💬｜general-chat

[warm junco]

for 2.1 models you also need a .yaml file

[lime violet]

Yeah I was trying to find the .yaml file for the default 2.1 online but it's like finding a yaml in a haystack 😄

But the 2.1 based custom models don't actually need the default 2.1 model installed right?

[warm junco]

https://github.com/Stability-AI/stablediffusion/tree/main/configs/stable-diffusion
For 768x768 its the v2-inference-v and for 512x512 the v2-inference

[warm junco]

nope

[lime violet]

Ah awesome thanks, got confused becasue the yaml file names are different than the sd2.1 model file names

[warm junco]

yea you need to rename then

[silent gust]

Probably a dumb question, but how do I use loras on top of a model in a1111?

[sleek raven]

urgh

I made my model with 2.0

now I cannot use control net

I feel so frustrated

[rigid surge]

Hey there, can someone help me out

I downloaded a1111 on a new pc

but now it doesn't open the small web-ui element before going straight into stable

like the place where you can adjust settings for launch/add parameters

[warm junco]

right click and edit the webui-user.bat to set launch paramters

[rigid surge]

It sais

it can't find python

even though I have python installed

[fervent thunder]

Someone care to share an up-to-date step-by-step tutorial on how to install Shivam's Dreambooth locally?

[lime violet]

There is a one-click installer for Auto1111 that installs all that stuff for you

[rigid surge]

I understand

However on my previous pc whenever I would launch the webui

I would get a small menu with options

On my new pc I do not

I'm trying to figure out why/how to get it back

It would have launch options etc

[rigid surge]

Damn I went from a 2060 super to 4070 ti and I feel like it really didn't make much of a different in speed. Only doing 3.7 it

[brazen stream]

What is LoRA?

[bold lark]

how do you upres images created with controlnet in specific pose?\

[vestal dew]

Different version / install

I have the no launch window version now too

[vestal dew]

I am just buying a 4070TI. It's not fast?

[rigid surge]

I mean I figured the improvement would be huge

The 4070ti is definitely fast

but maybe not specifically for AI things?

I see that its only using 20% of my gpu tho

and its hitting 3.9-4 it

nevermind

https://www.reddit.com/r/StableDiffusion/comments/y71q5k/4090_cudnn_performancespeed_fix_automatic1111/

Did these changes and now I'm hitting 11.6 🙂

[astral goblet]

yeah i was about to tell you about that exact post lol. i got a 4080 and it made a HUGE diff

about double

[waxen mirage]

I loaded a SD 2.1 model into google colab and the image generation time is taking way too long (5 min). Can someone please help explain why this might be happening? I have the 2.1 model in a folder that has 1.5 models in it. Could that be the problem?

[normal tusk]

I went from 10-12 on stock 4090 to 30 with cudnn fix + xformers when I did it a long time ago

I tried to get pytorch 2.0 running a well with new xformers and ru triton for apparently possible 40, but that didnt work out

without a proper guide, I bricked 3 installs and some hours trieng for 30->40 jump

the ones I got now is Cuda 12, cuDNN 8.8 beta running perfect

[astral goblet]

lol i was attempting pytorch 2 for a bit too, but the updates that happeend to cuda 11.7 from 11.3 were helpful enough

[normal tusk]

and the older 0.16 cxformers instead of the whacky 0.17 for triton and torch 2

[astral goblet]

i'm using a nightly of 0.17 lately. not sure if it makes any difference but no harm is good

[cedar glen]

does anyone know if controlnet is planning to support 2.1 at all?

[astral goblet]

there's already support, but people need to spend time to train the models. there is an example training project in the main github for it

i think someone has trained an openpose model for 2.1 already too, but its rough i've heard

[cedar glen]

hm i guess maybe i am not understanding, when I use controlnet I can't use the base 2.1 model correct?

cuz i always get errors when i have 2.1 selected

sad that most of the community is sleeping on 2.1 and still stuck on 1.5 imo

[vestal dew]

Bookmarked for when my 4070TI arrives

[astral goblet]

thats correct. the code for controlnet supports 2.1, but no controlnet model is trained yet. you know how you have to select openpose, hed, scribble, etc.. all of those are different controlnet models. they're all made for 1.5 specific, but controlnet models can be trained for 2.1. they just haven't come out yet. takes time to train. i think each of the originals was trained for a week on a single 3090ti, but i might be off on that info.

[cedar glen]

oooh ok i see, so if i have 2.1 selected i will get errors when using canny, hed etc..

ty

[astral goblet]

yeah. until 2.1 versions of those are released, and then you'd load those in tandem with a 2.1 base or refined model

[cedar glen]

wow i hope people start training on 2.1, i really dont get all the hate for 2.1 tbh, obviously i know the criticisms but i still feel the pros largely outweigh the cons when comparing 2.1 to 1.5

[astral goblet]

https://huggingface.co/alfredplpl/ControlNetForSD2 it's rough by the looks of it, but these are coming along

[cedar glen]

oh nice! ty

[astral goblet]

it's just one openpose version, and its.. its rough

[cedar glen]

i can wait haha

[astral goblet]

yeah i just noticed, author even says its a proof of concept

[cedar glen]

cool well hopefully we'll see something usable soon

[astral goblet]

yeah i can feel it coming, in the air, tonite, oh noohoh

[vale creek]

I am in the process of choosing a computer to buy. There is a set of 3 used Tesla k80s for sale at a low price. Is it better to use them together or buy a 3060ti and use it?

[cedar glen]

https://tenor.com/view/drum-tyson-gif-18911438

aw

[vestal dew]

would 3 k80's be more powerful than the 3060ti? Certainly more VRAM

[cursive shoal]

More VRAM doesn't necessarily mean more powerful

[vast ingot]

you can be slower with more vram. 2 params are useful

Cuda core count/speed of making pictures for a given pixel count
VRAM => higher size pictures or more batch size possible.
The last thing, higher batch size, can result in some speed gain though

1 batch of 2 pictures is faster than 2 batches of 1, if you have enough VRAM to run it, with the same clock speed

[vestal dew]

correct, like 4070ti is faster, but for SD the 3090 has the strength of more VRAM

[noble lynx]

how do i see older #1023999442338201721 ? i can only see the winner and not the other submissions ?

[vestal dew]

So how would 3 k80's compare to a 3060ti?

I see

[heady pagoda]

hello , im new here , i was watching a video by corridor crew and saw that they were putting their own pictures and turning them into anime using AI and stable difusion ... my question is , does anyyone know what ai they were using ? most AI picture generators use just promts and i barely can input my own picture

[vestal dew]

link the video

[heady pagoda]

https://www.youtube.com/watch?v=_9LX9HSQkWo&t=10s

this one

[vast ingot]

yep, I watched all they did using SD. they doccumented it at the first of the video you just linked, but let's break it down

they use Stable Diffusion as main tech here

but they do a lot more

first they shot on green screen some scenes with good poses and movement

[vestal dew]

SD with A1111

[heady pagoda]

is SD a blender add on?

[vast ingot]

then, they have trained a model, taught the AI, on a style they want, here a specific anime style. or they use a pretrained model.

[vestal dew]

SD is AI image generatopm

generation

[vast ingot]

Then they use trick to stabilize the video in SD, and in other movie making software

SD is also a blender add on yes, but at its core, it's an AI that transforms text into pictures

[vestal dew]

what do you mean by this?

[heady pagoda]

they used deflicker filter in davinici resolve

[vast ingot]

multiple things. same seed, control net, and they have a deflicker tool in their software, resolve studio

[heady pagoda]

so how did they get their hands on that specific SD program?

[vast ingot]

it's here and free to get and install and personalise at home 😉

[vestal dew]

OK, but they only used SD to turn the images into what they wanted, right?

SD also works in Photoshop

[heady pagoda]

so they got stability AI?

[vast ingot]

it's the same program as we all use around here. they choose a nice model on something like "civitai" or they trained one, and then tinkered with the generations parameters, using their knowledge and other AI users around that have found tips to help progress in the animation sector

[vestal dew]

yes, like everyone here

[vast ingot]

yeah, we call it SD for short, Stable diffusion

it's the name of the AI itself

S.AI is the lab that published it

[heady pagoda]

nice!!!!

[vestal dew]

So there;'s a video addon or extension for SD?

[vast ingot]

deforum

or lots of scripts and extensions in automatic too

batch processing all frames in img2img mostly

you input a fixed seed, find the right controlnet setting, and batch the folder where you unpacked all your frames

then just ffmpeg the output into a new movie

then you can go with flowframes for interpolating more frame for example, or deflickering

[vestal dew]

I saw a video of a woman using SD to do motion capture of sorts. Made video of her, then turned it into whatever moving the same way - in SD

[vast ingot]

yep, open pose + canny double control net is very good in that setting

depth too

[vestal dew]

what video card is needed for that?

[vast ingot]

not that big, it will depend on what size of frame you do for sure, but the controlnet model's aren't that heavy

[vestal dew]

I hope the 4070TI I bought can do it

[vast ingot]

6 should be good imo, 8 almost sure (talking for 512x512 pics)

[vestal dew]

the parts are on the way, once I stopped depending on microcenter

[vast ingot]

[normal tusk]

is there a way to increase sample size for the UI in some config for auto1111 ui past 150?

[vast ingot]

yep

[normal tusk]

I have to resample so many times in img2img bc it didnt quite finish

esp with many prompts

with just 150

[vast ingot]

you can change all defaults, min, max, and steps of the UI in config-user.json

[normal tusk]

oo

gotta check that out

[vast ingot]

sorry

[normal tusk]

sweet thanks

[vast ingot]

wrong file

ui-config.json

[normal tusk]

does the sliders and stuff adjust

or do I have to edit their values seperatly

nvm ill just try later

[vast ingot]

the sliders and all will keep the same visual style

they will adjust their steps

[normal tusk]

so the scale just changes?

[vast ingot]

you need to reload the UI for that though, in the "settings" tab

[normal tusk]

but if I move it midway its half of max

thats important

[vast ingot]

yep

you can always click in the little space where there is the number itself, top right of a slider, and enter a value

and last thing, if you have 2 widely different uses of SD that would require different sliders, you can... have 2 different config files, and have 2 different bat files to start your UI. Depending on what bat you start, it would take the corresponding config (there is a startup option for that)

[normal tusk]

is there a doc for the actual max caps to set?

[vast ingot]

not on such things nope

most will work until your hardware can't handle it, or you go over the INT limit ^^

[normal tusk]

with the xformers it shouldnt be to hard on the hardware for upscaling

[cursive shoal]

still trying to figure out how to use stable diffusion properly

[normal tusk]

while without the 2048 is p. much really a lot as max

its like 1/3 or so? vram usage

[vast ingot]

some specifics that lost you ?

[normal tusk]

with xformers

[vast ingot]

it's about 1/3 less used, so 2/3 vram left

[cursive shoal]

i dont know what prompts to put in

i dont know what a proper prompt should look like

[normal tusk]

positive or negative prompts?

[cursive shoal]

positive

[fervent thunder]

anyone know when mj5 is release?

[normal tusk]

I think the fastest is to just scribe a landscape

to get a hang of basic things

how to add things and change them

[vast ingot]

well, prompt is quite a thing to master for sure. You want to try to describe the picture as it would be named if you encountered it in the wild on Internet basically.

[normal tusk]

do you know these crazy long japanese manga / anime names?

[vast ingot]

that will include framing keywords if you go for photgraphy, or real life events, ...

[normal tusk]

which are basically a full description by itself

yeah those

you plant all over it

[cursive shoal]

oh

[normal tusk]

with , on tags

[cursive shoal]

thats helpful haha

[normal tusk]

I like putting them in a txt editing below each other

that normal ui is kinda useless

except for pasting it in

[cursive shoal]

what tags should i mostly use?

[vast ingot]

also, looking at https://lexica.art/ can give some examples, even if a lot aren't the best prompting there isstill a very nice place to start

[lyric hamlet]

where can i download Stable Diffusion to my pc?

[cursive shoal]

thank you

[normal tusk]

nvm I just saw you can spam enter to add rows

didnt expect that, was searching for normal web things like a drag thing

[vast ingot]

check more in #🤝｜tech-support for that, but here https://github.com/AUTOMATIC1111/stable-diffusion-webui/ or here for example https://invoke-ai.github.io/InvokeAI/

[lyric hamlet]

thanks

[normal tusk]

landscape

, colorful
, highly saturated colors

it also helps a lot to use few

and check what they do

[vast ingot]

"rainbowshift"

works wonders

[last merlin]

Ultra detailed 8k

🤣

Fujifilm

[pallid flame]

hi , can anyone explain "model hash option" to me

[eternal oar]

Do anyone have a vps hoster for free(im highly broke)

[last merlin]

Mastepiece

[vast ingot]

a hash is like a file "signature". You can see the hash of a file online, before downloading, and checking it's the same when in your software, to make sure it's not broken or tempered with. it also acts as a "shortname" in automatic in some way

[pallid flame]

ok thank you

[azure yew]

gmmmm
has anyone used stable diffusion as a level designer for games yet?
https://twitter.com/rustyedrusted/status/1630677027522244608?s=46&t=z88V3Tcxq9Gxao3hXGPU3A

[blissful robin]

does anyone know if stability.ai charges for things that get NSFW filtered? seems to happen a lot for no good reason

"you have not been charged for images blurred".. great.. just needed to know if I had to charge users

[vestal dew]

Chilloutmix and basil...they seem to wreck hands more than other models I've used. Is there a vae or other way to get better hands with these two models?

[topaz blade]

Hello, new user here

Are there any educational articles/videos/tutorials for newbies?

[paper smelt]

One message removed from a suspended account.

One message removed from a suspended account.

[steep mortar]

ControlNET is freaking awesome

[topaz blade]

@steep mortar Thats what I hear. I just need to learn to use it, somehow 😄

[steep mortar]

Im only use the openpose model

my question is tho, is it also using my model ckpt or is it all going thru openpose model

Woohoo! your model file is controlling the file output

[quick bobcat]

Anyone have insight into seed length as it relates to variability? For example would 100/101 seeds have the same variability as 1000/1001?

[cursive shuttle]

I need help finding these LORAs in Chinese!!! <lora:画风 原神:0.4> <lora:画风 新海诚:0.8> <lora:画风 焦茶:0.3>

aanyone here knows about these loras?

[dreamy igloo]

There are a few filters on Playground that I'd like to get, I've been searching on Civitai and have not found anything like them: Dream Haven (the text prompt show up as lushill style), this one is sorta painterly/fantasy, I guess it can be replicated with another model/ti/lora, just have not found one like it yet. Geometrieva Style (text prompt show up as geo2009 style) its kind of cyberpunk but with a fixed palette of colors and not dark/night scene oriented. This one is hard to replicate with models/ti/loras I've seen. Perfume Style (kind of like Analog diffusion but with more glamour, I guess I can just use Analog diffusion and add glamor fashion or objects into the prompt to compensate).

[vapid basalt]

Is this the official stable diffusion discord? No suggestions or bug section?

[verbal osprey]

I posted this in the tech-support channel but that seems dead.

I'm getting an error when trying to use openpose and controlnet.

\stable-diffusion-webui\venv\lib\site-packages\ffmpy.py", line 98, in run
raise FFExecutableNotFoundError(
ffmpy.FFExecutableNotFoundError: Executable 'ffmpeg' not found

Anyone know how to fix this? I downloaded an extension and it must have messed with some settings because it used to work perfectly.

[vestal dew]

https://stackoverflow.com/questions/60561571/ffmpy-ffexecutablenotfounderror-executable-ffmpeg-not-found

[elder cradle]

Yes, this is the official server

That's why it has "Verified" badge

[topaz blade]

I must run SD with Mac Mini M1, sadly, so I am not running very fast.

But also, it seems I cannot make images bigger than 512x512 (1024x1024 with hi-res fix)

Is there a way to work around this? Any kind of setting that I could change to allow this?

[verbal osprey]

Thanks. I installed ffmpeg through the extension area but there's no exe there....

[gloomy hornet]

what ckpt is recommended for lora training of a real persona? The guides I've been reading are dealing with anime characters.
First try inbound and I'm hesitant to run on a pruned model and doubtful about the quality of base SD-1.5

[dreamy igloo]

are you training sd 1.5 or 2.1 lora? There are plenty of photoreal models that you can choose from on civitai. Pick anyone of those for non anime oriented loras that you want to train

you can select "photorealistic" on the bar above the model images to narrow down your searches

if you don't want your resultant lora model to always show NSFW content, select models that skew less towards those, like classic negative if you want cinematic look (2 versions for SD1.5 and 2.1), realistic vision (1.5), there is one called cmodel (forgot exact name) for 2.1, mangled mix for 2.1, but definitely tons more for SD 1.5 based photoreal model for you to choose from. Choose based on their aesthetics and creative capacity (how "wild" can their images get, based on the samples people create), maybe you are going for realism and want the fantasy elements out, or the other way around, you may want more alien/fantasy/wild artistic results.

[barren zodiac]

i dont know if anyone can help me but my automatic1111 has stopped working over my local network, works on my main pc but not any tablets. was working perfect up until today (using the --listen arg. in webui-user.bat)

[void citrus]

I've been working on this too but I never got it working in the first place. I use the other situation where you login with Gradio

I'm trying to make a chromebook connect locally with 127.0.0.1 but now my cmd window says 0.0.0.0:7680 and it doesn't work. Tried port forwarding I'm just over it the login works for 76 hours I guess

I guess I'm just bad at port forwarding

[sullen imp]

so, im completly new to this and yesterday i tried for my first time to upscale pictures with image2image, but always i go above 768x768 i get the error cuda out of memory even tough it says theres like 7 Gigs free (using 4070ti) any ideas how to fix this?

[warm junco]

Did you select the sd upscale script ?

[sullen imp]

hmm im not sure actually, do i have to specifically select it?

if so then i think i didnt

[warm junco]

Yes if you talking about upscaling

[gloomy hornet]

Thanks for the advice, I've decided for realisticVision now. The goal was for sd 1.5 because I hear there are some incompatabilities with 2.1 still. Good to hear that a mix is fine!

[warm junco]

Maybe you tried to just change the output resolution

[sullen imp]

i just selected img2img selected the resultion i want and clicked on generate

[warm junco]

Okay yea, but That should work with your gpu

Do you use --xformers ?

[sullen imp]

no

[gloomy hornet]

You don't use img2img to upscale, what you are trying to do is to diffuse a new image on top of the img you fed the gui, if you select a high resolution, the error makes sense.
You have three options, either generate your desired image again with its seed and check hires fix with .3~.7 denoise, or you choose the Extras tab for a very straight forward upscale method, or as CS1o suggest, you can get the 'ultimate SD upscale' script that's used under the img2img tab.

[warm junco]

Edit your webui-user.bat and add --xformers behind Commandline_ARGS= then restart

[sullen imp]

ok thx, gonna try that once im home

[warm junco]

Then it should work and yes for upscaling try the sd upscale script under scripts dropdown

Or create the image with highres fix

[wide juniper]

Does anyone still use instruct pix2pix? It wasn't effective when it first came out, and everyone is used in controlnet now

[quasi crescent]

Do anyone wana help with training a model using noise offset, I can do it on my own but I do not have a great gpu.
I want to train a lora model on the best images as now we can train images with proper style
Please Feel free to DM if you have high vram gpu and basic training experience

[fervent thunder]

Why should it?

I had no idea who that it before you mentioned him (I still don't really, but I see that's a person lol)
You can train AI on any person's photos if you want..

nah and I don't really care

[light garnet]

funny

[lean laurel]

Man, it's really amazing to see the training process of an embending

Like, in the start it's crap, but gradually looking better and better until it's just like you imagined it

[elder cradle]

Oh, Andrew Tate? You may want to fine-tune using images floating around google

Anyways, imagine MJ server allowing non-members to chat in other channels, that would be a massive shitshow considering 7m users (or higher) likely contains a bunch of anti-ai dudes

[halcyon granite]

if I have a embed file to insert in the negative prompt. Do I format it like this [filename] ?

[bold lark]

how do you upres images made with controlnet in specific poses or with specific facial features? I usually generate images in 512 pixels and then take it to img2img and increase the resolution to 1024 with 0.6 cfg scale. It introduces a lot of new details and corrects a lot of mistakes, but the pose or facial features gets changed. What process do you use for upressing?

[green ingot]

hye

[vast ingot]

Hey ! welcome around

[astral goblet]

you're teaching a machine with education! wild right?

[fervent thunder]

hey hey how do i use Stable Diffusion Version 2?

[pallid stream]

anybody have experience with gammac dreambooth?

[uneven heath]

yoink stable diffusion 2 model in your models folder along with the .yaml file

take care of prompting, and be careful in using negative embeddings too

1.5 might not have required a lot of them but 2.1 does, you're doing more sculpting here

best resolution to work with is 768 x 1024, 1024 x 1024....pretty much high res images

[hallow scaffold]

https://media.discordapp.net/attachments/1065347477554860032/1080541804404932678/ksanghera_a_colourful_human_body_dancing_in_a_graceful_ballet_a_88cb968b-1e3f-4bab-9155-968196b16d53.png

[uneven heath]

amazing result

[vestal dew]

with mangled hands

[uneven heath]

pads

[pallid stream]

where do i go if im having issues getting dreambooth to work? dont want to bother here

[lean laurel]

Yeah! Feels like my own child lol

[proud nova]

Can somebody help me with canny controlnet? How does that work? My results aren't those black and white images but something else. Model installed, using canny preprocessor

[vestal dew]

if you're only getting an image and not the second one for the model, then controlnet isn't running

what is your GPU?

[proud nova]

I am using google colab

Oh wait. I think I figured it out

[upbeat tangle]

Guys how can i use control net?

[proud nova]

What exactly do you want to do with it?

[fervent thunder]

Where can I find sample training input images of different man/woman for training Dreambooth? Photos of the same person (man/woman) in different poses, backgrounds, etc. I don't want to use my photos and, since it's for a production website, it should have privacy of those people in mind, so it should be something in public domain or so. Any ideas? Is there a 'sample input images' folder for Dreambooth somewhere that I could use?

[proud nova]

Do you want using openpose?

[thick epoch]

yo

how do i reopen my instance of stable diffusion locally

[little yoke]

Click on the .bat file in the installation folder

[little yoke]

It will open a little black command window and will run a few things. Takes a minute. Then open SD in your browser

[thick epoch]

Thanks

[little yoke]

Click on the .bat file in the installation folder

[thick epoch]

I think I got kicked out of the midjourney server for asking how to rerun stable diffusion LMAO

No warnings lmao, alright

[little yoke]

Not cool. If I were you I would create a shortcut for that .bat file to have on the desktop

[thick epoch]

Alright

I’m not at my PC right now but, I only had a short time to use stable diffusion yesterday, and it wasn’t generating images

It would say “waiting” and then never make it

And then when I switched to DDIM, the progress bar went halfway and then didn’t make anything

Using 1.4 btw, when I get back to my PC, I’ll try to fix the issue, or if anyone can help me that would be great

[azure cargo]

Anyone knows of a way to have weighted prompts in Automatic1111? I am trying to create a mixed race, polynesian and scandinavian portrait and nothing so far works.

[normal coral]

@azure cargo highlight the part you want weighted then press ctrl + arrow up/down

[azure cargo]

That doesn't work with spreading the weight though between the two different races. Best I got so far was Polynesian|Scandinavian but is not quite the result I had in mind.

[dry sail]

Hi i'm new to coding and AI, is there any good articles or videos. i can watch/read. To get started

With stable dif

[forest brook]

https://civitai.com/models/14605/howls-moving-castle-interior-scenery-lora-ghibli-style new LoRA for scenery style of Howls Moving Castle

[vast ingot]

Hey ! what are you trying to learn to do ?
you say coding, but there is no coding to use the AI. a little more complex install than usual programs for sure, but you shouldn't need to code anything now.
#1072220168534642768 is a good place to check, as well as #1072229020520947753 , then you can read the faq I'm about to link on how to access the AI.
There are lots of videos on youtube explaining Stable diffusion, if that is what you prefer as format, any search will give a lot right now, but just starting by testing it is usually the best

[surreal perch]

FAQ: I'm new here, how do I generate images ? Where is the bot ?

Welcome ! There is no bot currently to generate your images on discord. You may want to start by taking a look at the #1072220168534642768 channel. You can access Stable diffusion in different ways : 1️⃣ the official website, https://beta.dreamstudio.ai/. The easiest and fastest way to access Stable diffusion with 200 free credits. For any question on it, you can find help in the #1025467151206854736 channel. 2️⃣ Installing Stable diffusion on your computer. There are numerous projects that let you do that, and you will find help in the #🤝｜tech-support channel. 3️⃣ Running Stable diffusion in the cloud, through rented GPU services, using notebooks. You can find lots of them shared and discussed over in the #1011228442399883294 channel.

[hidden lynx]

Where can I find the 1.5 download?

[vast ingot]

https://huggingface.co/runwayml/stable-diffusion-v1-5

model link directly https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.safetensors

but you'll need to accept the terms first

[hidden lynx]

You are awesome thank you much

[rotund quiver]

Anyone know why the xformers arg makes images produce different results even with the exact same prompt settings? For example I can click generate, let it finish, click generate again without touching anything else and the second image will be different from the first. xformers is so much faster than default but the randomness it introduces is not very desirable.

[warm junco]

Is the seed the same number or -1 ?

[rotund quiver]

Exact same seed and everything. Same seed, same prompt, same seed delta, same method, steps, scale, etc... literally nothing changed between generations.

[warm junco]

Can you post an example?

[rotund quiver]

I can upload some example images

Gimme a few mins to generate some more I deleted them

Posted in #🏞｜general-with-images

[vestal dew]

I want to see this too

[fervent thunder]

Where do I share robotic insect images?

[vestal dew]

hmmm, Im getting nothing but purple blobs today

[junior comet]

hey guys

is there a model that can generate line arts (sketches)

i tried it with midjourney and it could make them but i want to know if SD can its better

[gentle otter]

#1070829655638151208

[vestal dew]

everything fine yesterday, today I get purple boxes

[arctic rock]

whats the prompt to get a word written on a subject's body?

for science

[fervent thunder]

Yea , that's issue with xformers, you don't do xformers if you want consistency.

It's knows issue

[young shale]

people there is some command to give averages like height, weight and the classic 90 60 90

[leaden mural]

Guys

[vast ingot]

what's happening ^^

[vestal dew]

https://www.youtube.com/watch?v=8V8QMbLd2f8&ab_channel=RetroAlexander

[vestal dew]

Ive never heard of that

[young shale]

that's why I ask if there is any order to do that 😛

[vestal dew]

controlnet does a lot, but I haven't heard of being able to input height and weight

[young shale]

good to keep experimenting

[vestal dew]

I tried height and weight in SD...no effect

[barren zodiac]

anyone get this trying to install extensions? (AssertionError: extension access disabled because of command line flags)

[hybrid coral]

Hola gente acabo de inagurar un server de Shitposting y memes em general aquel el que quiera acceder que me mande un mensaje asi le paso la invitación
Los espero jsjsjsj

[fervent thunder]

do loras work for 1.5 or 2.0? i assume not

havent used them just yet lol

[hybrid coral]

Es de memes

[vestal dew]

Control Net about to get old. MultiDiffusion Region Control is the new toy.

[fervent thunder]

new extension?

[vestal dew]

yes

not sure if it works with anything yet

[fervent thunder]

nice.

oh

[vestal dew]

https://www.reddit.com/r/StableDiffusion/comments/11fmo4g/more_control_than_controlnet_code_is_out_for/

there's a beta for 2.1, and in the Reddit thread someone says it's available for 1.5

https://huggingface.co/spaces/weizmannscience/multidiffusion-region-based

https://multidiffusion.github.io/

[fervent thunder]

hopefully auto adds it soon

looks incredible

Region control...is it what I think it is?

yea...kinda, nice, been waiting for something a bit more official for a while

[fervent thunder]

compositional control is a huuuge leap forward. prompt bleed has been an issue since day 1 and even clever use of inline prompt editing only did so much to mitigate it. there was an extension for automatic1111 that allowed you to divide the image into sections and then prompt for each portion called "two-shot". it was a pretty big deal that flew under most people's radar, but it was overly convoluted and very difficult to get desired results on more complex stuff. this is the next step above that and looks to be a LOT better

[lilac reef]

Adding bones to the flow is a HUGE step forward, but, sadly, not for 2.x yet.

[fervent thunder]

this feels more like adding bones to it. and yeah, i agree lol https://www.youtube.com/watch?v=ptEZQrKgHAg
genuinely crazy how far this stuff has come in just 6 months

[lilac reef]

When I used to do 3d bones came in like this, looked like this, and was a game changer.

I think this is alright but we need to stay within our own eco system not be required to load in a 3d program.

like Blender

Mind you, this is just an evolutionary step and I suspect we will not need it soon enough.

[fervent thunder]

agreed, that's why i didn't link it when i saw the video premier about 13ish hours ago. it's not far off though, i bet. at least blender is also opensource. either way it's a big deal and it'll be awesome

[lilac reef]

What is sort of cool is that a day will coome (soon) that working in 3d will be for a niche item and even raytraced, GI, SSS, etc... will all be done in the AI. In my eyes that will be amazing. Think of the render farms that will suddenly no longer be needed.

[fervent thunder]

when all this stuff released to the public last year, i was telling family that i wouldn't be surprised to see ai capable of generating shortform movies/video clips in like 10 years. after watching this for the last 6 months, i now feel like i -severely- underestimated it. honestly wouldn't be that surprised to see coherent output in a year from now

[lilac reef]

I have been saying 1 generation to a max of two generations of video cards I can sit in my bedroom and make a full on film/movie in under a month. That is for voice work, and the generating. 2 gens is about 4 years, may 5 max.

One thing we are missing that once hits it is all over is temporal cohension.

[vestal dew]

If months are like years...

[lilac reef]

I am basing what I say not only on the tech

[fervent thunder]

consistency is another one of those day 1 issues, but the advancements with dreambooth/hypernetworks/embeddings/loras makes me feel like that's not gonna be as big of an issue in the future either

[vestal dew]

I made the comment on the wrong one. I mean it on CustomCombo's comment of 10 years

[lilac reef]

temporal cohesion is what we lack.

what that does is look back at the previous frame, or two, and go from there so no glitches. It simply flows.

we are damn close btw

[vestal dew]

it's amazing

yet when I tell it dressed like a cowboy, it puts the guy in shorts and a t-shirt

[waxen mirage]

https://www.reddit.com/r/StableDiffusion/comments/11fsdtq/controlnet_offsetnoise_lora_stable_diffusion_30/?utm_source=share&utm_medium=ios_app&utm_name=iossmf

[fervent thunder]

try "in a cowboy costume" instead. i've found "costume" works pretty well for getting the desired result if trying to put someone in a specific outfit

[vestal dew]

I will try that

maybe I'll have more luck with that than with chatGPT right now

we are now trying costume. It takes my feeble GTX 970 some time

my 4070TI was delivered Wednesday, but instead of a GPU it was a lens protector

[fervent thunder]

oof. yeah. i was on a 1060 6gb until just last week. it's rough for low end cards

[vestal dew]

1060 6 is much better than GTX 970

[cursive plume]

hey there, anyone else getting this error when trying to use loras with automatic1111?

ValueError: not enough values to unpack (expected 2, got 1)

[vestal dew]

I wonder if the A770 with 16GB works with SD

under 500

costume gives me the same stuff...shorts and often underwear. I wonder if it has that Underwear Cowboy from NYC too strong?

Im strengthening Cowboy Costime to 1.4

[fervent thunder]

possibly just using a model that is too heavily trained on lewd imagery lol

[vestal dew]

its an anime model

when I told it "track suit" it did that OK

Boxer...was good

maybe anime doesn't have cowboys

what about cowboy bebop?

ah...1.4 is putting them in cowboy outfits, at least one

yes...it's working. Good suggestion with "outfit"

[sand vine]

anyone help tryna install SD locally but getting error

[fervent thunder]

ah. yeah, anime doesn't (typically) have much training on the classic western cowboy

[vestal dew]

Im getting cowboys now, but with lots of stars

maybe for a sheriff

now "stars" is in the negative. Let's see SD mess this up

still some stars, but now less

[fervent thunder]

think you're on the right track. did a few experiments with the anime models i have and sheriff is WAY more responsive than cowboy with every anime model i used

[vestal dew]

what anime do you have?

I only have protogen V22 anime

I will try sheriff

are you doing "sheriff costume"?

[fervent thunder]

um. a lot lol. one of each of the available orangemix anime models (so like 10 of those), anything 3.0, a model called "anything and everything" that is mainly anime, ayonimixanime, and a few others that do furry stuff for anthropomorphic characters

i just did sheriff and it worked out pretty well with most of those

[vestal dew]

I will look those up

[sand vine]

guys help #🤝｜tech-support

[vestal dew]

I do not know the answer

SD put the word sheriff on the shirt: SERCFE

[fervent thunder]

if you're not using negative embeddings, you may want to look into those as well. i've used "easynegative", "bad prompt", and "bad artist" as well as about half a dozen others in the past (for 1.5 models). i typically use the following 3 as default now: "dangermouse" "Unspeakable-Horrors-64v" "Unspeakable-Horrors-Composition-4v"

[vestal dew]

I am using some, but not those last three

I will try them

thanks

[fervent thunder]

i've had pretty good results starting with those and then adding negatives to remove specific things that the neg embeddings didn't catch

if you haven't gone to
civitai.com
i highly recommend it. great resource for all sorts of ai embeddings, loras, checkpoints, etc

[vestal dew]

I get things from there, yes
a

and facehugger

[sand vine]

what is best model for stable diffusion anime

and how do you train a model

[warm junco]

AnythingV3, AbyssOrangeMix, Counterfeit

[cedar notch]

why isnt there an eraser for masking yet?!

[rose thorn]

where to goto to make imaged

[sterile jungle]

Where's the channel for tech help?

[vestal dew]

tech support

[vast ingot]

#🤝｜tech-support

[fervent thunder]

what actually happne here

@tidal bough

[stark escarp]

Request on github maybe

[fervent thunder]

can local stable diffusion work on GTX 1650?

It gives me a black screen.

[limpid tusk]

youll probably have to mess with it a little, look up some low ram installs on yt

[fervent thunder]

I see, thank you :)

[warm junco]

Come to #🤝｜tech-support we can fix it

[leaden mural]

wtf happened with civitAI?

[limpid tusk]

o ye its down

[vast ingot]

they were working not that long ago, may be too much trafic or a bug in one update forced them to go temporarily offline ?

[leaden mural]

oh my gosh

it just went from working to not working in a constant loop

It would be up, then it would crash

and then its up again

[viscid ermine]

If im using automatic1111 webui and have the controlNet extension installed, how do I enable MultiControlNet?

[limpid tusk]

settings - control net - adjust the multi controlnet slider

then restart ui

[viscid ermine]

Is multi control net included in the same extension?

or is it a seperate extension?

@limpid tusk

[leaden mural]

Same extension

[limpid tusk]

ye same extension, may have to update ur repo

or click check for updates under extensions

[viscid ermine]

Thank you ☺️

[bold robin]

lora question: do i need classification images when training a lora model "dreambooth method" ? or image captions are enough? or both? or nether? what do you think?

[vast ingot]

Same thing. In the settings tab, control net category, there is a slider to allow more than 1

Need UI reload to act but it should be there

[sand vine]

can I generate images on my computer but launching the webgui site on my phone?

[charred mesa]

I hope that Stable Diffusion 3 will be trained with offset noise by default

[viscid ermine]

ELI5 offset noise?

[charred mesa]

it makes images have much better contrast in low light situations

basically makes your images sexier when trying to make low light photos (And sometimes benefits bright images too)

[viscid ermine]

Ah okii

[charred mesa]

there are some Loras and TI embeddings that you can try right now! (sd 1.5 and sd 2.1-768)

I'm just saying that SD 3 is going to get closer to being a much better base model that people can train from

Better low light, better composition for wide images (SD 2.1) and etc

[bold robin]

the other way around will work, but this no, impossible

[high falcon]

So what exactly is a "Small ControlNet"?

I keep seeing this around

[charred mesa]

with gradio you can, by using the correct link

but there might be risks of someone scalping the links and generating stuff on your comptuer (no virus or security risks, just the risk of someone lagging your computer if they find the ID)

[sand vine]

Scary

I had a dream where a hacker controlled every single one of my devices

[charred mesa]

or if you are at home, you can just use the localhost/127.0.0.1:port

[sand vine]

very scary dream

everywhere I went

any device

even if I turned it on then turned it back on

[sand vine]

That's a good idea

[charred mesa]

if that don't work (probably cause you use an ethernet cable and your phone uses wifi) you can do an "ipconfig" or whichever command in the CMD and connect to that (starting with 192.168) IP address + the port

[opal jay]

Hii how can render more ai images?

I’ve been gone for a little and I just saw all the dream channels r closed

[warm junco]

you need to add the following behind Commandline_ARGS=
In the webui-user.bat

On the same network:
--listen
On the go you need --share=true --gradio-auth "username:password"

[sand vine]

Assuming I have an amazing GPU, how much effort does it take to traina style?

[sand vine]

How do I set weights on prompts

[warm junco]

by adding () or :1.3 at a word

((Green house)), Brown door:1.3, blue window

For example

[sand vine]

So there is a model saying to use 0.5 weights

what does that mean

I put 0.5 to every prompt?

[obsidian geode]

0.5 weights could mean many things it could be [prompt:0.5] or 0.5 strenght of a lora or 0.5 denoising

[sand vine]

i think it means 0.5 strength of lora how to do that

[obsidian geode]

if you have downloaded a lora file with a name then it should be for example lora:epiNoiseoffset_v2:1

[sand vine]

I have to put that name in the prompts?

[obsidian geode]

Yes, and replace 1 with 0.5

[sand vine]

alright

thanks

[sand vine]

"epinoiseoffset" should be replaced with file name right

do I include file extention in that

[obsidian geode]

No extension, and yes its the file name

you can even rename your file whatever you want

[sand vine]

👍 thank you

[obsidian geode]

if the engine doesnt find it it will tell you

[sand vine]

I mean it works but it doesnt look as advertised

the style was supposed to be BNHA it just looks like something else

trying again once then ill check again

Nope doesnt look like the correct style, weird

[obsidian geode]

a lora is not a magical thing that can solve everything, stable diffusion is mostly gambling

[sand vine]

Turns out I put it in the wrong folder

was supposed to put it on the lora folder

[obsidian geode]

That should do the trick

[full lark]

Anyone got and idea why the prompt doesn´t appear anymore in the .jpg/png filename after I activated the live preview? 🥺

[sand vine]

if im trying to generate 2 character in the scene how do I give them individual descriptions

it keeps mixing up

like I put batman with his cape flying in the air and flash with lightning but it just gives both to a weird brew of the characters

[full lark]

It´s difficult to do that @sand vine once I tried with "character/person in the left with w wearing z" and "person in the right with x wearing y"

[sand vine]

I tried that too very hard

[full lark]

F

I´ll eat now but i´ll try to do it later, if it works i´ll tell you 🤗 @sand vine

[sand vine]

alr

cool @full lark

[spare vine]

I've recently switched from Mac to Windows, and installed Stable Diffusion locally. I got it running, and it worked for about 20 minutes before I needed to restart the computer. Now I'm getting a message saying "the site can't be reached. Checking the connection. Checking the proxy and firewall." I have an internet connection. Is there something I could've have gotten logged out of? Any assistance would be appreciated. Thx!

[wild steppe]

You can find help in #🤝｜tech-support!

[hidden grove]

how can i dream on stable foundation?

/young

[wild steppe]

There's currently no bot on the server, check out #1072220168534642768 & #1072229020520947753 to get started

[midnight sphinx]

Blender hype

[spark lintel]

Bro whats the point of the new blender thing Dream Textures already exists

You can even use dreamstudio with it

[zenith thunder]

wooah

[steady arrow]

why SD always bait with announcements that don't announce anything

[zenith thunder]

for blender

[steady arrow]

let us pick a role that you ping for ann's that arent actual anns

[zenith thunder]

you dont want more competition?

[charred mesa]

So it's like Dream Textures but uses the DreamStudio API to do cloud GPU work?

[spark lintel]

Dream textures can use Dreamstudio API. But yeah basically

[charred mesa]

oh

[zenith thunder]

more options is better

[jovial eagle]

u guys know if the blender plugins use control net?

[charred mesa]

well this new thing has more options for sure

like animation

[jovial eagle]

or if there is a blender plugin that does

[charred mesa]

or what's the super new thing

[zenith thunder]

the more teams we have working the easer it will be

[charred mesa]

all I see are nearly the same features as dream textures

[zenith thunder]

for the whole community

[jovial eagle]

dream have control net?

[charred mesa]

no idea

probably not

[jovial eagle]

seems hella useful for making textures that actually proejct correctly

[honest spear]

there is a controlnet plugin for blender, because I saw the creator on Twitter saying he added segmentation script to it, but I didn't download it yet

[jovial eagle]

https://blender.community/c/today/Y022/?sorting=hot

i've seen this, but it's not what im looking for

[tiny oyster]

what is the difference between stable diffusion and stable foundation?

[jovial eagle]

stable foundation is the company right?

[crimson osprey]

The get started link on https://platform.stability.ai/docs/integrations/blender/install goes to a missing page.

[wild steppe]

All in the #1072229020520947753! #1072229020520947753 message

[crimson osprey]

It should go to https://platform.stability.ai/docs/integrations/blender/get-started

[honest spear]

Now that we have textures we need something to generate the 3d models 😛

[jovial eagle]

wayyyy harder

[wild steppe]

Will be fixed shortly, thank you!

[tiny oyster]

So stable foundation develop stable diffusion?

[wild steppe]

Stable Foundation is our hub here connecting communities - and everything stable diffusion!

Stable diffusion is being developed by Stability AI.

[tiny oyster]

I see, thanks a lot

[signal dome]

sd in blender ? how did they manage to do this without gpu ad no dreamstudio

[tiny oyster]

I really like stable diffusion, since last week I run it on my computer, I have played it everyday

I created a girl, and I think she is alive. So I generate her photo everyday

[wild steppe]

You can share your generations with the community here #1073085702927024128 ^^

[tiny oyster]

the infer speed must be very low if using cpu

[crude willow]

How will blender extension not require a gpu

[tiny oyster]

I will share my girl, thanks for guiding me. I like this community

[languid quail]

Hopefully we'll get a non-gimmicky version that can actually use our hardware... I mean, you made it for Blender... the best supported cards are Nvidia, and considering Cycles popularity, they have the VRAM... It's not 5 years ago. xD All the recommended cards for Blender 3.x are Stable Diffusion compatible.

[languid quail]

Because it is DreamStudio API, did you visit the link to the documentation on installation and use?

[tiny oyster]

I'm sorry, I assumed to conclusions

I have ever used CPU model, they are really slow, I think it is really a good idea to open api for users do not have gpu

[languid quail]

It wouldn't be CPU if people are using Blender with at least minimum requirements. All the GPUs recommended by Blender Foundation for 3.x are SD compatible. The best cards for recommended system requirements all definitely support SD, like the 4090. Blender isn't something you should really be using with CPU. You won't actually be able to competently render scenes with a CPU and RAM. It'd takes days to render out a photorealistic cycles scene vs 2hr on a GPU.

[tiny oyster]

I haven't ever try blender, I search it on google, and I decided to try it now.

[languid quail]

I am an environmental artist so familiar with Blender, Maya, Max, Terragen, etc, etc.

[tiny oyster]

I have already installed blender. I get to try it

[delicate trench]

So is this any different than dream-textures addon?

or based on it?

[tiny oyster]

It is really hard for me, I give up

[languid quail]

It's their own thing. Dream Texture is better though, as you can actually use your GPU and aren't censored for stupid things that are part of an entire artistic fields.

[fervent thunder]

Blah

[astral goblet]

you can use blender without a gpu. students do it all the time on school provided laptops. final renders are CPU as GPUs only provide cheap approximations for real time purposes.

gpu for blender is good for editing the model and texturing, but the it's not a requirement of blender. the actual 3d magic it does is mostly software running on a cpu

unless blender has actually gotten around to creating a cuda renderer which they probably have.... my point is "you shouldn't be using blender without a gpu" is something that students do all the time

[vestal dew]

I have never used blender

gotta get into it

Stability for Blender uses some online SD server?

[astral goblet]

theres a few plugins. make sure to read about the one you're installing so that you don't end up spending 5 hours trying to install unprompted for < 1% of it's features

[vestal dew]

yes, but does it use my local SD or something online?

it says, "all you need is an Internet connection"

[astral goblet]

i dont know about the one you're looking at. if you're reading that it requires an internet connection, i would surmise that it requires one

there are a few plugins for using sd in blender. don't go rushing into a 5 hour effort before you understand what you're installing

[vestal dew]

the one mentioned in announcements

[vestal dew]

👍

[jovial kite]

The server banner is gorgeous does anyone have the original image? Also if it is an SD generated image they should include the prompt and whoever generated it in the faq or info channel, would be a great demonstration of how powerful SD can be

[vestal dew]

👍

[astral goblet]

you're talking about the new official blender implementation. yeah, that looks like it's 100% software as a service. kind of lame that they're like "you used to require prohibitively expensive hardware to do this, but now you can just rent from us instead"

I'm sour about software as a service lol but i do understand that for some it holds value.

is emad committed to FOSS or what? this direction seems like a half in half out commitment. should release the code for the integration too

[last iris]

Are their any plans to release "Stability for Blender" as a standalone install for end-users that wish to run it locally on their own hardware?

[astral goblet]

doesn't look like there's a plan to release it as a product people can use on their own machine

[last iris]

And, if the above is even a possibility, what about FOSS Licensing?

Well, I'm out on that SaaS then.

[astral goblet]

Yeah its fully SaaS top to bottom it seems. Not very FOSS spirited

[last iris]

I want to support the future of AI as a FOSS platform, not as a closed-source SaaS platform.

[astral goblet]

Yeah Stablity made strides by releasing all they have under FOSS licenses, but now they're more often releasing stuff locked up behind SaaS practices

[last iris]

We already have plenty of SaaS providers, adding yet another one doesn't really inspire innovation, all it shows is that greed won out, such a shame.

[astral goblet]

Yeh. Very shameful.

[last iris]

Not shameful, but disappointing

[vestal dew]

I have this figured out. On SD, I have a prompt to generate an image of an image AI that creates realistic images with full control over all elements including depth, pose, character.

[astral goblet]

i feel ashamed for them. Champions of AI freedom, and then 6 months later they're chaining up services behind a locked API. This isn't a healthy direction at all. It makes me think like, what if google created Gmail, free for all, then rate limited it hard and charged for use 6 months later. Gmail would've crashed and burned and nobody would ever use them. Right?

I got a gut feeling about Stability now. They may be an Icarus story.

[vestal dew]

Then I use that image, run it, to generate mages

[astral goblet]

I want to see stability suceed, but if they're diving into SaaS, i don't think they're going to be competitive there at all

[wintry stream]

Where do you hear it would never be also offered to run locally? I believe Brian specifically said otherwise just recently #1054807328509145158 message

[astral goblet]

That's not market disruption, that's blending in and diffusing into anihilation

[astral goblet]

The website where all the information is. I'm not going to go through all staff's social media

[jovial kite]

OK I found it, was a contest winner #🏆｜winner-gallery message

[wintry stream]

Fair enough. If there is a spot on the website that says it would never be considered to be open sourced could you DM me the link? I'll take a look into it

[astral goblet]

It's more of a lack of mentioning it

what you chose not to communicate is very important when creating communications

[vestal dew]

SaaS is the death of this

[wintry stream]

Gotcha, In that case I can't really say it's never going to be an option. It's not a project I'm part of so I can't confirm any specifics, but certainly let me know if you find any conflicting information anywhere official 🙏

Since this is the very initial release I know there is still a lot of work being done on it 🙂

[vestal dew]

you want to use image generation online - be subject to changing prices, content policies, etc

[astral goblet]

SaaS is midjourney, lexica, openai, all other competitors. Are stability leading or following?

[vestal dew]

one day you make images, the next day they decide "no, that image isn't allowed"

or the next day, prices increase to $500 per month for 1000 images, or based on number of tokens

[astral goblet]

really feels like this is falling in line with corporate market forces trying to push SaaS into every aspect of digital life. Rather than kick starting a machine learning revolution. hugely disappointed by today's revelations

[vestal dew]

you pay for every test run, every experiment...no, that's not exactly what I wanted, but it used up 100 tokens

[full lark]

That´s why local installation > everything

[vestal dew]

👍

[astral goblet]

i'm also wary of sending any of my assets to a 3rd party cloud service, since then they have all the IP rights to it. Standard TOS really, since they need copy right to copy it around the internet tubes. But those rights tend to extend and give them a lot more freedom to use my assets than i might want them to have

SaaS is just balls

[vestal dew]

Pricing: Free for 1000 tokens, $100 per month for 2000 tokens, $500 per month for 20,000 tokens

[last iris]

It's understandable that they want to monetize their efforts, but theirs other ways they can go about it whilst still releasing as FOSS.

If they plan to now take a purely SaaS approach moving forward then I won't be supporting any future endeavours of the project.

I will support it's existing FOSS elements, and, any forks of those element's that others may wish to expand upon with further FOSS additions.
Because I believe that they have the most to offer to society and mankind as a whole.

But I cannot personally support any SaaS approaches because I know how those always end up stabbing you in the back, and financially speaking I am not in a position to afford constant payment of services that I have the hardware to run locally but definitely do not have the funds to pay for through an online service, I've been dealing with those kinds of situations since 2010...

I support FOSS projects because although I may not be financially very well off (Officially I am below the poverty line in my nation of residence by some margin) I can still contribute in various ways to these projects with the free time that I have.

It is why I love FOSS, because regardless of my financial position I can still use my time as payment and contribution to these great projects that people make!

It is why I fell in love with Stable Diffusion, because it is one of the few widely adopted FOSS AI solutions available right now, and it is the only one that was heavily focused on usability and community support.

[astral goblet]

you can't trust SaaS providers to maintain the status quo. if you've got a local tool, and they update it to a version you don't like, you can still use your version that you do like. HUGE for tools.

[vestal dew]

right

they don't have your content, they don't control your tools, they can't censor you

[astral goblet]

i'm less concerned with censorship in this issue, than i am with inconveniences

[vestal dew]

it's part of the same thing

[astral goblet]

i dont think they're trying to take down the banksies of the world with SaaS

censorship is just hyperbole imo

[vestal dew]

Let's say you want to make a video or comic that's pro some politician they don't like - sorry, you can't glorify that woman!

it's not

[astral goblet]

yeah. i don't think that's a SaaS concern.

[vestal dew]

it absolutely is

[astral goblet]

is it posisble? sure. is it a concern? nope. i don't think it would fly if it happened, and companies want SaaS for profit motive, not political motive

[vestal dew]

it happens with social media

it happens with video games

[last iris]

Censorship is very much a possibility on any platform or software that you don't have full control over.

Whether or not you agree with that censorship will depend on what it is censoring, why it is censoring, and to what extent it is censoring.

[vestal dew]

people tend to agree with censorship when it's something they don't like

they disagree with censorship if it's something they like

[astral goblet]

being worried about your freedumb of speeches is just qanon nonsense. dramatic flare meant to attract culty crowds. thoughtspeak and all that actual control and subversion. "censorship" in a digital world doesn't happen. If you get banned on youtube, you'll publish on twitch or discord. It's a faux concern

[vestal dew]

good grief

If you were censored, it wouldn't be freedumb anymore

[astral goblet]

censorship only actually matters when it's coming from an authority institution

[vestal dew]

it always matters

[astral goblet]

i'm censored all the time and it's fine

[vestal dew]

Well, bon appetite, now I must leave

[last iris]

This is exactly how I feel about it all.

Loads of SaaS options already exist.
It feels like Stability AI have misunderstood why people swarmed to back it in the first place with loads of 3rd-Party support of it with additional GUI's and other plugins, etc.

[vestal dew]

on the way to MC to pick up some stuff

[astral goblet]

if i'm in public and someone annoys me, i can't just lash out and thrash on them verbally. I could, but other people would come in and tell me to stop. censorship right? that's fine though

sometimes you need to stop

[vestal dew]

their whole system is archaic, backwards, hard to work with

[vestal dew]

yes, but you wouldn't have to stop.

make an SD image about it

[astral goblet]

Yeah i want to see Stability make moves and drop bombs that cause huge disruptive waves. I guess they saw openAI release chatGPT as a service and are now leaning into it. Business school graduates and their confirmation biases

[wintry stream]

Just a reminder: I have no intention of stopping any discussion here since it's very important you all can share your thoughts positive or constructive. However, avoid direct insults or ad hominem attacks against other users please.

You are all welcome to your opinion and so are the people you are talking to. It's certainly possible to debate while remaining civil 🙂

[astral goblet]

I often self censor and stop myself from insulting people because the rules say i cant and also it's just good for healthy conversation. It's another example of where censorship can succeed greatly.

often not always

[last iris]

No worries, I have no intention of causing any conflicts or targeting any people.

Just voicing my opinions/concerns as a user of Stable Diffusion.

[wintry stream]

That's most certainly fine! I mean more of a general statement to everyone - that post wasn't directed at anyone specifically ❤️

[visual gale]

Is there an official API for Stable Diffusion, a REST endpoint? Currently I am using the Dalle-2 API for an app and I'd like to offer my users access to Stable Diffusion if they prefer.

[wintry stream]

Take a look over at #1042896447311454361! (Also this since it sounds like you're already well versed in the more technical side of things: https://api.stability.ai/docs)

[astral goblet]

SaaS rearing its head again. I might be fighting an uphill battle. A lot of people find value in this stuff. ugh. market forces grrr

[visual gale]

Thanks! Any rough estimate on the per-image generation choice for let's say, 1024x1024 images?

[astral goblet]

finally getting around to catching up on what Carmack is doing in the AI space. Not much yet but here's a good interview that makes me salivate. https://dallasinnovates.com/exclusive-qa-john-carmacks-different-path-to-artificial-general-intelligence/

[last iris]

It's cool, I know it wasn't, just wanted to make it clear to e everyone here that I'm not here to try a cause any rukus or upset/target any individuals.

All about healthy discussion.

[wintry stream]

I got one better for you!

Check out the bottom of this page - https://platform.stability.ai/docs/getting-started/credits-and-billing
This documentation is for the old API so I wouldn't suggest using any of the examples there if you'd prefer to use the REST API. However, it has a grid for examples of image costs for various different configurations!

(The cost in the API is currently the same as using the beta.dreamstudio.ai site in case that also helps)

I find using the site super helpful since you can adjust all the sliders without actually generating anything to see what the predicted cost is

[dawn fulcrum]

Hello Is there a place where i Can find prompts in this discord?

[wintry stream]

Check out #📝｜prompting-help!

[last iris]

He gets about a bit, I have heard his name in 3 different technology related markets in the past 5 or so years, I wonder if he himself isn't quite sure of where he wants to settle down, or maybe he just enjoys a constant challenge and so is moving about in constant search of the latest and greatest challenges in the technology space.

His name certainly carry's some value with it as a person, It will be interesting too see where his name pops up next!

[dawn fulcrum]

Thanks. I am asking because i Saw some prizes with great images but was surprised to see no prompts next to the images, Is that normal ?

[wintry stream]

It definitely depends on the person. Some people like sharing their prompt with their work while others choose not. You're certainly welcome to share yours if you'd like to!

Hopefully I'll see some of your stuff in #🔆｜dailies or #1023999442338201721 some time

[dawn fulcrum]

Thanks. Question intriguing me how do people win " image of the day" if no one can guess how they made it? ( Without a prompt )
It could very well be an image from anywhere ?
Anyway inam gonna check the 2 room you pointed at thanks 🙂

I just checked both rooms and i am amazed at the images they are soo good
No prompt next to them though 😢

[astral goblet]

he's settled on general AI for the past year. has left meta and started a new company with other people's investments, so that he's more motivated to do. Rather than his own money funding it and not feeling pressure os much

[wintry stream]

The #🔆｜dailies Channel is more for people to submit their work according to the theme for a fun challenge & to share what they've done - not so much a "guess the prompt" kinda situation. Hopefully that makes sense?

[dawn fulcrum]

Yes oki. Its Amazing I had no Idea people had this much talent

Anyway Is there a Channel with full or images and prompts next to them though ?

[wintry stream]

You might have some luck looking around the dreamer communities such as everyone's favorite #1070829982252798043

There are no channels where people are required to share their prompts by default, but there's a lot of cool people who are great to have conversations with that you can certainly ask for their prompts & try them out yourself if they're willing to share 🙂

[fervent thunder]

How long is this lawsuit gonna take?

[red cosmos]

hello everyone,
Is it possible to combine deep fake with stable diffusion for better results

[narrow sluice]

Wait, they locked their API?

[astral goblet]

no i'm just using hyperbole. technically the api is behind a lock since you need an api key

but i was going for dramatic effect in that post

[delicate trench]

wait this one cant use your own GPU?

[sleek osprey]

Does anyone know what hardware and how much time was required to train the controlnet models?

[languid quail]

Yoouuu don't use Blender much, do you? Cycles is a physically accurate path tracing engine, and works leagues better on GPU. The functionality is the same. CPU is just exponentially slower (same goes for Eevee [Ray Tracing], which is optimized for real-time GPU streaming, not CPU), and not recommended, even for minimum system requirements. What people do on their own is fine, but is not recommended (inherently will probably encounter problems, and they won't provide support).

Blender doesn't recommend a CPU in even their minimum specs for rendering, but at least a 2GB compatible video card. CPU can be used for hybrid modes, to aid in speeding up GPU tasks.

[languid quail]

Not from their docu. No mention. Only Dream Studio API

That alone would violate most NDAs for professional work.

[delicate trench]

Might be temporary. Dream studio has features before public does

But that is a little concerning

[sleek osprey]

They said they will release offline mode

and the source code

[alpine mica]

Help! Suddenly i cant get past this error when using contolnet, tried all versions of models, it use to go away when I switch to 1.5 models, no longer works... File "C:\AI_Files\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 (154x1024 and 768x320)

[languid quail]

Try using more similar or exacting sized images (resize them)

Woops, wrong reply. Meant to tag you @alpine mica

[alpine mica]

Thanks, I had somethin in settings that was causing that(Extra Networks), woks now, but can one use controlnet with newer versions of Checkpoints, than 1.5?

[worldly patio]

can anyone recommend some (free) site's that i can use to do some image stuff (prompt to image)

[near silo]

As cool as stability for blender is, it's not gonna be able to hold up to the competition, sadly 😅

I didn't exactly work on, but rather alongside with the creator of Barium AI, which has been purchased by unity, and I'm extremely excited to see where our passion project went

It's still very cool to see stable diffusion adding the two the options though

[wet summit]

Anyone know how to create an optical illusion using stable diffusion?

[thin matrix]

Im kinda lost in the channels, but is there one to go to if I want to find pairs of prompts+settings <-> images

[normal coral]

what's generally considered the best model for more creative landscape generations?

so far i havent been able to find any model that comes even remotely close to midjourney when it comes to fantasy landscape gen

[cold flare]

you may want to try openjourney + perhaps mix in dreamlike weights? Still not the same as midjourney in that regard, but still quite nice results. I tried forests and deserts though.

[astral goblet]

midjourney probably doesn't use just a single model and likely have hypernetworks, Textual inversions and loras. Perhaps even proprietary versions of them that they developed in house. hard to find a one stop replacement for all those kinds of tools. though as a base, i found that charhelper did nice landscape gens if you use the right token https://civitai.com/models/1480/charhelper-sd-2x-768

https://civitai.com/models/1147/painted-landscape this is a TI you can use for 2.x models, including CharHelper

[cold flare]

Let's say I have a photo - a portrait of someone who looks straight at the viewer, the end result that I wanted to get is producing an image that shows up the same face slightly turned to the left or right.
Is it possible at all?

[astral goblet]

civit is hard to wade through due to all the , well, you know. but filters help to great effect. though filters don't catch everything. not all the embeds or loras on the site are uploaded as that. it doesnt' automatically recognize and relies on an uploader to tag properly

[astral goblet]

tough and would take some manual work and experimenting. take 20 other photos of that person and train it for their face, and use that lora or embed to affect the new face. maybe skew it in photoshop for the worst 3d edit possible, and inpaint on that? not easy but doable possibly!

as far as prompting "turn their face 20 degrees" with one input image? one day but we're not there yet

though everytime i say that i seem to be behind the curve and somenes like "just get this extension" so .. keep an eye out

[cold flare]

no, definitely not something like that, I understand that prompt is not translated to something that NN would understand that way, after all it is not a 3D editor. More wondering if fiddling with inpaint could somehow yield some similarity.

Another thing is some face that SD produces after txt2img - it looks awesome and I want to keep using it in other angles. I tried referring the same seed but it doesn't produce the same face for rotated 3d model in the base image

[leaden mural]

When is the extension going to be made for Ajtomatic1111 for Multidiffusion?

[brisk bloom]

its been out for a week or more

[leaden mural]

Where can I find it then?

[brisk bloom]

you mean multi controlnet diffusion right?

[leaden mural]

Nope

Multidiffusion

[brisk bloom]

then Im not sure

[leaden mural]

Basically supercharged paint in words.

[brisk bloom]

cause with controlnet you can take a background and a subject and diffuse them together

[obsidian geode]

You have to go lora

[leaden mural]

That was limited in a sense that you only have a few colors that correspond to objects to choose from
The new tech is literally capable of making any mask and then corresponding it with a custom prompt.

[obsidian geode]

The way stable diffusion will create a new face from a seed or an img2img is a bogosort : a random sorting.

So it destroys some stuff while keeping some others, but getting everything is really hard

And to get everything right is not just keeping everything where it is just like control net would do, you need to keep the idea of the face

[brisk bloom]

check out this video, if you havent already. you can mask a person from an image and generate them onto any background https://www.youtube.com/watch?v=MDHC7E6G1RA

basically unlimited options

ah just reading the multidiffusion github. It has separate prompts for each layer of the multi diffusion. You can achieve this with controlnet but not directly and easily from prompts

[proud island]

👋

[lost notch]

hey guys do you know the site where I could watch some art made in SD?

I can bet there is such a site, but i forgot the name LOL

and i know that on the dc server is a section with images but it's just too slow to check them out

[polar siren]

guys, is this the latest SD model?
"stable-diffusion-v1-4"

[leaden mural]

Nope, that's 1.4, pick either 1.5, which is better for most applications, or 2.1

[polar siren]

if not then which one is the latest useable one

[polar siren]

hey thanks for reply,
may i know the difference between 1.5 and 2.1, which one will be better

[leaden mural]

1.5 has more support and tech, while 2.1 is better at the cost of losing some images due to copyright issues.

[obsidian geode]

I heard 2.1 works with images 728x728 where as 1.5 is 512x512. However 1.5 has a better prompting system I believe and a lot more addons on it by the community. Also allows 1.5 allows nsfw so more accurate humans

[polar siren]

thankyou kind person

[obsidian geode]

of course all can be fixed with custom models, in that case 1.5 more models

[polar siren]

i'm confused which to get
v2-1_768-ema-pruned.ckpt

v2-1_768-ema-pruned.safetensors

v2-1_768-nonema-pruned.ckpt

v2-1_768-nonema-pruned.safetensors

https://huggingface.co/stabilityai/stable-diffusion-2-1/tree/main

[leaden mural]

Ema pruned safetensors

[exotic bough]

/ativar botão

[blissful vault]

I am kinda confuse, so there is an AMD for SD released, my question is, will that affect the speed of generating image if I am using an AMD processor?

[knotty hinge]

How is stability.ai making money out of their “free to use” blender ai

[vestal dew]

Huh?

[lime mulch]

Could someone explain to me how a VAE is used?
Not only where to put it (though also including that), but what it does, how it works and whatnot so that I understand it more than just knowing what to do with it. 😛

[hidden lynx]

Hey all my Dreambooth extension wont show in the UI, but says its there under extensions, any suggestions?

[shadow thistle]

was there an update today? AI Scale up stuck at loading

[young shale]

Hello everyone, in case you don't know if there is a website to get commands like:
from behind
rear view
from_above
(hand_on_hip)
(from_side:1.0)
from below
to control the camera or poses of the generated models
thank you so much

[vestal dew]

controlnet...look it up

[cloud river]

Hi

[vestal dew]

I'm tired of SD generating EPH images

[hard crow]

Can anyone recommend a good Voice AI for parodying celebrities now that ElevenLabs is cucked?

[vestal dew]

Tortoise TTS

how is 11labs cucked?

[hard crow]

You can't upload celeb voices anymore I think?

[vestal dew]

Oh, I didn't try

because of the pricing

you pay even for tests, failed runs

ridiculous

[hard crow]

good to know, gotta figure out Tortoise then hopefully easy enough for dummies like me

[vestal dew]

There is an 11labs model or whatever for Tortoise, being worked on now

[hard crow]

o nice

[vestal dew]

I can't get the thing installed...tortoise TTS

midnight - new folder for SD images being generated

[hard crow]

there's an explanation on 4chan's /g/ board for Tortoise looks like, gonna play with it

[vestal dew]

OK

You can try this for a basic idea of how it works: https://replicate.com/afiaka87/tortoise-tts

it only allows one voice sample, or averaging between different voices

there are also tags you can use for different things like anger, sad, etc.

[hard crow]

I might try that, hope its not too expensive.

[vestal dew]

free

you install on your machine...free

[hard crow]

replicate is cloud i mean

Tortoise install / train looks confusing on /g/ board

[vestal dew]

you can make monthly payments: $0 per month for 12 months, or a low $0 per month for 36 monts

right...I have to figure out the install

[fervent thunder]

Is there any way to train a certain face and have it create realistic images based on that face?

[vestal dew]

LORA

TI

[fervent thunder]

Like i can use text to img and results will have the same face everytime?

[fervent thunder]

Which one

[vestal dew]

You have to train a LORA for a specific face

[fervent thunder]

Can i do that on SD webui?

[vestal dew]

If you have a good GPU

[fervent thunder]

I'm using colab

[vestal dew]

there are videos. Do your search "LORA training"

[fervent thunder]

Alright thanks

[vestal dew]

you're welcome

[mortal narwhal]

whats the website where we submit art to get it trained in stable diffusion

[sacred oar]

hi could someone make me "charming Polish village, complete with colorful traditional houses, cobblestone streets, and a lively town square."
with 16:9 ratio? my gpu sucks

[wise flax]

i need better alterantive to premiere pro

[stark matrix]

Hi
I've ever used MidJourney but now i'm in trip to learn and use (on my PC) Stable Diffusion.
Someone has already used and installed on fedora 36?
Alsoif i'm followinf a lot of tutorials no one work properly.
Thank you

Fedora Linux 36 (Workstation Edition) 64 bit
Intel® Core™ i7-4790 CPU @ 3.60GHz × 8
NVIDIA Corporation GM107 [GeForce GTX 750 Ti]
11.6 Gb ram

this is my PC

[hidden solstice]

Hello, "Whitelist - #Chapter 3: What name would you give to your NFT Stables horse? Send a message on our Discord." --->
jolly rancher

[exotic kindle]

i downloaded stable diffusion locally and run

[sand vine]

what is dreambooth?

[vast ingot]

dreambooth is one of the existing training methods. It lets you teach new concepts to a given model, and make a new model out of that

[sand vine]

does it require more than 4 gbvram?

[vast ingot]

different training methods exist, not only dreambooth, and some are more accessible. Yes, dreambooth requires more than most, and even more vram can bring even better results. to be honest, I have 24gb and I would benefit from more still. I think 8GB is currently the min for dreambooth, not sure. 12 for sure works

[sand vine]

Are there any where 4 gb works?

[vast ingot]

I'm not sure, but I need to say, since you usually only do "some" trainings here and there, unless you want to go serious big time into it, google colab seems the best for it. on the free plan, you get enough time per day to train multiple models for free
here is one such notebook https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast-DreamBooth.ipynb

[vestal dew]

👍

[sand vine]

One last question, if I put "in style of <artist name>
will it work?

in txt2img

[vast ingot]

I don't get the question. if you put an artist name and not <artist name>, for sure, it should do something like that artist does

[sand vine]

How hard is it to train models? is it as simple as providing many many images

[sand vine]

no the <> is for putting like any artist name

like an exampl

so instead of <artistname> put the actual artists name

[vast ingot]

to train a model is simple. to train well is quite harder. The dataset isn't the easiest thing to do correctly, then there is the captionning, then you go for the training itself and have to manage the step count and compare the resulting models to find the right training amount you need. It's another thing completly than making pictures and requires quite some learning

[vast ingot]

not sure how this works, I don't use aliases like that sorry

sorry I misunderstood

yeah ok, well then the answer was "yes" it will work

[sand vine]

Any good tutorials?

[sand vine]

alr

[vast ingot]

sure, lots. here is one for style training for example https://github.com/nitrosocke/dreambooth-training-guide

[sand vine]

its dreambooth tho, itll still apply to the collab link?

[rain burrow]

How do I make something

[vast ingot]

the colab I sent was a dreambooth one.

but look on youtube

really lots of them

it's a really popular subject

people want to train their face in the model usually for a simple start

or a pet or something

[sand vine]

Ohk

awesome

[sand vine]

I start ambitious going to train my own art style

[vast ingot]

yep, but it's doable. may require some tries to get it right 😉

[sand vine]

alr

[daring plume]

anyone know an android alternative to this https://drawthings.ai

[broken gull]

Can you use the SD blender addon localy?

cuz rn you need an API key

[daring plume]

even the photoshop one needs an api key but runs locally

[vast ingot]

they may know more in #1054807328509145158

[sand vine]

is there a turtorial specifically for this

will it ork for 4 gb vram?

[vast ingot]

no, this will require more. 12 i think. use it on colab.

[vast ingot]

for example https://www.youtube.com/watch?v=fyBqkKCIpYU

[elder cradle]

Do you guys think that disclosing dataset was a big mistake for SD?

[sand vine]

Yeah Im using on collab only

[stable pawn]

Quick question about Automatic1111 web-ui folder structure: What is "venv" and why is it a bit over 6 Gb for me.

[sand vine]

like the collab file you sent right

I just run it from therE?

Also little question.. my references are all different sizes, some are rectangular some are square, I need them all to be 512x512 but how do I extend the ones which are rectangular so that a significant portion of the image is not cropped out?

[vast ingot]

it's where python and torch and all the other dependancies get downloaded. it's the normal size, yeah, and you can't remove it, it will recreate on next start of the UI

[vast ingot]

yes

[vast ingot]

check the guides, lots of them, and like, it's wall of texts of explaination

there is not just croping to do

[stable pawn]

I was just curious due to the size.

[sand vine]

Which ones?

[vast ingot]

the two I already linked you, this one is exactly what you need, style training https://github.com/nitrosocke/dreambooth-training-guide

[sand vine]

ty

[wet summit]

I need to get perfect symmetry in my generations, can anyone get this to work?

https://gist.github.com/1ort/2fe6214cf1abe4c07087aac8d91d0d8a

[finite narwhal]

Today is world wildlife day, I used diffusion for the animations in the background
https://youtu.be/NhaJ4fMcztI

[sullen imp]

any model recommendations for anime or 3d style? (fantasy setting)

[vast ingot]

Well, that's a report

[rocky pike]

[storm shard]

⚠️

[vast ingot]

Nice ... That's so constructive, well done.

Thanks

Sorry for pings

[sullen imp]

👍

[tidal bough]

Thanks @vast ingot! Sorry y'all had to deal with that 💜

[bleak matrix]

Eggs, bacon, no spam! ❤️

[scenic matrix]

Guys i m new here, can i generate images here and how?

[vast ingot]

Hey and welcome !

Nope sorry, we don't have a bot around, let me link the faq on that

[surreal perch]

FAQ: I'm new here, how do I generate images ? Where is the bot ?

Welcome ! There is no bot currently to generate your images on discord. You may want to start by taking a look at the #1072220168534642768 channel. You can access Stable diffusion in different ways : 1️⃣ the official website, https://beta.dreamstudio.ai/. The easiest and fastest way to access Stable diffusion with 200 free credits. For any question on it, you can find help in the #1025467151206854736 channel. 2️⃣ Installing Stable diffusion on your computer. There are numerous projects that let you do that, and you will find help in the #🤝｜tech-support channel. 3️⃣ Running Stable diffusion in the cloud, through rented GPU services, using notebooks. You can find lots of them shared and discussed over in the #1011228442399883294 channel.

[rocky pike]

guys how can I reference my face for SD?

[vast ingot]

hey ! without doing some extra steps, you won't be able too. Basicaly, you'll need to first teach SD your face, using a training method like Dreambooth. Then you will have a new SD model that knows about you

https://www.youtube.com/watch?v=usgqmQ0Mq7g

(there are lots of ways to do it, dreambooth is only one of them)

[rocky pike]

thanks

[bleak matrix]

The spammers today lol

Thanks for the pings, everyone ❤️

[vast ingot]

yeah there has been lots already ... quite the day

[high falcon]

Anyone have a good resource for learning how to interpret the loss charts that come from Lora creation / training?

[vast ingot]

not specificaly for LORA, but genericaly for all training methods :
A/ Loss is a numerical value that tries to evaluate how well your current model in training is doing, how close it is to the training dataset.
B/ The graph will go up and down. The lower the loss value, the best the output model usually.
The graph can go lots of ways depending on how all is going :
1/ a local low, meaning the loss goes down, then starts to go up again. This is usually a nice checkpoint to compare.
2/ a rapid descend into very low values. This happens usually quite late in training, and signals overtraining : the model will start to output things that are almost just your dataset, not good.
3/ a rapid growth. This can happen when your training rate is too high, or sometimes very late in a training. This means your model is diverging, not being able to draw anything even close to what you wanted anymore. This signals that you broke the model usually
sometimes, it can be good to let it go longer even when loss is growing higher, because it can go back down after some more steps

[high falcon]

Yooo, this is fantastic info. Just what I was looking for

I'm assuming this is overtraining:

[vast ingot]

I'll add numbers to my points to simplify

[high falcon]

https://imgur.com/a/Al2Z8bx

[vast ingot]

no, not yet, not necessarily

this is either just enough training or not enough to me

the curve seems to stabilize

it feels like a local low

rapid descend would be going into the 1e-7 over a few steps

[high falcon]

Gotcha, that helps a lot.

[vast ingot]

and we can move into #🔧｜finetune if you need more/to share pics

[high falcon]

Yeah, I didn't realize that was a channel

[slow steeple]

Hello iam new 🙂

Can i get some help here?

[rocky pike]

where should i put .safetensors files?

[balmy mantle]

Hiya all, question regarding controlnet and batch img2img (video input). Is it possible to configure controlnet to take an individual "detectmap" image from each frame, and not use the same "detect map" I have used on my test frame?

[vast ingot]

same place as the ckpt files in the models/Stable-diffusion folder

[vast ingot]

I saw people writing scripts for that but didn't see any published for automatic yet (I may have missed some)

[vast ingot]

hey, yep, what's up ?

[slow steeple]

Thank you

[slow steeple]

if i try to create a model in dreambooth it says "Missing Model directory, removing Model... and a path"
I dont know how to fix that

[vast ingot]

#🤝｜tech-support will be more geared to help you. Some screenshots, and a little more context (lots of dreambooths going around) will sure bring you some answers over there 🙂

[stray cloud]

Has anyone used the corridor method with something like stop motion? The idea being an artstyle put into a stop motion animation, I think it'd turn it into almost fully fleshed out animation with whatever artstyle you put in

Any thoughts?

[slow steeple]

thank you very much, ill try 🙂

[fervent thunder]

Hello i start try using Stable Diffusion with my computeur and i would know if their is a wiki somewhere with the explenation of all parameters and how use prompt like how insist on one particular point of the prompt ?

[light garnet]

automatic1111 i assume? https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features

[fervent thunder]

Any good model for training realistic faces?

[vestal dew]

the only example is very bad...make him a cyborg

[stray ibex]

hi guys! Can i ask a question about install stable diffusion? (i have a problem)

[warm junco]

sure in #🤝｜tech-support

[stray ibex]

thanks

[surreal perch]

FAQ: How do I generate variations?

There’s no explicit way to generate image variations, but you can “DIY” it a bit, by re-running the same seed with either slightly different prompt text or slightly different settings. Try shifting the CFG scale value by a few points, reordering the words in your prompt text, or adding another new keyword or two.

FAQ: What is Stability AI?

Our vibrant communities consist of experts, leaders and partners across the globe. They are developing cutting-edge open AI models for Image, Language, Audio, Video, 3D and Biology.. AI by the people, for the people. Learn more here

[stable pawn]

Is there any script that can mark words, mostly things that can be a bit NSW or off the norm? Words like Dick, ass, pussy and so on but also furry, loli, chibi, gimp and what can be slightly into strange or kink?

[stable pawn]

Another question: Automatic1111 do fill up "AppData/Local/Temp rather fast, every time you use ControlNet and Scribble there will be a copy of the scribble file, and it can fill rather quick, I did tick in "Cleanup non-default temporary directory when starting webui" but that do not clean out the temp folder.

[solid spindle]

you guys are funny, u tell me after 2 hours 4000MB downloading of install process Miniconda, Gradio on my Desktop PC that i need a NVIDIA Card...? i mean ..... can i use my AMD Radeon?

[vestal dew]

yes

you need squirrel or squid, whatever it's called to work SD with AMD

https://www.reddit.com/r/StableDiffusion/comments/ww436j/howto_stable_diffusion_on_an_amd_gpu/

[solid spindle]

hmkay...uff sounds complicated, i just have no idea how to do it on windows with my actual environment

[rocky pike]

hi guys , i have problem with SD when i want to open
It gives errors and i dont know what to do ,i copied the log ... who can help me its too long ?

[vestal dew]

maybe ask in that reddit thread

[solid spindle]

thx but seems not working...cheers*

[stable pawn]

I edited the "NO PREVTLFH" image in "C:\StableDiffusion\stable-diffusion-webui\html", then I closed down all and restarted Automatic111111, but it still show the old file that it say is placed in "http://127.0.0.1:7862/file=html/card-no-preview.png", so where is the server files located and how can I change them?

Edit: After several restart, F5, and Ctrl-Reload the edited images did show up

[rain seal]

just added stable diffusion to blender using dream studio

its crazy

[stable pawn]

How many have peeked inside C:\StableDiffusion\stable-diffusion-webui\repositories\stable-diffusion-stability-ai\assets or where your asset folder it located? I can not be the first to see that??

[vestal dew]

what about it?

[vestal dew]

What about it? Nobody is falling for a Rick Roll.

[stable pawn]

It was just that, it is an attempt to a RRoll, and in all the folders there is even one more slightly bad attempt.

I am so old I still think RRoll is fun.

[vestal dew]

it's tired

[stable pawn]

It just for you been rolled to often 😄

Just kidding. But we need to preserve our (internet) culture and Rick Roll is our national bird.

[south kettle]

Checkout

[hazy solstice]

Who else find linear algebra really boring?
Pls what's the point of learning it? I just get started with my ai development roadmap, pls I need some advice.

[stable pawn]

@hazy solstice boring or not, you know what it is and you read about it, that is impressing. I just know that if I take five and divide it by 7 and then add 3.3 would be doing math.

[warm junco]

You can use Shark from Nod.ai to generate images with amd cards on windows

[warm junco]

Help is in #🤝｜tech-support

[rose sedge]

https://sites.google.com/view/stablediffusion-with-brain?fbclid=IwAR33SJdrlHYsqU1JBzK3-qdBJB_hBsaNza_kPOJuEW6sPzBpKrWrhnn4tbY

this is getting wild

stable diffusion with brain activity?

yup

[hazy solstice]

Do I need to fully understand it before moving to the next step? Which is data frame and series.

Roadmap is here: https://i.am.ai/roadmap/#fundamentals

[high falcon]

So, if I wanted to train a Lora to be a specific art style, do you generate regularization images using the words "in the style of" or what? Anyone been down this road already?

nvm, it looks like the general consensus is to lean into the words "art style" or "illustration style"

[urban barn]

Hey Stability AI, I would appreciate it if you brought back thispersondoesnotexsist.com

I know you bought the domain

[late ocean]

anyone seen that latent couple helper script? It seems it would help but looks a bit suspect so unsure about running it

[urban barn]

Now because of this, I can’t use thispersondoesnotexsist.com and the other AI face generator sites using thispersondoesnotexsist.com’s ai

There no way thispersondoesnotexsist should be redirected to stability.ai

[high falcon]

Ma'am, this is a Wendy's.

[urban barn]

Oh in that case I’ll take your Bacon Double Stack with some fries and a root beer! 😋

[high falcon]

lol in all seriousness though, you can use the beta.dreamstudio.ai site and try some different prompts for generating fake people, or you can do an image2image generation by uploading a photo of someone else and getting a fake person back

[urban barn]

Hehe, I understood the joke XD but yeah thanks for the suggestions! I’ll try those methods. I was kinda hurt when thispersondoesnotexsist was gone but according a user on its Reddit, the domain expired.

[high falcon]

Thanks for going with it 😄 I didn't mean anything personal by it. Just saw an opportunity to meme and I'm a simple girl.

[fervent thunder]

What’s the best model to generate attractive photo realistic people?

[urban barn]

Bwahaha no problem 😁 I don’t mean to take any offense haha I’m a memer and a simple girl myself! Also, I was hungry when I wrote the reply hahaha so yeah

[high falcon]

There are several models available for this, stable-diffusion can do it. The more important part is your prompts, IMO

[sonic turtle]

what model is best to use fore realism, i'm using rpg_v4 right now but sometimes it gives a clay like result, is there an other that would be beter to get realistic pictures?

[dire siren]

keyerror

[sonic turtle]

Thx

[warm tiger]

does anyone happen to know if automatic 11111 has it's own discord?

[vestal dew]

it was created in Oct, but it moved to the StableDiffusion discord

[dawn plume]

Hey guys! I am looking for developers who are interested in ai generated art, people who are trying to host their own models. Anyone interested in a chat?

[vestal dew]

find someone who can eliminate the hands/feet problem. Then you have something.

[warm tiger]

say i have a photo and i want to remove a large object. like a painting on a wall. what options in inpaint should i use? cant seem to get it to remove the object and just generate wall textures.