe yew
willow coyote
#

Aha, nice. I'm 39.

Have you tried improving interiors with controlnet-hough?

fringe yew
#

whats that

willow coyote
#

Can do some crazy shit with that

#

Lemme find an example

fringe yew
#

i use autocad , 3d max

#

and soon vray

willow coyote
rancid quail
flint compass
#

Hi !

ancient epoch
#

heya does anyone know how can i run automatic1111 as an api that i can connect to and request images

fervent thunder
#

(4 GB Graphics/AMD Radeon RX6500M/144 Hz) How well is this graphics card? And is it integrated or dedicated?

stable bison
#

Hi everyone! I have a question I couldn't find an answer. Is it possible to ask for stable diffusion to generate objects inside an image i provide?

For example, suppose I have a photo of grass field. Can I ask stable diffusion or other model to ""draw"" a tire laying on this grass field?

frank spruce
fallen vault
#

Why I can't see any bot channels?)

wild steppe
fallen vault
#

I yes if I click on it but in my channels list I don't see any of the bot channels

willow coyote
#

Any of y'all good with Zeroscape? All I've managed to do is add some barely noticable car movement. I need the camera to move around and stuff, how is that done?

https://i.imgur.com/xtqwr5K.mp4

fallen vault
#

If I then click add channel to the list then I see it there looks like by default they are unavailable or hidden...

wild steppe
#

They shouldn't be hidden by default thinking

fallen vault
#

Hm I don't have browse channels option there

#

I see events and then center stage category

fervent thunder
wild steppe
fallen vault
#

Oh found it, thank you very much))

willow coyote
misty linden
#

hey is there a way to make img to img take a non tiled image and turn it into a tilable seamless image? i tried using the Asymmetric tiling extension but that didnt work aswell as the normal tiling enable button... kinda new to stable diffision as im a 3d level artist but i know a decent amount
[1:45 PM]
also im using the texture diffusion checkpoint to help it make 2d texturemaps instead of 3d images.

soft moss
#

anyone has experience with google colab? the "compute units" scare me a bit. dont know if they are alot or to little

pearl fog
#

I think it's about 6 hours before they boot you, but that was back when the free tier wasn't so restricted

#

*6 hours a day

soft moss
#

i was thinking on paying for the middle tier. do i still get booted out after 6hs ? of usage? or inactivity ?

pearl fog
#

I believe it's usage. If you're paying for it, they shouldn't boot you if you're inactive. They will boot you on free if you aren't wiggling the mouse every so often

#

Or if paying customers need your server

forest trout
#

I miss all the free compute that you got with a basic Google Collab subscription 🥲

#

But I've got a 3090 now so it's fine I guess.

steep star
#

@earnest lichen can I dm regarding your training?

earnest lichen
#

sure

#

you can also ask in sdxl chat though - training talk is ok'ed by devs

autumn onyx
steep star
frank spruce
fervent thunder
#

Ah k

frank spruce
#

can i change the weight of a lora halfway thru generation or are they fixed

#

it didnt work

#

i think it's immune T,T

boreal cosmos
#

can intel UHD or amd vega 7 do image gen

#

maybe with directML?

warm junco
warm junco
frank spruce
#

i like playing with variation seeds

#

even 0.01 weight gives big changes

warm junco
#

You can also use the X/Y/Z script to compare

craggy musk
#

hey! i'm struggling with inpainting to get the results that I want :c any tips?

boreal cosmos
#

if its just a regular model then you might need to convert it to an inpainting model first

craggy musk
#

ohh! that might be th eproblem then :o

boreal cosmos
#

note that this does degrade the model in my experience, so if the model already can somewhat do inpaint (such as revanimated) then I would just stick with the original model and generate a ton of different inpaint results and pick the best one

#

I would select 'Generate Forever' and come back 10 mins later to pick best result

craggy musk
boreal cosmos
#

right click the Generate button

#

you are using a1111?

craggy musk
#

I sure do have that option :) thanks

boreal cosmos
#

yw

craggy musk
#

When I do the inpainting, do I change my prompt to only include the new stuff? or keep my old prompt and add something there ?, I'm currently using only new things and added a drawing in the inpaint sketch mode. I also tried the same in the inpaint mode but I am unsure what I should be using ^^"

boreal cosmos
#

I would add the new stuff to the same prompt and hope for the best, removing the original prompt break things so I dont do that unless the area I'm inpainting is small

boreal cosmos
#

although it seems to take 40 minutes to generate a single 512x512 image of a banana

wooden briar
craggy musk
#

I'm trying to create my own inpaint model with the link you provided just now, but the reddit posts says to put sd-inpainting in slot a, my model in slot b and pruned in slot c, while choosing the add difference option to teach model b the difference between a and c ?
However the checkpoint merger states :
The difference between the last two models will be added to the first. Requires three models; A, B and C. The result is calculated as A + (B - C) * M

Do I need to put my model in slot A and inpainting/ pruned in b and c?

boreal cosmos
#

do as the OP said

fleet ferry
#

Hey guys. I'm wanting to generate sticker pack type images. Sanrio style would be a good example. I'm able to do it quite successfully in midjourney but haven't achieved nearly as much in SD. Not finding anything similar on Lexica either. Advice?

fervent thunder
#

how does somebody on a smart phone access negative prompt/aspect/style

stable pawn
#

In the Bot-1 channel there is bots posting same prompts over and over again, I think there is 3 bots, can some understand what the point with that is? I do not think it brake any rules, but there have to be a reason for posting same prompt several hundreds of times per day.

high lichen
#

Anyone know how to enable multiple gpu usage. I have a few gpu’s connected and looking to run them all when using stable diffusion. How?

fervent thunder
#

those are not bots

fervent thunder
#

im doing it right now

#

because the images generated are so random with so many glitches etc you have to do many many passes if you are trying to get it to make something specific that looks how you want

stable pawn
fervent thunder
#

yeah or they are just making images for their own use for any number of reasons

#

like im in bot 4 trying to make an image for a t shirt

#

and my friend is trying to make a logo for his business

willow coyote
#

swin2sr is definitely my new go-to upscaler 👍

wide tangle
#

Hello!

I am new to all of this, so bear with me.

What stability ai models can be used for commercial use, if any?

wooden briar
#

All of them

tulip yoke
wooden briar
#

Laws are imaginary

tulip yoke
tulip yoke
wooden briar
#

But really I don't know the answer, sorry. I would do it regardless

weak sundial
#

Nobody knows the answer

wooden briar
#

Yeah, these AI tools really highlight the problem with IP

#

I hope AI pulls the whole IP house of cards down tbh

tulip yoke
#

?

quaint garnet
#

Does anyone know how to cancel a command?

sly burrow
warm junco
grizzled dagger
#

I want to do a walpaper for my pc but the quality was never good, i put 8k quality but it's doesn't work, do you have a solution guys ?

potent spire
#

Since laws are "imaginary"

potent spire
#

Whats the problem with IP.

#

?

warm junco
potent spire
#

One good and big reason why professionals and companies do and would avoid generative AI or at least and especially Stable Diffusion or Midjourney for example

#

for that id recommend Adobe Firefly when its out of beta

#

But ofc you wont have copyright protection unless you put work in

cyan stream
#

Hi all, is there any interest in porting these models to Julia language ecosystem?

#

Particularly LLMs ?

hollow minnow
#

Can i make something using img2img?

fervent thunder
#

hi

#

i need help

#

OSError: [WinError 126] The specified module could not be found. Error loading "F:\AI\Stable\stable-diffusion-webui\venv\lib\site-packages\torch\lib\torch_python.dll" or one of its dependencies.

bleak matrix
#

Good morning, everyone!

#

I hope you all are doing well this morning 🙂

dark escarp
#

can anyone tell me where to put commandline arguments in comfyai

grizzled dagger
strong elbow
#

helloo, does anyone know how to create a lora? I have been trying for a while now and i have been watching a lot of tutorials but they are not helping at all

wooden briar
wooden briar
potent spire
#

A more extreme one

wooden briar
#

And do laws prevent you from murdering?

potent spire
#

But i personally hold more to IP of mine or many others than to life of many people

potent spire
#

As with a lot of people

#

But what exactly is so bad about IP in your opinion?

wooden briar
wooden briar
pearl fog
#

An issue I see with copyright is how long it lasts. A certain company ||Disney|| keeps pushing to extend it so that none of their stuff will enter the public domain, even things from the 1920s. At some point, you're losing stories that are a part of history just so that some company can keep making money off of it.

boreal cosmos
potent spire
ebon stratus
#

What can be the budget for an optimal pc for ai

#

Which includes 3060 12g

steep star
#

if you are from us

potent spire
steep star
#

you have many used options in us

ebon stratus
#

I'm not from us

steep star
#

not sure about other countries

potent spire
#

If lets say my grandpa was an artist and he puts the copyright and ownership of the artworks to me and my name and when i get older i give those to my children and so on

#

Legacy

ebon stratus
#

I can get a decent pc for 500 dollars here

#

I'm not sure about the specs

#

Is it better to buy a prebuilt or should I build it myself

pearl fog
#

Building it yourself is usually cheaper

ebon stratus
#

Got it

pearl fog
#

I have a build saved here but the prices are US

ebon stratus
#

Ok thanks I'll compare it

wooden briar
potent spire
#

Oh, well move along then

wooden briar
warm junco
potent spire
#

Nice try commie

wooden briar
wooden briar
#

Take care

potent spire
#

Those werent my words

wooden briar
#

I literally couldn’t care less

#

Move on

potent spire
sly burrow
boreal cosmos
#

I did not test cpu as the branch didnt have cpu option

shadow swan
#

is there a way to setup automatic1111 to only restore the largest face? if there's people in a blurry bokeh background it tries to restore those faces too which i don't want

boreal cosmos
shadow swan
#

is there a way to do it automatically? i'm doing a bunch of batch images

boreal cosmos
#

maybe use a remove background extension to remove all the background of your batches

#

then after that, batch overlay the new restored face images over the old images

#

(the remove background works on the batch menu)

potent spire
frank spruce
#

ami crazy or does every lora drastically change the style

ebon stratus
#

Which factor is the most important for a gpu

#

Is it like vram

#

Or something like that

tough bough
#

Depends on what you are doing. Generally you want the fastest processor with the amount of VRAM you need. So if you are not doing training, you don't necessarily need massive amounts of VRAM. Though models are getting more VRAM hungry.

#

Someone else would be better for actual numbers.

frank spruce
#

I can barely get by normal image generation with 4GB vram

ebon stratus
#

So is rtx 3060 12 gb ram enough

#

I do some training

#

Tho

frank spruce
#

i have wanted to train a lora just to make a particular image

tough bough
#

That should be enough for generating at least, and I think for a decent amount of training as well. I know I have seen some debate between that and 3070 for which is better. I genuinely don't know the answer, and it may depend on what your goals are.

frank spruce
#

your goals will change if youre like me

ebon stratus
#

Ok 👍

#

That looks good to me

potent spire
ebon stratus
#

I'll start with one

potent spire
#

ideally an AI specialised GPU but a gaming one works well too

ebon stratus
#

Then I'll upgrade it with two gpu

#

How much gpu does colab provide?

#

I'll go with the 3060 for now

potent spire
#

I will probably go for RTX 4090 next year

#

If 50 ones come out maybe 5090

fervent thunder
#

how do I generate the normal maps used for control nets?

fervent thunder
#

Hey guys how can I use this AI to edit the images

hybrid ivy
#

When SD 2 coming out?

jade hamlet
#

Hey everyone, its been a while since I tried installing SD, but I was hoping to give it another try. I was hooing someone might have a tutorial I might actually be able to understand.

#

Every time I try following a tutorial I end up getting the morst images and I can't get https://civitai.com/ models to work.

regal sundial
#

hi everyone, im looking for help to see if its possible to implement SDXL into Invoke AI

vernal ore
#

Hi everyone

undone garden
#

Should I sell my 4070 ti for a 3090?

vernal ore
#

Does the 3090 perform better?

undone garden
#

That's what I'm asking. It does have double the vram

vernal ore
fervent thunder
#

then use those tags to increase quality in your prompt

blazing pasture
#

Holy crap, I just figured out how to use LoRA's

velvet plover
#

Where can I ask questions related to ClipDrop?

craggy musk
#

My generations always get stuck at 97-100% and stay there for quite some time :c the console tells me that the generation is already finished though :c any ideas on how to fix that? saw something about changing some VAE settings to sequentiell, but I couldn't find that in the settings

craggy musk
warm junco
craggy musk
#

512x768, just generating one image, 20 sampling steps, 20 high res steps, 0.7 cfg ^^" :)

warm junco
craggy musk
#

no adblock enabled :)

#

Okay! It is probably a VRAM issue as it spikes when the image finished generating :c

warm junco
craggy musk
#

rtx 2070 / --xformers

warm junco
ocean mango
craggy musk
#

just did a fresh install 2 days ago :c i'll test around some more :) thank you very much for the help!

warm junco
craggy musk
#

that makes sense ^^" must have missed the channel when I looked earlier :c

warm junco
humble panther
#

is dreamstudio still the best way to access sd?

#

ive been gone since september

pearl fog
#

You can download it if you have a good GPU

humble panther
#

I do!

#

4090 😄

pearl fog
# humble panther I do!

Sorry it took a while to respond. Automatic1111's webui is the most popular version. Here's the install instructions

#

A lot of people here are using ComfyUI instead, if you like node based setups

humble panther
#

should i follow the "installation on windows with nvidia gpu" or the "automatic installation on windows" directions in the first link?

#

and no worries! i appreciate you even stopping to help

pearl fog
#

Automatic installation seems to work better for people

humble panther
#

whatever this process is, does it stay updated throughout when things release through sd, like whatever SDXL is?

pearl fog
#

You're going to have to download the new model when it officially released

#

ComfyUI is supposed to release a recommended setup for SDXL once it drops, since the node stuff can be complicated

#

I dunno about automatic but I assume he'll add an SDXL update too

humble panther
#

Oh gotcha. I've also heard about something called Loras

pearl fog
#

Yeah they're like tiny add-ons that can add new concepts to a model

#

So like if you have an anime model, you can make a Lora for a new anime character that isn't in the base model

humble panther
#

Alrighty cool, if its compatible let me do these initial steps and try getting this to work first

#

oh wait a sec is this like a local UI for using stable?

#

so you dont need to type into a command prompt

#

like how you had to back in the day

pearl fog
#

Yep!

humble panther
#

Wooooow

#

So its like running something similar to dreamstudio with having an actual visual UI for settings and all that

#

i guess things change after 10 months XD

#

amazing lets get this working

balmy hemlock
#

Hi, any expert of lora training? ( kohya )

golden tree
#

hey guys i'm new to playing with AI, i wanted to know if there's a way to scan safetensor files, i downloaded picklescan all ok with the models, but safetensors can't be scanned by it from what i've seen, i wanted to know if there's any way to scan them, sorry any error I'm using the translator

pearl fog
#

Safetensors files are supposed to be safe, you can't get pickled by them like ckpt files

finite cloak
#

if your translator isn't getting it, "safetensors" starts with "safe" ie it's specifically named for the fact that it doesn't have the malware risks that pickles do

humble panther
#

okay pashy

#

it says v1-5

#

is that ilke, good?

#

i've gotten the UI to appear in my web browser

pearl fog
#

Yeah that's the most popular base model. You can download other models too

humble panther
#

i saw there was 2.1

pearl fog
#

Did you try generating anything? Nothing complicated, just to see if it works

#

Yeah a lot of people don't like it as much because it's different to prompt for

humble panther
#

oh gosh

#

yeah it made... something alright

#

this did appear in command prompt though

#

>URLError: &lt;urlopen error [WinError 10060] A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond&gt;</div>"]

#

but an image did come out

#

it did not get xformers though

pearl fog
#

Oh you need to add that in your command arguments

#

I don't think it's necessary for a high end card like yours but it makes stuff faster

humble panther
#

what about esrgan?

pearl fog
#

Upscalers

humble panther
#

im looking at the github for it not quite getting how to include that into this

pearl fog
#

Did you have the old install on your computer still? The Sept one?

humble panther
#

nor the xformer

#

no that was a separate pc

pearl fog
#

Find the stable-diffusion-webui folder and open it

#

(you'll want to close out the webui and command prompt for this)

humble panther
#

okay i have the initial folder sd.webui and in that folder is webui, im in that second webui folder now

pearl fog
#

You'll want to find webui-user.bat

humble panther
#

see it

pearl fog
#

Right click and edit it

humble panther
#

alrighty this is whats inside
`@echo off

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

call webui.bat`

pearl fog
#

On the same line as set COMMANDLINE_ARGS=, add this:
--xformers

humble panther
#

should i put a space after =

pearl fog
#

Like this COMMANDLINE_ARGS= --xformers

humble panther
#

yee

#

okay that's done

pearl fog
#

Space doesn't matter

#

Save it and close

humble panther
#

is that how i'd put in something like esrgan

pearl fog
#

I believe that's supposed to install automatically

#

But DoubleClick webui-user to run it again

humble panther
#

would it be an extension?

#

gotcha

#

webui-user bat?

#

not run.bat? thats how i initially ran it

pearl fog
#

No webui-user is the file you run

humble panther
#

oh im guessing its because i did the other install method

pearl fog
#

Yeah

humble panther
#

but the command prompt did load xformers it said

pearl fog
#

It should give you a URL when it's done

humble panther
#

yup! in the ui now

#

trying to see if the gan is in here

pearl fog
#

Yay

humble panther
#

would that be under extensions?

pearl fog
#

Upscalers

#

The file itself should be in the models\ESRGAN\ folder

humble panther
#

hm if i open esrgan its empty

#

in fact, most are

#

esrgan, gfpgan, lora, codeformer

rain sierra
#

hey

humble panther
#

hiddenitems are viewable at this time as well

#

ello void o/

pearl fog
#

What steps did you follow for the install?

rain sierra
#

does stable diffusion work on linux

pearl fog
#

Yep!

rain sierra
#

cool

humble panther
pearl fog
#

If you're on AMD it's better on Linux

#

Cool the one that seems to break stuff

humble panther
#

ff

#

ill restart it!

#

i have good internet should be no problem. just delete the folder entirely?

#

sd.webui

pearl fog
#

Do you have a venv folder?

humble panther
#

where would i find that out?

pearl fog
#

Windows search

#

It usually makes stuff there

humble panther
#

hm typing venv into search pulls nothing up

pearl fog
#

I'm not sure if that version of the install will put git on your computer

humble panther
#

would it appear in downloads?

pearl fog
#

Under programs

#

It should have installed python at least

humble panther
#

yeah appdata - local - microsoft - windowsapps - i see ptyhon.exe and python3.exe

pearl fog
#

So... Yeah. If it's being dumb you might have to delete the SD folder and try again.

humble panther
#

Im fine with that

pearl fog
#

Is it just not working at all or is it just esrgan that's missing?

#

Because you could download the file separately

humble panther
#

well now that i know that's the one that breaks things, who knows what else is missing or lacking asides from esrgan. the gfpgan is empty, codeformer is empty, and a few others

pearl fog
#

I don't know why automatic even keeps those instructions up

#

The other version works fine

humble panther
#

i guess i might be confused then. i thought you said automatic works more often for most people

#

oh wait did you mean use this one

#

`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.`
pearl fog
#

Using the Automatic Installation on Windows installation

#

Yes

humble panther
#

gotcha, i did this one
`Automatic Installation
Windows (method 1)

A very basic guide to get Stable Diffusion web UI up and running on Windows 10/11 NVIDIA GPU.

Download the sd.webui.zip from here and extract the zip file.
Double click the update.bat script to update web UI to the latest version, wait till finish then close the window.
Double click the run.bat script to launch web UI, during the first launch it will download large amounts of files. After everything has been downloaded and installed correctly, you should see a message "Running on local URL:  http://127.0.0.1:7860", opening the link will present you with the web UI interface.`
pearl fog
#

You should have python so you can skip that

humble panther
#

ill download git then (searchign git in explorer showed nothing)

pearl fog
#

So install git, then git clone

humble panther
#

ill delete the original sd folder for safe measure

#

i installed git, but i do not run the git.exe that appeared in downloads right

#

or do i run that first and that is the install

pearl fog
#

If it's installed you should be able to do the git clone

humble panther
#

into command prompt yeah

#

okay copy pasting this into command prompt

#

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

#

'git' is not recognized as an internal or external command, operable program or batch file.

pearl fog
#

So I guess it wasn't installed. Run the git exe

#

No special conditions; just keep clicking "next" until you're done.

humble panther
#

anything that should be toggled on/off for the install?

#

ah okay

#

wow, that was a lot of nexts lol

#

okay that finished

#

step 4 says Run webui-user.bat from Windows Explorer as normal, non-administrator, user.

#

should i see that in a file somewhere?

pearl fog
#

It would be in your stable-diffusion-webui

#

Folder

#

You git cloned right?

humble panther
#

ah i had to find the folder i see it now

pearl fog
#

You can add xformers now or you can run it

#

Your card probably doesn't need xformers

humble panther
#

fair enough! doesn't hurt, already added

pearl fog
#

Cool, cool

humble panther
#

uhh hmmm

#

python is DEF installed

#

but

#

`Couldn't launch python

exit code: 9009

stderr:
Python was not found; run without arguments to install from the Microsoft Store, or disable this shortcut from Settings > Manage App Execution Aliases.

Launch unsuccessful. Exiting.
Press any key to continue . . .`

pearl fog
#

I bet it's the path

humble panther
#

its in microsoft - windowsapps

#

lets just move em to downloads see if that works

pearl fog
#

When you install it, you need to select "add to PATH" in the installer.

humble panther
#

ohh, see i never manually installed it the previous method did it for me

pearl fog
#

I wonder if you could run the exe again and select that

humble panther
#

yeah lets try

#

python.exe or python3.exe

#

ah probably 3

#

since that'd likely be 3.10.6

pearl fog
#

Yeah it's supposed to be python 3.10. 6 version

humble panther
#

wth did that other method do

pearl fog
#

It's just there for the amusement of the computer demons

humble panther
#

i think i need to just go install 3.10.6

#

manually

#

i downloaded this from the link
python-3.10.6-amd64.exe

#

weird it says AMD

pearl fog
#

I think it's referring more to 64 bit computing with that

humble panther
#

okay ive clicking webui-user.bat again

#

stuff is happening

#

sheesh headache and a half there

#

ill wait for it to complete but regardless ty for sticking with me through this process

#

says something about a new pip update is available, in the command prompt that is

pearl fog
#

It always says that, not necessary

humble panther
#

coo coo!

#

do i have to worry about addiing --autolaunch or --update-check anywhere?

pearl fog
#

It's personal preference. Some people like to avoid updates in case something breaks

humble panther
#

so are the folders in models supposed to be empty?

#

esrgan, gfpgan, codeformer, lora, ldsr, among others are empty

#

most say folder empty except for a select few like stable-diffusion, vae-approx, karlo

pearl fog
#

Did you try running webui-user again?

humble panther
#

yup its up and running

pearl fog
#

Is the webpage loaded?

#

Stuff like loras will be empty because you haven't downloaded any

humble panther
#

Yeah webpage has been loaded

pearl fog
#

Try to generate something and see if it does something

humble panther
#

oof i just input this to test it out
Realistic image. a futuristic city, design Frank Lloyd Wright. futuristic helicopter with vertical propellers. that would be like a village that would not have more than 10,000 people, where each family would receive 2.5 hectares and that water would be a public resource, while food, energy and other goods would be produced locally. The main design feature of this city is that industrial spaces would be replaced by urban. Style: photography
my goodness does it come out awful xD

#

but yeah something did generate at least

pearl fog
#

Upscalers should be under the img2img tab

jolly heart
#

hi

humble panther
#

where could i find it in the tab if you know?

#

ello gztar o/

pearl fog
#

Upper left. Look for txt2img and img2img should be next to it

humble panther
#

oh sry i meant im in img2img now

#

would it be in sampling method?

pearl fog
#

SD upscale under scripts

humble panther
#

i got the following

#

lanczos, nearest, esrgan_4x, ldsr, r-esrgan 4x+, r-esrgan 4x+ anime6b, scunet gan, scunet psnr, swinlr 4x

#

would gfpgan be in there if i had it?

pearl fog
#

Yeah those are different upscalers. I like swinIR but it depends on what you like

humble panther
#

so if i wanted to make these images a little more coherent i should get loras?

#

oh i guess the sampling method matters too

#

any recommendation haha

#

there's so many now!

pearl fog
#

Dpm++ 2S is popular

#

Everyone has their own opinions though.

humble panther
#

2s a or 2s a karra

#

yeah

#

im sure ill explore em more in time

pearl fog
#

Karras never seem to cooperate with me, but they're supposed to make stuff better

humble panther
#

so loras are things you can install that just overall help produce images a little more clearly?

golden tree
pearl fog
#

They add a new concept or style to it

humble panther
#

yeah i see there's a "style" box that's empty

pearl fog
#

Oh, and there's a new website full of custom models

humble panther
#

and these aren't loras?

#

interesting

pearl fog
#

Check if it's a lora or a full model. Models are 1.99GB+ while loras are tiny

humble panther
#

its funny you said 1.99

#

thats how big this one is lol

#

so i download this one for example

pearl fog
#

Yeah loras are like 200MB or less

#

There should be prompts on there too. If you see a picture you like, click the ℹ️at the bottom

#

And you can copy their settings to tedt

humble panther
#

dreamshaper_7.safetensors is in the download folder, do i just move those over to the models folder in sd?

pearl fog
#

Goes instable-diffusion-webui/models/Stable-diffusion folder

humble panther
#

ahhh

pearl fog
#

Loras go in the Lora folder.

humble panther
#

is that the site i'd find loras?

pearl fog
#

If you mess it up it will give you some error like All NaNs produced

#

Yep! That's where all the loras are

#

The search kinda sucks so I just do a Google search "whatever Lora site:civitai.com"

humble panther
#

could you give an example? is a lora like a narrow focused specific thing? like there's these models, but then there's loras whihc would hone in on something specific? like a superman lora?

pearl fog
#

Full model: comic book style
Lora: Superman comic by X artist

#

Usually loras are for specific things that won't be in the main model

#

The full model is kind of like a full game, while the Lora is like DLC adding a special small item to it

humble panther
#

gooootcha. is there a model you typically like using?

#

for just generalized things

pearl fog
#

I like deliberate

humble panther
#

this is difficult

#

im trying to copy what someone else did

pearl fog
#

So they're using two textual inversions in the negative prompts you'll need those too

humble panther
#

i input that into negative prompt yeah

pearl fog
#

If you want it exact (it might not be 100% exact though because it's dependent on hardware)

#

They're calling two files with those trigger words

#

Those files go into the stable-diffusion-webui\embeddings folder

#

Of course they used all the fancy stuff for their fancy example pic 😂

#

You would need to restart to get the effect of them

#

But, yeah, results will be different if they're using a TI and you aren't

humble panther
#

ill do a restart!

#

i did find deliberate btw

#

there are two files i had to download for it

#

one is .ckpt and the other .safetensors

pearl fog
#

You only download one of them

#

The .safetensors are better because they can't contain malware and they're faster

#

.ckpt is only there for compatibility issues with really old stuff

humble panther
#

does styles dropbox get used?

pearl fog
#

It lets you save your prompt so that you can reuse it

sterile raven
#

We need better web design on these models

humble panther
#

dang there's a lot to this!

#

what does BadDream and FastNegative do as a negative prompt?

pearl fog
#

They're supposed to improve image quality

#

People usually have a long set of negative prompts, so turning it into a textual inversions saves them from having to type a bunch of stuff

humble panther
#

whoa!

timid vigil
humble panther
#

the problem is gonna be which checkpoint to use, which sampling method to use, and which upscaler to use........

#

is there a way to make it prioritize certain things

#

in that persons prompt

#

holding a magical greatstaff, 1girl, (medium shot:1.4), european, caucasian, long hair, hair in motion, wind-swept hair, wind-blown hair, twilight blue, sparkling silver, galaxy purple, black velvet, standing, hands on hips, hands on waist, (enchanted sorceress robes:1.3), (glowing necklace medallion:1.3), golden glowing jeweled circlet, intricate stitchwork, ornate patterns, elegant, (noon, sun, sunshine, blue sky:1.25), fcDetailPortrait, (masterpiece:1.2), soft lighting, subsurface scattering, heavy shadow, (high quality:1.4), golden ratio, (intricate, high detail:1.2), soft focus, in a forest canyon, epic view, grand scenery, coniferous montane forests, tall trees, (mountainous glaciers:1.2), redhead, stoic, battleweary, determined, (glowing blue eyes:1.5),

#

i can't get the initial 4 words to come through

#

"holding a magical greatstaff"

pearl fog
#

You can add weight to them, that's what the :1.5 stuff is

#

A word is usually 1, but you can increase it by adding the number at the end

#

(cat:1.5) if you want more cat (cat:0.7) if you want less cat

humble panther
#

so would it be "holding a magical greatstaff (1.5) or would the beginning chain for the whole prompt to (holding a magical greatstaff:1.5)

pearl fog
#

Some stuff it gets confused by, though

#

(holding a magical great staff:1.5)

#

But it might not know what a great staff is

#

Maybe just staff

humble panther
#

true

#

im def having an issue with img2img inpainting as well. im drawing a mask over this extra hand thats coming out of the dress and generating and nothing occurs

young ivy
# humble panther `holding a magical greatstaff, 1girl, (medium shot:1.4), european, caucasian, lo...

A sorceress with glowing cobalt blue eyes and long flowing wind swept red hair, holding a magical great-staff and wearing her enchanted robe of intricate stitchwork in dark violet colors contrasted by silver, exploring a river canyon at dusk, with tall forest trees topped by grand mountains with majestic glaciers, one hand on hip and the other wielding her staff, soft lighting, subsurface scattering, ambient occlusion, dark in background, natural lighting, ultra wide angle, HDR, 8K ?

#

don't bother with high quality and masterpiece

#

they don't really do anything unless the model sucks and has decided to name everything good masterpiece

#

also depends if you're using a model responding to natural language (CLIP) or danbooru tagging

lofty atlas
#

Hello which diffusion model can i train with a 3070? I am trying controlnet but it needs more than 8 gb of vram

coarse warren
finite cloak
coarse warren
young ivy
#

uhuh

#

what is it supposed to be?

#

and why is another website charging for their front/backend anything new?

young ivy
#

esp if you're charging for it. Then nothing has changed

coarse warren
#

Which one do you think people would prefer to use between the cloud-inference extension and rundiffusion?

young ivy
#

what? I think people prefer the free open source extension or webui they can install locally, or even a gui or docker

#

if you want to make something that costs, you need to put in a lot of time and effort into making an entire UI, not a webui, but a software

#

you better work hard to make it bette,

#

smoother and create a workflow that is uninterrupted

coarse warren
#

Previously, we believed that users could use cheap gpu like the GTX 3060 to generate 512x512 images locally and then seamlessly switch to the cloud to generate higher-resolution images and perform Lora training.

young ivy
#

cause if all you're doing is charging for an API or something....Then thats just not it

#

people already have an API for the official website

#

and an extension with a switch could easily have people moving back and forth

coarse warren
#

web-ui seems to have become a fact, so we don't want to disrupt users' habits.

young ivy
#

no webui is just something that's easy to make because of gradio

#

so it's the developers that have the habit

coarse warren
#

The stable diffusion api is a bit awkward as it seems to be only usable for developers. However, developers will still go on to develop products that compete with webUI, such as "getimg."

young ivy
#

as a user my flow wouldn't be dosriåted by an actual software

#

nah the API works for a plugin for photoshop and one for Krita to my knowledge

#

it also works on some extension, don't know the name of it

coarse warren
#

However, people seem to hope that automatic1111 becomes the Photoshop of this era.

young ivy
#

it won't

#

people aren't waiting for automatic, they're waiting for it to work on less vram

#

that's it

coarse warren
#

Most people find it difficult to afford, so we designed this extension that can generate 70,000 images for just $100. hope it can help everyone.

young ivy
#

I mean it's great that you're thinking forward and want to be an entrepreneur but I still don't understand what it does

#

you haven't told me

#

all I've gathered is that it's a webservice with powerful gpus

coarse warren
young ivy
#

and you're charging for the extension?

coarse warren
# young ivy and you're charging for the extension?

Currently, the plan is to charge based on image generation requests as we utilize cloud-based GPUs. However, in the future, we also have plans to develop a universal extension to support more stable diffusion API platforms.

steep iris
#

guys i have a question
i want to make Photograph with one person but sd always makes photos with two person
how i fix this?i tried writing solo or 1man but always gives 2 person photos

coarse warren
young ivy
#

So you're basically just coding your API into an extension?

coarse warren
young ivy
#

Hmmmm

burnt kernel
#

Hi! I'm new to this sector and I have a specific question for a project.
Can Stable Diffusion overlay one object on another?
I give examples:
If I have a t-shirt and a logo, can the AI put the logo on the t-shirt?
If I have a F1 car and a drawing, can the AI print the drawing on the car but taking into account the shapes of the car?
Let's say that there will be always a static image (base image) and another that need to be un top of it.

delicate tiger
burnt kernel
#

Both images i'll provide
I only need the tool to generate the final image

#

@delicate tiger It's possible to do that in Stable Diffusion?

delicate tiger
#

with the t-shirt you're better off with photoshop imo

#

the car might be difficult, needs controlnet probably

burnt kernel
#

🙃

fluid charm
#

Hi guys, I'm new to SD. The faces on my characters look off and I don't know how to fix it, anybody that could help?

stable pawn
#

I ask this for I had a stroke and still have some trouble sorting information in a logical way, I did hear that a real 0.9 is out now, is it any point to get it or metter to wait for 1.0 and if it can be run locaaly can some point me to where I find the correct info to do so

fervent thunder
#

I got this error when tried to run controlnet scribble

winged hearth
#

Is there any APIs for convert video from text and image. ??

#

Please help if anyone knows..

balmy hemlock
#

Anyone can help me with a problem I meet with Kohya ?

sharp mortar
#

it's stfo server?

craggy musk
#

for some reason my images end up only having half a head, the rest gets cropped by the image border :c

stray trout
#

Are bot channels cleared for mentions after a set amount of time?

#

I’m not able to search in this server for the bot mentioning me. No results are found.

fervent thunder
#

can someone help me out?

raven flicker
flat steeple
#

Does anyone know what a very fast sampler is? I'm generating the class images for dreambooth -- and it looks like class images don't matter that much for dreambooth

#

so I'm trying to use a fast sampler to generate my class images faster

alpine bear
#

Although in my experience, UniPC has needed more steps to get good results.

#

and my experience is limited to a few specific models.

fervent thunder
#

Hey anyone know what direction to point me in if I want to make my own safetensor lora model?

fervent thunder
#

What other positive prompts would you guys recommend other than (masterpiece:1.2) for generally better image quality?

wooden briar
#

High-res fix is now generating 2x figures for me when it used to not. What setting can I change to decrease that?

wooden briar
flat steeple
#

Is there some website that lists all the available VAEs / ranks them by quality?

#

/ what is the recommended prompt website for finding good quality prompts -- I'm trying to make very-realistic faces

young ivy
#

a good model shouldn't requiire masterpiece or bestquality

#

they normally don't do anything

chrome jacinth
#

if you take an image from google and use it as an image to image prompt and then you use the image that stable diffusion generated, have you stolen an image if you remove the image from google afterwards?

fervent thunder
#

I would like to know about copyright. I read your terms of use, but I don't understand much. Do I own the images I create, can I do what I want with them (send them to publishers, use them for commercial purposes, etc.)? Please answer me if there is a competent person who knows

potent spire
chrome jacinth
#

if I use it for a game that I want to sell

potent spire
#

as soon as you share it everyone can use it how they want

#

so technically you can use it for commercial purposes for example but the consequences are another thing

#

plus depending on result you may be sued but that one depends

potent spire
chrome jacinth
#

so stable diffusion images cannot be used to for professional use?

potent spire
#

because as said you dont have the copyright on it unless you have significant human input in it AND it isnt similar to the google image

chrome jacinth
#

oh

potent spire
#

otherwise you can be sued and everyone can take those images

#

because no copyright protection

chrome jacinth
#

but it would be fine if I made my own image prompt from scratch right?

potent spire
#

even then there is potential risk although smaller

chrome jacinth
#

even though it would be very different

potent spire
#

since you dont directly use img2img

#

and chancecs of getting a copyrighted output is smaller

#

chances

chrome jacinth
#

but the images are generated by me tho

potent spire
#

but you still wont have copyright if that matters to you, copyright to the image you generate...not your game 😄

chrome jacinth
#

so how can I be sued for something that I made?

potent spire
chrome jacinth
#

oh

potent spire
#

thats one of the big reasons companies avoid this, i mean for them ofc the potential risk is much higher

fervent thunder
chrome jacinth
#

but then what's the point of using AI image generators if you can't use the images for anything?

#

is it just for fun?

potent spire
#

not might, there would be

#

but thats just an example

potent spire
#

it really depends

#

Adobe goes the commercial safe route and Nvidia apparently too

#

but even those that dont, people are still using generative AI images for several purposes

#

even selling them

chrome jacinth
#

so I might be at a risk if I sell my game on steam cuz 90% of the graphics are made by AI

potent spire
#

many use gen AI just for fun tho

#

or as a replacement of doing art themselves

potent spire
chrome jacinth
#

really?

potent spire
#

gotta look it up again

#

so dont take it yet