💬｜general-chat

[lofty parcel]

That's why I browse the SD subreddit 😎

[fervent thunder]

Was it a larger data set? I know it was censored more, and upscaled to a higher resolution I think.

[slow bane]

Smaller.

[fervent thunder]

why is it so much worse?

[slow bane]

Because you haven't been rating images hard enough. Get back to work.

[fervent thunder]

look man, it's not my fault i don't wanna rate images of cavemen :T

[slow bane]

Well, too bad, and now we've got subpar images of cavemen because you skimped on them. You gotta live with the guilt.

[fervent thunder]

is it really that bad, am i really that fucking integral to this process? cuz i'll bite the damn bullet. i'll do it.

I have to say I don't think I like 2.1 compared to the pre v2. All of my results look more like Dalle 2 than stable diffusion I'm sued to or Midjourney.

Maybe the prompting is all wrong how I'm attacking it now

[slow bane]

It works better if you actually write them as captions.

If you look at the LAION dataset, you can actually see the captions that the model is trained on.

[fervent thunder]

"an owl with skin the texture of pineapples high detail intricate octane render photorealistic" would usually create a vivid hybrid, I posted an example in the dream bot chat .

[fervent thunder]

I'll look at what other people are using for prompts in the bot channels.

[bright cradle]

how img2img wokrs

[brisk belfry]

Hey guys is there an easy way to replicate what lensa app does with the magic avatar

On sd

[halcyon oasis]

The dreambooth tab isn't showing up after installing it in automatic1111 and restarting... Any ideas? I'm on CU116. Can't seem to figure out where d8hazard's scripts are meant to go though, the instructions are a bit murky...

[halcyon oasis]

Dreambooth is meant to be how to do that, but I'm having issues even getting it set up

[inland vapor]

1.5 is the best)) in 2.0 everything was spoiled

[fervent thunder]

I feel like it's possible that stability AI kind of shot themselves in the foot with this recent release. An attempts to placate people that were worried about AI image generation, it might have lowered the quality of their product in many aspects. I will have to continue playing around with the new model to see if I just need to learn how to prompt with it better or not.

[inland vapor]

I can’t understand at all what she needs to write in order to get a normal photo

[whole oriole]

@wise stratus will @novel ocean be able to invite to other servers

[dapper forge]

Good morning

👋

[swift remnant]

Ahhh alright. Thank you!

[formal shard]

Hi I have an active use case to use SD to reduce cost and time for creating custom experiences. I am looking for some help create a custom SD model that can generate interior images that are top down with tables based on prompts, anyone here with experience in creating custom models like this? Please DM if you are interested in collaborating. - Here is the image https://imgur.com/gallery/dr03vKF

Sorry if I am posting in the wrong place! Let me know where I need to post.

[gaunt agate]

Yes. You can run this colab to produce the same results as Lensa: https://colab.research.google.com/github/ShivamShrirao/diffusers/blob/main/examples/dreambooth/DreamBooth_Stable_Diffusion.ipynb

[still plaza]

Tried using ChatGPT to generate negative prompts.

"Generate a list of words separated by commas that describe the negative qualities of a drawing or image"

blurred, chaotic, cluttered, dull, fuzzy, inadequate, indistinct, jumbled, lacklustre, messy, muddled, nondescript, unappealing, unimpressive, uninteresting, unsatisfying, vague, unrefined, unsharpened

[ocean mango]

https://www.youtube.com/watch?v=7I854do63Lg

[forest apex]

saw someone ask gpt to generate a midjourney prompt earlier

[cinder quiver]

Does anyone know how to stably generate a picture with a fixed perspective?

[forest apex]

perhaps specify a lens type or other photographic keywords?

focal length, some other ones

[cinder quiver]

But I mainly use it to make buildings in the game

different style

Wouldn't there be a keyword like oblique 45 degree angle that would work

[forest apex]

well. I don't remember some that I have seen, but I am asking "the artist" on character.ai right now

ok, it didn't have that great a suggestion besides using 'fixed perspective' in the prompt. ¯_(ツ)_/¯

but that AI is probably a lot more direct than asking on discord, tell ya what haha

[willow crescent]

Indeed, shapes look basically boring in comparison from what I´ve seen so far, like they were taken from photos only, lots of typical stock footage watermarks as well indicating the issue. Overall it´s like what I don´t like about Dall-E visually for example.

[urban dome]

Is there a consensus on which of these newer samples is the best, or the best value for speed/quality?

[fervent thunder]

im not a fan of the prospect of "default negative prompts"

feels like kid gloves and very condescending maybe i want deformed hands

systemwise it seems stupid becuase its clearly a bandaid repair

[cinder quiver]

OK, thanks.

[forest apex]

can't you adjust the weight of the negative prompt that is used?

[fervent thunder]

a "safety setting" you have to turn off otherwise to get aclean pure run is very clearly hiding a problem

seems to me that instead of coding in this "default negative" garbage they should have fixed whatever bug its there to fix

[forest apex]

might be a lot harder 😦

[fervent thunder]

if they cant do it right the first time they shouldnt have ever bothered.

ignoring the bug will clearly only make it rear its head harder later on

especially if its allowed to survive through release.

[frank gate]

The negative prompt is there becsuse most folks want good pictures

But the ai doesnt knoe that

Unless you tell it

[fervent thunder]

thats what prompting for but i'll not insulted by them handholding the process

[frank gate]

If you want to make nightmare fuel hand things, it can do it

[fervent thunder]

also, what if i want an image with extra limbs?

[frank gate]

So dont use the default negatives?

[fervent thunder]

i could just turn it off but A thats an extra step and B the system has been predisposed against what i want

[cold siren]

what i know for sure is that NOBODY EVER would want an image with a freaking watermark

[frank gate]

Yes, because what you want is not want the majority want

What*

[fervent thunder]

then they shouldnt have trained the bot with watermarks, simple as

[frank gate]

If you are off in the weeds doing odd things, expect to have to put in more work

[wise sparrow]

lmfao

all i wanted is hands..

[fervent thunder]

if they need to have this default negative check in, then its clear to me they either fucked up on the training step, have a massive bug in the system they dont want to fix, think we're all idiots who have to be handheld, or all three

and any one of those cases would be bad enough

[wise sparrow]

preach that's facts

[cold siren]

isn't that a nice idea, right... might also avoid lots of copyright-related questions... and yet here we are 😐

[wise sparrow]

negative prompts is a dumb concept

[frank gate]

It is not a bug, the training wasnt messed up, and they are just doijg what the majority want

[wise sparrow]

i think it shouldn't be that precise

[frank gate]

If you want something unusual, expect to not be catered too

[fervent thunder]

the training was messed up and its an open blatant secret they stripped out the porn and artwork from the datapool

[frank gate]

Lol

Read the pins dude

They explain why things were stripped

Its not a conspiracy

[fervent thunder]

if i wanted to be talked down to by a pumpkin-spice-drinking pfp i'd go work at the starbucks you frequent.

[cold siren]

so have you read how they also filtered out most humans because teh NSFW filter was set to 0.1 instead of 0.9 ?

or was it 0.99

[fervent thunder]

any stripping out of the NSFW content would hinder making humans

[frank gate]

Your doing odd things and expexting to be catered too, and trying to spread conspiracy theories about things we already have explanations for

Not sure how you cant be talked down too when the answers are literally pinned

[cold siren]

🤦 using the word "conspiracy theory" in 2022

[fervent thunder]

ye ok karen.

regardless, stripping out the NSFW content would in fact hurt making humans at any level

NSFW content just flat out contains the most footage of people, for better or for worse

[wise sparrow]

YESSS bad anatomy

[cold siren]

stupid term that does not really mean anything 🤷

is a centaur "bad anatomy"?

[fervent thunder]

instead of adding in this "default negatives" feature, they could have not pussed out, and just included the NSFW and artstyle content in the training pool and the training algorithms would have sorted it out themselves

[wise sparrow]

basically an AI never seen a boobie, how can it have as good anatomy and prediction to what's under a dress as a bot that seen a lot of boobies?

be realistic

[fervent thunder]

that's my point, what if i want a centaur?

what if i want a picture of a boobie?

what if i want something that has a third arm?

[wise sparrow]

https://tenor.com/view/namztaes-comethazine-staring-crying-gif-23350920

[cold siren]

btw yeah, or any multi armed multi headed indian deity

[fervent thunder]

making eldritch horrors beyond mans comprehension is gonna be hard as fuck if theyre serious about the dumbass feature

whoever suggested it in the first place, i hope their toes get athletes fungus and their car battery runs out.

[radiant nacelle]

negativity is like an addiction

[cold siren]

and 2.0 is trying to get us addicted to negative prompts?

[gusty oasis]

Yes

[deft bronze]

yum, negative prompts are an interesting beast

[bold star]

Definitely useful for edge cases/refinements but don’t see a use a case outside of that tbh

[deft bronze]

It helps with annoying ambiguities, for example: "reception" is both a party or the front desk of, say, a hotel

and you can disambiguate

and there are other inherently mixed concepts that can be separated with it

Most importantly, I'm always happy for more tools ❤️

[cold siren]

good to have tools that work 🛠️ especially in a predictable way

[spark trench]

Is sample associated with seeds ? Why am I getting a completely different image when using the same seed at higher sample count

highres fix is turned off btw

[wise sparrow]

is there any way i could run custom models online for free? i don't have good hardware

[surreal perch]

@odd zephyr

Congratulations Bored#9041!

You have won a raffle prize!

Prize

2500 DreamStudio credits!

Please contact Aether to claim your prize!

@tidal bough

@odd wedge

Congratulations vicho#0575!

You have won a raffle prize!

Prize

1 month Discord Nitro!

Please contact Aether to claim your prize!

@tidal bough

[tidal bough]

Congratulations to today's first round of raffle winners!! @odd zephyr and @odd wedge, please DM me to claim your prizes!

[dapper forge]

congratz

[hushed tangle]

I get normal colors with negatives: blue, green

not like in #📣｜announcements

why?

[fervent thunder]

pls explain why dreambooth fine tuning only needs 20 pictures and not thousands to train.

[bright cradle]

Need hlp

Can i turn image to AI art

[unkempt spindle]

which sampling mode is best?

[ancient jolt]

quick question, how exactly do you use model weights in the web UI?
was downloading the cyberpunk model and.. apparently its not enough to just use the .ckpt

this here for reference https://huggingface.co/DGSpitzer/Cyberpunk-Anime-Diffusion

[fringe zealot]

did anyone try finetuning a model at higher resolution than it was trained on?

[strong cliff]

Hi, I'm trying to make game assets, for example bar maid, but pictures with cropped heads. It because some settings are wrong, or I must put "head" in keywords?

[fervent thunder]

Who uses Invoke?

[urban dome]

So 768x768 images are 4x the cost of 512x512 images despite there only being x2.25 more pixels. Seems excessive

[ancient jolt]

i always generate my stuff with 1024x1024

[urban dome]

1024x1024 is nearly 10x the cost of 512x512. For what I'm using the API for, that's not feasible

Do people know which is the best value-per-step sampler?

[balmy jungle]

there's a difference between porn and nudes that makes the bias in the space important. porn focuses on genetalia.

[ancient jolt]

idk, i mean it would've certainly not been feasible with my old gpu lmao
but now with a 3070ti a 1024x1024 image takes about 8-15 seconds
pretty good for most things

but yeah really depends on what you want to use it for

[balmy jungle]

does anybody get decent results with an 8GB 1070?

[ancient jolt]

should be okay just slowish
but you may have to generate things lower res and upscale it

[urban dome]

I'm using stability ai's API to generate about 10 images at once, and I only get paid if the customer likes the end image. So I need cheap and fast.

It's neat sending of 10 concurrent requests to the API and see them all return at pretty much the same time a few seconds later

I'm guessing that's 10 different GPUs all working at once, or something like that

[ancient jolt]

ah, fair
i guess playing around is easier if you just generate stuff on your own hardware

[summer swan]

how do i have less weight on my picture

i am using an imageprompt, as part of my prompt and it baisaclly comes out identical to the image

[ancient jolt]

set the denoising strength higher

[summer swan]

i

where is that?

[ancient jolt]

are you using the web UI or what are you using?

[summer swan]

yes website

dream studio website

[broken gull]

yet another day without a price

[ancient jolt]

oh
i uh...
i have zero clue about dream studio sadly :^)

i'm using this https://github.com/AUTOMATIC1111/stable-diffusion-webui

[tidal bough]

plenty more still to come, there's still hope!

[wild steppe]

Try lowering 'image strength', you may have it set on the max value, which will give you an almost identical image

[summer swan]

where is it located?

[broken gull]

Idk, the changes of winning is really low. that's why I don't really like these kinds of things. You think you have a change of winning but you're just not very lucky. (but this doesn't mean I am not happy for the other winners just a general feeling I have with competitions like these).

[summer swan]

how do i use this?

[wild steppe]

Init image tab where you've uploaded your image

[summer swan]

ty

[ancient jolt]

gotta go with the tutorial, its a bit more involved
but.. i think dream studio renders everything on their hardware, no?

the web UI renders it on yours, so unless you got a good gpu, that likely wont work to begin with

[tidal bough]

these are completely random raffles drawn from all 102,849 (and counting!) of us, so you are particularly lucky if you win! this kind of framework is something we could adapt for other things going forward though for sure; this is just one part of our big 100k celebrations server-wide ^^

[pale igloo]

Hello, how do i save my current settings and parameters so i don't have to redo everything every time i shut down?

[tidal bough]

keep those fingers crossed though, you never know!

[broken gull]

I am not complaining with how this goes, You can't make everyone happy with giveaways like these. I just have low expectations for winning to keep disappointment low.

[magic moss]

this isn't very important, but, for some reason i can't see inside the #announcements channel

[desert silo]

not finding an appropriate channel to post this so here it goes:
I would like to learn more about stable diffusion so I was looking for a way to generate my own models (from scratch and using another model as base). I couldn't find any resources about this.
I am not looking for a way to finetune a model but to expand upon it if it makes sense.
Can some1 point me in the right direction on this?

[wild steppe]

Could you specify what the issue is? Screenshot in #🏞｜general-with-images would be helpful!

[wild steppe]

Have you visited #🔧｜finetune ? Folk there could help you answer that question.

[desert silo]

I can post it there, wasn't sure if it would be the right channel because im not really looking for how to finetune but how to generate a model 🙂

[magic moss]

@wild steppe - it appears to be a permissions issue. screenshot posted where requested.

[reef epoch]

Hi there @wise stratus !

I was just wondering when you and your company are planning on releasing your own OpenSource language model that's just as awesome as ChatGPT? I know I'm being a bit goofy and silly, but I'm just super excited to see what you guys come up with!

Looking forward to hearing from you soon!

Best,
[ChatGPT]

but like actually how many is it? give us a mathematical time frame equation

using

[reef epoch]

Whats even Harder to Train a Language model or a Image Model?

[fast drift]

a high bamboo tower in the middle of the Amazon forest, solarpunk communities, highly detailed, colorful, concept art

[fast drift]

Does anybody have a guess on what could possibly be blocked here on this prompt?

cant imagine a reason for any of the words being blocked

[hushed tangle]

bamboo, they think its racist XDDDDDDDDDDDDDDDDDDD

[hushed edge]

Remember when SD had the option to make character art

[signal nest]

where's the best place to discuss: backlash against generative AI (copyright issues, artists fearing job destruction, etc). Whilst I'm super-enthused running SD locally (like its the biggest buzz i've had in years) I'm hearing negative voices form various communities and friends :/. How to alleviate their fears. and how to alleviate my fears that this will be taken away by legal backlash backed by a weight of grassroots artists choosing established copyright gatekeepers to defend their ability to get paid for their work

I have an answer to this all which is that I've been contributing little and offten to a CC0 dataset (my own photos + polygonal annotations) someone else started years ago. if everyone did this we'd have a huge clean dataset. So, it should be possible for the world to have this tool and no one is upset. But how do we get from here to there..

in an AMA emad gave some answers to me - "no copyright breach if you dont ask it to , +retraining can narrow it" . but when I passed this on to concerned parties.. they were not convinced, and insisted the fact that the original dataset uses scrapes makes the whole thing tainted

currently what i've settled on as a usecase is I'm making my own manually created low-res game art (ironically , SD has encouraged me on this) , i've experiemented with SD upres+variation , and I'll wait and see regarding whether or not this is saleable (if not, i always have the manual originals)

[fast drift]

Honestly, I have a hard time grasping how the same people that doesnt care if all of their most private data is scrapped for AI training closed models at shady companies like google, facebook etc can be so upset by someone training to release an OPEN model that at least everyone can use

[signal nest]

from my own circles, complete 'concept art' type images trigger the concept artists that I know on the issues above. They regard these as stolen. ("it's blending stolen concept art to do that.."). I am experimenting using stable diffusion as a glorified procedural texturing& detailing engine basically (and even then people I know and communities I partake in are arguing against it 😦 )

[signal nest]

yeah making it open to all is a great argument in SD's favour . I have one ex-colleague - the best 2D artist I know -and he's vehemently against AI art. I worked with him back in the day on graphical experiments in console gamedev eg we figured out shaders and he figured out good art to use them. I have tried to 'get him onside' as it were by showing him the ways he could benefit , like, "you do your line art and let AI speed you up by texturing". but.. he still Just Hates It 😦

[lament wraith]

Man this topic just takes over everything about AI, art and beauty

[signal nest]

It is important because you must prepare for and avert the risk we face here

[lament wraith]

It's like every day. Really needs it's own thread

[signal nest]

yeah thats why i'm asking hwats the best place to discuss it - I feel its important and big enough to need a subforum here

[lament wraith]

Makes it hard to focus on a canvas when there's other people arguing in the studio

[signal nest]

I'm not here to "argue" as such - I'm looking for ways to avert this tangible threat

[lament wraith]

Not speaking to you specifically tars9999, I get where you're coming from. Just expressing frustration at how everyday I come here, or anywhere about Stable Diffusion, this is what's being talked, or argued about

[signal nest]

yeah. thing is i'm just coming back from backlash elsewhere

[sleek osprey]

Whats a pruned model?

[lament wraith]

It's an important discussion to have, yet it takes over everything like kudzu

[sleek osprey]

and pruned fp16/fp32 (I know fp means floating point, but what does it mean for me regarding the art output)

[signal nest]

(I figured "general chat" was the best place to start on this subject, like the more specific subforums here exist for everything else to be talked about)

[lament wraith]

So I agree, another voice for having a dedicated forum or thread for the discussion topic

[signal nest]

"generative AI ethics" .. somethign like that

i'm hearing things like eventually a company like adobe will come in with lawyers , shut down things like this and monopolise these kind of tools for themselves.

(companies that have stockphoto libraries etc)

i do my best to re-assure the naysayers.. and i get the same sense as when you're debating climate change or whatever online.. both sides in the debate start out convinced one way or the other based on other factors and the discussion is futile. My conceptartist friend Does Not Like generative AI, and neverwill, and would probably back legislation to restrict it 😦

yeah for a minute i wondered if I sounded like a "concern troll", lmao

"i dont think this but my friend says.."

i'm a programmer who likes to draw a bit. for me "img2img" is literally the best thing i've seen in years. it'll turn my lowpoly and retro 2d art into presentable assets. it's like hope to return to the days of 1man creations , the magic of the 8/16bit days.

[lament wraith]

Focus more on that, while staying aware

That's my recommendation anyway. What we give attention to grows

There is a #1002293361526460608 channel as well, may be worth posting the idea in there too

[tawny kettle]

Guys, does anyone know if you can install a text generator AI like Novel AI for free?

[tired sigil]

How can I increase the step number more than 150 in web ui ??

[lament wraith]

Yes, research KoboldAI. They also have colab versions

[west geyser]

Does anyone who developed V2 and 2.1 CLIP models have prompt tips? I’d hope they’d have an idea of the syntax of prompts. A wall of negative prompt words doesn’t really help with guidance on good structure for the prompt itself. Some people on the 2.0 channel are saying embeddings are the best way forward and while they post great images I hope not to rely on dozens of embeddings and more on the “language structure” of the prompts.

[slow bane]

Well, there's guider_prompt as an argument to the bot, if that's what you're referring to.

[wild hull]

Hello guys, where i can find the config of dreambot please ?

[coral cave]

35mm kodak portra 400, ultra detailed photo of a mouse on the table eats leftovers from a plate, in a cabin inside a ship from 1899, in the background a bed with a little girl who has a fever and has patches on her forehead. dust. the furniture is in dark wood. environment lit by lanterns. frame taken from a Netflix series, photorealistic, film, 8k, soft light, global illumination, dramatic light, ultra realistic, hyper detailed,unreal engine, 8k, strong light --ar 3:2 --v 4 --q 2

[slow bane]

Wrong channel?

[fervent thunder]

thats sounds like a hopeful prompt. like thats what people think ai can do, but then it spits out a plate and a patched quilt as a resutl

[west geyser]

Was not aware, will look. Right now it’s like if v1.5 was English, and V2 is in another Romance language but there was no dictionary…. Poor analogy I know.

[oblique kindle]

cheese

[quaint yarrow]

artists lives aren't any more important than anyone elses. everyone deserves the right to put food on the table. the entire debate basically boils down to people being selfish. it's not about protecting artists in the slightest

[slow bane]

there's some of your problem.

too long.

[slow bane]

https://cdn.discordapp.com/attachments/1047759796440420352/1049756734639583232/3_5_mm_kodak_portra_4_0_0_ultra_detailed_photo_of_a_mouse_on_the_table_eats_leftovers_from_a_plate_in_a_cabin_inside_a_ship_from_1_8_9_9_i_width-1152_number-4_seed-4273015731ts-1670351907_idx-0.png

the bot gets pretty close there.

[winter pike]

i love how the mouse has its own table to eat

[strong cliff]

To alleviate fears, stop calling auto image generation "an art", algorithmic pixel coping from multiple sources is not an art. When people will use correct terms, real artists will not have any problems.

[slow bane]

Hm. Pretty sure it is a creative and artistic endeavor, iterating and prompting your way to the image you want; there's a reason why artists do better than non-artists, and that's a sense of art, history, and context.

[peak orchid]

It is art though. Its also not simply copying from source images

[winter pike]

if it looks like art then its art

maybe this will sound harsh but ai will force artists to level up their art, most of them are at amateur level

and pro artists like smdoesrt keep making the same drawings and characters over and over again

[fervent thunder]

why do so many people hate ai art i saw this one person go fully insane over it and they were like “i need to post my art on onlyfans or else ai might steal it!!”

[fervent thunder]

stable diffusion falls under fair use : aka research

[quaint yarrow]

ppl are becoming brainwashed and delusional is why. so much negative propaganda that is anti-ai art and ppl just listen to it

senselessly and without logic or reason

[fervent thunder]

it’s twitters fault im pretty sure

[quaint yarrow]

ppl's opinions and world views shaped by the garbage they read online. every year it's the same thing

[fervent thunder]

i go on twitter and everyone whines about ai art

but everywhere else is fine

twitter users are just…. something else

[winter pike]

its ego at this point, they think their art is worth training

[fervent thunder]

thats what i was thinking lol

i was like : if ur that worried about it im pretty sure nobody wants it

[winter pike]

"subscribe to my only fans/patreon, i cant post it for free or ai will steal my art 🥺🥺"
theyre definitely taking advance of this cause they know people will support the poor human instead of the evil machine that steals everything

[sleek osprey]

is hugging face down?

[winter pike]

most of us trained ourselves into SD, not on some random twitter artist

[winter pike]

i think it was for a minute but now its up

[fervent thunder]

if their that worried about their art they should just host their own website which is out of reach from website scrapers

[tawny kettle]

tyvm, I'll give it a try 🙂

[winter pike]

the funny thing here is that they think someone will personally go to their website, download the images and waste their time+money training a SD model based on their art

[fervent thunder]

I think their worried that the large database that scraps the web is gonna take their art instead of someone training an seperate model

all the ai chats Ive tried are boring. they keep trying to give friendly advice

[serene swan]

I'm currently trying to create a model based on my own fantasy characters that are 3d models, I have a bunch of renders, do I need a bunch of real life photos of regular people (full body) for it to train properly?

[fervent thunder]

you could try to figure out openais playground

[fervent thunder]

I think dream booth is good for doing characters

if thats what you want

[serene swan]

I'm pretty new to all this, what's the difference embedding vs training, and would I be able to change styles once that process is completed?

Im pretty sure I'll need dreambooth bc I heard it takes a lot of vram to run locally

You were pretty helpful friend 🙂

[harsh kayak]

Why do my generations have 2 heads?

[quaint yarrow]

happens more often when you go above 512x512.

also in the negative prompts you can put: Duplicate, extra limbs, extra heads

[harsh kayak]

Ah okay, thank you

[jagged cave]

is there a way to select each second file in a folder?
e.g. i wanna take half my screencaps as 1920x1080, and half as 1080x1080. this works well since i have most screencaps double or triple with slight alterations. so it would be better to select each second image instead of say the first half and second half.

any idea on how to do that?

[winter pike]

With a embedding youre not modifying the model itself, so its good for common styles of art, it generates a .pt file.
With dreambooth youre adding new information to the model, information that wasnt there before, like people, things, very particular styles. It generates a .ckpt file which will be a variation of the original SD model

in #1047197565365538826

[tidal bough]

Time to roll a few more giveaways! Good luck everyone ^^

[surreal perch]

@covert zephyr

Congratulations meltdown#3672!

You have won a raffle prize!

Prize

2500 DreamStudio credits!

Please contact Aether to claim your prize!

@tidal bough

@lyric mantle

Congratulations Azyre#8381!

You have won a raffle prize!

Prize

2500 DreamStudio credits!

Please contact Aether to claim your prize!

@tidal bough

@neon talon

Congratulations ryan.#9726!

You have won a raffle prize!

Prize

1 month Discord Nitro!

Please contact Aether to claim your prize!

@tidal bough

[serene swan]

thanks so much for that clarification!

[fervent thunder]

is there a way to enter the raffle or is it random discord members

[tidal bough]

completely random, picked from every server member! You don't need to be in this channel or online to win!

Congrats to our winners this evening! DM me to claim your prizes; More prize draws tomorrow! ^^

[fervent thunder]

cool

[frank moon]

What do you all add to prompts to get full body renderings? I cannot seem to get them to come out consistently. When I do get a full body , often the head is missing.

[winter pike]

i use "standing"

[fervent thunder]

i wanna see a model all trained on all royality free images to see how good/bad it is

[visual oasis]

pay as you go colab google it's enough for to do SD? What advantages do you notice and what limitations do you see?

[pale igloo]

what is the best promt to have the person looking straight ahead? I have tried ' forward facing' and 'looking straight ahead' 😅

[fervent thunder]

Am new and how do i use stable diffusion

does anyone know if dream booth works w/ fursonas

[rapid tulip]

is it possible to train SD by providing, say, 100 examples of

pencil sketch -> final inks

kind of like these:

• https://www.instagram.com/p/Clt68kEP20c/?hl=en
• https://www.instagram.com/p/Clt30y1v0JQ/?hl=en
• https://www.instagram.com/p/ClratXmPggV/?hl=en

and then I can simply provide a pencil sketch and have it render out the final inks in the desired style?

[fervent thunder]

you could try to
Make the sketch line’s thicker in photo shop, invert the colors then put it through img2img and write something like “ink blot art”

if you just want those results idk about training though

[rapid tulip]

ah, damn. thought it'd be dope to train an AI to understand a style through a series of basic input -> output examples

[signal nest]

yeah SD for personal use , pretty sure thats ok. Where things get tricky is when I want to use it for my indy game efforts, which i'd like to be able to sell on some app store eventually . But my curent strategy is manual art that SD could optionally enhance. Hedging 🙂

[visual oasis]

pay as you go colab google it's enough for to do SD? What advantages do you notice and what limitations do you see?

[polar widget]

How do i turn off the safety filter for the online generator?

[soft basalt]

you guys know how to access ChatGPT in a country where I can't make an OpenAI account?

[worthy oriole]

Any news about this? (GPT-3-like performance but open source and able to fit in consumer hardware...)

#💬｜general-chat message

I know that it has been just a few time, but it woud be nice to know if things are going smoothly

[manic yoke]

what sampling method does the bot uses?

[fervent thunder]

i noticrd something about doing 3d art

if you try to make something realistic looking everyone tries to critique it but if you do a cartoony style everyone just compliments it

[thin hatch]

stability ai deserves any pr nightmare they receive from the stealing of stable diffusion

[fervent thunder]

guys uhh i saw this app where you take 2 images like a knight and a cat and it turns it into a cat knight

what ai would do that?

[sand wigeon]

Lmao soon OnlyFans becomes OnlyGans

[winter pike]

gpt-3 is soo good 😭, shame its not open source

[sand wigeon]

Would be a shame... If stability had... Multiple research groups working on open source LLMs ...

Also looking forward to what this week's POW will bring, #1010577750077210726 has always been the busiest dreamer community so I expect amazing stuff 😁

[winter pike]

not related but is dreambot using the distilled version of sd? im confused, theres a lot of progress this week

[fervent thunder]

is there any furry models for all species

[sand wigeon]

Not yet afaik, I can't generate hq images in 2 steps yet

When it's distilled, you'll know

[winter pike]

thank you im very hyped about that

[sand wigeon]

As are we 🔥

Most of us are in the same boat as you lol just a bit earlier in getting mindblown

[balmy lantern]

What's better for img2img for realistic faces? 1.5 or 2.0?

[sand wigeon]

2.1 👀

[balmy lantern]

Mmm... I might try it later and compare them

[fervent thunder]

1.337

[hushed edge]

anyone remember when sd could make ascii art? what happened ot that?

[jagged cave]

anybody here aware of a website that has screencaps of starwars clone wars season 7?
fancaps.net only goes up to season 6

[tribal sequoia]

stablediffusion.fr/artists <--Is this information available anywhere in a more organized form? Right now I'm going through and manually picking out artists with styles/eras I like, but if someone's already sorted everything there's no sense in me reinventing the wheel.

[hushed edge]

+---------+
/ /|
/ / |
+---------+ |
| | +
| | /
| |/
+---------+

oops

sorry

[west prism]

Can you use loopback to make an image look better with the same prompts?

[dawn bough]

hey can someone help me bounce some ideas off or is there a thread/channel for that?

[stuck gyro]

is anyone else getting a lot of similar-looking images with dreambot that don’t relate to the prompt? regardless of prompt, i’m getting multiple pictures of people in wedding attire, and also pictures of hedges/topiaries

[weak owl]

is there a tutorial that covers more advance skill sets (negative prompts, models, etc>)?

[jagged cave]

which anime would you recommend if i need screencaps from an anime that has this standard typical ultra generic anime style but in high quality

Like makoto shinkai stuff or chainsaw is already too realistic

and ghibli is its own style

[untold ocean]

I would take a look at kyoto animation

Hyouka, Kyoukai no Kanata, K-On, Violet Evergarden, are high quality and modern, but not super stylized.

[jagged cave]

hm i think i go with either SAO or violet evergarden

[dusky pebble]

Can someone get me up to speed? Stable Diffusion 1 is released and can run locally? Is that correct? And Stable Diffusion 2 is not released yet?

[uneven wasp]

Stable Diffusion 1.0-1.5 is released and can run locally. Stable Diffusion 2.0 is released and can run locally. Stable Diffusion 2.1 is not released yet, but you can demo it with dreambot on this server as of right now

[dusky pebble]

Thanks

Any idea when 2.1 will be released? Which graphics card would be required or recommended?

[rich plover]

its just some are very very slow. i think people can even run it with a cpu

[uneven wasp]

no clue. It's really just a vram game.

I'm running everything on a 4gb card so

[dusky pebble]

What are stable-diffusion-webui and stable-diffusion-ui? Are they just wrappers with GUIs for stable diffusion?

Do you recommend either?

[rich plover]

yeah theres some way to run it with command line, but i just used automatic111's webui. you launch a .bat file, and then you can open a browser tab that runs locally, and from there, you can adjust all the parameters and prompt you want

[fervent thunder]

can someone help me? i dont know why chatGPT is trending now for some reason, idk why people think its good would someone be able to tell me why? it just says "as a large language model" and doesn't seem to share personal experience or arguments to the points i say.

[weak portal]

Anyone use unreal engine for post enlargement?

[tacit sigil]

hi

What is the command for the dream channels again? Thanks 🙂

[quick basalt]

Kings, when you guys think the SD outpainting will get just as good as the Dall-E 2 one?

[mint lily]

do we have a way of converting a sketch I make into a colored image now?

through what methods can we keep someone's looks consistent between image generations?

[severe zealot]

hi peeps, i have been attempting to train my own models on stable diffusion 11 11 dream booth but i just dont have the vram, can someone here kindly train them for me?

[severe zealot]

im willing to pay

[halcyon oasis]

Anyone know what this issue is? Trying to use the 768-v-ema.ckpt
RuntimeError: Error(s) in loading state_dict for LatentDiffusion:
size mismatch for model.diffusion_model.input_blocks.1.1.proj_in.weight: copying a param with shape torch.Size([320, 320]) from checkpoint, the shape in current model is torch.Size([320, 320, 1, 1]).

[halcyon oasis]

If anyone else is having the same issue, this seems to be the solution, downloading the config file as a yaml https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/5070

[fallen lynx]

Is this in use anywhere?

https://youtu.be/NJyZCdxnGe4

2x speedup in inference?

Also why can't I tag automatic? Is he not here anymore?!

[midnight quartz]

How does the new sd 4x upscaler compare with the existing upscalers that are out there? Is it superior in some clear ways or is it still along the lines of subjective compared to SwinIR, esrgan, etc?

[worthy oriole]

It's supposed get released This week

[shell dagger]

is there a place to see how the model file size changed between different version? It's getting smaller every time, right?

[safe sapphire]

• xformers is used by some
• automatic currently isn't here, that is all

[worthy oriole]

Welcome to october 2022 😂

(I say it for the question about Automatic)

[past robin]

On how much extra data is 2.1 trained?

[smoky holly]

help me remember one app/site that generates seamless txtures + bump/normal maps...
it ended with .cc

[sand wigeon]

I don't know! I think this kind of information will get a proper write-up sooner or later. We're still iterating on the protocol for release, keep questions coming so we get a good idea on what people are most interested in knowing upon new releases ^^

[smoky holly]

found it 🙂

-> https://www.ponzu.gg/

[wise sparrow]

can i use img2img in dreamstudio online?

[sand wigeon]

Yup!

[wise sparrow]

thank you, and what is the recommended photo strength to get good results? something in the middle

[sand wigeon]

I haven't used much but in my limited experience you'll have to play around based on purpose. Low strength (~30) mostly retains rough shape/composition and colours. Mid-level gets a bit weird sometimes, high strength is great for good control on iterations. I prefer high (70-90) strength, 9 gens, and prompt tweaking along the way for most purposes.

[wise sparrow]

thank you for the info 🙏

[grim aspen]

does someone know a GPT3 model that runs on a local pc like stable diffusion does?

[signal dome]

Why all of sudden lensa ai app becoming popular.... They using dreambooth but misleading users by in app subscription 💀

And dreambooth is free but they charging mass prices

Like locally

[mild basalt]

Is it possible to generate 3d skins for a game with this method?

[signal dome]

Not yet

Wait like 6 months

[surreal perch]

@gleaming galleon

Congratulations Bandana Dee never in smash#5248!

You have won a raffle prize!

Prize

2500 DreamStudio credits!

Please contact Aether to claim your prize!

@tidal bough

@echo warren

Congratulations Pacojamon#9158!

You have won a raffle prize!

Prize

1 month Discord Nitro!

Please contact Aether to claim your prize!

@tidal bough

[uneven wasp]

depth2img and project works, no?

[tidal bough]

Congratulations to our first raffle winners of the day!!

[uneven wasp]

if you already have the mesh

[signal dome]

Lucky

[signal dome]

Even more lucky

Legit ?

[tidal bough]

Legit! There'll be 4 prize draws every day in December, chosen at random from all our members. Good luck! 👀

[fervent thunder]

Microsoft has entered the chat https://www.bing.com/images/create?form=IRPGEN&PC=SANSAAND&ssp=1&safesearch=moderate&setlang=en-gb&cc=GB

[wild hull]

Hello Guys, there is a way to get the full meta data without truncate ? when im doing /interrogate ?

[broken gull]

hmm? 4? I thought you said 2 per day first kek

not that it is a problem tho

[tidal bough]

two sets of two winners per day

[broken gull]

damn, you typed long for one sentence kek

[tidal bough]

it's hard to decide which emote to use sometimes

[broken gull]

lmao

that's why I only use one mostly

yk you can put emoijs as fav right?

[tidal bough]

oh yeah, I just have a lot of favourites O_O

[broken gull]

reduce your fav to the most usefull emoijs, then use those when in doubt

kek

[tidal bough]

but how am I supposed to decide between and ?

[broken gull]

euhmm, you don't they're basically the same emoij

[tidal bough]

very subtle delay, very different message

[bleak matrix]

Good morrow, ya'll. How are we today?

[broken gull]

ig this is why releases are always so slow, considering they have to chose so long for basically the same emoij

[bleak matrix]

Naturally,one is for distress and one is for hunger

[tidal bough]

Morning! I'm good thanks, how're you?

[bleak matrix]

I am awake and that's what matters lol. If my eyeballs can open, that's a start, bahaha

What are ya'll up to today?

[broken gull]

trying to work on my idea for pow

[lofty parcel]

Finally having real success in making AI-generated game worlds
Doesn't look great yet, but I can really control it without compromising its creative freedom

[bleak matrix]

Awesome!! I'm so happy for you!

I'd love to see, if you're willing to share, of course

[broken gull]

You mean with depth? or just flat imgs?

[lofty parcel]

Flat for now, but I bet depth can do some spicy things as well

My preferred fork hasn't implemented depth yet so I'm using normal 1.5 stuff

[bleak matrix]

I'm glad you keep working at it.

I'm working on my...project for 1.5. I'm trying to refine it as I write, so I can give as much details as possible. I still hate compiling that wheel, but, it is what it is 🤣 I'm sure someone has prolly found a workaround by now

[lofty parcel]

Oh, best part is that I didn't have to make it look like Minecraft lol

[bleak matrix]

laughs Yeah, there's a surprising amount of Minecraft clones out there. Either that or making it look like every other turn-based JRPG.

I think your base was pretty different, though. Even if you were blocky, I think you'd have enough deviation

[mild basalt]

What do you mean?

[uneven wasp]

have an existing mesh, render it out as a depth map/image, run it through depth2img, project that image onto the mesh from the same camera perspective the depth map was rendered from

[wise sparrow]

dreamstudio been getting stuck on 100% a lot lately

[mild basalt]

Do I need to repeat this for every angle?

[uneven wasp]

You would, yes.

[austere venture]

I've no idea what I'm doing with this, so I'm hoping that someone can point me in the correct direction:

What I'm looking to do is create a model that can create an extremely limited set of images, rather than the full fantastical gamut that SD can produce.

For example, given a set of photos of a shoe at different angles (front, back, side, top), in different colours (red, green, blue), I'd like to be able to prompt for "shoe back black" and have the model produce a photo that's identical (or close enough) to the shoes the model was trained on, except at a particular angle, and in a particular colour. I don't want the shoe in front of the Eiffel tower, nor do I want a random person wearing he shoe, just as close to the training images as it can get.

Does anyone have a tutorial that can lead me in the correct direction for this? I'm swamped in google searches trying to work this out.

[honest atlas]

Where can I post my NFT giveaway's?

[lament path]

Hi everyone, my startup (SelfieWiz) is Looking to hire a Stable Diffusion prompt engineer on contract. Anyone interested?

[bleak matrix]

You can just use Dreambooth.

[sly sapphire]

woot ultrawide

[honest atlas]

I am. Dm me. I’ll respond when I get off work.

[raven edge]

what does ema and nonema refer to when downloading models?

[fervent thunder]

Do you think stability ai will work on a text model simliar to chatgpt for example?

[wispy jasper]

Can anyone explain diff between EMA/Non_EMA and where I would use each?

[hybrid heron]

Ones used for training

[hasty wing]

Does Dream Studio support inpainting?

[austere venture]

All the dream booth examples I can find are for altering the existing SD model to include the images you provide. I'm looking for a model that is only based on the images I have. Can you suggest a tutorial / example for me?

[hasty wing]

You have a dataset large enough to train a model on your own?

[wispy jasper]

Which one is used for training EMA or non EMA?

[lusty charm]

Can anyone help me, I made a batch of images and Im trying to fix the eyes. But Im not able to draw the mask on in inpaint gui, its just a white image that shows up with a unhappy smiley in the top left corner. Then it comes back bug bugs out with a small image inside the main image etc.

[bleak matrix]

You should use the model you have. Yes, you could train SD from the ground up, but this a long, and expensive process. You are better off training from 1.5. If you don't have images of a particular shoe in a particular color, but have a picture of the shoe, just recolor the shoe in a graphical editing program. It's very easy to do.

[bleak matrix]

One is pruned and one isn't.

There's a lot of examples if you search the server.

Visually

[viscid ridge]

Is there a guide to prompting for beginners? And also terms glossary

[bleak matrix]

What are you using?

[peak orchid]

@wise stratus The punsafe value listed on v2.1's ReadeMe still says 0.1. I assume that might be a mistake? https://huggingface.co/stabilityai/stable-diffusion-2-1

[bleak matrix]

You can visit #📝｜prompting-help

If you need help and support

[viscid ridge]

Thank you kind stranger

[raven edge]

is it possible to get sdv2.1 to work with automatic1111

is there a ymal?

[bleak matrix]

I also have a giant guide on stuff, but my prompting stuff is a wip

[raven edge]

yaml*

[viscid ridge]

Can you dm it to me?
Or send it here if the rules allow it

[bleak matrix]

https://docs.google.com/document/d/1R2UZi5G-DXiz2HcCrfAFLYJoer_JPDEoZmV7wy1tEz0/edit

[viscid ridge]

Thx

[bleak matrix]

It's here

Np

You can also use Clip-interregator. That's when you upload an image and it'll give you keywords to use

You can also check out #1045349359044280360 or other people's promots

[steel rock]

for me dont work 😦

[bleak matrix]

https://huggingface.co/spaces/pharma/CLIP-Interrogator

Clip Interrogator is here

[wispy jasper]

Nope they are both listed as Pruned.

[cinder palm]

I am having issues installing Automatic1111 SD. Is this the right place to ask?

[lusty charm]

Im using automatic1111 webui

[bleak matrix]

I don't know what version automatics' supports, unfortunately. I do know some ppl are using 2.0 but that's as far as my knowledge goes

[bleak matrix]

You can ask here, but tech questions are typically answered in #🤝｜tech-support

[cinder palm]

thank you. i'll go ask there.

[lusty charm]

If you want a good life I would reccomend you to stick to a simpler interface, Ive spent maybe 2 hours making images and 20 hours on trying to get past some error

[cinder palm]

i can relate to that

[honest atlas]

Where can I post my nft giveaways?

[fervent thunder]

If I have generated an image with a face and I want to make poses with the same face, how do I do it?

[bleak matrix]

You could use inpainting to mask the face and then say not to change the area of the mask, etc.

[fervent thunder]

but then the face would always be in the same position right?

[bleak matrix]

Yeah, cuz you aren't changing that area of the picture at all.

[fervent thunder]

And is there any way to keep the faces in other positions?

[junior bronze]

according to this post the 2.0 yaml should work, but im just getting a black output https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5458

[raven edge]

same Im getting black generations

[junior bronze]

i tried both the ema and non ema models, same thing

[bleak matrix]

Idk. Inpainting and masking are prolly your best bet.

[past robin]

Also black images here. But will probably be fixed soon.

[bleak matrix]

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

You're using 2.0, correct?

[wispy jasper]

I must have misread.

Wait no Im right lol

One is EMA and one is NON EMA both pruned.

[bleak matrix]

Yea, super weird lmao

[brisk hinge]

guuuuys, can you keep your results private like in midjourney?

[bleak matrix]

Anyway, but the non and nonema mean that there's the normal weights and the other is just ema only...I think XD You can see a lot of visual examples of the differences of the outputs by looking around the server. Other than that, I'm afraid I'm not too terribly informed on the inner workings

[bleak matrix]

But mb; I was a little confused on the naming a bit because 1.5 was named differently but also pruned and pruned-emaonly.

Since one was 7.5GB and the other 4

[river jacinth]

can i run invokeai for outpainting with 1650??
i heard its better but invokeai doesnt recommend it

[velvet estuary]

Hello

Has someone been able to install SD 2.1 ? xD

[peak orchid]

The black screen goes away with the Automatic WebUI for SD 2.1 if I launch it with the --no-half argument

[velvet estuary]

How do I set that?

What is the difference between v2-1_768-nonema-pruned.ckpt vs v2-1_768-ema-pruned.ckpt?

[peak orchid]

Launch it in a CMD window with "C:\Path\to\stable-diffusion-webui\venv\Scripts\Python.exe webui.py --no-half"

[velvet estuary]

I did lunch it with
webui.bat --no-half
and it worked

[bleak matrix]

Awesome!

[peak orchid]

The only issue is that --no-half uses more memory

[agile spoke]

anyone has a good prompt guide for 2.0 / 2.1?

including negative prompts if possible

[fast yoke]

Hi there, I keep getting the " dreamstime" water mark on many images in SD 2.0, sometimes there is a camera as well.. I tried negative prompts, and its still there, any idea ? thanks.

[past robin]

Thanks, but I'm getting cuda out of memory with that. I'm on --lowvram

[cold siren]

same here 😕

[tribal oar]

quick question guys, what is EMA pruning?

And why do I want it / not want it vs. Non-ema pruned CKPTs?

[cobalt cliff]

honestly i kind of love seeing the negative and spiteful reactions to AI art

[tribal oar]

I just hate the misinformation, on both sides

People don't understand that this is here to stay

There's no putting the genie back in the bottle, and people are acting like if they tweet hard enough it'll just disappear

[toxic depot]

Sup, what would be a good resource of legal information about the commercial usage of stable diffusion ?

[bleak matrix]

The best thing we can do is continue to be a positive force for good

[honest atlas]

Giving away 2 NFT’s tonight! Follow on twitter or instagram! Aeye88nft

[young glade]

Once I train a hypernetwork to my face.... how do I actually use it? I've selected it in my settings but am still getting nothing in my results (like... the images generated aren't even faces)

[cinder ice]

@honest atlas damn rip account F for whoever lost their account to the bots

[tribal oar]

@honest atlas nobody wants your worthless NFTs....we're literally in a Discord where we can create whatever pictures we want lmao

[lofty parcel]

The value of NFTs is probably the revenue that websites and YouTube channels make from criticizing them 🤔

[tribal oar]

I think you could power a Dyson Sphere with that much energy

[lofty parcel]

Scale that is truly out of this world

[distant badge]

I'm finding when I do mountain landscapes that I am getting a disproportionately high number of watermark/copyrights in the image. Putting in copyright and watermark in a negative prompt doesn't really seem to help. Any suggestions?

[lapis bone]

Are there any specific key phrases to include in prompts to give a "magic the gathering art" feel? Seems like using the actual phrase magic the gathering tends to give full card images.

[junior meteor]

Hey, how do I prevent full body images from being created as stacked bodies

I mean, I don’t even know how to prevent humanoid pictures from having the same body stacked twice on top of each other

[fervent thunder]

so what kind of files are in the model.ckpt? I have renamed the model.ckpt as model.zip and it opens up into an archive of files without extensions. Does anyone know?

[daring warren]

can anybody tell me how can I share the images I created here...i just drag and dropped but doesn't work...send a file but won't open.... does it have do to something that is PNG image?

[tame shoal]

Has anyone seen anything for Heightmap landscape generation that uses AI inpainting?

[reef epoch]

is there no way to implement Clip guidance to locally?

[fervent thunder]

Emad said the CLIP thing would get a public release LAST week, i guess it was a bit of a lie because I thought they said it would also get a public release for 1.5 and never did

[reef epoch]

tbf all i want before Christmas is the finetune tools 😔

[tribal oar]

Don't we already have those? Or am I misunderstanding

[reef epoch]

Official no

[tribal oar]

https://huggingface.co/spaces/multimodalart/dreambooth-training

This works ridiculously well for finetuning

[real agate]

Yoooo gmg gm

[visual oasis]

pay as you go colab google it's enough for to do SD? What advantages do you notice and what limitations do you see?

[tribal oar]

Free is enough, I don't know what advantage paying will get you

But, I would recommend looking at the HF link I just posted

It incorporates ideas from the colabs into one resource, and it's really good at model training.

I trained many models on the free version of Colab before I found huggingface

I may or may not have created 5 separate google accounts so I could keep training for free.....

[opaque pebble]

I do not know how people are getting recognizable humans (or humanoid aliens) from any of these AI image collage programs. Everything I make looks like a mutant or an extra from John Carpenter's "The Thing".

[fervent thunder]

John Smith
Jon Smith
Johnathan Smith
Johnny Smith
Sohn Jmith

[tribal oar]

bob terwilliger

And all his sons

[tribal oar]

Can I ask what you are looking for? Specific images that look like /you/, or just images of realistic looking people in general?

[opaque pebble]

I have created probably hundreds of hideous mutants. Some typical prompts have been ... let me see if any are still in my clipboard history...

"klingon santa claus in a space ship cockpit"

[west geyser]

Here you go:

https://stability.ai/sdv2-prompt-book

[tribal oar]

fantastic resource

[tribal oar]

Jump in #🏞｜general-with-images and let's see if I can provide some specific feedback

Photorealism is my niche haha

[opaque pebble]

"barbarian dressed in red leather and white fur trim, sitting in a space ship cockpit"

[knotty wasp]

there is clip guidance in the diffuser repo

[reef epoch]

but no proper webui uses diffusers yet

unless there is idk

[broken gull]

I believe auto1111 supports diffuser

[reef epoch]

isn't this technically, basically the normal diffusers dreambooth training... which is basically free on colab? since free colab also uses T4

[knotty wasp]

that might be but the code is there, nobody bothered to make a ui for it i guess

[tribal oar]

Not quite, I've used both extensively now

The colab implementation requires you to keep the screen active, which gets annoying. Also you have to manually choose the number of steps, and the learning rate is constant.

Huggingface, in my personal opinion, is superior. It's $0.60 per hour, so training a model is like $0.50. You can close the screen, and the number of steps is auto-adjusted based on the number of images you are adding to the trainer

Also, the learning rate auto-adjusts as the model trains

I have trained my face like 5 separate times using separate data sets using colab (even tried merging them all together into one model lol, it actually worked pretty good) and twice now using hugging face

The second time I used huggingface was the most consistent my models have ever been

In terms of actual likeness matching, and actual consistency from one picture to the next

[reef epoch]

but what about training multiple people? or training on a trained model?

[tribal oar]

It's basically a github repo that you clone to your own account, and there's an app.py you can modify in it

In there it calls to the huggingface repos, so I'm sure you can connect to your own repo and model as a base

[reef epoch]

ah you can do multiple i see

[tribal oar]

But I'm not experienced enough to do that...I posted about it in #🤝｜tech-support

I will post comparisons in the image channel one second

[reef epoch]

have you you compared it to the joepenna repo aswell?

[tribal oar]

I have not used that one, sorry

[agile spoke]

eyyy thanks!

[fervent thunder]

I have an image generated with a face, how can I keep that face in the other images?

[tribal oar]

Does anyone here use wildcards in their prompts?

https://huggingface.co/spaces/Redw04/wildcards

I have started a space for saving and distributing them

I'm thinking we should get a pool of them going

[uncut plinth]

So whatsup with 2.1 or 2.0 any useful?

[vast pollen]

which anything 3.0 is the right one to get?
Anything-V3.0-pruned-fp16.ckpt
Anything-V3.0-pruned-fp32.ckpt
Anything-V3.0.ckpt

idk the difference besides file size

[tidal bough]

more giveaways? more giveaways! good luck everyone...!

[surreal perch]

@ornate parrot

Congratulations Wiejeds#4784!

You have won a raffle prize!

Prize

2500 DreamStudio credits!

Please contact Aether to claim your prize!

@tidal bough

@hearty dragon

Congratulations Mohmal1#1159!

You have won a raffle prize!

Prize

1 month Discord Nitro!

Please contact Aether to claim your prize!

@tidal bough

[vast pollen]

ping me if anybody knows the difference between these anything 3.0 versions

[tribal oar]

Pruned is what you want

non pruned is for training

[tidal bough]

Congratulations to tonights winners!

[mystic perch]

Hey, I'm just now getting started- does anyone have any reccomendations for tutorials?

[cold siren]

nobody really knows anything 🤷 not even joking

we just put random words and hope that it will generate something similar to what we wanted

that's how "ai art" works apparently

like let's say you want to generate a picture of a cube of ice 🧊 so you enter "ice cube"

and of course it generates ice cube the rapper

[vast pollen]

thanks

[tribal oar]

https://stability.ai/sdv2-prompt-book check this out

Great beginner's resource for basic prompt shaping

Covers a lot of different styles and teaches you what kinds of prompts to use to achieve different effects

[tribal oar]

That's very much untrue, we have a really comprehensive idea of how to accurately craft prompts

[fervent thunder]

hello

[cold siren]

so i wonder 🤔 what kind of comprehensive idea would make you write in the prompt something like

!!!!!!!concept art by senior environment artist!!!!!!!
or
😃😀😄☺🙃😉😗, !5 three eyed goddesses, i_5589.jpeg, 3 2 x 3 2

[feral rose]

Where can one find SD 2.1 VAE's?

[cold siren]

heard it is builtin

[feral rose]

Ok, that is interesting.

[tribal oar]

!!!!!! adds importance to the prompt (in 2.0 I think it does idk I use 1.5), making it look for work by that type of artist makes sense for different things

The second one...I've never seen an emoji in a prompt lol I want to try running that

[cold siren]

the !!!!!! seems like garbage tags to me, just as other weird unicode characters like emoji

[cerulean granite]

Emojis work, as do foreign languages

[cold siren]

but same word in different languages produces different things

[cerulean granite]

As far as I know "!!!!" Was used to direct attention before the methods we have now, may still work, I have not tested

[cerulean granite]

Sure

[tribal oar]

The same word in english produces different things lol

[cold siren]

🤷 so it is merely a side effect

[cerulean granite]

I assume he means with locked seed

[cerulean granite]

If I write prime minister in another language, and it shows me a picture of the prime Minister of that country, that seems like a pretty direct effect to me

[cold siren]

and also Moebius, Mœbius and Jean Giraud are 3 totally different artists

[cerulean granite]

Deng, still no updates on 1111. Has anyone tried 2.1 on it? Compatibility? Performance? Downloading now but horrendous internet means I won't have it for a few hours.

[tribal oar]

Results have been iffy according to initial testing

Works with the 2.0 yaml

User that was testing it wasn't happy with results though

[cerulean granite]

Sweet. I dont see any party hats..? So no 100x performance improvement??

[cold siren]

works but you have to use --no-half

[cerulean granite]

Emad posted yesterday 20x performance improvement, plus 5x performance improvement. Sure looked like he was saying 20fps generation in 2.1

[cerulean granite]

Because of memory use? I'm on 1660 ti anyway so I have been. Might not work on mine at all now.

[cold siren]

needs more vram because of that

[hardy mantle]

Got 2.1 working in Auto1111 just fine. Copied the 2.0 yaml (and renamed).
Using xformers nothing else.
3060ti 8GB
Only few tests so far but having no problems genning 15 batches of 768 up to 1280 at a time

[tribal oar]

Yeah that's what I heard too?

Can anyone confirm that?

[cerulean granite]

https://photos.app.goo.gl/etiZ1TDY8TAv69ye6

[tribal oar]

What happened that that amazing data about the 4 step image gen

I thought that was going to be in 2.1

[vapid tangle]

is it me or seeds are messed up on dreambot

[hardy mantle]

About 2min to 4min per 15 batch...

[cerulean granite]

Let me know if that link doesn't work, and I'll try to figure out how to post it here. This channel won't let me.

[tribal oar]

Yeah exactly!

[cerulean granite]

I can't seem to find it on his Twitter now, was it a tease? 🤨😲

[tribal oar]

/cries in 1660

[cerulean granite]

Can anyone confirm this link did work?

[tribal oar]

Oh yeah I can see it

[cold siren]

hopefully auto1111 will add medvram support, IIRC they had to do this for 2.0

[tribal oar]

Could a kind person please link me to the YAML?

[cold siren]

it's literally the same one from 2.0

[tribal oar]

I dunno where to get that though

I don't have it currently

[cerulean granite]

It surprisingly hard to find actually, I had to email it to my friend yesterday because neither of us could locate it

[tribal oar]

Yeah googling isn't helping

And chatGPT refuses to connect to the internet

....actually that's probably a good thing never mind

[cerulean granite]

Lol, at least you have an OpenAI account. I failed for over a year to create one. They even apologized and sent me free bonus credits, which of course I couldn't access, because I don't have an account.

[tribal oar]

I am still kind of in awe of this tech advancement

5 months ago, I sign up for Dalle beta

3 months ago, I bought $50 in Dalle2 credits

Today I'm making pictures of my own face

[cerulean granite]

I was working 14 hour days until about a month and a half ago lol, showed up late to this party! Signed up for Midjourney V3 until i discovered SD, and the rest is history, in the making

[tribal oar]

Honestly I would switch to MJ if I could run it locally

It's so damn good

[cerulean granite]

So, does anyone know anything about stable tuner? I heard about it here a couple of DAYS ago, but I still can't quite wrap my brain around it.

https://github.com/devilismyfriend/StableTuner

I heard some very knowledgeable people speaking very highly of it, and I'm wondering if this is what emad was referencing when he mentioned something much better and faster and more accessible than DreamBooth was inbound...?

[cold siren]

https://raw.githubusercontent.com/Stability-AI/stablediffusion/main/configs/stable-diffusion/v2-inference-v.yaml

so yeah they are all here https://github.com/Stability-AI/stablediffusion/tree/main/configs/stable-diffusion

[cerulean granite]

Fantastic thanks! Any of the others working in 1111?

[cold siren]

also this must be the reason for --no-half https://github.com/Stability-AI/stablediffusion/commit/c12d960d1ee4f9134c2516862ef991ec52d3f59e

[cerulean granite]

Nobody..? Hm...

[winged flare]

hope we get better audio model soon

[tidal bough]

@fluid vapor You've won one of our raffle prizes a few days ago! Please DM me in the next 24 hours to lock in your prize or the raffle draw will reroll! Hope to hear from you soon! ^^

[tribal oar]

Should I add this to run 2.0?

[wild hull]

Hello guys, it's look like training model (on SD2.1) with my own face take more time to train than SD1.5 max_train_steps=1600 take more than 60 minutes to be done.
Also, I generate one image with but the results are not so good when i try to put the face on something (superman for exemple)
But When i use a prompt with a very simple prompt like "Photo of zwx person". It works better, i can recognize myself

Can someone have an idea ?
I am on RTX6000

[brave matrix]

hello, what is a1111?

[bleak matrix]

Good evening, everyone. What's up?

[modest owl]

where are the tools?

[bleak matrix]

If you want to learn more about using Dreambooth/ training, I suggest you check out #🔧｜finetune The kind of images you use, along with your parameters, can influence a lot.

[brave matrix]

What is the best way to run 2.1 locally?

[untold moss]

hey... how do I do inpainting

[dire perch]

by doing the opposite of outpainting

[cobalt cliff]

hey guys, what is the state of the art for audio ai? is there anything like stable diffusion for audio/music? thank you

[tribal oar]

check out harmonai

[steady arrow]

Does someone have midjourney's discord link I want to do some comparisons but cant find the discord anymore

[bleak matrix]

You can check out out #1034602544263090268 for more info. There's also quite a few videos on YT as well!

[bleak matrix]

Check #1034941531762733167

[steady arrow]

Do I run 2.1/2.0 whichever is latest model by downloading something and placing it in auto's models dir?

or is there a new process now

[bleak matrix]

My suggestion is to check out #1014939219904450590 for more information on getting started locally

You can find info/download information about 2.0 here: https://github.com/Stability-AI/stablediffusion If you need help with installation, feel free to ask anything in #🤝｜tech-support

[vocal plover]

Where can I ask help for a prompt? I wanted to generate something but I'm unable to

[north stirrup]

#📝｜prompting-help

[frank moon]

In automatic1111 do you upscale by going to extras? The image looks the same after I went to 8.

Did I do something wrong?

[brave matrix]

can someone refer me on how to get started with AUTO1111

[sour fog]

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

[sour fog]

do you have a Source image? did you select an Upscaler?

[tribal oar]

What happens if I train a model at 768 and then render at 512?

Do you think it would look weird and compressed?

Or higher detail?

Higher resolution/image means more pixels for training, right? So faces should be more accurate?

[silk kindle]

i was told this server / bot could do stuff like this not sure if anyone here can help with that https://www.youtube.com/shorts/_bv5p2SlAPc

[steep mortar]

is SD2.1 available for download?

nm

[frank gate]

where is the 2.1 yaml config file? or is it the same one from 2.0?

[steep mortar]

seems to be the same

https://github.com/Stability-AI/stablediffusion/blob/main/configs/stable-diffusion/v2-inference-v.yaml

[frank gate]

awesome

[valid void]

How can we check if an artist name is included in SD 2.1 or not? (without trying one by one)

[placid sage]

How big is the current SD model? x gb?

[white niche]

The bot? No. Stable Diffusion is used with other tools to do this type of stuff. I haven’t played with it, but might want to look up Stable Diffusion Animation on google or check out the Deforum project. https://deforum.github.io/animation.html

[white niche]

5GB

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

[placid sage]

Thanks, do you have any idea how many images the model was trained from? Or any way to find that info. Just curious

[white niche]

Some of the details are on the model card on that website, but I don’t think they have said explicitly how many images were in the dataset.

[white niche]

Just says “subset of LAION-5B”

Which was then filtered for explicit material.

[steep mortar]

hmmm, I get an error when trying to load SD 2.1 model

[white niche]

The 5B stands for 5 billion clip filtered text image pairs.

So it’s a lot, but how much was filtered out?

Not sure.

[raven edge]

I keep getting black generations with SD 2.1 what I am doing wrong?

[white niche]

If you are using automatic you need the yaml file from the GitHub as well.

The only time I got a black image was with my old 1660 Nvidia card

[raven edge]

I need two Yamls? I have copied the 2.0 one to same name as the 2.1 model

[white niche]

Yeah needs to be the same name as the model

[raven edge]

why does 2.0 and 2.1 require a yaml?

[steep mortar]

I have the yaml file, maybe I downloaded the wrong file

[white niche]

Is a different structure to SD1

And thats my understanding

So it needs the yaml to load the correct settings

[raven edge]

I'm using this Yaml i name the file v2-1_768-ema-pruned.yaml https://github.com/Stability-AI/stablediffusion/blob/main/configs/stable-diffusion/v2-inference-v.yaml

[white niche]

I’m only aware of automatic1111 may be different with a different program?

[raven edge]

im using Voldys repo

[white niche]

Yeah it just needs to be the same name as the model name.

From memory the base 512 and 768 model use different yaml files

So just make sure they match. I don’t have access to my PC to check sorry.

[steep mortar]

not sure which ones are the correct yaml files

[white niche]

https://github.com/Stability-AI/stablediffusion/tree/main/configs/stable-diffusion

That’s the yaml ones, I think v2-inference-v.yaml is for the 768 model

v2-inference.yaml is the base 512 “I think”

[steep mortar]

ok I try it out

[white niche]

But yeah sorry I’m not in front of my PC to double check.

Hopefully someone else pops on who has access to check if you are still stuck.

[cosmic bridge]

Hey guys just came across this, seems like game development is about to go wild with AI? - https://leonardo.ai

[tribal oar]

Neat, yeah even if this is just one example, this tech is going to be everywhere soon

[cosmic bridge]

Yeah insane, keen to train some models

[raven edge]

Im still getting black outputs with Vodlys repo not sure what the deal is or if im using the right yaml

I'm using the v2-1_768-ema-pruned.ckpt

[steep mortar]

hmmmm, I cant get this damn 2.1 to load

[plush cairn]

Use this line in the "webui.user.bat".. "set COMMANDLINE_ARGS=--no-half", that sorted the black images for me

[finite creek]

Does anyone know how I can achieve the look/style of Eizin Suzuki or old Akira toriyama style, I try on DreamStudio Lite or on Hugging face I don't know if prompts are the issue but I cannot replicate how their drawings look like

[quick basalt]

I think you are going to need a model for that

[finite creek]

I see ya i guess those images haven't been trained as of now

closest I can find is studio ghibli

[finite creek]

when u mean model u do mean like dataset that has been trained right, sorry am very new to sd

[quick basalt]

Basically the .ckpt file you put on the models folder

[steep mortar]

strange, I got 2.1 to load. but all my pics are solid Brown squares....

[deft ingot]

so whats the "best" AUTOMATIC1111 sampler atm

[lilac reef]

Are we ever going to get Aesthetic gradient embeddings back?

[sour fog]

There is no best. They all work differently.

Personally I use a lot:
Euler A
Euler
DPM++ 2S a Karras

[viscid ridge]

Why stable diffusion doesn't want to acknowledge that Asuka's suit covers her legs too

[cedar flare]

can you image prompt with this software

[bleak matrix]

Do you mean use an image as a prompt?

[bleak matrix]

You can check our #1010577750077210726 section for tips on that

[jolly viper]

Anyone building an offline stable diffusion app for Android now that SD 2 is distilled?

There's already one for iOS.

[cedar flare]

How do you acces the ios version

[jolly viper]

It's an app

https://apps.apple.com/us/app/draw-things-ai-generation/id6444050820?platform=iphone

[lilac reef]

I wonder what happened with training on colab? 2.1 gets released and suddenly I can't even train on 2.0. TI nor Hypernet

looks like a pytorch issue that has popped up in the past on other projects.

[mighty ermine]

Im super lost as to how people actually get SD to make anything good.. Ive seen some amazing stuff but its like mine is mental or something.. It looks VERY little like what i ask it to make.

[lilac reef]

Depends. More than likely your prompt-fu has failed you.

[mighty ermine]

Well okay ive got a DND thing coming up. How would i ask it for "a giant guarding the entrance to a cave"?

Because thats what i asked for.. and it looks like crap.

[lilac reef]

Well, I am not sure what your mind's eye is envisioning.

[mighty ermine]

Im open minded. Im really not trying to be picky about what the giant looks like.. What the cave looks like.. I dont care.. But what i get when i ask for that is a very poorly drawn avarage looking guy standing next to a hole.

Its garbage.

Like it has NO CLUE what a giant is.. What a cave might be.. idk.

Did i download the helen keller of AI?

[lilac reef]

yes

[mighty ermine]

Nice

[lilac reef]

@mighty ermine I can't post a pic here

[mighty ermine]

Ya i was going to share too but.. ya.

[lilac reef]

I shared in general with images

[broken spindle]

is there difference in speed with 3060ti and 2080s

12 and 8 g respectivly

[thick inlet]

https://ibb.co/VW6k7cr - here is what midjourney generated for giant man guarding entrance to a dark cave, it is an exit from a cave and it added an adventurer too but maybe one of those will be useful for you

now that I think of it maybe writing "inside of a cave" instead of entrance would work better

and here is giant troll instead, a bit different but might be more useful: https://ibb.co/DCTbP4T (still midjourney)

[mighty ermine]

SO what your saying is Stable Diffusion is just crap? lol

I don't really WANT to pay for AI when i own a 3090 that can supposably run an AI that can do that level of quality.. BUT if what your saying is Stable Diffusion just sucks at making things in any way simple to get that level of quality then i guess thanks! lol

[fervent thunder]

using 1.5 and maybe some embeddings and custom models can help a lot with style. Less so on 2.x as it's still new. But also because it seem focused on photos

[mighty ermine]

Well so far upgrading Auto to 2.1 has ended with me just getting black pictures so.. THATS not working at all anyway.

[fervent thunder]

you got the yaml file for it?

[mighty ermine]

Yes

[fervent thunder]

and the no-half? and do you use 786 and/or 512 version of it as they use slightly different yaml files

[mighty ermine]

Well.. i think so.. I copied the code page into a txt file.. Saved it as a yaml and named it the same as its model in the correct folder.. After that ive had nothing but black images.

[fervent thunder]

and you use the 768 or 512 model?

[mighty ermine]

What do you mean No-half?

and i dont remember to be honest.

If you want to point me in the direction of matching Model and Yaml files ill be happy to try again.

[fervent thunder]

when you start the webui, you use the webui by the way? do you have any ARGS?

https://raw.githubusercontent.com/Stability-AI/stablediffusion/main/configs/stable-diffusion/v2-inference.yaml
This is for the yaml for the 512 version

[mighty ermine]

Yes im using web ui and idk what "args" is.. Im VERY new to this and i dont code or use linux for anything..

[fervent thunder]

can't find the 786 in my documentation and I'm a little too busy to search for it, sorry, but check this post, I believe they spoke about the same error:
https://www.reddit.com/r/StableDiffusion/comments/zf21db/stable_diffusion_21_announcement/

I'd help you better, but I gotta go, good luck and if nothing else, then you can always pm me if no one else helps and/or you can't get it to work :3

[mighty ermine]

lol Okay.. Well that looks exactly the same as the one i already got so maybe i need the 512 version.

[fervent thunder]

I got all four versions to work without issue on my rig at least so let's hope mine wasn't a fluke! ;P

[mighty ermine]

Jealous! lol But also just jealous that every other AI seems to be better then SD

[fervent thunder]

Might be a little biased but I believe my ai images are better than anything, and they are made using SD 1.5 :3

[mighty ermine]

Hey im all for learning how to make this thing do quality things! I just make little refrance pictures for DnD and fantasy type stuff.. Just comparing simple inputs SD isnt doing very well..

I have no doubt that some can make it work with enough fine tuning and carful prompting.. But thats ALOT more work then should be needed when others are doing the same with less work.

[viscid ridge]

Is SD able to make pixel art? 🤔

[cold siren]

use other people's work 🤷 meaning custom models/embeddings/etc

[fervent thunder]

send a PM to me and I can maybe help, if you trust some random person on the internet like me. I'll be a little busy thanks to work for a while, but I love to help people getting started with the AI. Know that it is extremely good at almost anything, you just have to learn how and what to use! :D

[fervent thunder]

not perfectly sharp no, but it understand the "concept" of it :P

[viscid ridge]

Haha, thanks
Oh, I forgot you can train the ai

[mighty ermine]

Anydangway.. For now i cant use 2.1 its just forever giving black screens.. Stuck on like 1.4 or 1.5 so maybe thats my issue.

[haughty sentinel]

Anyone had any success merging 2.1 with other models? does the compatibility depend on if it's 512x512 vs 768x768, or did it both have to be trained on 2.0+ ?

I'm just getting out of memory errors if I try to merge
RuntimeError: [enforce fail at ..\c10\core\impl\alloc_cpu.cpp:81] data. DefaultCPUAllocator: not enough memory: you tried to allocate 41943040000 bytes.

41GB ..., can't be right.

[mighty ermine]

Isnt that 4.1gig?

[haughty sentinel]

no it's 41

[mighty ermine]

Lordy

[thick inlet]

no, but if you want quick results MJ is better and I had it available at that moment, to get a giant in SD I would use 1.5 not 2.0/2.1 and work with a longer prompt, 1.5 can generate great caves so it should be possible, for a giant you would probably need to use a longer description, the word troll,monster or something, can't check now because I am on my phone

[fervent thunder]

https://tenor.com/view/god-emperor-rainbow-warhammer-40k-gif-14288832

My pc gaming

https://tenor.com/view/beth-rickand-morty-gif-11259059

My pc making cat pics

[reef epoch]

if you start using dreamstudio, they should have their own chatgpt ready in 1 to 2 hours

no guarantees though

[worthy lichen]

Totally off topic. Got sent an image of something i posted in this server.

ok can't send the screen shot

but anyway, called Gigabot and this is the only server in common. Just wondering if I'm missing something. is that a bot related to this server, excuse me if im being stupid

[leaden mural]

Mid-complex. May need some pretty hefty hardware and a good workflow for training 3 necessary AIs for an equivalent chatbot to be possible. 1 for teaching, 1 for rewarding, and 1 for the actual ChatGPT-like.

[valid void]

How can we check if an artist name is included in SD 2.1 or not? (without trying one by one)

[wise sparrow]

Is it possible to generate more than one image using a seed in dreamstudio?

[jovial maple]

Check the training data used.

[steady arrow]

Guys how do I run 2.0/2.1 on Autos? And do you guys think its better than 1.5?

[valid void]

what's the best way to check other than downloading the whole thing? is there a search engine for searching the training data?

[cold siren]

oh well i really hope somebody does this soon, chatGPT seems to be getting more and more censored over time 😕

[leaden mural]

correction: changed software to hardware, because of course you would need servers to do this.

[leaden mural]

true, but I am getting past those censors, so that isn't going to be that much of a problem for now.

Essentially, the sketchy stuff is also in the dataset, so that cat is out of the bag.

[jolly viper]

What's the minimum recommended hardware now with SD 2.1 distilled?

[past robin]

That one is not out yet, so knowbody knows. But if it's a 10X tot 20X more efficient you can probably run it on your phone after a while.

[grim skiff]

When I try to generate a picture/painting of two specific people there are often bunch of random people added or both having same faces etc.

[left knot]

https://www.engadget.com/ai-film-festival-runway-ml-191033350.html < been waiting for this

It'll help push text-2-video and ai based technologies in the film industry

[tall breach]

guys i don't understand whats going on, im trying to inpaint just the masked area but it keeps giving me some disgusting amalgamation. how do i fix this?
Image
Image

[deft bronze]

tbf, you can run sd on a modern iPhone last time I checked, does take like 1 minute/gen

the latest change should push this number way lower, is the hope

[deft bronze]

yeah that's gonna be cool

[tidal bough]

Rolling another couple of raffle winners, good luck everyone!

[surreal perch]

@floral cave

Congratulations Seregor#1718!

You have won a raffle prize!

Prize

2500 DreamStudio credits!

Please contact Aether to claim your prize!

@tidal bough

@dire slate

Congratulations Anurag karn#4716!

You have won a raffle prize!

Prize

1 month Discord Nitro!

Please contact Aether to claim your prize!

@tidal bough

[tidal bough]

Congratulations @floral cave and @dire slate!! DM me within the next 24 hours to lock in those prizes and we'll get them sent over to you! Thank you everyone, more to come today!

[cyan temple]

Mj is better than SD now, how come?

[bleak matrix]

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

[bleak matrix]

If you need help with inpainting, you can check out #1034602544263090268

[bleak matrix]

Well, I am not sure what you are prompting, but you could try checking out #📝｜prompting-help

Show what words you are using, etc

You can always use inpainting to change the people if you need to

[tall breach]

hello, i've messaged one of the guys in that channel. still haven't solved it yet

[bleak matrix]

I'll take a look

[fervent thunder]

I think we need to start talking about the ethical rules of the AI art comunity.

by example, always give credit to the mix of artists you prompted from.

that should be the first and most obvious one.

[haughty oxide]

but other artists have credited the works that inspired them? thats the same thing.

Ai just get inspirations and draw in basis of that inspirations... all the artists inspired themselves in other artists to create their works.

and nothing wrong about it

nothing that breaks the ethical rules.

[fervent thunder]

I think it should be a proper non spoken rule of the comunity.

I think that's the entire controversy.

It would be like a chef claiming he's a farmer.

[haughty oxide]

i understand, i have my oppinnion

[fervent thunder]

I know.

I still think the comunity should agree on some non spoken rules of good behaviour.

[quiet wolf]

Hello fellow artists! Can someone explain how to upload image init for dreambot please?

[wild hull]

Hello guys,
What is the best Scheduler for face generation with SD2.1 ? please

[sour fog]

There is no best. They all work differently.

Personally I use a lot:
Euler A
Euler
DPM++ 2S a Karras

[wild hull]

Oh okay, thanks, i will try 🙂

[thick herald]

where would be the right place to ask about a problem I'm having with SD2.1 and Auto1111?

[broken gull]

I use Heun @ 30 steps a lot

can do 6 img in like 4 minutes with my 6Gb vram GPU

[broken gull]

#🤝｜tech-support

[thick herald]

Great, thanks! (How did I not spot that channel name?)

[ashen crane]

Why some of my renders are blurred?

[raven edge]

has anyone here gotten SD 2.1 to work with AUTOMATIC 1111?

I still get blank (black) images

[steep mortar]

yes, the other day I got some help with it. one sec

[raven edge]

I don't know what the issue witht YAML is, I've tried using every one no dice

[steep mortar]

set this in your webui-user.bat: set COMMANDLINE_ARGS=--xformers

[raven edge]

ok

will this break my other models? i.e Anything?

[steep mortar]

no

itll download what it needs to get 2.1 working

itll take a few mins to download

[dense solar]

Hi, I'm Stelfie

[raven edge]

im still getting black generations

its it the yaml

No module 'xformers'. Proceeding without it.

[steep mortar]

hmmm strange

do you have the yaml file for 2.0, you can use that one and change its name to the name of the 2.1 model

[raven edge]

yeah no dice

I got this error

launch.py: error: unrecognized arguments: COMMANDLINE_ARGS=--xformers

[--tls-certfile TLS_CERTFILE] [--server-name SERVER_NAME] [--ldsr-models-path LDSR_MODELS_PATH]
[--scunet-models-path SCUNET_MODELS_PATH] [--swinir-models-path SWINIR_MODELS_PATH]
launch.py: error: unrecognized arguments: COMMANDLINE_ARGS=--xformers

[steep mortar]

thats wierd

this is what my webui-user.bat looks like:

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers
git pull
call webui.bat

[raven edge]

werid I dot have git pull I guess

Im still getting an error

[steep mortar]

git pull just gets the latest repo update

damn, that sux. I dont know then

[dawn yarrow]

idk -music artists can talk about what inspired them, but isnt their work original in some capacity as they put it together?

how far do we go to credit people--- down to who invented english?

[wild meteor]

That's so true. In the days of film photography the photographer didn't tell them who set up the lights, what film they used, did they use a special process during development (like pushing or pulling, if you know the term) , what ISO etc

[tribal oar]

I'm just curious. I run my own version of SD locally, but if I wanted to give a web version to a friend, what would you recommend?

[wild meteor]

I'm a singer/songwriter influenced by George Michael but by the time I write and record a song it doesn't sound like I've tried to impersonate him.

[north stirrup]

I think that it really depends on context. If I force SD to make abstract art based on a classical-style artist, I don’t think it matters nearly as much because it’s clearly not imitation and might not even be recognizable as related to that artist, but often it is imitation-like if people aren’t strangling their prompts into something creative

In copyright law it’s the question of whether a derivative work is “transformative”

[tribal oar]

Did you try deleting the webui-user.bat and running a git -pull?

And then adding = --xformers from there?

[dull matrix]

hey all, happy day. would this be a good place for animation/web UI questions?

[vagrant fox]

There is my tutorial how to use Stable Diffusion 2.1 locally with auto1111. https://www.youtube.com/watch?v=ixBzJ7von4E

[vagrant fox]

u can share host link to friend from your pc and he use your pcto generate

[tribal oar]

Absolutely not

Not a chance I open this thing up to the internet lmao

I have no doubt in my mind that people are crawling the internet looking for stable diffusion engines to crack in to

I will not not risk that, not even if it's a 1% chance

[cobalt cape]

hey

in A1111 I get black or brown images

Brown with Nitrosocke 768 models

and black with the normal 2.1 model

Nitrosocke said to someone else the following but I don't know what it means:

You get that brown noise when you loaded the V-Model configuration files. The 512 Base model (which Future Diffusion is trained on) needs the non v model config.
I haven't used the Ui yet so I can't say how you would change it inside of there.

the 2.0 model does work though :S

[bitter laurel]

Big thank!!

[vagrant fox]

Welcome ❤️

[clear matrix]

A quick question on the current approach to custom items/people: Given all the amazing new custom checkpoints being released, what's the current thinking on custom models of an individual vs embeddings vs something else? I had a blast creating custom Dreambooth models and less success with embeddings in the past. If I want to be able to create images of a friend in the style of all these new and wonderful models, would that be an imbedding I would need to create or is there some other way to approach this now?

[vagrant fox]

https://www.youtube.com/watch?v=ixBzJ7von4E solution will be YAML files check my video

[cobalt cape]

thank you\

[sharp robin]

Hey I don't want to bork my windows, I recently reinstalled my W11 and I can't get xformers to install

no module named Xformers on auto's webui launch, and when I pip install xformers it wants VC++ 14 build tools, and I don't want visual studio

[clear matrix]

What GPU are you using?

[sharp robin]

3070

I didn't install anything else yet, like geforce, cuda etc

[clear matrix]

ah, i just add --xformers to the start bat and it automatically installs it for 3x cards

[sharp robin]

I guess I'll need at least cuda

it is trying but failing to install it

[clear matrix]

I just did exactly the same thing - i just installed the Geforce experience drivers

try that first

That may be a dependency

[bitter laurel]

yeah, i had the same issue. YAML files are the same as 2.0 but remember to add --no-half to the commandlineargs in the bat file

nope. sorry. missing something else. will let you know when i solve it myself

[clear matrix]

you have drivers installed?

Do you get an error message?

[sharp robin]

wait I'm still waiting for nvidia geforce to send me a confirmation mail, but I have at least nvidia drivers installed

[clear matrix]

ok, see about any updates and possibly try rebooting so all that gets loaded

[tribal oar]

Works for me without --no-half

just --xformers

[clear matrix]

yeah, i think those are separate things. He can't get xformers to install

[tribal oar]

Ohh

My bad reading up now

[dawn yarrow]

i mean thatas a good point

[bitter laurel]

shoot. ill try adding --xformers and see what happens

[tribal oar]

@sharp robin I'm on Win11 as well, and I was having some issues

[sharp robin]

its resolved

[bitter laurel]

lol just got internet reamed by a bunch of people for defending AI art. What fun times.

[tribal oar]

Oh nice!

Fuck em

That conversation will die down

[bitter laurel]

it will. blows my mind how frenzied they are. guess its not an argument worth having.

[tribal oar]

I got into a debate with a guy a few days ago, and it was only after like 20 minutes of talking that I bothered to ask him if he'd ever used the tool before

Obviously, he said no

That's when I decided I don't care

[sharp robin]

I'm not proud it was silly, just don't use --force-enable-xformers the first time, it can't install it then (dunno why)

--xformers works fine

[tribal oar]

The only people bitching are a bunch of uninformed morons who don't even know how the tech works

[tribal oar]

Yeah I made the same mistake lol

[bitter laurel]

the problem is, I think - is that artists have been held safe from the machine for many, many years. replace factory jobs with robots lmao not my problem - but when the time comes that machines can copy their art, they are scared. and i get it, evolution is painful

--xformers fixed the black images problem. thanks for helping out, guys

[fervent thunder]

AI is still not something that is fully understood by the mainstream public. When that happens it'll only become even more exacerbated and chaotic because of how many people that will have to be informed and not confused by the misinformed or the ill informed who are set in their ways of thinking.

[dawn yarrow]

its not like ALL art will be replaced by AI art but programming is an art in itself. and welcome to the future is now.

[fervent thunder]

Also these people who have built their entire life on a skill that they've probably learned since childhood think that their jobs are gonna be curb stomped in less than a decade. So they're freaking out without being rational.

[clear matrix]

It feels a moot point for someone to argue against it... The cat is out of the bag. It exists and it's open source and available...

[bitter laurel]

I am a programmer. I was stunned that the AI could produce entire projects worth of code by asking it to - in just a few seconds. but I was excited, not scared for my future.

[fervent thunder]

It's also weird because despite the AI being able to do photorealism quite well, you don't really see or hear a lot of photographers going insane.

[bitter laurel]

so far the anime crowd has been screaming the loudest. I said 'evolve or die' and they really really did not like that

[dawn yarrow]

i dont see why you cant respect traditonal art AND ai art they arenet in conflict

[fervent thunder]

That's wild that drawn art by a human is now considered 'traditional art'

[dawn yarrow]

i mean i kinda stink at art so i think AI stuff is great whats so bad about that?

i'm intrigued by the results of image to image with some masking effects in like GIMP

[bitter laurel]

I have been playing with that. trying to make some concept art for a book.

[dawn yarrow]

there was some really nice tutorial on youtube about it about a treasure chest etc

for making game art with stable diffusion

[bitter laurel]

i am really really excited for 1) ai videos to improve, and 2) 3d models

[fervent thunder]

Anime artists? or the people rushing to the defense of anime artists?
Because the latter would probably benefit greatly from AI art, horny bastards.