#💬｜general-chat

1 messages · Page 77 of 1

covert sand
#

what did you mean let it fail to load

pale latch
#

you've got nvidiia so it'll work. i was just saying it would just not load if you tried in the case of getting an amd card.

if you're going from nvidia to nvidia, i'd just refresh drivers and see if it works

covert sand
vestal dew
#

2 monrha with SDXL, still no Controlnet. What gives?

teal jackal
#

Hello

#

Who knows when we can download the open source version of stable audio?

alpine tulip
#

thank you for your reply. I have similar thoughts

tranquil grotto
#

HI guys, what do you think of Stable Audio. As an audio producer it's really impressing me. I hope the audio quality improves soon, the tracks all sound distorted. It would be good to hear your thoughts.

small turret
#

hi guys, i have a problem with the controlnet in the img2img tab, i canno't preview the preview, what can i do?

vale steeple
#

wait

#

when there is Stable Audio?

#

I think I only heard of MusicGEN months ago

plucky shoal
#

it is there already now.

tranquil grotto
#

Stable Audio techies, we need the music output to be "flat" i.e. flat EQ, and check that output volume, bring it down a bit.

fervent thunder
#

I was ruining my images. Clipdrop downsamples when you use the eraser. The site upscale.media upscales with a higher quality than clipdrop for free

plucky shoal
#

you never know what is free. You can upscale on your owns in super quality and proper models. Using some this way oriented software

slate heron
#

hi .. using leonardo,.... I want to create two characters .. one with short hair and his girlfriend with long hair, now my prompt is this : "show two characters walking next to each other, max the adventurous boy with short dark brown hair, mia his girlfriend, orange red long hair" I am wasting really a lot of token generating two the same looking characters .. both have same hair lenght .. their faces are often the same and its ignoring my description .... could you tell me please what I do wrong ? ( I tried more descriptive prompts but nothing works ) ... help :))

plucky shoal
#

maybe 1boy with short hair and 1girl with long hair. But you need probably perform it many times. If you were using A1111 i would suggest you regional prompter, where you divide screen, and then tell what is in what part. @slate heron

slate heron
#

thanks @plucky shoal

plucky shoal
#

for nothing 🙂 hope it will work. I know it works for some those 1boy and 1girl, but bit of luck is needed

slate heron
#

ugh that worked at once ! ... GREAT !! thanks again @plucky shoal

plucky shoal
#

np happy it helps

graceful spade
#

Yo

#

need to know something, is it possible to create a video (movie /reel kind of using deforum sd? auto111?

suppose i have a story "turtle jumps from parachute" or "naruto escaping from jail" (in a cinematic style?)

#

@plucky shoal hi, i remember you've helped me a lot with sd, is it possible^

plucky shoal
#

Hi i am not sure, i only did one prompt to another prompt to another prompt. Like zoom in bit, or like fly througth.

graceful spade
#

any other option? if not deforum

plucky shoal
#

better to ask somebody with video exp. There is site that doing very short videos but with say wink eye, so i think it is not still possible.
Will search for that site, somehow pico or so...

limpid cobalt
cyan comet
#

whats the prompt for the ai to use literal text?
as in i want "ice cream" to be written out in full in the image

pale latch
#

stable diffusion really sucks at doing text. sdxl has some capabilities, but to prompt specifically for written text you probably thinking of deep floyd

cyan comet
#

isnt there a bracket mix that induces the gen to write out the text in full?

pale latch
#

in deep floyd yes

cyan comet
#

whats that?

pale latch
#

another model architecture seperate from stable diffusion

#

funded by stability though

cyan comet
#

ok

pale latch
torpid ether
#

hello everyone，I'm working full-time on the open-source project SolidUI, which can generate any graph with just one sentence. I welcome friends to participate and contribute together.

wild steppe
neat rapids
#

Hello. I consistently get BIAS words in negative embedding. This is BAD. Is there a way I can learn how to generate my own negative embedding?

mild timber
#

What is Center Stage?

vestal dew
#

Anyone know what Würstchen is and how to use it?

#

The huggingface page has this: "Otherwise, the model is available through the Diffusers Library, so you can use the interface you are already familiar with. For example, this is how to run inference using the AutoPipeline:"
what in the world does that mean?

tame swallow
#

Ok i need some help

#

Can you change the perspective of a character's face if you use img2img?

simple cairn
#

Hi!
How many times I can use the bot to create an arts?

#

Or it's unlimited?

ocean crypt
#

I have a very weird request... I am attempting to use image to image to turn a... strange... photo into a renaissance painting that maintains the composition and detail of the original image and just makes it renaissance style... so far been unable to achieve that... Any tips for checkpoints or Lora that could help me out here?

#

The image in question is a uh... banana with a face Cringe making it as a joke for my friend. And yes I mean an actual banana.. well what looks like a porceline one or something but

#

Tried uisng img2img with no luck,,, keeps turning the banan into a human

#

Instead of what im after im getting some nightmare fuel Wheeze

pseudo mica
jolly crystal
#

Is there another way to use stable diffusion on your cell phone without collab?

#

I'm trying in all colabs but I'm not succeeding, I enter but 5 minutes later it gives an error

subtle vine
#

hi guys, I'm using Local stable diffusion installation and I have a problem with generating images because I wanted to generate image in 2560x2560 resolution but the highest resolution I can go for is 1024x1024 for img2img, every value over that gives me an error:

OutOfMemoryError: CUDA out of memory. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF.

I have INNO3D GeForce RTX 3060 Twin X2 LHR 12GB

What local installation of Stable Diffusion should I use if i wanna generate 2560x2560 resolution images and why img2img version has maximum 2048x2048 resolution. And what is the best way to get highres images without upscaling it. Or maybe the is good upscaling option I don't know about, because I was trying to upscale my 1024x1024 images but it's not that good as highres generated images.

small birch
#

As I'm a beginner to SD I'm a little bit confused... I'v installed the A1111 and some different models and LoRAs from civitai. But even if I use the same parameters, seed and so one from the example pictures my portraits look like mutants... need some help here 😦

shy pumice
shy pumice
small birch
shy pumice
#

post in the general-with-images channel

small birch
proud plover
#

I've been messing around with Stable Diffusion for a few days and I can't seem to get my images to not look like crap. Body parts everywhere, art style looks like a kindergartner drew it even if I copy detailed prompts from a site

#

Any chance someone can chat me while I do a screen share and walk me through what I might be doing wrong? Certain settings etc

shadow junco
#

Upscaling black-and-white line art with R-ESRGAN 4x+. The upscaling is fairly flawless. I haven't really played around with other upscalers, but have noticed that there are more options now than there were six months ago. Are there any that you prefer instead of R-ESRGAN 4x+ for upscaling black-and-white line art?

shadow junco
fervent thunder
#

Where in easy diffusion can i select the strength for image to image, of how much it looks like the original pic?

proud plover
shadow junco
#

Are you using Automatic1111 or another platform?

proud plover
#

I downloaded the checkpoint v1-5-pruned-emaonly.ckpt that the guide recommended

#

A1111 yes

#

although I don't know how to verify that, I just followed the steps for an a1111 guide

#

I don't know how to tell, is A1111 the model?

shadow junco
#

A1111 is the graphic user interface for the general program stable diffusion. You can use standard models or you can use other models.

#

The checkpoints are models.

proud plover
#

I'm not sure what that model is then, I guess a generic one? v1-5-pruned-emaonly.ckpt

fervent thunder
#

is lama cleaner slow? I'm using accelerate and cuda. With SD1.5 model it takes so long to erase the moon from the sky. 2% 4% 6% ....

shadow junco
#

They're available on civitai.com, along with LORA's which are like plugin "helper" models that are smaller.

proud plover
shadow junco
#

Best overall? Tie between Dreamshaper 8 and Realistic Vision 5.

shadow junco
shy pumice
#

stuff you see on lexica could be from 1.2 or 1.4

#

so the same prompt on 1.5 wouldn't give you the same result

shadow junco
fervent thunder
#

is gpu faster than cpu to erase using AI?

shy pumice
#

yes

fervent thunder
#

took 7 minutes to erase the moon

proud plover
fervent thunder
#

Where in easy diffusion can i select the strength for image to image, of how much it looks like the original pic?

shadow junco
# proud plover Do you have a text file with common negative prompts? Or do the models have the ...

Typical negative prompt example:

(deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, crossed eyes, bad irises, bad pupils, bad eyebrows, umbrella, long body, too many arms, umbrellas:5, stretched body:5

to reduce the likelihood of text I also use this in the negative prompt:

Kanji, Hiragana, Katakana, alphabet, syllabet, Cuneiform, Heiroglyphs, Ivrit, Rashi, Chinese Characters, Hebrew Characters, Hieratic, Demotic

This is mostly useful for 1.5 based models. Less so for XL models.

shadow junco
proud plover
shadow junco
# proud plover And so do you keep text handy in an editor to copy and paste each time, or are t...

Definitely keep notes and save them. There are also styles within the interface under the big orange generate button. That's a whole thing you could explore because it's very useful for quickly adding positive and negative prompts. You can also set A1111 to add the prompts you've been using to your saved pngs. If you lose your prompt, just find a previous image, drag it to the "PNG Info" tab, click either "send to txt2img" or the like. It will populate all the field for you. Remember to click the "random" (dice icon) button so you generate new stuff rather than that seed.

fervent thunder
#

So nobody knows how to set the strength of how much the image to image should look like the original, in easy diffusion?

shadow junco
fervent thunder
#

Is automatic still that annoying to install. Or can i skip most steps because i already installed easy diffusion?

pale latch
pale latch
# fervent thunder So nobody knows how to set the strength of how much the image to image should lo...

i dont know how easy diffusion works, it's acronym, ED, is unfortunate though. What i do know is how img2img works. it's not a matter of making an image look closer to the original or less. It's how much noise you're adding to the image before processing it. on other UI's you see a denoise setting. the higher you tell it to denoise an image, the more noise it adds into it at the start. So higher denoise setting means less of the original image will be there.

#

there's a sweet spot around 0.4 - 0.6 denoising where images retain a lot of their original qualitys while still allowing diffusion to do some magic

fervent thunder
#

Using online stable diffusion models i could set in % how close it would look to the original

subtle vine
#

@pale latch so I need to generate in resolution for example 1024x1024 and then upscale my image with some extentions?

pale latch
#

2048x2048 is the near limit of my 16gb card

jolly crystal
pale latch
#

Run it on another remote machien/service and connect your phone to that interface

#

or dreambooth or civitai or a hundred other web service SD pages

jolly crystal
pale latch
#

starts with being willing to pay money

jolly crystal
#

Isn't there a free way?

pale latch
#

collab was the free service provided by google, but 1000s of horny teens abused it for porn and now the free lunch is over

#

100,000s more accurately

jolly crystal
#

I did hentai XD

pale latch
#

google probably would've kept it free if people were just using the service for learning like it's intended

#

but, people did not

#

i myself benefitted from collab a lot while i was leraning. i didn't use the UIs though. They weren't available on collab yet. I used scripts and just edited them to see what they did. Cool stuff. I ended up getting it working on my AMD card in linux and then buying hardware once i had figured things out enough

#

point is, once i had learned enough, i stopped using collab. that's what it's there for

#

all the people who just abused the shit out of the service instead of using it for learning purposes, are the reason google is banning accounts that use webui's now

#

it's quite a shame that all new users now can't enjoy that boost google was offering before

jolly crystal
#

I think I'm a little to blame for this too, I used stable diffusion to create hentai, not just hentai but I created

pale latch
#

https://youtu.be/MaSRXvM05x4 here's a cool service being offered by civit. hook their generation servers up to your own local installed A1111 webui. but youv'e gotta pay

#

"Pornography drives markets" is often heard a lot, but in this case it actually ruined all the free access people had before. Nobody wants to offer free generation services or even touch the idea with a 100 foot pole

jolly crystal
pale latch
#

The short window where this was a free hobby is over. You could try looking into something like "stable horde", a group of volunteers providing generation services to people for free

#

but it's a token based system and they're treated like crypto currency that the people who run the network centrally control

#

so basically, a paid saas like any other service

jolly crystal
#

Is there a way to use free stable diffusion on PC?

pale latch
#

own the gpu first, and then its free

#
  • electrical costs
potent spire
pale latch
#

thats right

#

its still free if you use a gpu you own, but that's a cost of entry situation

potent spire
#

oh, yes thats true

#

but thats where Midjourney comes into the game as well

#

and for another target groups, Adobe Firefly

fervent thunder
#

Weird with webui stable diffusion i get much better and faster results than with easy stable diffusion, even thru i use the same models

pale latch
fervent thunder
#

How i get control net on the webui?

pale latch
#

extensions my dude

potent spire
#

although im not a pro yet but still pay for 3ds Max and other non-free programs

#

but usually hobbyists are running for free services...at first

#

on the other hand, a lot of the people if not the absolute majority on Midjourney are hobbyists as well

#

part of them tries to earn money ofc with MJ

pale latch
#

i used to be a pro in the graphic design field. i left and started working in other fields. now this has happeend so i'm a hobbyist with pro experience now. i started on the free collab scripts, no ui. then i got automatic1111 working in linux with an amd 8gb card. that worked out so well and i learned so much that when it really boosted me when i bought a new 4080

potent spire
#

understand and im always interested in how people got to their workflow and what their workflow is in the first place 😄

#

i started with Crayion/DALL-E Mini back then XD

#

i mean gen AI

fervent thunder
#

is this a scam?

I’m interested in buying a few of your artworks for my exhibition, about 5-10 artworks. But I’m only interested in buying them as NFTs.
I’ll offer $2500 for each of these artworks I purchase from you as I’d love to purchase these artworks as digital artworks(NFTs).

2500$ for randomly generated images for free with SDXL sounds way too good to be true.

potent spire
#

scam

fervent thunder
#

out of the blue someone in deviant art wants to buy my AI generated artwork offering over 1000$ ? wtf?

#

How would i install a sdxl model? Do i just make a new folder and put it in?

#

Or do i need to do something to even have sd xl?

distant rampart
#

anyone here know if there are any large collection of Styles.cvs files freely available?

fallow pine
#

tell me a common fix for the anime hands in different angles

#

without using controlnet

jaunty pike
#

Does anyone here know how to speed up the iterations per second on a AMD GPU (RX 6800XT), using The Automatic1111 SD setup?

fervent thunder
#

What do people on reddit mean anything but 1.5 is censored?

#

I tried some XL models, and i can make anything with them?

pale latch
#

those same people have also started to shut up a lot more and only the dimmer ones are still going on about xl being censored

trim magnet
fervent thunder
#

You can do nude people tho

trim magnet
#

nudity is not porn

fervent thunder
#

I see

pale latch
#

sdxl is version 1.0 too. sd 1.5 is a few iterations of the base. People can use the same dataset that created sd 1.5 to refine sdxl

fervent thunder
#

So what is even special about XL?

pale latch
fervent thunder
#

Attention resolution?

pale latch
trim magnet
#

yes we can we just need to replace the NAI dataset that contained like 5million images with a dataset of the same size so we just need a bunch of H100 or A100's to finetune sdxl

#

easy money

pale latch
#

sd 1.5 is trained with 512x512 resolution, so it has that much attention space.

#

xl is 1024x1024

fervent thunder
#

But it won't make it easier for me to make pics bigger than 512?

pale latch
trim magnet
#

never said nai made 1.5 i said nai leak that had like 5million or 17million images is what made 1.5 anime models capable of doing the variety of anime styles it can do

pale latch
#

the pornography is censored loudmouths are so silly. y'all paid unstability.ai to make all these community porno models and they're squandering all the cash on their generation service they sell. Their obvious trademark infringement shows that they're unwilling to comply with etiquette and rules, yet they raise nealry half a million

fervent thunder
#

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.

#

What that means?

trim magnet
#

either add --no-half or change vae

pale latch
#

personally i think xl has better anime capabilities out of the gate than nai has

#

and novelai used a publical booru dataset too

fervent thunder
# 
RuntimeError: mat1 and mat2 shapes cannot be multiplied (616x2048 and 768x320)
trim magnet
#

yea the only difference is size of the dataset if we could finetune on a 3090 we would probably reach the same levels of quality but too expensive to finetune sdxl

pale latch
#

Model refining on a 3090 is expensive for 1.5 too. That's why its done on clusters generally.

Anyone can do this. I bet someone legitimately doing a project with real research that sought to bring superior anime capability could petition stability to donate cluster time. A lot of community good will would be available to them if they proved they were doing the work.

fervent thunder
#

When i use control net, what impact does denoising strength have?

pale latch
#

nobody is going to give money without seeing a dataset being worked on at this point. Not after Unstability swindled those good will patrons.

#

The opportunity is there for anyone to explore. I'd even expect they take some of the donations for themselves to cover the expenses of their time and energy. Would build a good brand for themselves too

#

Sucks about unstabiliyt, but they at least proved just how much money is out there willing to donate to this kind of cause

fervent thunder
#

Like if i put normal map in net control, and check pixel perfect, what use does that have if it won't really follow the normal map if i put denoising to 100%?

pale latch
#

pixelperfect is just for the preprocessor's benefit. if you're putting a normal map directly into controlnet, don't select any preprocessor

fervent thunder
#

How you use the recolor net control thing?

#

and do i have to select any model for it?

pale latch
#

these are advanced topics. helps to follow guides.

#

maybe don't jump into all the contorlnet models immediatly. learn the limits of prompting

fervent thunder
#

Where i find some guide or something that explains all the 14 control net models?

trim magnet
fervent thunder
#

So for example i use lineart, it guesses the lineart from a reference picture? But does it also make sense to select lineart if i want to turn already existing lineart into a picture?

trim magnet
#

lineart is only for coloring existing lineart into a picture

fervent thunder
#

But the article says "Line Art renders the outline of an image. It attempts to convert it to a simple drawing."

#

And when you use it, it does that?

trim magnet
fervent thunder
#

Like i just used it, and it also gave me the lineart pic with the generated pics

#

But what if i want to turn lineart into a picture instead?

trim magnet
#

u can use either canny or lineart try with both

fervent thunder
#

So they both spit out lineart guessed from a picture for me, but they also can turn lineart into a picture?

trim magnet
#

just put the lineart pic into controlnet and enable it and play with the weights to see which gives u better results

#

yea u can also use a 2nd picture with color to give it that color to the generated image

pale latch
#

any model can take a raw input that you've given it. preprocess just works the reference image into what the model knows

#

try the lineart model with coloring book pages

fervent thunder
#

I noticed that when the bot says that I need to vote more, if I try another channel, it usually works

proud plover
#

I want to take one picture of one person, and one picture of another person, and use img2img or something similar to put them in an image together. is this possible?

fervent thunder
#

I got this QR code control network thing, but somehow the reference thing i put in barley has any effect on the output image

covert sand
#

how can i stop sd from saving grids in the output folder?

fervent thunder
#

Settings

fallow pine
#

as stable diffusion programs can generate exactly same batch again and again so I think it would be great to have an option to set like a "batch range".. I mean like if i'm generating 20 images in a batch now later I can continue from 21 to upwards [like 21 to 40] instead of starting the batch from 0 to 20 same ones again. it would save us lots of time and power to gather new generations.

pure canopy
#

Hey sorry to disturb, I am a SD nerd , do u know a way to change emotions in a existing image? I want it for a yt video thumbnail and dont know enough photoshop

near pulsar
#

hello. I am trying to invite friend in this server, but the link is not working

tight pewter
#

does anyone the name of the extension that adds a vae dropdown menu next to the checkpoint dropdown menu at the top of a1111 webui?

fallow pine
near pulsar
tight pewter
tight pewter
fallow pine
fervent thunder
#

How do people get 20 it/s on the gtx 4070 ti? I only get around 1-5. What do i do wrong?

#

Weird earlier i got 10

#

Ok on base i get 14 it/s

#

Not sure if its worth it optimizing for the last 4-5, and then might get worse results

plucky shoal
#

maybe resolution, what UI are you using? ComfyUI or A1111?

pseudo mica
#

can anyone explain how can you train AI with other pictures (i got a folder ready and installed SD) but i do not know how to use it

fervent thunder
#

Oh automatic

#

I thought i got comfy

#

Does sdxl not work with image to image yet?

#

Same with netcontrol?

fallow pine
plucky shoal
#

@fervent thunder Have you --xformers. in your webui-user.bat?
Als speed depends on used sampler

fervent thunder
fervent thunder
plucky shoal
#

add --xformers in set COMMANDLINE_ARGS=
set COMMANDLINE_ARGS= --xformers

fervent thunder
#

What will that do?

#

Do i have to edit the shell webui user file too?

#

Also i just delete set comandline args and replace it with
"--xformers in set COMMANDLINE_ARGS=
set COMMANDLINE_ARGS= --xformers" ?

plucky shoal
#

speedup 🙂

#

no, just what i wrote in second line

fervent thunder
#

Or do i just ad xformers after the equal symbol?

plucky shoal
#

yes

fervent thunder
#

How does it archive that speedup?

#

Does it make the results worse?

plucky shoal
#

no

#

delete you can anytime, but almost all using it

fervent thunder
#

Do i have to start it thru the bat file then? Or is the bat file just the settings for it?

#

Oh its installing xformers

#

Thanks

plucky shoal
#

just restart if you are in. Run ...
Seems o.k. then, report what gain you have 🙂

fervent thunder
#

It jumped from 14 baseline to 20

#

hi

plucky shoal
#

🙂

fervent thunder
#

Not bad

#

can i use stable diffusion 1.0 on api ?

undone garden
#

Does stability ai have a public roadmap for the next few years?

hazy frigate
#

hey, on DreamStudio I get the message: Something isn't quite right with your prompts.

with this prompt:

A bright-eyed, innocent young girl with chestnut hair in two pigtails, holding tightly onto a tattered plushy bear as she navigates the impossibly large and cave. The mood is mysterious, but her blooming resolve shines stronger than her fear. Illuminated by the faint glow of bio-luminescent mushrooms, the scene presents a narrative of bravery.

negative prompt: modern urban background ,adult characters, darker tones ,gore, warfare elements

prompt strength: 1/auto

generation steps:25

model: SDLX v0.9 and 1.0.

Can someone more knowledgable tell me why ?

fervent thunder
#

Is it normal that once you turn hires fix on you get sometimes a entirly different pic, even with same seed?

#

What is the solution, have the images right from the start be done with hires, to get the end results already?

plucky shoal
#

yes it is normal. Solution is using resizers from extra tab.

knotty turtle
#

Is it possible to create SDXL textual inversion embeddings?

edgy ridge
#

Would it be okay if someone could extend an image I have to fit into a phone wallpaper

real crane
#

How do I upload a custom image and generative fill it in the desired directions?

fervent thunder
#

Can you already integrate animategif in a1111?

warm junco
proud plover
#

I want to take one picture of one person, and one picture of another person, and use img2img or something similar to put them in an image together. is this possible?

fervent thunder
#

RuntimeError: Expected weight to be a vector of size equal to the number of channels in input, but got weight of shape [1280] and input of shape [2, 2560, 8, 8]

#

Whats that? I put a image in animated diff, and i get the error, and after that i get the error every time now

fervent thunder
#

I get that bug where on second run of trying to make a gif, it does that

#

Im not sure what to do. Did disable xformers, and the models are 100% in the right folder

#

On animatediff

polar seal
#

Hello everyone, I need to make a cover for my Facebook, what measurements should I use in the image prompt?

warm junco
polar seal
sleek cliff
#

HellO!

worn wagon
#

Hi, I have a question. I downloaded one of the models and I want to use it with ComfyUI, and I have a pretty decent PC. My question is, is there a chance it will damage my PC/GPU? from overheating or something?

brisk vault
tender cove
tender cove
#

The AI Horde section in #1080946152318443610 links to ArtBot. AI Horde, Stable Horde, Kobold Horde, Haidra are all the same thing (yes, there's too many names).

brisk vault
#

how to close comfy UI, do i just close the browser and cmd?

forest trout
#

I think you just close the terminal.

crude willow
#

Hey, so I just learned about this software called glaze, which artists are using to "protect" their images from lora and ti style training. Its an adversarial network that makes the minimum amount of visual change to an image, while throwing ai networks off of recognizing what style that image is. the idea is that when prompting, styles typed in will not correspond to the image output.

#

does anyone familiar know if this works or not?

brisk vault
#

where do i put samplers i downloaded?

crude willow
#

there is a folder

brisk vault
#

is it the one called "extra_samplers"?

crude willow
#

honestly i dont know

brisk vault
#

C:\ComfyUI_windows_portable\ComfyUI\comfy\extra_samplers

crude willow
#

I find that there are enough options out of the box

dark iron
#

`did they lower the output res on the images?

#

used to be higher right?

#

672x384 lol

fervent thunder
fresh barn
#

Where could I ask for a prompt for a specific photo?

simple wharf
#

How do you upload a sketch and make it create a good image from it? I uploaded a pic but all it does is create a completely different brand new image

lime dock
#

Is the teaching atlas of problem-solving ideas and steps for generating comic style leetcode problems using LLM and Stablediffusion?

swift juniper
#

hello. just a quick question. in dreambooth, is steps per image more important than total steps? like if i did a 10 image dataset with 100 steps per image for 1000 total steps and im happy with the aesthetic. but i want more flexibility so I add 10 more images. at 100 steps per image thats now 2000 total steps. is that necessary for a similar level of quality or can i halve the steps per image and do 1000 total steps and get a good result?

lament cedar
#

I'm trying to upscale using img2img and I'm getting this error:

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

I already tried activating the "upcast cross attention layer to float32" option but I still get this error. I'm running an NVIDIA GeForce GTX 1660 Super. Anyone have a clue what might be causing this, and how to fix it?

warm junco
lament cedar
lament cedar
warm junco
lament cedar
#

ok, do I disable the "Upcast cross attention layer to float32" option in settings?

warm junco
lament cedar
#

ok I disabled it, saved settings, tried again, but I still got the same error 😦

deft sentinel
#

hlo what is black theme name in automatic 1111

warm junco
lament cedar
elder maple
#

hi everybody, i have a question I am hoping somebody would have some ideas for.

#

i am using text to image and am feeding a summarized version of text through to be the prompt used to generate the image. The images being generated im using 50 sample steps and they seem to produce a nice job however the text being printed in the images are very difficult to read, even gibberish for most cases. Is there a way to make the text being generated within the images more clear? are we not prompting it someway that we should be so that text can be read clearly?

#

here is my example. hope somebody can help?

jaunty beacon
elder maple
#

okay, well i wont be able to do that in my case as I am generating images on the fly and wont have time for manual intervention on the images.

#

sometimes it gets really close to producing excellent results and other times all the text is gibberish.

jade citrus
#

Does anyone use comfy ui, ? Need some help with installing the reactor node, it's frying Mt brain

fervent thunder
#

Does anyone have a good doc for training self portraits with Lora or another way?

jade citrus
#

Is am1111 easy to install? I love comfy ui, but I'm wanting to use face swap

torpid ether
#

hello everyone，I'm working full-time on the open-source project SolidUI, which can generate any graph with just one sentence. I welcome friends to participate and contribute together.

#

You can also click on my avatar; there's a join link inside.

#

w00t welcome！！！

foggy shell
eager copper
#

hey yall im new here

#

hello

distant flint
#

I'm struggling so hard to render a robot without a head, I tried "mech", "robot", "automaton", various "headless", and negative "head" and "eyes" in prompt...etc but it's rare I can ever get a robot without a head. You guys have any idea?

#

"mech" gives the most reliable results without a head but then it makes for a very bulky torso, which I haven't been able to prompt out

eager copper
#

try vice versa big head small torso gives the mech more stability and movment

distant flint
#

I've also played around a lot tried to get a "slim" "agile" "runner build"...etc

eager copper
#

like if the torso is small the rest of the body will be able to move more easily and a bigger head of ur chose will give the mech balance u know

#

unless u dont want a head

distant flint
#

Yeah my struggle is that I don't want a head

eager copper
#

then remove the head and keep the torso small big torsos make a mech to unstable the mech could actually crush its self if moved in the wrong derection

distant flint
#

What do you mean "remove the head" 😆 I've tried all manners of negative "top heavy"...etc aswell

eager copper
#

the mech is unstable bc its metal

#

ok

eager copper
plucky shoal
#

i found something in my past, but not sure it is what you want + prompts seems to be not for headless robots.

distant flint
#

I'm trying to generate a fantasy magically-animated robot made of wood actually, but yeah I understand the logic of what I want, I'm just struggling to hint SDXL to it

distant flint
plucky shoal
#

i cant here, image is in general with images
Do you want large or small?

#

Have you tried headless?

eager copper
#

i love talking and working on robotics

distant flint
plucky shoal
#

robot and pistol is only international words with our origin 🙂

eager copper
#

true

torn mist
#

I've played around with a few Discord and web image generators which say they use SD as their engine, and got fairly good results. So I installed SD itself on my Win11 desktop today.

So I had a look around the web for any articles to guide a total noob like me - but pasting prompts from those articles into my SD returned images that were nothing at all like the ones in those articles. I'm doing something wrong, or not understanding something.

Can anyone suggest anywhere I can look for patient guidance for a beginner on how to use SD properly?

plucky shoal
#

you should have same model, same seed, and dont forgot to have vae. Can you post your pictures? @torn mist ?

#

what model are you using? 1.5, 2+ or SDXL?

#

and what UI are you using? A1111 or Comfui? there are other, but those are most common

torn mist
#

Thanks @plucky shoal . Sorry to display the depths of my ignorance, but how do I find out what model I'm using? Or which UI (I did see AUTOMATIC1111 flash by somewhere)

plucky shoal
#

A1111 probably you are using. Model is in left top corner if A1111

#

@torn mist most probably default 1,5 version

torn mist
#

Its a very simple promot, asking fo a mouse (I thought) but no mouse in the image. This actually one of the better pictures tbh.

#

Is that UI A1111?

plucky shoal
#

you have not any model (checkpoint)

torn mist
#

OK... how do I get one?

plucky shoal
#

but it shouldnt work at all

torn mist
#

(I feel so dumb, sorry)

plucky shoal
#

check folder Model in A1111

#

models

#

and here stable diffusion

torn mist
#

I don't seem to have a A1111 folder. This is a shot of my SD models folder ...

plucky shoal
#

post your pictures in general with images please. Hope it is o.k.

#

@torn mist here what is in folder stable diffusion

torn mist
#

I posted an example screenshot a few lines back - will that do?

#

It's SFW 🙂

plucky shoal
#

yes image you posted, folders, what is in folder Stable diffusion there.
Here cant be posted images.

torn mist
#

Going to continue in "general with images" ...

plucky shoal
#

yes

covert sand
#

what is the purpose of saving grids? so you can look back and see how your prompts affected the images and such?

fervent thunder
#

whats the newest stable diffusion?

wild steppe
median eagle
#

Not sure if this belongs here or on another channel. #✍🏼｜rules-and-tos says nothing about self promotion being not allowed on this server so I hope this is fine. If not, I'll move this message elsewhere.

I created a storybook generator that uses ChatGPT and stable diffusion. I created a survey to evaluate the system. You'll look at the images generated for 10 different stories and vote yes/no for if the images have a consistent style or not.

Shouldn't take longer than 10 minutes, will probably take less time. Here's the survey - https://forms.gle/ddPz2hje29G24Jz68

sinful viper
#

Hello guys, does anyone know how I can give the instruction to write a specific text in an image creation? I have already spent many credits trying without results

median eagle
sinful viper
median eagle
sinful viper
severe ferry
#

Yoyo everyone, I'm new here.

#

Is this free to use??

#

Or a free trial?

tender cove
#

Free

dusky stream
#

What chat to share models?

wise aspen
#

Guys, whats is ur fav/best anime model for 1.5/SDXL?

warm junco
jovial wraith
fervent thunder
#

yall ever heard of tuneflow the melody AI generator?

sour needle
#

I have a question..

#

Could i get banned if i use google colab??

uneven arch
#

Hey Guys,

I'm wanting to train a model on a specific pixar-like character. I've generated 20 images of the sam(ish) character and I want to be able to prompt that character via dreambooth or something similar. Any tips for how I'd be able to do that?

unique cairn
#

what's the announcement

pale latch
#

its not an announcement. It's just pitching stable audio again

#

exact same annoucnement as last week, just fresh on the notifications . if that's how @karmic brook going to do the annoucnements page i'll mute em

#

it takes all my effort to fight notification spam

distant flint
#

Considering the amount of people using depthmaps with ControlNet Depth, is there any way to do the opposite, and generate a depth map from a render?

#

or to generate depthmaps in the first place?

copper crystal
#

any 3d renderer can output a depth map of the scene

distant flint
#

by "render" I mean generation, not a 3D render. I don't have the 3D information, it's what I'm looking to extract 🙂

broken gull
#

any easy installations for local use for StableAudio? I have only 6GB vram tho but maybe vram isn't that important since we are not talking about images

broken gull
distant flint
broken gull
distant flint
#

a1111 doesn't work for me for some reason, is there something similar for ComfyUI?

#

I was hoping for something based on image diffusion, I know there's a lot a tools out of there that can do a decent job (even Davinci Resolve has a node), but considering the amount of depth -> img generations I was hoping something trained a model for that

broken gull
#

there was this opensource program that also did it although idk if it was opensource it was free anyways where you also can generate displacements maps etc etc

distant flint
#

Materialize?

broken gull
#

oh ye could be

distant flint
#

Tool looks nice but is mostly for texture. I've used tools like that in the past but they don't really behave so well

broken gull
#

ye, but a depth map is basically a height map same data

distant flint
#

I know, I'm not looking to use that for flat textures though

broken gull
#

a height map creates depth

#

you can basically create a 3D mesh with them like those plane things that you use for 3D printers

distant flint
#

Right, depthmaps tend to not (always) be orthogonal projections though, and they're rarely linear, but I'm saying something like Materialize couldn't help me because it's not made to generate depthmaps from random images, it's made to generate heightmaps for diffuse textures, normals...etc

#

And as much as I like elegant algorithmic solutions, I don't think the tool is nearly as good as what image diffusion could do

sharp panther
#

Hello everyone, new SD noob here, I need some assistance regarding controlNet, I keep getting this same msg every time I try to generate a new image, "RuntimeError: mat1 and mat2 shapes cannot be multiplied". If anyone got any spare time to explain how to resolve this issue

cerulean star
#

Has anyone tinkered with extensions through stable diffusion api?

#

I'm trying to use root extension through api with "alwayson_scripts" field but there is no documentation around it

scarlet falcon
#

anyone know why my automatic1111 reloads the model every time I change my prompt?

#

extremely frustrating

plucky shoal
#

@sharp panther it is because you are using things for some version for different version.

vestal dew
#

SD is so dead right now

thin abyss
#

I keep getting the not able to finish the dream error

vestal dew
#

as a doorknell

vapid valley
#

Hello! Anyone willing to help an SD newbie out?

Have only been dabbling with midjourney and now venturing into SD to make my generations "pop" even more. I'm trying to upscale and sharpen my image via "img2img" with the "Ultimate SD upscale" script.

I've tried different settings but however I go about it I just end up with an image that is less sharp than the original

cobalt cape
#

im trying to use controlnet with a 3080ti but it takes 20 minutes per render, am I doing something wrong or is that how long it takes for everyone? using A1111

warm junco
#

What are your settings?

cobalt cape
#

1024 x 1024

#

thanks forthe response

potent spire
lunar bramble
#

hello
I am new at stable diffusion I was using stable diffusion 3 month ago - I had only 4 gb of vram back them but I have upgraded my pc

I have purchased rtx 4090 to learn all the concepts of stable diffusion

I have installed auto 1111

and I have also installed sdxl model

I don't know how to setup refiner and I have also added mikubill controlnet extension and downloaded depth - canny - openpose models for controlnet

how to activate it?

fallow veldt
#

then it is loaded as a regular model, and you can generate. But it is best use (I guess) if you do it in the generation process, right bellow the Sampling steps.

#

so it switches between the base model and the refiner, you can the % of steps each one is going to generate

lunar bramble
#

now the problem is how can I have installed extension of controlnet but its not showing

#

😦

grave kayak
#

Hey, I'm using Automatic1111 now, other webuis I've used before had previews of styles (like "photorealism" or some artist) and when clicking the preview the style would automatically be added to my prompt. Does something like this also exist for automatic111?

exotic valve
#

Hi everyone,

I'm looking for someone who can help me with AI art. I've tried to get into it myself, but I work a full-time job and film and edit videos for YouTube and I am a single dad, so I just don't have the time to dedicate to learning everything I need to know. I also have some hardware limitations.

I'm looking for someone who can generate realistic art and who is familiar with ControllNet. If you're interested in helping me out, please send me a message.

In return, I can offer to promote your social links on my YouTube channel and in my Discord server. I have a fairly active community, so this would be a great opportunity for you to get your work seen by more people.

Thanks!

vestal dew
fervent thunder
#

do you know if SDXL is trained with names of plants, species and such? I tried to generate "empress tree", which is a name for a tree with purple flowers. It instead generated a woman with empress clothes in all green fashion

potent spire
#

I mean people would use SD without controlnet as well.

peak torrent
#

How do i generate pictures?

hollow cove
#

Where can I find the best tutorial? For how many images I need to create a character in SDXL? What's an ideal number of images to use for training?

bright breach
#

Hi, is stable diffusion banned on google colab? It keeps disconnecting me while using it, is there like a way to stop it from doing that?

spice plank
#

I click the image preview while its being drawn but it switches to the previous image.

#

Is there a way to get it to preview the current image even though its not complete?

cobalt cliff
#

it does impeccable text too

potent spire
#

it looks promising, but lets see 😄

spice plank
pale latch
#

the old internet of yore coming back to haunt us

#

it was supposed to be a paypal competitor and they got acquired. it's how elon got bootstrapped. Now he's rebranded twitter as x and resurrected it

spice plank
#

I convinced that some new update around 1.6 made webui a large amount slower... It went from 5min per gen to around 8

pale latch
#

likely a localized issue or you're skewing your own memory

#

5min per gen tells me theres issues right out of the gate. that sounds like CPU mode

potent spire
#

okay it really looks amazing

spice plank
#

Using a 1660ti and outputting 512x768 with 2x hires fix should not take 8 minutes

potent spire
#

but lets see it in practice

spice plank
#

Is dpm 2s a karras slower than other ones?

rancid steeple
#

2x than euler a

pale latch
# potent spire okay it really looks amazing

if that's real world usaged. Remember when videogames surged from a small to ALL ENCOMPASSING market? we got a decade worth of cinematics for commercials instead of actual gameplay

spice plank
#

I moreso meant slower than the other dpm++ samplers

pale latch
#

openAI are likely staging these conversations. Training specifically for them. Hand crafting larry before feeding it into the training data. Trying 100 different reconstructions before they took the clips that looked like a natural conversation. Who knows how they actually did it. It's such a highly edited video that it's certain it's not real UI use

potent spire
#

but it can eventually accompany one of those

#

since people already pay for ChatGPT Plus

pale latch
#

openAI are partnered with Microsoft. This is just a marketing blitz. They aren't worried about MJ or SD

potent spire
#

or use it otherwise

pale latch
#

All windows 11 have chatgpt licensed now. They got their funding secured

potent spire
#

well i guess they dont sh*t in their pants because of the competition in AI art area

pale latch
#

chatgpt api access is more about teaching the market these tools can be used. It's a halo product.

potent spire
pale latch
#

it's their biggest success

potent spire
#

actually i forgot that Adobe and Google partnered as well and Bard will have the same thing as OpenAI is doing with DALL E 3 now but in their case it will be with Firefly

#

and Bard is free...for now

pale latch
#

bard isn't in canada since google hates canada's new laws

potent spire
#

still not available there?

#

in EU we had to wait as well

pale latch
#

yeah they're pissed about our new media laws so they're holding back services

potent spire
#

interesting

potent spire
#

broooo the price

#

97$/month

glossy compass
#

hey, is there a good website that showcases some great stable diffusion made images with their models and prompts?

copper crystal
#

$100 a month makes it a huge scam. It doesn't even implement inpainting very well. I can tell there's no controlnet models at all.

#

and that camera angle control? 100% fake controls. Your service probably signs up for auto renewel and is hard to cancel too

potent spire
#

@copper crystali mean even if not a scam, the price tag is bigger than Adobe CC all apps plan

#

this one is supposed to be alternative to that

#

maybe it draws credits from GPT-3.5 or 4 or Dreamstudio API

vestal dew
#

SD went backwards

floral umbra
#

I am so unsure if i should just say F it and get a budget nvme, seeing as even comfyui doesn't load models faster than 1.3GB's on a gen 3 that can read at 3GB's anyways, or get a "more durable gen 4".

potent spire
vestal dew
#

I don't know. When SDXL was released, there was no controlnet for it. Now there still isn't.

floral umbra
#

NEver mind. Didn't read the damn SDXL part eugh

trim magnet
#

there is a good reason why theres no controlnet for sdxl but luckily we have controlnet loras, thats all we will get for now

potent spire
#

although im not sure since i used SDXL for 2 weeks before i deinstalled

pale latch
#

diffusers team released a bunch

trim magnet
#

just the same canny depth

pale latch
#

how you mean the same? you said none a second ago

trim magnet
#

more of the same as controlloras depth,canny,inpaiting

pale latch
#

but that isn't none?

#

"luckily we have controlnet loras, thats all we will get for now" << why say wrong things then argue that they're right things?

trim magnet
#

oh yea u right luckily we have an english bot here to correct when something goes wrong

pale latch
#

hmm

#

people sure do love to spread misinformation and lies. i wonder what that's all about

trim magnet
#

not everyone can be as smart and informed on current topics as you,as humans we sometimes miss updates on current themes

pale latch
#

but, you knew about the diffusers controlnets. you told me what they were..... holy moly what even is this

trim magnet
#

sounds like a redditor thread would you like to continue i can switch to a more agressive LLM if you want

pale latch
#

before i came in everyone was talking about how controlnet is non existent. before i could even finish mentioning which ones diffusers had released, you told me that doesn't count

#

there was a team effort saying that controlnets don't exist here.

long thunder
#

hi! question for any software developers, m2 macbook air 15 or m2 macbook pro 14... and is 16gb ram enough to not feel strained

pale latch
#

This isn't the first time i've come in and said "y'all are wrong and this is why" and people get mad at me and act like victims because english isn't the primary language.

trim magnet
#

how could someone get mad at you

#

where we would be without you

pale latch
#

passive aggression is aggression and always rooted in anger

trim magnet
#

i wouldnt be the person i am without you

pale latch
#

being wrong is a good thing. it's your opportunity to be right next time. Don't be so sulky when you find out how wrong you are. instead celebrate

trim magnet
#

of course we celebrate,we have you

valid beacon
#

so this means we're still stuck on the same controlora level of releases cany depth and inpaiting?

trim magnet
#

aye ma'am i see a recent openpose release too u tried that?

valid beacon
#

i get errors with openpose sdxl ill try to ask in support chat 😊

vestal dew
#

Without it, everything must be made into a LORA

vestal dew
pale latch
#

literally just covered that there is controlnet models for sdxl

#

did some youtuber declare that it will never happen or something? People keep coming up with this idea that it's non existent for sdxl. i see it over and over

valid beacon
pale latch
#

all the ones marked sdxl

vestal dew
slender ledge
#

Hello, i'm installing stable diffusion on local and i have some questions, which channel should i ask?

copper crystal
vestal dew
#

it doesn't work in sdnext

#

I guess Ill install a1111 again, try it there

copper crystal
#

update sdnext. it just uses the contorlnet extension for it's purpose

plucky shoal
#

@frail prism hi, thank you for your stickers workflow! I have one question, as iam not good at workflow. Whats are parameters to get out of aliased outlines? I played with it a bit but cant get good result. Played with feather and so.

frail prism
plucky shoal
#

borders bud jagged? if proper word in english
mmnt

frail prism
plucky shoal
#

i want not jagged, I mean antialiased i want 🙂

#

@frail prism i think it is because rmbg cropping it not smoothly

frail prism
#

Oh, well, increase sigma parameter in mentioned "Mask Smooth Region" node. The mask will become smoother.

#

@plucky shoal Or you can draw mask manually in Mask Editor and use lower nodes line (with blue node at the end).

plucky shoal
#

thank you! @frail prism will try!

frail prism
#

@plucky shoal I can't paste image here, but note that there are two "Mask Smooth Region" nodes. One used for automatic masks (higher), and one for manual masks (lower).
And to use manual mask, connect blue node "Mask Erode Region" to blue node "Convert Mask to Image".

plucky shoal
#

o.k. i played only with bottom i think

thorn citrus
warm junco
fervent thunder
#

guys what are the dimensions for sd1.5 models

warm junco
candid kindle
#

dear fellas, u guys know a free method to geneare music video?

fervent thunder
covert sand
#

can someone explain the spaghetti thing to me?

warm junco
fervent thunder
teal wren
#

I had to apply --no-half to my startup to get SD to do sketch inpaint. It's inredibly slow. Is it generally that much slower than the other operations or is it related to the no-half thing?

cobalt crag
#

I wonder why when I generate image about hald body portrait...

#

it look beautiful

#

but when I try to generate full body

#

It's suck, ugly as hell

teal wren
#

The details are a lower resolution. This affects two things: one is that the lower resolution already looks bad. The other is that the images are generated in little chunks that are lower in resolution than the final pixel resolution and they guess what the pixels in between are supposed to look like. This gets very hairy at low resolutions.

bleak matrix
#

Good afternoon, everyone!

torpid ether
#

hello everyone，I'm working full-time on the open-source project SolidUI, which can generate any graph with just one sentence. I welcome friends to participate and contribute together.

#

You can also click on my avatar; there's a join link inside.

#

welcome！！

ivory nexus
#

Does someone know how to make good anime-style logo using stable difussion?

#

Hello!

blazing rose
pale latch
#

with real esrgan... 🙄

#

would be better off with a nearest neighbor algorithm for video

sly quail
#

Anyone having issues with using Stableaudio? It's not letting me login, the "try it out" button does nothing. :/

plucky shoal
#

i think this is common as service is very used. Try it and it let you login

fervent thunder
#

are there any prompts to generate high quality vectors?

spice plank
#

There a furry model that does good fur? Most furry models I find lean more towards humanoid instead of fur

#

I like the look of soft plush fur

#

Fur you can run your hands through

#

Like they use conditioner

#

Like very fluffy

#

Nice fur

#

Fur

#

I love you

#

Be my wives husband please

spice plank
#

Ive tried yiffymix and that doesnt do good fur, it is just humanoid with a "x" animal head

#

Kemonofurry model does better a bit, it does nice tails

forest trout
#

The model I prefer is EasyFluff but it's a vpred model. It would require cfg rescale.

obsidian geode
#

hello guys, do I need sdxl vae when dreambooth training?

#

for sdxl

#

as an option to kohya_ss, I believe I dont need but I ask here just to be sure

spice plank
forest trout
spice plank
#

Interestinv

floral umbra
#

Made a bat file for running comfy in one click from desktop for those who wants it thinky Just paste in notepad, save as "comfyui.bat" for example and run.
replace "user" with what your pc's user name is.
Replace "G:\Stablediff\Comfyuimanual\ComfyUI" with the path to your comfyui's root folder

set conda_path=%windir%\System32\cmd.exe "/K" C:\Users\user\AppData\Local\miniconda3\Scripts\activate.bat C:\Users\user\AppData\Local\miniconda3

rem Change directory to "G:\Stablediff\Comfyuimanual\ComfyUI"
cd /d G:\Stablediff\Comfyuimanual\ComfyUI

rem Run "main.py"
python main.py
valid river
#

do you think controlnet will ever release scribble for SDXL?

finite cloak
nimble shard
#

hi anybody has ever trained stable diffusion for video generation?

tall gorge
#

Any idea when we will have any big improvements in prompt following in open source models? Dalle-3 is insanely good at it

golden plank
#

Hello everyone. I'm trying to create a set of 2D monsters for an indie RPG video game project. Non-animated, 2D full frontal view of monsters (oldschool 🙂 ). I just started using SD and noticed that what is probably currently preventing me from achieving that with any kind of consistency is the model i've tried using (sd 1.5 and RV). Do you have any model in mind that is better at this task ? Any tips or things I should look into 🙂 ? Thank you in advance !

flint harbor
#

⌨️👁️

rapid adder
#

Hi all im using SDXL offline, i see there are bots here on discord are these bots generating images and are we aloud to use them?

warm junco
tall coral
fervent thunder
#

aku Shy sayang 👹 kamu skullreaction juga X_nerd sayang cutecat

floral umbra
#

there's workloads that is sorta tedious that can do longer, but needs batches of X amount of frames, and have to manyally be post processed per batch

valid river
#

same with any preprocessor

floral umbra
#

Hmm, can't decide whether to try cheaper storage being Crucial P3 for 70 euro, or just say F it and get 970 evo plus on sale. for 94 euros. As it's only purpose will be to load models, so dram is not needed at all.

onyx mason
fervent thunder
#

hi

nimble shard
lofty shell
#

hi friends.

#

somebody knew good prompts for octoberfest?

bright breach
lofty shell
soft sparrow
#

is ther some way to tell if an AI image generator is diffusion based on the output?

onyx mason
bright breach
indigo ermine
#

How do I tell Discord not to notify me about stuff in the #1072240143521554592 ? I already set the notifications on it to "Nothing" but it still shows me a badge whenever they tag @challengers in there...

flint harbor
#

⌨️👁️

blissful bobcat
#

i need help...

#

venv "C:\Users\new user\Desktop\A1111\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.6.0
Commit hash: 5ef669de080814067961f28357256e8fe27544f4
Launching Web UI with arguments: --xformers --autolaunch --medvram
Loading weights [5aacd1c8c1] from C:\Users\new user\Desktop\A1111\stable-diffusion-webui\models\Stable-diffusion\ParchartXL-1.2.1.safetensors
Running on local URL: http://127.0.0.1:7860
Creating model from config: C:\Users\new user\Desktop\A1111\stable-diffusion-webui\configs\v1-inference.yaml

To create a public link, set share=True in launch().
Startup time: 33.2s (prepare environment: 10.4s, import torch: 7.5s, import gradio: 3.5s, setup paths: 3.6s, initialize shared: 0.5s, other imports: 3.3s, setup codeformer: 0.2s, load scripts: 1.7s, create ui: 1.4s, gradio launch: 1.3s).
Applying attention optimization: xformers... done.
Model loaded in 16.6s (load weights from disk: 2.4s, create model: 2.5s, apply weights to model: 0.1s, apply half(): 3.8s, calculate empty prompt: 7.8s).
0%| | 0/16 [00:06<?, ?it/s]
*** Error completing request
*** Arguments: ('task(iq9fv8j43ns4ivq)', 0, '', '', [], <PIL.Image.Image image mode=RGBA size=820x616 at 0x2649ED15390>, None, None, None, None, None, None, 20, 'DPM++ 2M Karras', 4, 0, 1, 1, 1, 7, 1.5, 0.75, 0, 512, 512, 1, 0, 0, 32, 0, '', '', '', [], False, [], '', <gradio.routes.Request object at 0x000002649E8522F0>, 0, False, '', 0.8, -1, False, -1, 0, 0, 0, '* CFG Scale should be 2 or lower.', True, True, '', '', True, 50, True, 1, 0, False, 4, 0.5, 'Linear', 'None', '<p style="margin-bottom:0.75em">Recommended settings: Sampling Steps: 80-100, Sampler: Euler a, Denoising strength: 0.8</p>', 128, 8, ['left', 'right', 'up', 'down'], 1, 0.05, 128, 4, 0, ['left', 'right', 'up', 'down'], False, False, 'positive', 'comma', 0, F

#

i can't get stable diffsion to do anything. i have no idea what the problem it

warm junco
fervent thunder
#

how do I avoid generating images with the camera too close to the ground and out of focus background?

finite cloak
feral quarry
#

hey guys, is this server bot better than the stable difussion app itself?7

neat cypress
rapid adder
forest trout
#

Very little, but bot 1 has controlnet available.

fringe latch
#

Hey I'm Dhruv and I would try my best to nurture the friendship with any of you.
My energies are always there for somebody's help, usually in python.

robust spade
#

Hey chaps - wanna make basic 90s style rave vj loop style images. Like early Bryce 3d. Are there any models I could use!? Key prompts u could suggest?

left vector
#

hey im new to stable diffusion but ive been using MJ for a while. are there any outstanding fundamental differences in image generation between the two?

#

Like MJ is aesthetically pleasing and gets so many things right, all without having any understanding of what its creating or how anything is supposed to function. Does Stable Diffusion have the same issues?

warm junco
#

But the SDXL model for example is pretty good with short prompts

strange shell
#

guys can anyone tell me prompt for Guts from berserk lying in thunderstorm image expressing extreme sadness and solitude in manga styled artwork

potent spire
#

But still you cant train your own models etc. Not yet at least

#

And now DALL-E 3 seems to become a serious threat for MJ, for SD not soo much because many in the SD community prefer some of the SD advantages

worthy lichen
#

Hi guys, I updated automatic 1111 on my mac, and now it won't recognise my .safetensors models. can anyone please point me in the right direction to solve this. I've tried downgrading python

warm junco
spark trench
#

Anyone installed the new nvidia driver?

#

Does it still negatively impact the performance of SD?

#

I had to roll back to the one in May

fervent thunder
#

hi

vast tree
#

Hi guys, I have been using some LORA models for certain specific actions coupled with LORA models for celeberty faces. However, I am kinda getting the worst of both worlds, the ressemblance is a lot less and the pose always turned out to be kinda just standing still and looking at camera. I have tried messing around with different weight and the result is similar. Is there any way to make the 2 LORA work better together?

feral eagle
#

So I’m getting into machine learning in my new job at work which led me to finding stable diffusion. Setup on my amd 6700 XT gaming rig last night but as with gaming, I see the issue of not having tensor cores , I’m a hardcore amd junkie but if I am getting into ML and SD , I assume it’s time to ditch the 6700 XT ? If I built a cheap rig just for ml sd what’s the cheapest nvidia card to go with? And will this work with dual gpu’s?

worthy lichen
worthy lichen
main lake
#

Tell me prompt to generate strawberry elephant anoano wilkum whatsapp status

warm junco
orchid mason
#

discarding roop, with what other tool can I do face swap on A1111?, to be able to do something like remini

olive dagger
#

what's the best-best gpu for stable diffusion?

olive dagger
olive dagger
cursive hornet
#

Im completely new to using ai image generation. I have low resolution images that i want to upscale. Ive heard about img2img, controlnet, and real-esrgan but i dont really understand what they are and they relevance they play into upscaling. Also, does the input image resolution affect the quality of the final product (i have the same image as 640x480 and also as 960x720). In my head, it would make sense that the latter would be better to upscale with but idk if these programs work better with smaller images or not

lofty shell
#

What is your most used Checkpoint? Realistic Vision, epiCRealism, epiCPhotoGasm, epiCDream for best creations?

forest trout
#

Dreamshaper is still pretty good.

tender cove
#

🤔 I'm gonna use random messages here as prompts

spice plank
#

How do people go about making those loras for the world morphs?

woeful tartan
#

Welp. Solid pdf prompt book....for the rookies 😉

thin beacon
#

I need to know how I can engage two different characters in one image doing romance?

tribal pond
#

How to make stable diffusion work again after copy pasting the folder on another pc?

thin beacon
#

Is there any extension i can use in automatic1111 where I can engage two characters in one image with two different seeds and prompts?

worthy lichen
worthy lichen
warm junco
worthy lichen
plucky shoal
#

can you make screen with directory where models are? @worthy lichen curious, what can be problem. Only i can imagine suffix is somehow wrong...

worthy lichen
#

Yeah sure. I was trying to put in a screen shot but don't think I can put images here.

#

Directory is: Automatic1111/stable-diffusion-webui/models/Stable-diffusion

plucky shoal
#

put it in that chatt bellow

worthy lichen
#

The weird thing is that it was working, and then after update just didn't recognise them.

plucky shoal
#

seems o.k. i just have / other way round

#

probably some extension, have you not some sort of civitai loras or similar?

worthy lichen
#

yeha thats a mac thing

plucky shoal
#

pictures in general with images

worthy lichen
plucky shoal
#

can imagine

#

and all extensions up to date?

quiet kernel
#

Hello guys, I've searched the conversation but I have not found a solution to the issue I encountered after installing SDXL, It works to some degree but my controlnet seems broken 9/10 times.

this is what I have in the log:

*** Error running postprocess: D:\AIstuff\Image\Automatic1111\New folder\stable-diffusion-webui\extensions\sd-webui-refiner\scripts\refiner.py███████████████| 20/20 [00:12<00:00, 1.48it/s]
Traceback (most recent call last):
File "D:\AIstuff\Image\Automatic1111\New folder\stable-diffusion-webui\modules\scripts.py", line 651, in postprocess
script.postprocess(p, processed, *script_args)
TypeError: Refiner.postprocess() missing 3 required positional arguments: 'enable', 'checkpoint', and 'steps'

worthy lichen
# plucky shoal and all extensions up to date?

yeah. Reinstall a couple times. Thanks so much for your help. I rented an online Auto install. Graviti. Seems to work well. And using invoke locally. bought this mac before the AI boom, and now really wish i had a PC

plucky shoal
#

if related there seems to be something to try

worthy lichen
plucky shoal
#

👍

quiet kernel
plucky shoal
#

would try to disable refiner @quiet kernel but i got realy bad ideas

quiet kernel
#

thanks I'll try righ now

plucky shoal
#

@worthy lichen it seems updating python to 3.11 which on windows wont work is one of solution

lavish current
#

Chat bots are literally better than Discord

quiet kernel
#

same error after unchecking refiner

plucky shoal
#

on windows only python 3.10.6 -> 3.10.11 or so

quiet kernel
plucky shoal
#

You can, shouldnt be hard. Or try delete only VENV folder and run webui user.bat again

viral inlet
#

I have 2 GPU and It's use my GPU 1 not the GPU 0

#

How to select it ?

quiet kernel
warm junco
lavish current
#

Chat bots

#

Are literally

#

Better

#

Than

#

Discord

worthy lichen
plucky shoal
#

and other things do not helps as well?

worthy lichen
#

Doesn't seem like it. deleted the venv. So weird. Is there an extension for importing models using the gui. maybe i google that

reef tartan
#

been looking through all the resource stuff for hours, and i guess im just dumb or blind, but i cannot find any documentation or instructions on how to download and install stable diffusion on my own PC, only source code from the github-alike and how to run it from a cloud service.

#

can someone please point me to a link or a set of instructions on how to just download and install this onto my own PC?

#

really wanna try this

worthy lichen
#

there are quite a few guides. It is a bit manual.

#

its paid tho

plucky shoal
#

@worthy lichen i think it is easy to follow. It is probably best how to install it.

worthy lichen
#

Thanks, I agree. PS found a backup of old Auto install. and it works. Damn. wasn't as stupid as i thought.

grim aspen
#

will stable audio be installable on your own pc like SD?

worthy lichen
#

Now I broke Invoke AI. yay

shell quail
#

hi guys

#

i need help finding R-ESRGAN-4x+ upscaler, where can i get it? i searched everywhere

reef tartan
frosty pasture
#

Does comfyui give better results than a1111 whaddayathink

alpine egret
#

I use stable diffusion A1111 …. Are there any other free app that are as good as SD? I think Dalle and clipdrop are both paid, right?

gleaming wharf
frosty pasture
#

I was told confyui takes the prompting differently

#

That images come out "better"

reef tartan
quiet kernel
#

this is where I'm at now:
ule.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "D:\AIstuff\Image\Automatic1111\New folder\stable-diffusion-webui\modules\sd_hijack_optimizations.py", line 486, in xformers_attention_forward
k_in = self.to_k(context_k)
File "D:\AIstuff\Image\Automatic1111\New folder\stable-diffusion-webui\venv\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "D:\AIstuff\Image\Automatic1111\New folder\stable-diffusion-webui\extensions-builtin\Lora\networks.py", line 429, in network_Linear_forward
return originals.Linear_forward(self, input)
File "D:\AIstuff\Image\Automatic1111\New folder\stable-diffusion-webui\venv\lib\site-packages\torch\nn\modules\linear.py", line 114, in forward
return F.linear(input, self.weight, self.bias)
RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x1280 and 768x320)

#

can't copy the whole log

frosty pasture
#

Can someone eli5 about the hi-res fix? I find text about it everywhere but not what files i should download or how to get it to show up as the upscaler version some people have in their A1111's UI....

hoary furnace
#

Hey, friends,

I've been playing around with training Lora models for a long time, so I decided to get some experience training Dreambooth models. As a test, I created a model from about 500 images, set to 42 steps and 10 epochs, with a learning rate of 0.000005, with the intention of teaching it more images and styles later... Unfortunately, I found that in Kohya_ss for Dreambooth training, the option to build on previous training which is present in Lora training is completely missing.

So my question is, how can I train the model further? Should I select it as a source model and train new images on it? (if so, should I include the original images or just the new images?) or do I just have to train a completely new model, leave sdxl_base as the source model and then merge these models together or is there another way?

Thank you in advance for the answer!

split herald
#

is there a way to get a heatmap of what parts of the image a lora is affecting the most?

lavish current
#

What are the best free websites for cloud models

#

Prodia for quick gens and Patience when I want to use SDXL or more detailed art

#

I should add I (still) can't use a local model because I'm on Chromebook unless there is some easy workaround for this

wary belfry
frank breach
# worthy lichen Hi guys, I updated automatic 1111 on my mac, and now it won't recognise my .saf...

There are a lot of potential causes for that, but to make things easier on yourself you may just want to use one of the installers from https://github.com/brkirch/stable-diffusion-webui/releases to create a new install. If you are using an Apple Silicon Mac then you can try out the installer for the latest experimental version (you can download the application directly too if you'd like but currently I'd recommend the installer as it directs you to install your SD models first - the application currently downloads the SDXL model without prompting on launch if no models exist in ~/Documents/Stable Diffusion Web UI/models/Stable-diffusion). If you're using an Intel Mac then try the v1.6.0 release (https://github.com/brkirch/stable-diffusion-webui/releases/tag/v1.6.0-mac-builds) for x64. Either way if you have any problems loading the safetensors models with the install generated by the installer then please create an issue and I'll take a look.

wary belfry
feral eagle
#

So I’m going to spin up an aws g4 instance to see if a T4 will work better than my amd 6700 XT . I put Ubuntu on it. So what’s easiest way to get SD up and running on it?

#

Else I’m considering replacing the 6700xt with a 3060 12gb . Getting as working on amd has been a pain

knotty turtle
#

Do you guys know how you call a LoRa model in Fooocus? Is it the same as in A1111 with the trigger word and the model name in <______:1.0>?

worthy lichen
hoary furnace
neat cypress
warm junco
bright cradle
#

Python not found fix?

warm junco
# bright cradle Python not found fix?

Install python with "add python to path" checked.
If you already installed python then you can run the installer again. Click modify and then select "add python to system variables"
After that delete the venv folder and rerun the webui-user.bat

neat elm
#

i have a question, if i want to generate an image that contains two trained subject, how would i do this? Lets say i have a model for each trained subject, can i merge them both? How would i merge these 2 models?

warm junco
stark pagoda
#

Hey
I have a question:
I wanted to try the new Toonyou Gif Animation
But everytime i try to generate i get this error

RuntimeError: CUDA error: invalid configuration argument Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

Can anyone help me with that ?

surreal pulsar
#

hello

#

has anyone successfully installed stable diffusion on macbook pro?
i am getting some errors please help

frosty pasture
#

Is it better to make a lora or a lyco 😮

warm junco
stark pagoda
brisk mauve
#

do i need to use prompts if i have a model trained on my own paintings, and i want to give my 3d renders a paint filter with image2image? ive seen some tutorials and i see people use prompts both when generating with image2image and when they train the model by describing every image

#

so if the 3d renders are very similar to the result i want, do i need the descriptive prompts of elements in the image except the style

primal quartz
#

Hey Gs, Which files should I download along with A1111?

#

@warm junco

#

what is the recommended to get higher resolution and video to video, can you help me plz

warm junco
warm junco
frosty pasture
#

Is it very hard to make a good lora/lyco...? 😮 I wanna make 1 of a celebrity that isn't covered but I'm anxious, it looks tough xD

jade path
#

Please I can’t find the site link please send me

covert sand
#

anyone have any idea why im getting like bright green accents in my img2img when i didn't prompt that color and that color is not in the original image. every model is doing it

plucky shoal
#

can you post example? In channel below?

covert sand
#

the green doesn't show up in the preview. the previews look nice and then when they're fully done the green appears

plucky shoal
#

o.k. seems VAE issue if on end

covert sand
#

can you turn vaes on and off

plucky shoal
#

what model are you using 1.5 2.1 or SDXL?

covert sand
#

1.5

plucky shoal
#

what UI are you using? Vae is important

covert sand
#

a1111

plucky shoal
#

go into settings, there show all, and search for quicksettings. Here choose Vae, it will display it in top part of UI and you will be able to switch them or use none or Automatic, which is probably best

covert sand
#

is there a vae in there by default, or did i dl something?

plucky shoal
#

it is called sd vae there

covert sand
#

it was on automatic though anyway

plucky shoal
#

it is usualy baked in model. You can choose with baked vae and use automatic.

#

or you can DL if not wrong, size should be about 300MB?

covert sand
#

download what

plucky shoal
#

vae-ft-mse-840000-ema-pruned.safetensors

Try find this file and put it in vae folder, if not wrong

covert sand
#

ye thats the only one i have

ashen kelp
#

I'm also interested in VAE's. There's an "Automatic" setting - is this generally a good idea or no?

covert sand
#

im using pure white backgrounds. idk if that could have anything to do with it

neat elm
#

Hi all, any AI artist or just artist in general here? I would like to gather some suggestion on an AI art platform mobile application that im building. I cant really go into much detail about the business model but its basically a platform where u create art for free, follow ur favorite artist and like their artwork. Free to create art and video and some other thing.

plucky shoal
#

it is good idea.
If it not happen in preview, and in last step, it is very probably VAE

neat elm
#

Please DM me, i really want to build a platform for any “artist” to freely unleash their creativity via art and connect and support with each other so i would really love to gather what artists truly need

covert sand
#

should i have the base model 1.5 checkpoint even if i only use other models, or does it not matter?

plucky shoal
#

i dont understand now. Vae should be per model
1.5 for 1.5 SDXL for SDXL and so on.
if you asked about vae model, i would switch it to auto

covert sand
#

idk if you were responding to me, but i wasn't talking about vaes. was kind of a side question

ashen kelp
#

If I understand correctly, you can only use one checkpoint at a time unless you are also doing refining - then you can use two - basically

#

(A1111)

ashen kelp
#

Question about CFG scale:
How come some renders become absolute trash if the scale is higher than "what it wants"??

knotty turtle
#

Do we have something like Shuffle for SDXL ControlNet?

knotty turtle
#

The reason I ask, I'm trying to figure out a way to generate an image using a very specific color. Do you know of anyway I could make that work? I have a image of just the color that I was trying with different ControlNets, but to no avail. Tried Reference, Recolor, and IP-Adapter

steel garden
#

how to install stable difussion xl on linux? pls ping

covert sand
#

is there a tool that automatically inpaints a whole area like a whole white background?

hasty hornet
#

that's interesting idea, huh

covert sand
#

i installed inpaintanything extension but cannot get it to show up as a tab on my ui

wicked glade
#

yo

feral eagle
#

How hard is sd on hard drive bandwidth? Does having a fast ssd make a difference? I assume it doesn’t cause no way sd can generate data thT fast

covert sand
#

since they are pretty big files

feral eagle
covert sand
soft moss
#

can you use a pen tablet with A1111 ? to make masks in the inpainting tab for example

feral eagle
soft moss
#

if it works, that would be awsome

feral eagle
#

I paint anything add on does auto masks not tried it yet

#

But don’t listen to me I’m a noob ha ha

mighty rivet
#

how do i get stability ai staff role and can i do infinite generations with it

#

who do i gotta ping 😵‍💫

potent bison
#

Would like to play around with stable diffusion, controlnet and qr monster to see what i can mimic from reddit. Is there any... guide ? I guess ? to follow so that I dont install it incorrectly? I guess Im asking like, I am not sure if i need to mess with automatic1111 for that, or not... or what method I need to go about installing it, if that is my end result need

covert sand
soft moss
#

that is awsome. thanks

covert sand
covert sand
#

can anyone help? im getting these out of place bright green highlights on my img2img gens, where there is no green or anything in my prompt or in my original image.

#

it also looks like the highlights turned yellow/gold in another model

#

it will be like a thin spray-like mark around the edges of some objects

lavish current
#

Is there a Discord app that doesn't suck

#

Official one is super laggy on my device

soft moss
spice plank
#

How do I get a character to not be the main focus of an image? Trying to get some furry milfs on an alien planet. The milfs im the background and the general building/scenery in the foreground

#

Ive tried distant view and far away which did very little

covert sand
soft moss
soft moss
winged ember
#

I never used SDXL before so I wanted to try today, I got a model and a VAE but it won't let me change the VAE, it keeps switching back to the one I used for SD 1.5.
No idea why

soft moss
#

sdxl dosnt use vae as far as i know

winged ember
#

how do I use it?

#

I tried to leave that VAE field on automatic but it takes ages to generate an image

#

i got 16GB VRAM, so usually it generates fast

spice plank
#

Is there an extension that is like photoshop generative fill?

soft moss
#

you should be generating at 512x512 and then upscaling when the image is finished

spice plank
#

Hires fix

#

512x512 produces awful images without atleast hires fix. 768x768 is way better with hires fix but that uses almost all my vram.

trim magnet
#

how long it takes for u to generate a 768x 768 with high res fix

spice plank
#

Around 7 minutes with a 1660ti

trim magnet
#

thats just normal

spice plank
#

Yup

trim magnet
#

mine takes 6mins for 1080p ( 544x960 x2 high res fix ) im also on 6g vram

#

u could try to use 1.5 high res fix then upscale on img2img ultimate scale sd

spice plank
#

I can get a 1080p in roughly 10 minutes. I have to use tiles VAE tho.

#

Im surprised a 1660ti can do it at all tbh

trim magnet
#

hey at least you are not on amd 😬

spice plank
#

True, my last card was an rtx 480 8gb, thought this one would be a slight downgrade on the gaming side but its good for AI

trim magnet
#

yea 16xx series is not that good for ai 10xx series is faster despite being an older generation dunno why

fading cloud
#

I'm eyeing a rtx a2000 12gb... I know it's older tech but I can run it in my SFF... any limitations to older cards that I might want to know?

oak meteor
#

Hello

#

I am new, why there is so many version for the model?

#

does everybody download model from civitai?

fervent thunder
#

Hello, in which chat can I ask for help, advice on writing prom or other topics? (Nsfw)

plucky shoal
#

not sure about NSFW

young ivy
#

hmm can't decide if I should get a (both are) samsung 1tb M2 PCIE4 for 100 euro or a 2tb M2 PCIE4 for 170 euro

#

leaning towards 2tb

plucky shoal
#

win 11? System disk? @young ivy

young ivy
#

already have one for system

#

I need room for games

#

and games are starting to require ssd or even m2s

plucky shoal
#

in my case i cant use it as system disk, if you already have it then superb!

fervent thunder
#

Hello, how can I add a bot to my discord server?

warm junco
#

Thats what I got for games and ai

compact nimbus
#

yo

#

is there an api i could use?

civic birch
#

false

#

soon

static cape
#

Hey guys and gals.

Is there a way to load a Lora in A1 for the Hires Fix pass only?

bleak matrix
#

Good morning, everyone! How are we all today?

fervent thunder
#

When i will be able to pay to use it ?

covert sand
#

does anyone know why i have to change the model before i start using a1111? if i don't change the model ill just get solid grey images and errors

plucky shoal
#

what model gives you grey images with errors?

long grove
#

||∆||

covert sand
plucky shoal
#

try update your extensions and restart if it helps? Not sure at all

covert sand
#

na ive done those things

#

ive seen others w the problem

#

i had the problem on my last pc as well

#

a whole new build and fresh SD install and im still having it

sour creek
#

hey hey!

fervent thunder
#

Where can i ask a question about the software

#

@wild steppe sorry for interrupting but I got a question about sdxl

wild steppe
fervent thunder
#

Ok I was just wondering what the terms and conditions for the software

wild steppe
fervent thunder
#

can someone help me

formal stratus
#

Anyone of you already tried a ghost mannequin effect?

covert sand
#

where do i get the SD 1.5 base checkpoint model

#

just the vanilla base model

plucky shoal
#

in models/stable-diffusion

fervent thunder
#

Where can i download musicgen to run on my own pc?

digital fern
#

sorry for the interup, I'd type my prompt on channel bot, but its no response. And now I can't do any generate, what can I do?

ashen kelp
digital fern
fervent thunder
#

Hi all. Can I add a bot to my discord server?

#

Hello, can I make a video with my photos?

hot zephyr
#

Hey can someone recommend a good laptop that will run stable diffusion smoothly and fast enough?

Budget is 1k

trim magnet
#

Attention: If you or a loved one has been diagnosed with Mesothelioma you may to be entitled to financial compensation. Mesothelioma is a rare cancer linked to asbestos exposure. Exposure to asbestos in they Navy, shipyards, mills, heating, construction or the automotive industries may put you at risk.

jovial quail
#

Sorry, just ignore that, I didn't know it was fake

pine path
#

apparently google indexes bard conversation. You can view random ones just by searching for site:https://bard.google.com/share

#

needless to say, that's uh, not good

soft moss
#

you are told the conversations are not secret, or im reading your coment wrong ?

trim magnet
#

most private google product

spice plank
#

I can mix facial expressions by just adding more to the prompt? That what it feels like, not sure if its true.

solid coral
#

Hey folks, updated both Automatic1111 and openOutpaint and now there is no more "Send to openOutpaint" buttons on the PNG tab (or anywhere else for that matter). Anyone else having this issue? Any suggestions on how to get the button back? There doesn't seem to be a way to load a PNG directly into openOutpaint that I can see...

plucky shoal
#

are you not in incognito mode? I used to have issue i was unable to import images due this incognito mode.

#

so suggestion, try different browser if you have

solid coral
#

Opened it in MS edge but there was no difference

plucky shoal
#

hope it will work. 😦

#

i am in comfui, cant test. Sorry cant run both at once

solid coral
#

No worries and thank you for helping regardless (still haven't tried comfyui, should do that some day)

plucky shoal
#

some nodes are realy interesting

solid coral
#

Yeah I've heard good things about it!

#

Ok I resolved the issue. So... I updated automatic1111 like 2 days ago and removed git pull from the bat file because I want to be in control of when it updates. I didn't think that in the span of 2 days there would have been an update that broke it, so I didn't even try to update since 2 days ago. But I decided to try, put git pull back, it updates automatic1111 and voila! Now the button is BACK

#

So lesson learned: Always try the very very freshest update to see if that fixes things facepalms

plucky shoal
#

😄 superb it works.

solid coral
#

Sorry to bother you all with this. But perhaps if someone else runs into the same problem this might help them. At least when I searched I saw @undone garden had been asking the exact same thing in #🤝｜tech-support . So Ciccio if you see this message, try updating everything.

waxen gull
#

Im having problem with using animated diff with comfyui, I am following this workflow:
https://civitai.com/articles/2314

after downloading all the modules and checkpoints im getting this error

Error occurred when executing KSampler: mat1 and mat2 shapes cannot be multiplied (1232x768 and 1024x320)

Could anyone give me any guidance?

visual hill
#

Hi I am missing the SD VAE drop down on the top left corner for Stable diffusion. Is this an extension I have to dl? my sd version is 1.6.0

plucky shoal
#

mat1 and mat2 seems you are using 1.5 model for sdxl thing or other way round. @waxen gull

#

sd vae is posible to download on civitai.com for example. For model you are using. 1,5 for 1,5 sdxl for sdxl.

warm junco
#

In your link is a copy

steel garden
#

How can I increase the token count of the prompt in the sdxl model to 90 tokens?

from diffusers import StableDiffusionXLPipeline
import torch
from PIL import Image
import openai 
import re
import time
import os
        prompt_list = []
        for i, prompt in enumerate(prompts): 
            prompt_list.append(prompt) 
            print(prompt)

        pipeline = StableDiffusionXLPipeline.from_pretrained(
            "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
        ).to("cuda")

        for i, prompt in enumerate(prompt_list):
            generated_image = pipeline(prompt=prompt).images[0]
            filename_prompt = prompt[:100]

            output_path = f"{folder_path}/{filename_prompt}.jpg"
            generated_image.save(output_path)
            print(f"Generated image saved to {output_path}")

        counter += 1
        print(counter)```
#

my prompts looks like this:

A serene lakeside landscape at sunset, with a lone boat drifting on the water. It's a high-resolution photograph intended to feel natural, calming, and reflective. You can find it on Shutterstock or Getty Images, featuring extremely detailed and sharp focus. The image showcases warm, pastel colors and exudes an atmosphere of tranquility. The primary colors are rich golden tones with a hint of lavender. The image is accompanied by soft, warm sunlight casting long shadows.

young ivy
#

Off to my third cancer surgery, third time it's back and this time it's not a certain outcome...Wish me luck in a way that won't jinx me

odd arrow
#

which nvidia driver version do you use ? 531.79 ? or latest ? 537.42 ?

warm junco
tacit monolith
#

Yo I need help

#

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.

#

Do u know how to solve this?

warm junco
covert sand
#

I figured out why i was getting green highlights if anyone is interested

#

and why I had to change my model on UI start everytime to get an image

#

Still can't get my inpaintanything extension tab to show up. Anyone else had this issue?

elfin bramble
#

Hello I have a question : image created with stable diffusion are 100% unique ? And can i use it for commercial use ?

ripe kernel
#

a

brave pike
#

@hollow oriole i sent u dn

#

Dm

proud shell
#

hi guys
checking specs for a new pc, will these two give entirely different results ?
PNY GEFORCE RTX™ 4070Ti 12GB
PNY GeForce RTX™4080 16GB

Also is there a better or preffered company for AI stuff ?

plucky shoal
#

dont go with 4070ti

proud shell
#

what do you mean ?

plucky shoal
#

Not go with 4070ti, it has issues, and it will make difference for future for sure 16GB is more futureproof

proud shell
#

oh ok cool, good to know, any other models that could be better for the price ?

#

that's the specs they offered me for the 4080
ANTEC DF700 FLUX RGB
ASUS PRIME Z790-P WIFI Intel LGA1700 Z790 DDR5
INTEL I7-13700KF Tray No FAN/GPU 16Crs max
Corsair H150 RGB 360mm Liquid CPU Cooler
PNY GeForce RTX™4080 16GB VERTO™ Triple Fan
Corsair VENGEANCE 32GB 2x16GB DDR5 DRAM 5600MHz C36
Gigabyte AORUS Gen4 5000E SSD 1TB
Antec HCG GOLD 1000W Fully Modular 2X8Eps 8X8 PCI-E

around 3300$

covert sand
#

where do you download the controlnet models?

proud shell
covert sand
fast timber
#

guys, why is it that when I generate beautiful and high-quality art in Stable Diffusion, a new one appears on my half of the download lane, with defects and bugs? How to fix it?

warm junco
warm junco
proud shell
#

AI will keep it going up

warm junco
proud shell
#

then i'll probably go with this one lol