💬｜general-chat

[junior mirage]

Guys, there are any guides for download Stable from github?

[orchid wadi]

Lora is Dreambooth but reversed. Where Dreambooth works up to a solution, LoRA works down to a solution. But achieve same thing in practical sense, however can be used for different things. You can adjust the whole model with dream booth. LoRa adjusts only what is needed for that particular thing. Don't think it as "better or worse" than any other finetuning method, just another method. You can combine it with the other methods for greater effect. IT is just another tool in the toolcase.

The greatest effect can be gained by utilising all of the tools together.

[brazen orbit]

Has anyone used here SD for photomanipulations?

[trail ledge]

To change parts of a photo? Yes.

I've used inpainting to change photos

[brazen orbit]

To generate things needed

Also, what was the name of that site where you put an image

and it reverse search google

so you can get text from image iirc

honestly i odnt remember the purpose of that site but it was helpful

[next swan]

Can someone tell me where to find the info on HOW to install the software on my PC?
There was something to DL from GitHub and then install it in DOS....?
But i can;t find it,
Please HELP...

[verbal moth]

Hey guys if I'm having trouble with the install, where would be the best place to try and get some help?

[next swan]

hahaha, EXACTLY at the SAME time...lol

[brisk bloom]

Install guide in this video https://www.youtube.com/watch?v=VXEyhM3Djqg

[next swan]

Do you have the link for me to the install?
Then i can help you how to do it....
Im am a PRO in DOS....lol

[verbal moth]

I've installed Python, Git for Windows, SD 1.5, and Automatic 1111, but when I get to the part where I run the webui-user.bat file i'm getting some errors at the end

my local URL doesn't seem to work

[next swan]

Do you have the right GPU and Ram?

[next swan]

Thank you so MUCH, buddy...
WooooHoooo!!

[verbal moth]

"Runtime error: cannot add middleware after an application has started"

[prime cipher]

tell me please, here to generate as well as in midjourney?

[brisk bloom]

watch the video I linked. check you have done everything

[feral quail]

https://youtube.com/watch?v=-50NdPawLVY&feature=shares

[verbal moth]

Same issue... "This site can't be reached" when I type in my local URL

[versed yoke]

ok can we write in n #🤝｜tech-support ?

[verbal moth]

Not sure the best way to un-wind everything I just did and start over

[brisk bloom]

do a search in #🤝｜tech-support its probably been answered already. To unwind you go into windows Settings > Apps and uninstall

[eager shore]

Hello everyone! I’m excited to join this community and look forward to learning and the experience gained from this tech.

[fervent thunder]

welcome

[eager shore]

Great to be here thank you

[fervent thunder]

Great to have you here

[eager shore]

Are there archives for tech troubleshoots or way to look it up before posting? I can’t find info about getting my local SB to work on window running a discrete nvidia gpu

Can I just drop a question in there have not found an answer yet

[fervent thunder]

Im not sure. might want to ask in tech support in the stable diffusion section

[eager shore]

Appreciate it, how long have you been using this?

[trail ledge]

try searching #🤝｜tech-support for specific issues

[high tapir]

Question. What GPU's are best for Stable Diffusion? The RTX? The AMD? The professional QUADRO?

[vestal dew]

Qyadro, then RTX

[eager shore]

Why are people paying for sd services when it’s limitless on your computer?

Despite if you have lack of equipment etc

[little rampart]

Eran las 10 de la noche piloteaba mi nave

[eager shore]

Where did you take your ship?

[vestal dew]

if they have no GPU...

[eager shore]

I see do you think they will try to strip us of this in the future?

or Pandora box being open that it can never be closed?

[waxen swallow]

Running generation jobs in parallel is just quicker.

[trail ledge]

also, depending on your electricity cost, it may be cheaper to use someone else's computer

[little rampart]

Era mi taxi volswagen

[fervent thunder]

Era un dia de esos malos donde no hubo pasaje

[hardy junco]

i'm using img2img for change small part of a picture about a interactive game map, how can i keep the floor color?

[knotty dagger]

Anyone here try the Kohya GUI?

[trail ledge]

not yet, but probably tonight or tomorrow

[fast verge]

trying to get stable diffusion working but i keep getting runtime errors: cannot add middleware after an application has started

[trail ledge]

just answered that question in #🤝｜tech-support 🙂

[heavy knoll]

Do you guys know any model/Ai that is good at making dnd maps?

[trail ledge]

https://civitai.com/models/5012/dnd-map-generator 😄

[fringe minnow]

anyone know of a good virtual machine service to run the download version of this and what price to expect?

[trail ledge]

https://stable-diffusion-art.com/automatic1111-colab/

[fringe minnow]

thanks boo

[floral zealot]

hello guys somone can explain me what the :1.5 means? i know the max are 2

[signal dome]

its been very silent recently did sd stopped giving new models

i heard 3.0 but never came out

[stable pawn]

I think that 1.4 and 1.5 was early and before companies like Google was involved as much, Google have a rather religious moral and when they got more involved 2.0 and 2.1 did try to remove everything that could be "immoral" so the models become rather bad for folk that wanted to do more Anime like styles or even furriers, now when there is a bit of moral panic on the America based Internet I can guess 3.0 will only be able to generate images of birds except for Tits. (Read this ironical with a bit of troll humor).

[fervent thunder]

seems like it but idk

maybe theyre fine-tuning the hell out of something to really knock us out

[signal dome]

now we need chatbots from sd or language models cuz its in heat

trend

[fervent thunder]

yeah chatgpt is getting a lot of press thats for sure. and use

[light folio]

Laion is doing.

Open assist.

[signal dome]

open source ?

[light folio]

yes

[signal dome]

and consumer based gpu or like high ones

[fervent thunder]

I think I read chatgpt4 will be out soon too

I wonder if they ever updated Dalle since dalle2

[signal dome]

btw i didnt realized they changed the name from stable diffusion to stable diffusion

foundation*

[fervent thunder]

lol me either until just now

[light folio]

24gb vram

[light folio]

The press think Bing GPT is already using GPT-4.

[fervent thunder]

is 3/4 minutes per generation normal?

[light folio]

It would be better to call it Stability Foundation.

[fervent thunder]

depends on your card

[fervent thunder]

xtx7900

not really familiar with it. but if its a little older probably typical time

how much vram does it have?

it's high end but from amd, maybe like 2 months old gpu

is it nvidia card?

[light folio]

AMD doesn't support Stable Diffusion well yet.

[fervent thunder]

20gb

oh if its amd yeah i think then you need to run in like linux or something dont you

seems like you basically need nvidia card for ease of use

[fervent thunder]

yeah i know, i set it up using this tutorial https://www.youtube.com/watch?v=6MeJKnbv1ts&ab_channel=RoyalSkies

it works in general but seems not so fast

hmm it uses only cpu i think, its 70% in usage while my gpu is only at 10%

need to get that fixed 😄

[naive kraken]

I have RTX 3060 12 GB, which would be the best, latest diffusion model forks, image generation model should I download?

[plush forge]

Hello there, i'm sorry but i'm new to this, i installed Auto1111, does anybody know how to make generating images faster?

[forest apex]

@plush forge what kind of GPU do you have?

[plush forge]

Nvidia GTX 1050 Ti

(not the richest guy in the world ha)

[pastel fern]

so if i have a shitty old amd gpu should i still download stable diffusiom

[forest apex]

well make sure to add --xformers to your webui-user.bat file.. beyond that, a newer GPU.. I'm on a 1060 6GB, it's not too bad depending on how much vram you have

@pastel fern how old?

[plush forge]

where should I add it?

[fervent thunder]

is the 1080 TI the best option for 11GB vram? I know there are 12GB vram+ cards, but i can't afford those. unless there is one that works better with stable diffusion than the 1080TI that is under ~$200

[plush forge]

at the end? anywhere? between the last set and call?

[pastel fern]

amd 480 or smth

hmm i can buy a gpu

[forest apex]

ohh, I don't know but it says it's comparable to a 1060, but AMD support is different

[pastel fern]

what gpu can i buy thats budget amd good

[fervent thunder]

amd is bad for ai gens. you want cuda cores in the nvidia cards

[plush forge]

terbo where should I put the --xformers command in the file? at the end?

[fervent thunder]

set COMMANDLINE_ARGS=--xformers

[pastel fern]

what caRd would u recommend below 300＄

[forest apex]

a 3060.. seen some go for pretty cheap used

[plush forge]

Thanks! and if i want to use --lowvram i should add it after? like this: set COMMANDLINE_ARGS=--xformers--lowvram

[fervent thunder]

put a space between them, but yeah

[plush forge]

thanks!

[forest apex]

also make sure all your graphics drivers are updated, any other drivers

[pastel fern]

is it safe to buy used gpus

[neat condor]

i'd say throw in extra 100 and get used 3070 or 1080.

if u are paying 300 quids, it'd be a shame to get bare minimum.

[pastel fern]

i feel like ill be bottle necked by my cpu LOL

i see a post selling used 3070 for 410

but how do i know its safe to buy user

[forest apex]

that's the fun part.. usually older GPU's that have been run a bunch need to have the thermal pads reapplied.. but beyond that, being able to look at the board helps, no shorts or burns or anything

[pastel fern]

the seller is offering a 3 months personal warranty

am i more scared for under performance or nom perfromance

[forest apex]

what on ebay?

[neat condor]

from my experience it's very rare that nvidia GPU's have died on me

20+ years experience from 7 dead GPU's 6 were AMD so i never touch AMD anymore

[pastel fern]

2nd hand selling site popular in my country

[neat condor]

as long as it's nvidia GPU and not chinese ripoffs that sell under the names of "Afus", "giganbinte" etc, i'd be quite calm about the possibility of it being bad.

[forest apex]

so my EVGAA card isn't legit... hrm

[neat condor]

that's a fcking steal man. wish i could grab it in my country at such price. they sell here around 800-1000$.

[plush forge]

what are Textual Inversions, Aesthetic Gradients, and Hypernetworks?

[neat condor]

different ways to affect the generation.

[pastel fern]

thats what u said wtf

u said 400 for 3070 lol

[neat condor]

yeah if i had access to ebay 😄

probly should've added that too 😄

[plush forge]

ok, like lora's but how do each affect the generations? in general terms

[neat condor]

it's quite complex but i'll try to explain it on laymans terms with non-100% accuracy: TI's - are directives that are being applied to currently loaded model which results in generations to be nudged towards what TI's were trained on. no new diffusion info is added to the model whatsoever.

LORA-s are like attachments, compartments which include extra diffusion info that are attached ontop of currently loaded model.

hypernetworks - are what they say, hypernetworks that enhance the hypernetwork capacity of loaded brain, still no new diffusion info added but model becomes better at generating stuff the hypernetwork was trained on.

aesthetic gradients are basically poor mans TI, takes lot less time to train, results are much less adequate.

and styles are most unpredictable out of the bunch.

that's about it i think.

[plush forge]

okok

thanks!

'nother question, is there any way to change TI prompt keywords?

[neat condor]

rename TI lel?! 😄

[plush forge]

wait what?? seriously? is it just renaming the text inversion file?

[neat condor]

yup

[pastel fern]

with stable diffusion will i ever be cpu bottlenecked

[forest apex]

I'm running two generations on an i7... %30 cpu, so probably not

main bottle neck is... wishing I could have more generations running. pretty much

[pastel fern]

oh lol

so i dont even need to get a 3070

[plush forge]

@neat condor sorry to bother, is there any way to add prompts to lora's?

or delete/change

[neat condor]

u can train other lora and merge with current lora. or train out prompts from given lora.

no other way that doesn't require training.

[forest apex]

no that machine has 2 GPU's going simultaneously, is what I meant

[plush forge]

anyone managed to use the shirtlift lora succesfully?

[pastel fern]

do people mostly generate art and the postprocess then post

or just immediately post

[forest apex]

never immediately post

wait until the hard drive is full, that's what real artists do

then just post like one pic, feeling like scrooge sitting on a bed of gold coins

[fervent thunder]

I don't really post it anywhere , but usually I generate a good pic I like (out of hundreds) , then I process it...alot...
Change details , outpaint, fix issues ... and after few days give up.

[pastel fern]

cos its so misleading

how can i tell something posted is pure processing or edited

like i wanna see what should i be striving for pure SD

[fervent thunder]

Where exactly?
Usually if it has prompt included - it's just gen + upscale

[pastel fern]

hmm like on pixiv

[fervent thunder]

oh and maybe outpaint too, but most of them shouldn't be edited alot

[pastel fern]

do u have any examples of pure SD？

like a gallery

[fervent thunder]

nope =\

[wary pivot]

how to use it for creating images?

[uncut bough]

gm

[wary pivot]

HAI

[plush forge]

how to download models from hugging face?

like this one https://huggingface.co/Oosayam/UnstableSamInkDream/tree/main

[forest apex]

.safetensors are similar to .ckpt

[hidden wren]

anyone figured out how to generate words?

[plush forge]

yeah but i can't seem to be able to download from that page?

[forest apex]

click a .safetensor, then click the download link

[plush forge]

thank you!

[cerulean halo]

so, I'm preparing a research paper for arxiv and I don't know how to make the images look beautiful for the paper 😓

I've been using vanila sd 1.4 for all experiments, so probably it's that xD

can someone help me understand the modifications and additions that could be done so the examples look prettier?

wait, I should be doing this on general-with-images so I can send examples

[muted mural]

anybody knows how to fix the "training finished at 0 steps" bug when training with your own images?

I see other people have the same issue

[modest escarp]

sad that stable diffusion is no longer open source loooool

rip being the leaders of open source ai development

[forest apex]

?

[modest escarp]

1.5 is not going to be open source

[forest apex]

huh? 1.5 what?

[warm junco]

xD wtf

[forest apex]

you sound like you want unstable diffusion man ;>

[plush forge]

huh, didn't know that was a thing, i'll be on my lookout

[elder cradle]

You're trolling right?

If it isn't, how can you say that 1.5 is no longer "open source"?

[bleak matrix]

Good morning, everyone!

How are we fairing today?

[forest apex]

morning Sunny 🙂 but it's never morning here

[bleak matrix]

Sadly, never, yet here I am, sneezing, and my eyes are waking! Ah!

Hahaha!

What are you all up to today?

[bleak matrix]

Are you talking about 2.1?

[forest apex]

messing with mixture of diffusion

[bleak matrix]

Wooo!

There are many ways to use SD. If you can't run it locally, you can use Google Colab, or there's the demo on Hugging Face, which is limited but free: https://huggingface.co/spaces/stabilityai/stable-diffusion

Well, you can check out #📝｜prompting-help to get help with that. There's also a lot of good references in #1045349359044280360 as well, so you can check them out. Though Dreambot has gone to sleep, and is no longer running, I still have some tips in my Friendly Guide that you might find useful: https://docs.google.com/document/u/5/d/1aHJ9RBt_vlCwJQBVUUsb7VghKB-wynv7WGTxm9ozL1k/mobilebasic You can also check out my carrd with links here: https://atypicalconsortium.carrd.co/

I haven't updated my guides in awhile because I am working on a lot of things right now, but I do plan on making a 2.+ guide/transition/more info when I can.

Np! You can always check out the prompt book, etc, too.

And yeah, it's just in the browser

[elder cradle]

Use stable horde

[bleak matrix]

I understand that having a webui is easier to understand

But yeah, there's a lot of great community options

[cerulean halo]

thank you for your reply. I am currently using v1.4. As per a recent discussion on general-with-images, it looks like I can get much better results just by prompting

without relying that much on custom model checkpoints

[bleak matrix]

You absolutely can. The relationships in 2.0 are stronger--you just have to finesse them a bit, and iterate a little more sometimes. 🙂

[cerulean halo]

yeah. I don't know the keywords, so I've been doing all my research in 1.4 (also I think it's best to avoid too much customization in research results). But my results would look terrible in the paper without at least basic prompting xD

by the way, I received a link for some prompting guides but it was not what I was looking for. Do you have some list of "enhancing" keywords, like "esquisite detail" and "natural lighting", etc. that I can try?

there are some guides out there, but the ones I've seen are more about artists or choosing a specific style

[bleak matrix]

You can use CLIP Interrogator to help you understand if you need to

[cerulean halo]

with the images people post around here, right?

as in, get inspired by other people's keywords?

[bleak matrix]

(You can see my past research in the links in my carrd)

[bleak matrix]

You don't have to use images people post

Also, you'll want to look into using negative prompts as well (the info is in the Friendly WIP guide, as I mentioned)

But basically CLIP will give you a string of words you could potentially use for your prompt.

As for using exquisite detail, etc...It just depends. There are a wide variety of factors that go into making it work.

You can add these words, of course, and I reccomend that you experiment and try for yourself to see what combos you can find.

Lowering or upping the cfg will influence the output of these words for your final result. So even if you add this detail, etc, just keep in mind that there's a combination of factors going on--your steps, sampler, cfg, your prompt, etc.

Personally, when I write my prompts, I like to write them as "subject, environment, style."

This helps keep my prompt clean. I keep my grammar very simple, and group words that are important together.

I.e., I follow the rule if I am talking about a dress design or something, I will typically keep that to one line with all my key words. Like, "Woman with brown hair, blue dress with ruffles, in the style of traditional ink drawing."

[cerulean halo]

Oh I had missed the link. Just skimmed it right now, it's pretty detailed

[bleak matrix]

Np

My 1.5 guide is still relevant. The relationships may have changed by the principles are the same

(Volume 1 & 2)

[cerulean halo]

much of the advice in that link seems to revolve around the idea of words affecting each other in unexpected ways, is that correct?

like, lemme get an excerpt

""spheres of candy" interpreting it, at base value, as cloth, association with farms, balloons, fun, etc"

that's what you can relationships

that's pretty interesting to me, since that's one of the issues I'm tackling in this paper

but first I want to make sure I understood what you said in the guide haha

[bleak matrix]

Yup

There are a lot of different, very fun, and interesting interactions

There's a lot

More info in my Volumes LOL

Like 400 pages 🤣

And I have been writing so much more laughs on 2.0

[cerulean halo]

Huh hahaha

So much info

Like a hobby in itself

[bleak matrix]

It's like a textbook, ikr

I like writing, and I'm a nerd, what can I sayyy?

[cerulean halo]

Anyway, maybe you'd be interested in the thing I'm developing. One of the applications has to do with removing relationships, and I'd very much like to know what you think

[bleak matrix]

I study SD and shake it around like it's in a sack lol

[cerulean halo]

I wanna show two illustrative samples, should we go to general with images?

[bleak matrix]

Sure

[cerulean halo]

HAHAHA i know this obsessiveness feeling as well XD

[bleak matrix]

Hooray for nerdiness!

[fervent thunder]

Hi everyone!

Hi,
A1111 - is there a script to automatically change all my available Models with the same prompt to see where i could get the best result?

[rustic scroll]

Been out of the loop for 3 months.
What are your thoughts on 2.1?
I would guess they make 1.5 (normal + inpainting) obsolete, but it doesn't seem like it based on the msgs here.

[rustic scroll]

You can use the Scripts tab in the main page (in the bottom)
Select "X/Y/Z plot"
Then choose your 3 models for each letter

If you need to compare more models, you can use this script:

https://huggingface.co/Froddan/model_compare_script

[fervent thunder]

Sounds interesting, always wondered what it does, but never enough to click on it.
Thanks 🙂

[rustic scroll]

You're welcome 😄

[fervent thunder]

@supple wasp I'm sorry, I don't know what I did, I thought I could trust you to talk to me about it.

Anyone know why my 2060 super won't show up in my local stable diffusion 2

Peter my friend can't help either lol..

[wide juniper]

When i use a lora am I supposed to use the trigger word as well as the lora:exampletag:1?

[eager shore]

I see prompts with () or (()) around text does it result in different focus? Reading through chat, excuse me for not fully research on my own yet, but felt like asking

[fervent thunder]

It looks exactly like what i want ... but i get an error, not sure what to do about it.
AttributeError: module 'modules.hypernetworks.hypernetwork' has no attribute 'apply_strength'
Any ideas?

[charred mesa]

hmm no update from AUTO for a long time

[fervent thunder]

Position in text matters more but () adds weight, [] is the opposite.
Example:
woman, (man), [animal] - should generate more likely a woman than a man and is less likely to generate an animal
at least thats what i learned so far 😄

[fervent thunder]

PC restarted?

[ornate lodge]

What the fuck did you say to me you little bitch? I’ll have you know i graduated top of my class at navy seals and i have over 300 confirmed kills. You’re fucking dead, kiddo.

[orchid wadi]

That meme always make me think about the fact that the Veterans of my country, and every one I know that been a peace keeper and has confirmed kills have some degree mental health problems about the fact that they have killed people.

[cerulean halo]

That's quite.... sinister

[worn granite]

Is there any websites that have a more fine tuned stable model that is text to image?

[winged musk]

Does anyone know about how to get good images of two characters interacting? Without using img2img? I haven't seen any good SD images that I can recall with more than 1 subject doing anything other than standing next to each other not aware of each others' existence. Would this entail training a model with a bunch of pictures of people doing different stuff together?

Also what about training characters that exist in a ton of different styles or have been played by a bunch of different actors? Like if I wanted to train the joker- would I have to pick one specific joker from the same movie and train that?

[prime cipher]

please tell me, is the bot turned off for a long time?

[wintry stream]

#📣｜announcements message - Should have all the details you need!

[cold nexus]

How to train images with stable diffusion?

And in what does consist?

[waxen swallow]

Try out the deliberate model

[waxen swallow]

Community driven models have exploded in popularity and variety

[uncut plinth]

youtube was deleting Aitrepreneur video something about midjerney someone have this video or know what is about?

@golden lantern

[maiden crystal]

anyone using Invoke Ai ? Was it easy to install?

[uncut plinth]

https://www.youtube.com/watch?v=PCofkmr2x9g

[maiden crystal]

its just a different window to interface with SD i guess. Probably was out before Automatic1111

https://invoke-ai.github.io/InvokeAI/installation/010_INSTALL_AUTOMATED/

I got no clue about plugins or stuff.

[elfin pivot]

Which model do you think is best for cyberpunk sci fi kind of art style with cybernetics ?

[ashen agate]

SD can't use pics as models to generate other pics can it? i've seen img2img but that "alters" an image doesn't make a new on ne with the likeness of a model

[chilly sierra]

Hey there!

AI is the hottest market right now.

But most people don’t know where to learn and share.

That's why I curated 100+ AI Communities for you:

• Facebook Groups
• Subreddits
• Discord Groups

It's free for everyone.

Steal it here: https://theveller.gumroad.com/l/AIComunities-byTheVeller

[lament steeple]

There are a few ways to fine tune SD off of other image sets

Textual Inversion, Hypernetwork, LoRA, and fine tuning the model itself

[brisk bloom]

if you can describe your prompt as close to the image as possible you will get a closer resemblance, but never exact

[ashen agate]

That's one victory for midjourney

[brisk bloom]

using the denoise slider can do the same thing

[ashen agate]

My examaple has the AI taking an image as the model then producing a whole diferent one
i failed to get that
like you give it a picture of cloud strife, say I want this dude on a swing and you get cloud strife on a swing

[brisk bloom]

i posted a pic in #🏞｜general-with-images of an example. the far left in a character from cyberpunk that I created variations of

[ashen rampart]

Created my first AI animation using stable diffusion today👌

[fading turtle]

Are there any good models based on art from TTRPGs like Pathfinder (preferred) or D&D? If not, what's a good general purpose model to use and how can I edit the prompt to do that well?

[north orchid]

Hi i’am new here 👋🏻

[ashen agate]

My gui has no denoise slider :( i'm using Gradio

I guess it's the strength thing?

[brisk bloom]

yeah. its for how much of the image you want to change

[ashen agate]

Tried less but it's still not what i'm talking about
it just used the pic as "borders" to make variations

[brisk bloom]

that is at a low denoise. a higher denoise and more of the original changes. at 1 its a totally different image

maybe post an image of what you mean

[eager shore]

anyone here a coder?

[hollow pebble]

Hello guys

I've been using SD for curiosity. And now i have simple knowledge how it works. But this isn't what I'm gonna talk.

I'm an processes artist, and I feel afraid about AI
Am i not gonna be necessary in a near future?

If everyone can make an illustration, what's the purpose of considering themselves "Artist"

[wide juniper]

If I find two models aimed at creating specific characters and merge them together, will I get a model that can generate both characters, or will both characters be merged into one?

[alpine bear]

it'll more than likely be mixed. Your best bet for specific characters is using LoRAs.

[eager shore]

Why this can make your work 10x better and cut down on time. Just enjoy it while we have it.

[hollow sail]

wht room do i have to go to to make a image?

[plush forge]

does anybody know if it's ok to change the filename on SD models?

I've been trying to search for protogen but all the weird names like assdddfmodel1vadbyjohnythegoodboyV1-reality-profanrpgmodel, really confuse me

[brisk bloom]

yes you can

[elder cradle]

Imagine having a podcast in which both pro-ai and anti-ai dudes debating with each others

[crisp elbow]

Can we make video in stable diffusion

[lament steeple]

I would end up somewhere in the middle. I can see both sides of the arguement

[tepid sun]

was it this server or some other one which had channels to generate images?

[bold ridge]

Holy crap! I never saw a robot that makes pictures for you! This is amazing! Nyeheheh

[tepid sun]

lol

[lament steeple]

it's not this server, but I'm bored and need a break from the synthetic LoRA dataset I'm working on. Drop some prompts in general-with-images

[hexed summit]

Hi everyone,

I recently joined from Midjourney due to financial reasons. I've been following Olivio Sarikas' guides on Stable Diffusion on Youtube and I ran into some issues while setting it up. I installed local SD 1111, Invoker AI with "realisticVisionV13_v13.ckpt" and "v1-5-pruned-emaonly" models, but I encountered errors trying to train images for Laura with "realisticVisionV13_v13.ckpt" for Kohya_ss GUI.

My memory is low so local SD 1111 isn't working for me, so I'm planning to try the online version. My main question is: What's the best, all-in-one way to use Stable Diffusion? I'd prefer a workflow that doesn't require jumping between many GUIs. Can someone recommend the ultimate, optimal setup for Stable Diffusion?

Thanks in advance!

[hollow urchin]

Thank u

[lament steeple]

You can do most of what you need to do in automatic1111. You shouldn't need anything else unless you plan on training, but it does that mostly well too. Only thing it has issues with is LoRA training and that just wants to merge with the model right away instead of being stand alone

[hexed summit]

Thank you so much for the response 🙂

[twin mist]

susan is the new karen for me

[fervent thunder]

Why DM , I don't want DM , I want to make fun of scammers publicly

[lament steeple]

I'm giving away my left right toe to 20 people DM me to know how 😛

[astral pivot]

NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try using --no-half commandline argument to fix this.

any idea how to fix?

[lament steeple]

first check. Did you try to generate with an inpaint model. if no, add --no-half to your webui-user.bat and that should fix it

[astral pivot]

wdym inpaint model?

https://civitai.com/models/4219/one-piece-wano-saga-style-lora

im using this, not sure if its inpaint or whtvr

[lament steeple]

That's neither

What is the name of your checkpoint?

[astral pivot]

onePieceWanoSagaStyle_20

[astral pivot]

how?

[lament steeple]

you're using automatic1111?

[astral pivot]

i honestly dont know 💀

sent u a pic in dm pls check

[lament steeple]

OK found your problem. You have a LoRA selected for your checkpoint

that file is in the wrong folder. You shouldn't be able to select it there

[astral pivot]

where do i put it?

embeddings?

[echo tulip]

No there should be a folder name Lora

And you put the Lora file in there

[lament steeple]

We got it straightened out in IM

Would have helped here, but I was bouncing back and forth and needed to keep it in one window

[echo tulip]

No problem nice

[chilly valve]

morning all happy Monday

[lament steeple]

Is it morning already?

[empty barn]

Yeah, it's noon already.

[hollow cedar]

Hey people I am a newbie so I am starting and please forgive my basic question I want to input a small dataset like 300-500 images of random images. Then I want to let SD decide about an "average" picture, basically like, if there are trees on 20 images, weigh that higher than other and in the resulting image add rather a tree than anything less common pictured objects.

I only have an idea about the process but do not technically know where and how to start. Should I upload the pictures anywhere? Should I tag each image with the objects visible?

[magic citrus]

hey friends, a new post from me, this time I was experimenting with different captioning approaches and their impact when training a style. Let me know what you think, what else should I try, etc. The usual stuff by now
https://followfoxai.substack.com/p/impact-of-image-caption-on-style

[muted mural]

Anybody had problems with youtube when uploading AI content? I uploaded generic AI content to YT shorts like "football players as RPG characters", and I just received news that my Google account linked to my YT has been suspended. Also, they shadowbanned one of my shorts because, despite it being "public", it's not on my shorts page. Also comments on said video got deleted. Please help.

It makes 0 sense. It's family friendly content. 100% made by me, without any copyrighted material.

[fervent thunder]

how can we help? we're not youtube

[muted mural]

I was just asking if anybody has similar experiences

Probably doesnt even have anything to do with my content

[zenith thunder]

google is investing biillions into ai

i dont see why they would be blocking content

chat gpt threatens google search but its going to take years before its better than google search

everything is optimized to google search

[stable pawn]

In Automatic1111 setup I see an option to not add a secret hidden watermark in the images we create, what is this for and how bad am I if I disable it? Some say it was to prevent AI images to be used in learning, but what would happen if we trained a model with only AI images that we like, is that like divide by zero or 42?

[sudden ruin]

Im looking for some nice guides for training, ive been watching videos and stuff but i feel like none are really in depth q.q
Like how many steps, how many images in the dataset and so much more
Anyone has some nice resources for that stuff?

[young ivy]

I know how you can help me earn money. Just send it to me. You'll be the direct and only reason I have those money. Make sure it's a lot of it. Not crypto though. Paypal! I kinda need some dough

[echo beacon]

Hi using the Automatic 1111 for making some art

but i want to know

which SD model am i using?

[young ivy]

upper left corner

[echo beacon]

Can you come to #🏞｜general-with-images chat?

[young ivy]

if it's really quick, I'm not feeling well

[astral pivot]

a Lora i downloaded doesnt seem to work, i put the trigger words and tried everything but it just generates a random character, can someone help?

[fervent thunder]

hi guise, is there any way to generate an animation out of image generation process?

or an image sequence

like midjourney does

[wispy jasper]

Wowza.... https://twitter.com/lil__criminal/status/1624780032269844480

OpenAI wants a monopoly, and to clamp down on open source generative models

They really need to change their name.

[trail wing]

The influence operations they reference are the kind they don't approve of. So they are begging for daddy state actor to come help so it's biz as usual

can't wait for big search with ads to tank when someone introduces a sanitized search extension with limited consumer side model

[thick wharf]

MedARC (open-source medical AI research org supported by Stability AI) launch livestream happening now:
https://youtube.com/watch?v=PK2Sl3NRFH8

[fervent thunder]

Hi all, newbie here. I really enjoyed playing around with midjourney and now doing some research on how to get into this topic.
I‘m torn between just using an App like draw things/ starry ai/ diffusion bee or diving into this whole locally hosted self trained model expert thing I see you talking about. But thats probably on a level I am not ready for.
Any recommendation on the best way I should go and checkpoints to take along?

[fervent thunder]

just go on https://civitai.com/ and see what you prefer \ like , there are plenty of good models.

[warm junco]

if you want just generate images there are many possibilities, like simple apps for ios and mac os and on windows there are many webuis that you can install localy with tutorials and help from here. The Most features you will get with a local install, but you should have more than 4gb vram

[fervent thunder]

hey guys released a 90s buffy Lora, hope you enjoy it ❤️ https://civitai.com/models/8585/czeros-90s-buffy-the-vampire-sarah-michelle-gellar-lora

also @warm junco hi mate hope ur well

[warm junco]

hey captain 😄 yea all fine!

nice lora

[pseudo elm]

Any ideas for a love poem?

Someone

[fervent thunder]

thank you ❤️

[pseudo elm]

Please

C'mon

I need one idea at least

[warm junco]

ask ChatGPT 😛 jk

@pseudo elm what idea do you need ? like a poem for a dog and his toy or a man and his bike for example?

[pseudo elm]

A poem for a guy

Tomorrow is Valentine

SO a love one

[warm junco]

okay xD

[pseudo elm]

Just

An idea

NOt the whole poem

Cause I started one

[fervent thunder]

sorry guys i posted twice to the site updated link: https://civitai.com/models/8586/czeros-90s-buffy-the-vampire-sarah-michelle-gellar-lora my first person Lora is released

[fervent thunder]

I will check this out, thanks. But I still need to do some research to understand how I would actually add these models and set them up/use them.
So midjourney probably has their own model prepared that can be used with their cloud gpu power, right?

[fervent thunder]

To install it locally seems to be the Goal in the long run. Although an m2 mba (16gb) isn‘t maybe the best way to do it. But to learn about it, that should be Fine.
What would you recommend to run locally on a M mac? A1111 or something else?

[warm junco]

For all the features A1111 is recommended but to play around with you can use Diffusionbee too

[fervent thunder]

kinda , midjourney has their own AI and machines, yea, you're paying for a machine basically.
models you just put into

\stable-diffusion-webui\models\Stable-diffusion

folder when you install sd

[iron glen]

Sorry for the dumb question, but how does midjourney create such more refined images right away? are they essentially adding additional parameters behind the scene? or do they have their own models?

[fervent thunder]

They have their own thing

[cobalt cliff]

guys... where can i buy a stable diffusion t shirt

i want to wear this software

softwear

[river maple]

I hope Emad comments on this

[neat badger]

Anything out that can do animations kinda?

[brisk bloom]

Deforum

[neat badger]

Thanks!

[cedar plank]

When are models trained on 2.1 at 768 going to come out? I have a 3090, can I train my own?

[brisk bloom]

illuminati diffusion, redshift diffusion to name a couple

[astral goblet]

Dive right into running local. it's fun and you don't have to worry about the host stealing your created content at all.

investing in a great desktop to host from home is a worthy goal for this hobby. or new profession! whatever you make of it. it's so accessible to get into and its so exciting to train on your own machine. Though renting cloud compute specific for training is a worthy endeavor also. so many paths into this new field.

even though there are lot of models available, refining what exists to your own directions is also something you can do!

[deep sand]

What is the best way to generate an image of a person in a certain style (i.e. dreambooth but it actually generates the photos of a person as an anime and anime only)? Would you need to finetune a dreambooth model to a certain style and then finetune those model weights to a person or is there an easier wat to do it?

[astral goblet]

i've been having a lot of success training embeddings on a base model, and then using them in a refined model

[eager shore]

Totally agree, its fun and I like learning tinkering with it.

@fervent thunder Welcome to the community, Feel free to hit #🤝｜tech-support, Friendly and knowledge folks in there. Have fun with it and continue expanding your knowledge.

[hexed stump]

anyone know any good anime models that doesn't generate weird fingers

[valid bloom]

@hexed stump my good man, we all have weird fingers from time to time 😛

[hexed stump]

is there a way to fix it

it happens more than often

[valid bloom]

@hexed stump Majority of the time it will, its a SD thing, sometimes it gets it right, sometimes it doesn't, you can always try push prompts but there is always a 50/50 to be honest. Hell i'm still battling that one 😮

say positive prompting of (perfect fingers) and a negative prompt of something like (mutated fingers), (deformed hands) etc

[hexed stump]

hand's look even more funny now LOL

[valid bloom]

Yeah its a up hill battle lol

[hexed stump]

now Im getting an extra arm and fingers, arms funky ect...

[valid bloom]

Its ok, i just got a extra head, and for some reason a belly button on the cheek 😛 -shrug-

[hexed stump]

lol

I give up lol

[hexed stump]

for negative I use bad hand anatomy, bad finger anatomy

it worked more often-ish

[valid bloom]

Damn, perhaps ill give that a go 😮

[hexed stump]

for positive prompt roughly perfect hand proportions, roughly perfect hand diversity, roughly perfect thumb and fingers, roughly perfect hand range of motion works so well

[fervent thunder]

how it feels to use stable diffusion

https://youtu.be/KIXTNumrDc4

[hexed stump]

how to lose your mind trying to fix hand's roughly perfect hand proportions, perfect hand diversity, roughly perfect thumb and fingers, roughly perfect hand range of motion, roughly perfect metacarpus, roughly perfect manus

[livid depot]

So you guys keep regenerating the image until it comes out mostly normal looking?

[astral goblet]

i'm grabbing tehse controlnet models. an extension for webui updated with support for it

[fervent thunder]

i have a laptop would getting an external gpu help with this?

stable diffusion

[astral goblet]

yes, if the laptop has external gpu support

[subtle parcel]

umm..

[plush forge]

hi there, in a1111 what are styles? I noticed I could create one but i'm not sure what it does

[brisk bloom]

you can save your prompts and load them later

[umbral yacht]

jokes on me... I'm not using prompts right now, but I still want to have that style

[sonic idol]

when did Civitai decide to be dog water about logging in

[plush forge]

Nice

[astral pivot]

Lora isnt working for me, can someone help

[plush forge]

try using (lora: filenameoflora)

[astral pivot]

i did

[plush forge]

are you sure is the exact same filename?

[astral pivot]

i manually picked from the lora tab in the interface

so i dont think it can be wrong

i downloaded the extension for lora too

[plush forge]

which thingy are you using?

[astral pivot]

the thing is, i put it in the prompt but it just generates a random pic and not the lora character

[astral pivot]

??

[plush forge]

auto1111?

[astral pivot]

i really dont know

[pastel fern]

so is 3060 12gb or 3070 better

[astral pivot]

help 🥲

[lament steeple]

3070 is 8GB

[plush forge]

does anybody know why a1111 doesn't let me generate a 1024x1024 image?

[pastel fern]

i heard 3060 is better then?

[empty barn]

It probably throws an error if it fails, take a look at that.

[plush forge]

ok thanks!

also, does anybody know what should I train if I want a face?

For example I noticed there are some TI that are from people (usually models or smthng) and noticed that when prompting bobby fischer or magnus carlsen (or other ones) they usually don't appear/appear really bad

[cold siren]

heh yeah but it knows how all the important people look - biden, trump, batman....

[plush forge]

true, but if i want any specific person it starts getting apart that's why, do I have to train a model? a lora? a ti?

[cold siren]

i guess 🤷 unfortunately can't give any real advice, my old ass gpu is not good for training 😦

[lament steeple]

Yes you would need to train a model. There are a few guides on youtube for how to do it

[plush forge]

okok but is definitively a model, not a ti nor a lora, right?

[cold siren]

apprently TI/embedding or hypernetwork also could work

[lament steeple]

Haven't tried it myself yet, but I'm about to

ti or lora would do the trick

TI, LoRA, Hypernetwork, or Dreambooth would all get your training done. All depends on what you have for resoources

[cold siren]

but i wonder what to do if you want to train not a face but something more general... like, a specific pose or dunno, holding/using an item in a specific way

[lament steeple]

you would have different people doing it

[plush forge]

can you give me a rundown of what I need for each case?

[lament steeple]

Maybe after I get it to work?

I'm still preparing a dataset for my first run. All my knowledge is still book knowledge so far

^ understands the concepts, but hasn't actually done it yet

[lament steeple]

https://www.youtube.com/@OlivioSarikas/videos This is where I'm finding the tutorials I'm going off of

[fervent thunder]

How can I do prompts

[muted topaz]

hello everyone

[plush forge]

Aah ok sorry i thought you knew

Ill be sure do check the videos tho!

[fervent thunder]

What are the system requirements for stable diffusion

[astral pivot]

can someone help me for img2img? i tried to change some parts of the generated image but it just gives me the same thing when i run it through img2img

[olive wasp]

its too big probably for vram, highest i can do on my gou 3070 is 768x1024

[olive wasp]

you can probably do more with 3060 but its slower. You can probably train your own model while on 3070 I can only train embeddings and even that is hit and miss sometimes

[fervent thunder]

vram is the important part for ml/training stuff. 3060 is better than 3070

[lament steeple]

Make sure that if you are on inpaint, you actually inpaint something and check your noise tolerance

[lament steeple]

It's so much fun watching people hate on your paint brush 😢

[stable pawn]

It is time AI is getting better grammar and understanding of words, I am dyslexic but I know that "small" is not a synonym for "young" or "Child like", so why do Stable Diffusion try to make a person 'child like' when I add "holding a small teddy bear" and how can I avoid that words like "small", "tiny", "large" change the humans and not the object I want to make smaller?

[lament steeple]

add "child, childlike" to your negative prompt

[stable pawn]

Duh... Guess who feel stupid now? Thank you @lament steeple

[lament steeple]

anytime

[stable pawn]

I wish there was a function for saving and sharing negative prompts and set some as individual standard .

The only thing I add are to get rid of all Copyright markings in the images, "Copyright, watermark, brand logotype, signature, artist name, image footer".

[lament steeple]

you using automatic1111?

[astral pivot]

bro wtf does brush even do

[astral pivot]

i was on img2img but since u said, im trying impaint cuz i got curious

wait ill explain what i want to do

[lament steeple]

the inpaint brush tells stable diffusion where you want to change the picture

[astral pivot]

wait i cant send imgs

yay'

[lament steeple]

if you want to be an artist rather than an image generator, it's your best friend

[astral pivot]

see #🏞｜general-with-images

[blissful flame]

i whant buy iphon. Is it good idea?

[stable pawn]

@blissful flame If you want to and have the money and is in need of a new phone, why not, it is a popular brand.

[rocky marsh]

Is it possible to remove generation parameters from already generated images?

[empty barn]

Of course. It's just text in the exif.

[rocky marsh]

yeah, but is there a tool for it in Automatic web ui? Or do I have to make one such myself :D?

[empty barn]

I don't think Automatic does it. But there's tons of programs for reading/modifying EXIF out there.

[rocky marsh]

just went through few, and they just remove "personal data" which is predefined fields

[empty barn]

Dunno then. Might be easier to script it yourself if you know basic programming.

[rocky marsh]

never had to touch any images so yeah

will do now

[brave matrix]

Hey guys what is this pix2pix thing

Pic2pic

Idk

Img2img

[safe sapphire]

exiftool -all= /path/to/img.jpg empties metadata

so, basically you tell instructpix2pix which edits you want to the img

[brave matrix]

Instruct2pic

[olive wasp]

it transforms image to different one,or part of that image

male subject to female, blonde hair do black etc.

[rocky marsh]

just did it in c#

((System.Drawing.Bitmap)System.Drawing.Image.FromFile(file).Clone()).Save(System.IO.Path.ChangeExtension(file, null) + "_m.png");

[brave matrix]

How can I use it? In automatic?

[bleak cliff]

Gimme free rtx 3090

[fervent thunder]

Soon.

[chilly valve]

Morning!

[astral pivot]

is there a prompt i can use to get the default anime art style

[static cape]

Have there been any Updates on SD3 at all? V2.1 is still a horrible base and the community is still trying to get the best out of 1.5, but it's obviously at it's limits at this point.

[neon apex]

Is it ok that I’m here out of fascination more than anything?

[astral pivot]

same

ai scary

[neon apex]

just curious because I’m considering working in it in some way someday. specifically nlp
so just curious about development etc

[neon apex]

I have a very mixed view, can’t blame you

[thin heath]

I am making these stable diffusion tutorial videos. Any idea where people can have easier access to this tutorial?
https://youtu.be/vv4HqtDPOUU

[astral pivot]

man soon ppl will be able to animate pics or smthn

[fervent thunder]

I hope for animations and 3d models too

but well...3d models are alot more complex

but animating simple sprites will be huge progress for simple 2d dev too

[wet solstice]

Any chance I could get someone’s help? Been trying for two days to make an image.

[fervent thunder]

What kind of help?

[wet solstice]

I’d be willing to give a few bucks for the trouble.

[wet solstice]

I don’t know how to get the image I’m trying to get. I don’t know if it’s my prompts or just general stupidity but I’m really struggling to come up with anything halfway decent

[fervent thunder]

Go ahead , describe what you're trying to do , which model you're using , your prompt, maybe we can figure it out

[wet solstice]

I’m trying to create a hand with a broken knuckle, which is bleeding a river of psychedelic blood out of which is growing a forest of mushrooms. I’m getting nowhere close. My best attempt so far is in #🏞｜general-with-images

I’ve tried like 20 different websites and apps, every mixture of keyword I could think of

[fervent thunder]

Yea achieving it won't be easy in 1 prompt, I'd try to start with bleeding knuckle part and outpaint or image to image it further if you aren't getting what you want from gen

[wet solstice]

To be entirely honest I understood like half of what you said

I’d be more than happy to cashapp someone like $5 if they could help me get a decent image

[fervent thunder]

if that's the case, what you're trying to accomplish is unlikely to happen. you're trying to gen something that doesn't exist blending 2 very different topics that most models aren't going to be well trained on with a color scheme that is also unnatural. a single prompt isn't going to cut it lol. i'm not saying it can't be done, but you're gonna need a decent amount of knowledge to pull something like that off

So...there are few things you can do with AI

• you can generate image
• you can "outpaint" image , adding pixels to one of the sides, or even all of them
• do img-to-img , which can help you to change some parts of an image, this way you can add to your image whatever you want.

It'll take some attempts too

[wet solstice]

I can also skip the mushroom forest if that helps.

[fervent thunder]

So...what I'm trying to say, with this kind of prompt you will probably need to do both outpainting and img-2-img

• generate bleeding knuckle
• change bottom side to force it to make a river + background you want
• outpaint to increase size with more river and background or whatever will happen to be there

[wet solstice]

Alright. Is this accomplishable on mobile? My wife has the laptop at her sisters

[fervent thunder]

If you're using cloud services - yea, you probably can , never tried

or well if laptop is 8+gb vram

honestly, your best bet is something that is gonna probably be too advanced for you, but you can look into it: there's an extension for automatic1111 that allows you to assign specific areas of your gen to a different prompt. so you can have half the gen be the bleeding knuckle and the other half be the mushroom forest. can probably just get a style embedding for the psychedelic part, but that extension will probably simplify things. it's opparco/stable-diffusion-webui-two-shot on github if you wanna give it a shot and have a decent knowledge getting/running stuff from github. EDIT: forgot to mention it's not part of automatic1111 yet, which is why you need the github knowledge so you can modify the files necessary for it manually

oh yea , that's an option too , never tried it

[wet solstice]

I just figured I’d end up having to do two, and just photoshop the hand in place

My knowledge of creative apps begins and ends with ms paint lmao

[deft bronze]

sounds very specific, but trying to throw money at people will probably help 🙂

[fervent thunder]

That works too , I did some generations where I just put things I wanted on the image and added it to prompt , AI kinda figured out what I wanted, but...not every attempt was successfull...img2img is pain in the ass.

[wet solstice]

It usually does lol

[astral pivot]

is there a discord server for anime characters' lora?

[wet solstice]

Well if I could commission someone to make like 1-2 images for me that’d be awesome. Just trying to help out a friend and maybe do something for the wife.

[deft bronze]

unstable might know

also: if civitai has loras, I guess go there and browse lol

maybe someone did an a1 studio for you to use

or ano_hana or whatever style you want

[fervent thunder]

there are (waifu diffusion, unstable diffusion etc) , but you may be more interested in civit.ai for that type of thing.

[waxen pasture]

Does civitai have a discord server? I feel like I remember hearing they do

[fervent thunder]

civit definitely has loras. it's basically the only thing getting updated there nowadays lol

[waxen pasture]

Yeah i've noticed that, I haven't even really tried any loras. I've just been uploading TI's since I know how to do those and it's fun.

[fervent thunder]

they do

[waxen pasture]

Ah so I am not going crazy yet

My memory sucks

[fervent thunder]

it's linked on their mainpage

[lost granite]

Hello? This is my first time on discord. Stable Diffusion brought me here.

[waxen pasture]

Welcome!

[slender haven]

i'm new to ai stuff but not discord in general

[fervent thunder]

I can help you to learn how to do it the way I was editing image, for free.

I'm not going to do it myself tho, that's alot of actual work.

Just so you understand - when I was trying to make image look exactly how I wanted it to look, it took me hundreds of img2img generations , changing and fixing little details of an image.

Tool , mentioned in this message might and probably will make it easier, but I never tried it.
#💬｜general-chat message

I'm not trying to raise the price or whatever , I don't think you can even pay me, since I'm from Russia (let's not talk about that)
I want to help, but not by doing it myself.
So...I can show you how to change parts of image or how to "outpaint" an image and add something to it's sides.

[lost granite]

If anyone has really good guide I can follow to install a localized version of SD, I would really appreciate it. I was CS Major for a time and I have engineering background, so I should be able to follow. I have a single GPU with 11GB of RAM: 2080ti. I have a 32 core AMD CPU though I don't know of any models that utilize CPU, but I'm just getting into SD, so I don't really know. Any guidance to where I should read would be most appreciated. Thanks.

[waxen pasture]

Check out #1072220168534642768 , there should be some tutorials and resources there!

[lost granite]

Oh Excellent. Thanks.

[wet solstice]

I unfortunately don’t have the hardware

[fervent thunder]

You can use Collab or Kaggle

[proud nova]

Paperspace isn't bad too but their limited storage and Free GPU almost always out of capacity isn't as good as those mentioned above

[fervent thunder]

yea that's an option too , but it's veeeery slow, it's about same speed as my 1050ti , I don't even get how is it that slow with 8gb vram

But well, free and technically unlimited, so why not

[quick holly]

Guys, what happens if I merge two models?

[wide juniper]

When I use my laptop over my home's Wi-Fi network, is there any way I can use the local link to connect, rather than using the gradio link?

If it's http://127.0.0.1:7860/ on my PC, is there a way it can be translated to be used by another machine on the same network?

[proud nova]

Parts of those two models will merge into one. That means it might improve something but also can make something worse

For example lighting

[quick holly]

You mean it's random?

[proud nova]

You can choose ratio of how many parts of each models you want to have in the merged one. You can merge up to three models

[olive wasp]

yeah in automatic

this fella explains it well

https://www.youtube.com/watch?v=ln1RMEPKx3Q

[fervent thunder]

{light | dark} skin, should work as a prompt right?

[fervent thunder]

https://www.youtube.com/watch?v=IWz8ln4tkmg my first youtube tutorial ! you can control pose using stable diffussion now!

[rugged bough]

why this moon spawning multiple times "(1giant moon:1.5), "

😦

[rugged bough]

if you want between light:dark is the way

[fervent thunder]

ok heres what i want to do, make a batch of characters with variations of skin color

[rugged bough]

just add like multiple people as 10people or 10girl whatever and add brown skin, dark skin, white skin ....

it will assign automaticly i would do it like that

[fervent thunder]

that doesn't work with portraits though

[rugged bough]

hmm

[tall token]

how i train the a.i?

and how that work

[fervent thunder]

maybe too high gen resolution?

[rugged bough]

check the Aitrepreneur youtube

[rugged bough]

damn made 1 moon but now it i ssmall 😄

dang it

i think this model trainied only in squares

i am genereting wide image and it is freaking out

[fervent thunder]

Most models trained on 512x512 squares

some on 768x768

generate smaller and outpaint it further or try to use high res fix, might help too

[wide hare]

can someone give me some good VAEs?

[stray jasper]

Hi, Can I use the images that I generate in SD from my PC commercially?

[fervent thunder]

I think it depends from model you are using, some has it's own license

[stray jasper]

I am new to the subject, I installed SD1.9.1, where can I see those licenses for each generator?

[fervent thunder]

Look at each model description when you download it, if it doesn't have one , you should be fine as far as I understand

example , protogen
https://huggingface.co/dreamlike-art/dreamlike-photoreal-2.0/blob/main/LICENSE.md

[gloomy anvil]

Just dropped a new embed pack for 1.5, emotion controls - https://civitai.com/models/8860/scg-emotions-pack-simple-embeds-for-easily-controlling-a-variety-of-emotions-for-all-15-models-and-merges

[stray jasper]

ok thanks, i have homework to do here...

[harsh kayak]

Where do we type the #1074739806833217606 command?

[wild steppe]

Any chat channel!

[bleak whale]

Was about to ask the same thing.

[storm sluice]

Show me love 😘

[inland pewter]

🙂

[oblique jay]

Is there a way to train an embed online? My computer cannot handle training and I wanted to try to train an embed of myself

Or is this only something you can do locally

[fervent thunder]

Has anyone compiled an easy install without having to go to multiple sites and do a load of stuff?

[oblique jay]

There's one that I know of, but I don't know how advanced the program is.

[astral goblet]

launch with --listen, but then because that's on you wont' be able to update or install extensions

[tall token]

I Love myself

[astral goblet]

Saint Valentine is also the patron saint of beekeeping, so i always like to aspire to that aspect of the day

and BEEE kind to people

[fervent thunder]

I'm too impatient to download everything separately and then install it all 😭

[astral goblet]

i beelieve in you

[oblique jay]

https://github.com/cmdr2/stable-diffusion-ui This should be good for you then lmfao

Keep in mind that the downloads will take some time, so if it looks stuck, there's a chance it isn't.

[fervent thunder]

I'll download it later 😊

[oblique jay]

Hope it goes smoothly 🙂

There's always #🤝｜tech-support if you need any help, I've gotten plenty of assistance frmo there.

[high wedge]

Hey, question for people working hard to train amazing models 😊 what's your monetization strategy?

Do you think it's good to go for Midjourney style subscription, credits, selling models somehow or something else?

[lusty mason]

Happy valentines day stable diffusion people

[hazy lodge]

@midnight sphinx thank

[astral goblet]

You'll be happy to know that Saint Valentine is not only the patron saint of romantic courtship, but also beekeeping!

[lusty mason]

That's cool I suppose

[fervent thunder]

nice

[astral goblet]

🐝

[jaunty fiber]

Guys I am alone today, who can I give a sweet message? Even I want to send out some love

Else I’m targeting the announcement admin for being an awesome announcer

Stable diffusion

[astral goblet]

stability

well, unfortunately, it literally means it doesn't rock

but in a weird way, it does

[jaunty fiber]

ah yes

[restive fractal]

Hey all. Can anyone point me to where "clarity clarity 14" model could be found? I saw some great gens with it, but I'm unable to find it so far...

[jaunty fiber]

Sorry I’m not exactly sure which model that is 😅

[restive fractal]

On civitai it is noted as clarity_clarity14

[stuck furnace]

Stability releasing absolutely useful discord features😹

Useless

[frigid locust]

Happy Valentine's day

I'm crashing my car in a tree tonight

[jovial tinsel]

AUTOMATIC1111 if possible, please!

[wild steppe]

look at you guys go in #1074739806833217606

[frigid locust]

stfu

[wild steppe]

why not embrace the wholesome spirit

[fervent thunder]

Bonjour

[frigid locust]

Cause nobody has ever once loved me and this just makes me want to kill myself even more

[stuck furnace]

Lol

[frigid locust]

should be the bare fucking minimum to expect when putting hearts and shit everywhere, not everyone's doing well

[stuck furnace]

Lmao

[wild oasis]

what happened to the anime channel D:

[lapis gyro]

How can I find people who make AI art and can follow each other on Twitter?
I use the translation. Sorry it's hard to understand what it means.

[wild steppe]

moved over but still here #1072013871730131004

[wild oasis]

O

[wild steppe]

check pins for details!

[lapis ibex]

Hi

Wait so how do I use the dream bots again?

[jaunty fiber]

Ofcc!!

[drowsy robin]

From my knowledge, all of the current dream bots are archived.

[jaunty fiber]

🥹 great bot idea whoever came up with it

Kevy thought of it right?

Or made it

[lapis ibex]

Kk. Just wondering. As I see the dream bots but they seem to be locked.

[wild steppe]

Was a general consensus in the team to show appreciation in the community! Kevi coded it into our friendly neighborhood @surreal perch 💘

[wild steppe]

regarding dreambot: #📣｜announcements message

[drowsy robin]

That is because you have the Dreambot's Archive role.

[jaunty fiber]

Ahh amazing 🥺🥺

[drowsy robin]

If you wish, I could just direct you to a random online generator.

Like picfinder, or just a huggingface space.

[waxen crag]

How did you get these to get the expressions so well without it also copying the shape/skin/eye color of the face itself?

[gloomy anvil]

just a lot of testing

[waxen crag]

you use really descriptive filewords in the training?

[gloomy anvil]

I've had emotions controls for months now, just finally starting to get them out the door

next up is creature features 🙂 I've got a dozen or so different monster types, zombies, werewolves, vampires.. even hobos 😛

Also reworked my Apoc series for 512, that's coming too, though I'm still working through revisions on that one

[waxen crag]

I've seen other TI like that, but they usually cause the faces to look like the people in the training data... but these don't seem to have that issue

[gloomy anvil]

nah they won't, they're not trained. I vectorized them directly

[waxen crag]

Ah so you're using an embedding editor to make these?

[gloomy anvil]

yep

[waxen crag]

wow, that's insane

[gloomy anvil]

I do all 3, training, hybrid, and pure TI's. all the emotions and my toolkit are all pure TI's.

[waxen crag]

I guess I'll have to learn how to do that 🙂 the results speak for themself

thank you for sharing

[gritty gale]

Where is the dream bot?

[wintry stream]

You should be able to find all the context in #📣｜announcements or #1047610792226340935!

Direct link here to the post: #📣｜announcements message

[bold hemlock]

/showlove

[maiden crystal]

how can i use dream bot channels?

[nova sail]

Is anyone here using stable diff with blender?

[deep egret]

Do they still have the early member roles?

LFG they do

[static snow]

I'm trying to train an embedding to look like myself and it looks NOTHING like me after 3000 steps, any tips or tutorials?

[vestal dew]

what does it look like? Send it. Let people try it. What strength are you using i n your prompt?

[winter pike]

try dreambooth, embeddings are terrible for people

[glass vortex]

Hi, I do not see Dreambots, what I do wrong?

[wild steppe]

#💬｜general-chat message

[peak orchid]

Anyone seen the new discovery regarding solving the 50% gray issue with Stable Diffusion models? https://www.crosslabs.org/blog/diffusion-with-offset-noise

Might be a game changer for model training

[astral goblet]

i've been watching this and it's beginning to perculate into my gray matter

think i'm coming to an understanding

It'll be useful knowledge as i start building a dataset for my project

[plush forge]

Hi there! Is there any way I can access my auto1111 via phone? Or is there an azp like a1111 for phone?

(Android)

[blazing socket]

Is there a limit to how many people I can showlove to

[wild steppe]

Nope! Spread the love 💘

(you will only be entered for the raffle once/day so everyone who has shown or received love has an equal chance)

[astral goblet]

there's a raffle?

[wild steppe]

#1051259766976040960

[plush forge]

?

[blazing socket]

I've barely been in this server since I first learned SD, been in another server so I can't remember their name zzz

one of the mods I think

[fair hatch]

So there is no way to look a naked pictures or what.

Stupid thing blurred everything out. Sux bad.

On dream studio if I type naked angle it blurs all the pictures. Like it’s editing them. Like a parental lock.

But it has the pictures but there just blurred out

Yes blurry.

I type in what I want to see and it will generate pictures, but they are blurry as if they are censored

Yes, real people.

[fierce gorge]

Hey guys im new here. I think that when I generate Images below 512x512 the AI does what he wants. Anyone knows if I need to take the 512x512 pixels? I get very awful results when I try to generate an Image for example at 280x280

[fair hatch]

How do I send you a picture

[fierce gorge]

drag and drop it here @fair hatch

[fair hatch]

How can I post a picture here

Ok so do what again

[verbal osprey]

any new models popping off anytime soon?

[plush forge]

Hi there! Is there any way I can access my auto1111 via phone? Or is there an azp like a1111 for phone?

(Android)

[waxen pasture]

interested in models or embeddings?

because I have a few embeddings that are pretty good

[verbal osprey]

just official model releases. I'm not sure what embeddings are

[waxen pasture]

Oh, official models, the most recent is 2.1 I think

[frail isle]

Some1 gotta do art of the 1% https://www.oxfam.org/en/press-releases/richest-1-bag-nearly-twice-much-wealth-rest-world-put-together-over-past-two-years#:~:text=According to Credit Suisse%2C individuals,record-smashing peak in 2021.

Idk why state dont declare war to the industrial complex...

[verbal osprey]

The system has failed

[forest apex]

nah it's working perfectly

[pulsar carbon]

Hi peeps, i made my first lora model today. I had done dreambooth ones in colab before but this was a new experience for me. It turned out ok but i think it could be better. I'm looking for any advice or tips on how to improve my skills in this area, particularly about captioning, so i can improve my model. Can anyone help? Or can anyone direct me to some good resources to learn more? Thanks! 😄

[plush forge]

Oooh what about?

[waxen pasture]

I have like 7 embeddings I have made (some pretty popular even) that are pretty good and unique. All on civitai here: https://civitai.com/user/junglerally_

[ruby kindle]

🤖

[pulsar carbon]

Oh hey, you made the merfolk one, i downloaded that. Good stuff.

I gave you a follow.

Here's the lora model i did today.
https://civitai.com/models/8926/yuna-final-fantasy-10-character-lora

[waxen pasture]

Oh yeah that one! Honestly that one could have been better I just decided against improving it. I'm currently cooking up another one for superhero portraits, which should be cool. It's my first time trying training on a resolution other than 512x512 (512x640)

And thank you for following!

[pulsar carbon]

Interesting. I'd like to train one on mixed resolution but i'm too scared lol, i don't know what i'm doing and i've heard it can cause problems if you don't adjust settings for that accordingly. But i really need to do so i can get some more vertical aspect ratio images of characters which include the full costume/outfit, and aren't cropped.

[waxen pasture]

Well I'll see how this one goes, all I did was make sure everything was 512x640 and set the training resolution to that.

[astral pivot]

Ti's arent working for me, can someone help?

[brave matrix]

Hey guys anyone else having a hard time dealing with all these anti ai art neoluddites?

Maybe I should post this in off topic?

[plush forge]

Nice, im wanting to get into custom faces ti hope is not as hard as anything else

[plush forge]

The what?

[brave matrix]

the anti AI art people

[plush forge]

I mean most of their arguments are moot, dull, or easily destryed so mneh

[brave matrix]

i know

but its impossible to get through to them

they're so ignorant it makes my blood boil

[plush forge]

Yes, its called religion

[brave matrix]

they often have no idea how any of this even works

[plush forge]

Sometimes evidence or stuff that shatters your viewpoint of life is hard to accept man

[brave matrix]

meanwhile they're using technology and even AI on a daily basis on don't even realize how hypocritical they are

[empty barn]

Why do you care?

Arguing on the internet is usually pointless.

[plush forge]

This

[brave matrix]

you're right

its just annoying

[plush forge]

Just ignore and move on

[brave matrix]

thanks

[plush forge]

Like there surely mustve been people against cars

Planes

Trains

Fira

[brave matrix]

there's part of me that wants to go to the chats and tell them how dumb they are all

but that wont be productive lol

[plush forge]

Literally any technology

[plush forge]

I mean, many people are stupid

Aaand people who belong to a 'religion' even more

By religion I mean a strong set of beliefs based on either nothing or must their viewpoint and backed by a community of like minded individuals

Not actual religion as in faith or god

[brave matrix]

i think as the output of creative AI becomes better (ai gen videos, tv, movies, music, etc) it will be harder for people to ignore its value

right now they always just say,"oh if you zoom in close enough the ai images don't look good"

never realizing how far its come in just 6 or so months

[plush forge]

if you want to actually convince people

You have to gaslight them into thinking you agree with them

And ther change their mind from within

"Oh man this ai stuff is terrible imagine if everyone has access to do art wether or not they have the abilities"

"Have you heard of japans netfliv anime made by ai so terrible imagine if everyone could just prompt their script and get their movie thatd be so bad"

"People who dont normally have access to doing a movie (because of budget and personnel reasons) suddenly being able to express themselves? Nah mate, absolutely terrible"

[pulsar carbon]

Yo Tito, we better not have any trouble.

https://tenor.com/view/chuck-liddell-stare-ufc-look-kick-boxer-gif-17725529

[plush forge]

"People doing porn with ai generated images instead of abused porn actors in weird power dyramics usually embedded in sexual abuse? No m8 that actually sucks i love the incredibly dangerous dynamics of actual porn"

[plush forge]

Yo

What you doing here boy

[pulsar carbon]

lol

Iceman keepin' an eye on ya

[plush forge]

I swear we must be the most obscure (yet not incredibly niche) references hahaha

[pulsar carbon]

hehe

[plush forge]

@brave matrix just use this phrases while agreeing with them and being viciously anti AI

A drime will eventually drop

[plush forge]

Its actually funny

I chose tito cuz of an old ufc game I used to play and he hed the flame pants

Also @brave matrix there is a point against ai btw

A valid one I mean

[pulsar carbon]

Was it that one on PS1? I had that as a kid and didn't even know what UFC was back then.

[quiet shell]

[plush forge]

Sega Dreamcast, most probably we're talking about the same game though

if it is https://www.youtube.com/watch?v=EkrdmDLveUw this one, then yeah; again I had the dreamcast version though

I mean, most people download Auto1111 with SD so yeah

and mage.space kinda addreses al those points

If you have a good gpu, you can just run your own 'server' pretty easily

and pay no one

if you wan't the more accesible yet steramlined version then yeah

what? how would it either be compatible or non compatible with blender?

Also blender and photoshop are two different things? One is for 2d art, the other one for 3d

Oh ok, any way

Yea yea I didn't tak it as mean lol

https://www.reddit.com/r/stablediffusion/wiki/local/

@minor wadi welcome to AI

there you've got a link with what I think is necesary for you to enter down the rabbit hole

again, it's local so you must use a at least somewhat good gpu

otherwise it'll take long

about an hour or so, yes

auto1111 and SD are open source long ago

well, open to public

No idea if it's actually open source

but I mean, why would it matter

[empty barn]

It's a gradio-based front end.

Stable Diffusion is the method and you can download many models from sites like Huggingface and Civitai and run them using A1111's webui.

Which GPU do you have?

That's a CPU brand and a resolution. 😄

You are using Windows, I assume? Check the task manager.

And does it show a GPU?

RTX is fine, any RTX should work.

[proud nova]

1080? I struggled with CUDA on this one

[empty barn]

Most likely due to Windows, not the 1080.

[proud nova]

It's NVIDIA GPU - it should work so it's true ig

[cerulean kraken]

I can now basically turn anything into bone

[native meadow]

anyone can help explain something real quick about stable-diffusion

[delicate totem]

How long does stable diffusion take to generate "3d inpainted mesh of an image" in depth tab??

[cold siren]

imagine if anyone could produce any kind of entertainment, instead of paying zillions of money to a bunch of american companies force feeding the entire world with their superhero bs 🤔 sounds dangerous

[ancient epoch]

Hey all, hope you're good 🙂 I am looking for resources on SD for gamedev. Specifically, I find it impossible to keep consistency so to create 2D pose animations, or consistent enough to create 2D backgrounds (I am working on a Point'n'Click game). Any pointers would be great!

[wide hare]

Hey guys can anyone recommend me loras for good hands?

[plush forge]

hi there, I need help with a problem while trying to preprocess training,

when I try to preprocess it get's stuck and it says

||images return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking)
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 146.00 MiB (GPU 0; 4.00 GiB total capacity; 3.40 GiB already allocated; 0 bytes free; 3.43 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF||

what should I do?

[plush forge]

I don't know any but search in civitai?

[wide hare]

you don't have enough vram

[wide hare]

i did

[plush forge]

But how do I get to minimize the amount of ram needed?

[wide hare]

didn't find anything useful

[plush forge]

shouold I preprocess image by image?

[wide hare]

depends on the upscaling resolution and such

[wide hare]

yeah probably try that

[plush forge]

it's 512x512

[wide hare]

if you are doing more than 1 batch size try decreasing it to 1

[plush forge]

i'M JUST PREPROCESSING

[severe idol]

I won the giveaway

My first giveaway ever

I didn't expect to win

[plush forge]

NICE!

[wide hare]

damn!

[severe idol]

When I opened discord and saw the mention I was confused and then I saw that I have won!

[plush forge]

How can I create a public link? it says I have to set share=True in launch(). but i don't know what launch() is or where it's at

[plush forge]

or if there's a way I can access my a1111 through my phone

[wide hare]

add --share in the webui-user.bat file after the set COMMANDLINE_ARGS=
like this:
set COMMANDLINE_ARGS=--share

if you have --xformers just add a space and then --share

[plush forge]

Also it says there is No module 'xformers'. Proceeding without it.; shouldn't it be automatically installed?

[plush forge]

Thanks!

[wide hare]

no problem

[plush forge]

Do you know how to get the xformers module?

[wide hare]

it should install automatically

[plush forge]

huh ok

and do you know what You have unused kwarg parameters in Slider, please remove them: {'min': 0, 'max': 500} means?

[wide hare]

not really

XD

for xformers try reading this

it might help

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Xformers

[plush forge]

oh god

it's a nightmare

I have to install like 70 different things

ffs

[wide hare]

[plush forge]

Also how inconsistent is xformers in giving results?

I understand that it'll never give 2 images the same, but just how different are they?

[fervent thunder]

VRAM, that's GPU memory, not RAM
You can launch a1111 with "--medvram" or "--lowvram".

With 4gb I'd recommend always running with --medvram and --xformers , unless you need gen consistency.
(in this case remove xformers)

--lowvram might help achieve higher resolution or increase batch size, but it will decrease speed alot.
--medvram speed reduction isn't as significant.

also there are some settings that will move some stuff from your VRAM to your RAM, it might help too , just scroll through settings

[wide hare]

What he said XD

Hey soulless any chance that you have a lora or something for better hands?

[fervent thunder]

only if you'll add it as an argument

[fervent thunder]

I wish I had...I'm inpainting them manually 😦

[wide hare]

damn

I will try to make one then

[plush forge]

Has anyone here done a textual inversion?

Do you recommend when trying to get a specific face, to get rid of the other people?

[lament steeple]

Ask me again in about 3 hours 😛

[plush forge]

Loolll

[lament steeple]

Yes only one subject per image

[fervent thunder]

ping me if you'll achieve decent results in it

[plush forge]

@reminderbot remind me in 3 hopurs

[lament steeple]

I have one training now

[plush forge]

How should I remove the other people?

like, what's the best option?

[wide hare]

there are extensions for stuff like that but you should mostly only try to create images with one complex subject

[lament steeple]

The best option is to take another picture

[wide hare]

in the negative prompt put multiple people and such

[plush forge]

what?

[lament steeple]

other than that, try to inpaint or crop the other people out

[plush forge]

Ok! Thanks

[fervent thunder]

If it's not something you really like and want to keep - I'd suggest just doing another generation instead...
Inpainting , manipulating image is pain...

[lament steeple]

Tito is talking about the training set, not the generated image

[fervent thunder]

oh, sorry, not paying attention 😄

[wide hare]

multiple people, 2girls, multiple girls, multiple boys, crowd,

stuff like this

[plush forge]

I'm talking about the training set

[wide hare]

oh

yeah just saw

XD

nvm

[plush forge]

hahahaha

I was so confused by your comment

[fervent thunder]

I noticed doing numbers like 1girl , 2girls , mentioning amount of fingers most of the time achieves nothing , unfortunately 😦

[plush forge]

I thought maybe I could negative prompt into the description

[wide hare]

it really hates human hands

[plush forge]

try using (deformed fingers) as a negative thought

prompt

[lament steeple]

It would just get confused

[wide hare]

or at least non deformed human hands

[fervent thunder]

yea, deformed and disfigured is part of my default prompt

[lament steeple]

Neural nets aren't smart enough to know "This person, not that person"

[plush forge]

ok

[vernal wigeon]

sup y'all

❤️

[plush forge]

maybe I could use a prompt or something

like

a (subject i care about) next to a human

or smth like that

[wide hare]

hallo

[lament steeple]

you are making it harder for yourself. open the picture in gimp and take them out

[plush forge]

okok, i'll train this one as is and if the result is too bad I'll do another one

[plush forge]

Any tip for training textual inversions with 4gb of ram? I don't care if it takes long but it interrupts because it has no memoy