💬｜general-chat

[remote oasis]

Does this use my existing local install or a whole new instance of SD? I’ve noticed my PC doesn’t sound as lout using the web ui

[brazen fox]

I imagine it's an entirely new installation, complete with it's own venv- but it's chucked full of optimizations which I think your initial install might not have had

[potent spire]

Nah, disagree. They didnt lose their "special" status. I dont even know if they ever had one depending on what we are talking about.

But i agree that some actual artists may and do use SD, DALL-E and/or MJ to make references and ideas out

Many dont tho for several reasons

And for some this is just a toy and not even a tool for workspace

[fervent thunder]

very interesting! the wiring seems like it would be challenging. I find myself wondering what it'd sound like just in general lol. I'm primarily experienced with wood too so I get it. Life ain't straight in the wood world! Plenty of cheap spoil board never hurts though. Especially if you got a vacuum system under some MDF. perfect set up really.

There's so many variables with a CNC router it's absurd. I worked on an older one which was even more manual than the modern day ones.

The program ran on Windows XP and it was the "newer" set up lol

In the end it is just g-code but they got pretty involved with the interface programs back in the day

[placid sphinx]

Probably not relevant for diffusion models tbh. I've yet to get properly stuck into the fast.ai course but was planning to start there then maybe rekindle this idea. Force me to get back in the garage again too.

[fervent thunder]

I think it could be if the proper application was done.

What I've seen that fascinated me was this dude hooked up minecraft to his wood CNC

so if that sort of thing is within the realm of possiblities I don't see why we couldn't hook up a diffusion model to one.

It might not be able to do all of the things but it could do a lot.

I'll find the video

I've been obsessed with the idea since I saw it years ago but I didn't really ever have the set up to do anything with it.
https://youtu.be/G5eo5d8F5DU

this video is #1 on the list of videos I share with anyone who has any interest in CNC. I know a ton of software ppl but I'm in the process of reaching out to anyone with CNC knowledge. I have a decent amount of knowledge about it but I'd say in terms of like...raw G-code knowledge I'm still pretty green.

[halcyon granite]

I'm seriously tired of people post AI Vs anti AI art posts in Reddit. Everytime I open the app I see that nonsense

[full lark]

Hello people!, is it just me or does anyone else think the opt-out option on https://haveibeentrained.com makes no sense? Everyone can opt out any image from training, even if you are not the owner.

[drifting wind]

Sure, if I recall correctly (I don't save prompts) the prompt was "A photograph of a (human-like hippopotamus:1.5) (wearing a suit:1.4) (standing behind a desk:1.2)", negative prompt was "blur, haze, hazy, blurry", and the model was hassanblend1.4, 60 steps, 10.5 scale, eulera, 512x512, restore off, highres off, random seed. Side note: this prompt works pretty well for people, too, but you have to turn restore on. Sometimes highres fix is an improvement too, at around .3

In AUTO1111. what would happen to the model if I start training with one model, train half the steps on that model, interrupt the training and change the model, then finish the training with the new model?

[drifting wind]

The problem is, we're at the start of what could be a serious and far-reaching shift in the public's stance on AI art, and these conversations, complaints, rants and so forth need to happen, especially on the pro-AI side because the anti-AI group has momentum, numbers and money on their side. We're not nearly as vocal about the defense of AI as they are about attacking it, and it will only get worse.

[hollow scaffold]

Hi all, quick question: I've been trying out img2img inpainting for a while now but the results are bit disappointing. Does anyone have a tip for downloading a new model/dataset or anything similar to get better results?

[halcyon granite]

I think most of the posts are just serving to shoot AI art in the foot though. People think they're fighting traditional artists (like myself). Both sides are reinforcing black and white views

ultimately the people with money have a better chance if there's rabid antiart people getting triggered aswell

[hollow scaffold]

And if this is not the right place to ask, what is?

[drifting wind]

Totally, but without the "shooting themselves in the foot" posts, there's waaaay less visibility. Those kinds of posts do serve the purpose of raising the collective awareness, even if they're negative in scope and don't bring much to the actual conversation. It's not likely that there are many people on the side of "traditional artists" jumping on Twitter and telling people to stop complaining about AI art. The volume of posts and commentary is staggering on their side, and nowhere near as big on ours.

[drifting wind]

https://www.reddit.com/r/StableDiffusion/search/?q=inpainting&restrict_sr=1&sr_nsfw=&include_over_18=1

[silent locust]

Thank You ☺️

[halcyon granite]

I think less visibility is advantageous, it gives the community time to build (maybe it's too late now I suppose). I just see too many people fanning flames when people quietly making cool stuff in the background seems to me like a harder target to nail down.

[drifting wind]

BTW, just saw your website. Love your architectural drawing style.

[halcyon granite]

thanks, I'm actually making a cyberpunk game using AI art

I couldn't match my art style with AI but I managed to get something similar and really cool

[drifting wind]

Dope, hope it turns out great!

I saw the bioluminescent AI characters too, liked them a lot.

[halcyon granite]

ah my website doesn't have any relevant content to the game. It's not gonna be regular AI imagery but I won't give details away for now 😅

[frank moon]

What are some good sites to find good models? Civitai was suggested but, it does not always load.

[unique nacelle]

That's awesome

[dull hare]

is there anyway to make SVG format with AI ?

rather than generate it then draw it into illustrator ?

[short tapir]

Does anyone have a suggestion for a model is used for making pokémon specifically?

[hidden anchor]

I generated an image and later on closed Discord

is it lost forever?

😭

[dull hare]

on the right corner click Inbox --> mentions u will see it

[hidden anchor]

ooo, thanks legend

[fervent thunder]

looks cool

sweet! maybe I can access it via extension? I'll see. thanks for the share. I'm going to try it out now

[quick basalt]

My kings, I have a question, how to install hypernetworks? Is it like an embedding, you put in the folder and then use the word to trigger it in the prompt?

There are barely any information on the internet about them

[unique nacelle]

Why is it that images load blurry half the time for me?

[lofty parcel]

Probably the NSFW filter, it has a lot of false positives

[unique nacelle]

Ohh thank you

One of mine was just a faceswap between two male celebrities

Is there a way to input an image and ask it to generate the surroundings?

[vagrant solstice]

Hello, how are you? I would like to know if there is someone who has already deployed a refined gpt 3 model as a product. because I'm wanting to make a website that has an assistant that performs different tasks but I don't know how to do it, since I started in this world a month ago. xd any help is welcome. Ha and I also leave a page that I discovered has many apps that use AI registered, I hope the info is useful for you: https://www.futurepedia.io/leaderboard

[lofty parcel]

Yes, you can use image inpainting to mask out the edges of an image and use a prompt to describe it

[lofty parcel]

Since GPT-3 is closed source, you'll have to use the OpenAI API. It's pretty simple to use and you can finetune it if need be (for a price). Just make sure you don't underestimate your expenses and be careful with their ToS/Guidelines

[loud fulcrum]

Hey everyone, I am new. I was wondering... some people say sd 1.5 on dreambooth training is still better than dreambooth and 2.1. Is this true?

[loud fulcrum]

Hey, do we need a yaml file with other ckpt models?

[waxen swallow]

https://huggingface.co/spaces/lambdalabs/text-to-pokemon

[humble wigeon]

https://www.youtube.com/channel/UC32FbQWPO_VMAHcWqYIKG4g

[wispy jasper]

So what happened to releasing new models/tools every couple weeks? 🙂

[sinful mango]

Hello, I'm currently working on a video about art AI and copyright. It's mostly about potential legal challenges that are coming up, but it's also intended to provide a different (positive) perspective on the technology to artists.

Part of the video is showcasing some innovative tools and art processes, in order to try to show how the technology can expand human creative potential.

If you'd like a video of your artistic process or tool you've developed that uses Stable Diffusion included in the video, please feel free to send it my way. Cheers!

[vapid tangle]

i dont think so

[bleak wolf]

"potential legal challenges"

what they gonna do? make a lawsuit for each 5.85 billion image one by one?

well

they can do stuff

but i dont think, if google and many megacorporations are involved

it will be that effective

worst case scenario

we keep the software locally

[sinful mango]

i cover all that in the video, i'll be sure to share it here once it's done ;D

[bleak wolf]

Ty 😄

[sinful mango]

Copyright Alliance (partly led by Disney) recently released a paper on copyright and AI. they are definitely going to try to outlaw its use for anything beyond hobby projects

[bleak wolf]

partly led by Disney

shit...

[sinful mango]

yeah lol

the title of the video is AI vs. Blood Mouse

[bleak wolf]

Please beat Disney up

i never knew google would be my friend

[bleak wolf]

yesss

[sinful mango]

what makes you think Google will help us? curious to learn more

[bleak wolf]

It is investing in AI

Well

corporations that wants to profit off of AI art might be our friend here

[sinful mango]

ah, i see. well i hope they do something to help if necessary. basically, Copyright Alliance wants to try to legally enforce the use of "ethical AI", which means AI that is not trained on anything copyrighted (without permission).

if you see Google talking about "ethical AI" it's probably not a good sign

[bleak wolf]

or "ally" rather

if you see Google talking about "ethical AI" it's probably not a good sign

fr

AI that is not trained on anything copyrighted (without permission).

for me it is legally inconsiderable as stealing when each image's affect is less than 0.000000005%

It can be branded as "Inspiration" but thats me eh

[sinful mango]

yeah, i definitely agree that it's not theft. the sad thing is, many artists are being encouraged to be outraged, and now they're being directed to support the Copyright Alliance.

i just had a discussion with an artist who said he wants AI to be developed to scan, for example, art assets in indie video games, to prove they weren't generated with "unethical AI" models, or for developers to have to prove they used "ethical" models by providing cryptographic digital signatures for each art asset.

i go over some of this tech in the video, some of it actually does exist.

i fear that if the Copyright Alliance gets their way, this will be the future.

[bleak wolf]

...

Even in the worst case scenario

We can just brute force it

I own a small PC building bussiness, inshAllah aspiring to grow

Once compute becomes powerful enough

AI will probably be able to do it with just 1K images

so

there might be a recession at the worst case

but it will, in my opinion eventually take over

just by

raw firepower

[sinful mango]

hopefully we can defeat the Blood Mouse once and for all >:D

[bleak wolf]

Yes!

#sayYesToAI

[sinful mango]

personally i'm really interested in the development of AI tools that empower human artists.

[bleak wolf]

fr

[icy pendant]

ask your friend if it's ethical for disney to pay artists to make a ton of art based on samdoesart's style and then use it to animate a movie. that's "ethical" according to his rules but I bet he would hate it but that's the future he wants

[bleak wolf]

i cant draw because my hand shivers

[sinful mango]

in the end part of the video, i am going to show some clips of some tools i've stumbled across

[bleak wolf]

I will make sure to support your video as much as i can

[sinful mango]

the conversation was not very productive. he just kept accusing me of "ripping off" artists.

[icy pendant]

artists are so enraged they don't even respond to logic anymore it's just pure emotion

[bleak wolf]

same

[bleak wolf]

fr

i had convos with artists

defeated each of their arguments

then they just started making up stuff to make me look bad

[sinful mango]

cool! thanks very much. i actually started this video in November after seeing Steven Zapata's "end of art" video (which i thought was ridiculous), but then got sidetracked and decided not to finish it.

then a few days ago i got ultra ratiod on twitter after posting an AI/Disney meme, so decided that i will finish it after all. >:)

[bleak wolf]

AI makes art easily accessible to everyone

to those who do not have the means too

it makes people visualize their imagination

[sinful mango]

one thing that anti-AI artists never discuss is how it could possibly be enforced (outlawing the so-called "unethical" AI models)

well, guess what, Copyright Alliance has been scheming up all sorts of nightmare mode ways to try to enforce stuff like this.

in fact there's a new bill that was just introduced called the SMART Act that could help them do this.

[bleak wolf]

.....

[uneven wasp]

What in the SMART act affects AI regulation? It just requires agencies to put out a framework that measures the success of a rule with certain large scale effects https://www.congress.gov/bill/117th-congress/senate-bill/2801/text

neither Copyright nor AI is mentioned

and copyright law is dominated by statutory and case law, not agency rule making powers which this affects

[sinful mango]

according to the EFF, it would give the government the authority to mandate specific technology that internet/web service providers must use to police user content for copyright infringement.

so, if they did want to try to enforce some tools to identify material generated with "unethical" AI models, they could potentially use this to do it.

that's my understanding

[uneven wasp]

Ok but where does it say that

[sinful mango]

say what? here is the EFF article i read: https://www.eff.org/deeplinks/2022/03/new-filter-mandate-bill-unmitigated-disaster

[uneven wasp]

bruh

[sinful mango]

that is from March, so perhaps the bill has changed since then, i'm not sure

[uneven wasp]

they named two bills SMART

heres the copyright act

https://www.congress.gov/bill/117th-congress/senate-bill/3880/text

heres the SMART act

https://www.congress.gov/bill/117th-congress/senate-bill/2801/text/rs

dumbass senators

[sinful mango]

lol i'm confused

if i am mistaken about this please let me know

the above was just my interpretation based on that EFF article and a couple other articles i read about it

i also found this: https://riponadvance.com/stories/armstrong-unveils-bill-to-protect-copyrighted-work-of-creators/

The Strengthening Measures to Advance Rights Technologies Copyright Act of 2022, or the SMART Copyright Act, H.R. 9541, was introduced by Rep. Armstrong on Dec. 14

i believe this is it: https://www.congress.gov/bill/117th-congress/house-bill/9541

[uneven wasp]

yea I posted the link up there (in the pair of links); I've been reading it

[sinful mango]

if it's as bad as the EFF says, i hope it is widely opposed.

[uneven wasp]

who doesn't love a good old recursive definition:

“(A) is used by—

“(i) a copyright owner to identify or protect a copyrighted work; or

“(ii) a service provider to identify or manage a copyrighted work; and

“(B) may vary across types and sizes of service providers.

[sinful mango]

haha

[uneven wasp]

from random online law dictionary

technical measure means any device, product or component incorporated into a work that effectively prevents or inhibits the infringement of any copyright or related right;

[sinful mango]

a banana is: a banana that is a fruit that is a banana

[uneven wasp]

I don't believe that the publishing of an allegedly non transformative derivative work (which is the argument required for AI trained on copyrighted datasets to be illegal under current US copyright law) could be regulated by a "technical measure"

It's more of an expansion of existing DRM authority

[sinful mango]

according to the Copyright Alliance, they are going to try to argue that training AI models on copyrighted material is copyright infringement under a lot of scenarios, regardless of how transformative the output is

here is their recent position paper: https://copyrightalliance.org/policy/position-papers/artificial-intelligence/

check under the "Fair Use" section

[uneven wasp]

Yea that's an argument of its own--for reference the impact of the original market is one of the four factors listed in the law that defines fair use

[sinful mango]

There are some who believe that use of copyrighted works for AI ingestion will always qualify as a fair use under section 107 of the Copyright Act. That view is inaccurate. While there may be instances where ingestion and training qualify as a fair use under section 107, that likely would not be the case if a TDM license is available, the use is commercial, or the resulting AI generated work harms the actual or potential market for the ingested work.

[uneven wasp]

My point is that, even if the argument that it is wins, banning it under "technical measure" authority wouldn't fit

[sinful mango]

so, if they argue that image gen AI harms the market for illustrators, it could cause a big problem

[uneven wasp]

simply because that isn't a technical measure as the law defines technical measures

[uneven wasp]

yes theoretically if it didn't fall under fair use. Though a binding license that public-domains all outputs should make it legal again because then outputs couldn't compete

[sinful mango]

oh i see what you're saying.

i mean, i have no idea how they could possibly detect if an image was trained using an "unethical" model using a technical measure anyways

but who knows, maybe they could devise something somehow

[uneven wasp]

You can't really, regulation wise they can make it guilty until proven innocent if the power to regulate is given to an independent government agency

[sinful mango]

the "transparency" section of their paper sort of hints at something like that

[sinful mango]

what if a games studio used a model internally to generate, say, texture assets for 3d models. since the output of the diffusion model technically would be competing with artists who work in that domain, do you think it would be considered competition and thus infringement ?

[sinful mango]

btw, could you by chance help shed some light on what they mean by "if a TDM license is available"?

[uneven wasp]

I believe it would have to be a derivative work which competes with the holder of the underlying copyright for it to fall under that provision. I.e. you're contracting greg rutkowski, fine tune a model on all his public work, then fire greg rutkowski and use that model instead

[sinful mango]

LOL

the example made me lol. imagine if someone did that

[sinful mango]

don't you think they could argue that if an AI model is being used to generate concept art, if it was trained on concept art (without permission) it would be considered infringement?

[uneven wasp]

TDM license=Text and Data Mining license

afaik

[sinful mango]

yeah, i did look up what TDM means, but i couldn't tell exactly what they meant by "if a license is available." i guess it must be a license that forbids scraping the IP for data mining.

[uneven wasp]

While we're on the topic, a technical measure might be able to include a DO NOT TRAIN metadata tag

[sinful mango]

interesting

[uneven wasp]

if the outputs are being used to compete with/replace the concept artist in the dataset it might fail that category

[sinful mango]

btw note that they mentioned "commercial use" as a disqualifier for fair use. that is pretty serious, as it sounds like it means you couldn't use the AI model to generate anything for commercial use.

[uneven wasp]

yea no its a factor in fair use but not an automatic disqualification

[sinful mango]

oh okay, so they weigh all of the different points

are you lawyer btw?

[uneven wasp]

no, just a college student who picks up a little bit of legal knowledge everytime it comes up in public controversy

[sinful mango]

oh i see, so it has to specifically be the artist who produced the IP who could be competed with. do you think they could argue that harming the market for their profession in general is enough to qualify, without having to prove that one specific artist was harmed?

[uneven wasp]

premium news broadcasts, for example, can display and report on copyrighted material under fair use for the commercial use of selling their subscription

[cold island]

can you tell me is stable diffuser a website for a software

i was only able to find the blog page

so i joined the discord

can you tell me where is the actual thing

[sinful mango]

@uneven wasp i'm thinking of rewriting some of my video script now to add more clarity. do you mind if i DM you?

[sinful mango]

you might want to check out the Midjourney discord server if you just want to try playing with generating images quickly

[uneven wasp]

(A) if you're not in the dataset and you're harmed by it theres no grounds for copyright infringement bc you weren't in the dataset
(B) if you're in the dataset and you're harmed by it, then you would try to sue the company releasing/using a model instead of you on the basis of it being copyright infringement.

[uneven wasp]

sure

(C) copyright infringement like this tends to be a civil issue (piracy can be a criminal one irritatingly enough but that's not the fitting desc here), so enforcement would be done by one entity suing another unless regulatory powers get involved then it comes down to whatever that specific authority grants

[uneven wasp]

There's a bot in the channels below, dreamstudio (stability ai's platform) has some free credits, theres a couple other places that offer a free tier, or you can download the model (its open source) and install, but it requires some technical expertise

[sinful mango]

oh didn't realize there's a bot here too

[mortal narwhal]

Is there any style wiki where different art styles in world are documented. If possible can any one start a wiki page that would be so useful

[mortal cliff]

who think AI kill creativity?

[lofty zephyr]

AI inspires creativity imo

[steep wing]

In a sense. I haven't 3d modeled much since I got into Stable Diffusion. on the other hand I have created a lot more different types of art and trained many models and embeddings since getting into it, so I would say it just moved my creativity elsewhere

[ornate wren]

anyone interested in buying a fully functional cross-platform app branded on tattoos for this technology?

You can see it under the name "ai-tattoos" in all your stores and many languages. Details per PM. Serious requests only (this won't be 100 bucks). The app does make some bucks, but I'm not anymore interested in having two full time jobs, because of shifted life goals (family > money). IG / tik tok / twitter etc. handles ofc are part of the deal.

The app (rather backend) has fully secured DDoS protection. Is serverless and we have without trouble survived multiple attacks on the system already (i.e. fake in app purchases, high load attacks on infrastructure without our costs exploding).

Backend is nearly fully set up with terraform, i.e. you can spin up dev instances, develop & test stuff without affecting prod, and then proceed to deploy it to prod.

Runtime costs are REALLY small, fix costs are below 5€, variable costs are based on user load, i.e. gpus scale up and down. Backend has not been touched for >1 month. There was not a single outtage, not a single hiccup.

[tender saddle]

Anybody down to chat with a newbie

[fervent thunder]

They will LOCK up stable diffusion with silly legislation in the name of 'protecting artists'. It doesn't matter which side is right or makes sense anymore, whatever crowd is the loudest will win.

[upbeat tangle]

eh i doubt it,most people really dont care about AI that much

[fervent thunder]

They already started a gofundme with hundreds of thousands to get someone legal to speak abt it

[upbeat tangle]

thats a joke lol

200k is nothing when you wannadeal with copyright legislation

[carmine fulcrum]

Nah, every digital artist tool in the future will use AI for things like healing brushes and fixing seams and UVs. Everyone will use it. There's no way business is going to let artists work slower just to make a handful of artists happy. It will just be a few years of kicking and screaming but then everyone will use it like they use ZBrush and Substance or whatevs.

[steep wing]

Yeah ask any questions you may have

[upbeat tangle]

but i think the bigger problem is emad caring about SD

im not sure if he would keep releasing open source models

[split crescent]

The sad thing about ai art are the artists just have to live with it. https://i.4cdn.org/h/1672345549783451m.jpg

[tender saddle]

If want to install stability diffusion on my windows machine. Do I ignore the dream studio stuff… I’m assuming that it’s a web based system

[steep wing]

Yeah dream studio is their online platform that uses a credit system

[fervent thunder]

AI has the potential to enhance, rather than replace, traditional forms of art and the artists who create them.

We also believe that AI has the power to democratize the art world, making it more accessible and inclusive for a wider range of people. We hope to foster a more diverse and vibrant artistic community.

[steep wing]

@tender saddle
Easiest way to install SD is with this launcher/installer
https://www.reddit.com/r/StableDiffusion/comments/zpansd/automatic1111s_stable_diffusion_webui_easy/

[fervent thunder]

even AI agrees that AI art is good ^

[tender saddle]

Wow that’s amazing thank you very much!

[steep wing]

No problem! My dms are off, but feel free to add me if you run into any issues and want some help

[brazen fox]

I think the Disney copyright alliance is scarier- they are notoriously vicious when it comes to this stuff and have very deep pockets. As emad said, if they hire a lobbyist stability could hire three, that's not a problem

[iron stratus]

gm

[steep wing]

Just had Stable Diffusion explained this way to me by a custom chat AI,
"Stable Diffusion is a machine learning algorithm that uses latent space to generate artistic images and other types of media. In machine learning, a latent space refers to a set of underlying variables that are not directly observed, but that can be inferred from the data.

To generate art using stable diffusion, the algorithm first creates a latent space by training a machine learning model on a dataset of artistic images. This training process involves adjusting the model's parameters so that it can learn to recognize patterns and features in the images.

Once the latent space has been created, the algorithm can then use it to generate new images by sampling points from the latent space and using them as input to the model. The model will then generate an image based on the input it receives from the latent space, creating a new and unique piece of art.

One of the benefits of using stable diffusion to create art is that it allows for a high degree of control over the final product. By adjusting the parameters of the model and the sampling process, it is possible to create a wide range of different styles and types of art. Additionally, stable diffusion can be used to generate art that is highly realistic or more stylized, depending on the desired outcome."

[velvet oasis]

what's the difference in use-case between depth2image and img2img?

I was thinking about a project idea using depth2img but then realised I could just do that with the img2img anyways

[steep wing]

Depth2Image keeps the original structure of the image and replaces it with your prompt, trying to keep the shape as close to the original as possible

[steep wing]

I use depth2image for taking sketches or generations and then changing the subject while keeping the same composition

[velvet oasis]

aah, so basically img2img takes a looser interpretation of the picture

was there a parameter where you could say how much deviation a generated img2img could take from the original prompt? I know you can up CFG for text2img but I haven't used img2img at all 😓

thank you though!

I'll probably try using depth2img then; it's a really cool thing

[brazen fox]

img2img just adds new noise to the image, depending on your denoise strength, and tries to denoise it from that point onwards- it uses the image as a base for a new picture

depth2img can completely change the picture while still keeping the general composition- but colors and objects can be completely different

[candid shell]

Hi

[rugged mulch]

Hey everyone! Just joined this discord and feeling overwhelmed with all the content available.. can anyone please direct me to where I should go to start learning how to add images of myself to a model, train it, and then use myself in prompts. Big thanks in advance! (Saw this on Corridor Crew and would love to give it a go)

[velvet oasis]

I'd recommend using stuff like stable boost (or whatever it is) first, depending on your programming expertise

Best to google around and slowly get into it so you don't scare yourself 😳 (I started by just scrolling lexica art for way too long and now I wanna know how all this shit works properly)

[rugged mulch]

Appreciate the tip - I'll give that a look and start scrolling. I've done programming throughout Uni and a bit for work so I'm familiar with a little, but there's still so much to learn.

Will start with Stable Boost

[velvet oasis]

Automatic WEBUI is also a thing

[old raven]

hi guys , can some one help me out im having trouble acheiving realist faces given the source image both in a1111-sd and mj4

[ocean mango]

https://openai.fund/news/introducing-our-first-investments

[wary egret]

Guys, if it possible, when I downloaded the initial image to dreambot, to give a prompt to AI not strongly to modify the person?

[ocean mango]

https://lifearchitect.substack.com/p/the-memo-30dec2023 AI great info

[brazen fox]

wrong channel buddy, go to a channel like #1047760914008522782 and use the command others use

[bleak matrix]

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

[bleak matrix]

Good morning! I have a new users guide here: https://docs.google.com/document/d/1aHJ9RBt_vlCwJQBVUUsb7VghKB-wynv7WGTxm9ozL1k/edit?usp=drivesdk and don't forget to check #1014939219904450590

[dull hare]

Thank u , gonna test it out 🙂

[lofty parcel]

I found out that the thing I said yesterday about completely cracking controllable 2D worldgen with Stable Diffusion... was actually wrong and I have more work to do

[bleak matrix]

That's development for you!

But don't get discouraged

Keep moving forward

So long as you are chipping away at it, you'll be making progress

[lofty parcel]

I think there will always be failure cases because SD itself has failure cases
But hopefully I can start getting good accuracy

[lofty parcel]

My current dilemma is that I'm commissioning a real life artist and AI allows me to make super good, specific references but idk if they will be upset if I show them

But when you can't draw, being able to generate references of specific elements you want is a godsend

Img2img allows you to make mockups of compositions

[potent spire]

Many wont be mad in that case

You prompt a reference

And you respect artists if i see that correctly

[lofty parcel]

I think seeing all of the debate has made me fear that most people only see things in a black and white way with no room for nuance

[bleak wolf]

there are also many people in the middle ground

[keen nexus]

hello, how do you make a text that is respected ? thanks

[young ivy]

anyone know what might be causing my bracket () to turn into {} in automatic? It happens automatically and only in there

[lofty parcel]

Have you checked the Github issues? Usually when I have a specific problem on someone else's open source project, I look for a Github issue about it to see if anyone found a cause or solution

[young ivy]

i don't know how to search for a specific one

[lavish current]

Hopefully tomorrow I will have the day off to finally generate more art and do more AI stuff

I stopped for a while (probably due to work) but it's so much fun

Hopefully I will also get time to actually draw more of them lol

[summer nexus]

Does anyone know how to img2img each picture in a folder using the prompts in each png file (in automatic1111)? I would like to first create 100s of 512x512 images and then choose the best for img2img with higher resolution. Like hi-res fix, but in two parts and only to some images.

[unkempt hatch]

so is this correct in my webuserui.bat file?

`
@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --medvram --autolaunch --git pull --call webui.bat`

[crimson violet]

okay

I'm tired of staring at a white screen.

How do I make this enter dark mode?

[peak orchid]

Does the depth model require anything special to use effectively?

[ionic copper]

ayo whats good

[stable vortex]

the twitch is not smooth, any advices to rewatch the live

[lyric lance]

hello

[bold thicket]

Is there img2img in the discord bot

[lyric lance]

hey guys do you know any models that are good for 2048x2048 ?

[bleak wolf]

There arent any

Generate at low res

then upscale with R-ESRGAN

[lyric lance]

yea though so, but I wonder if there are any other nice models, do you guys know any good sources?

also why if i want to make a 2048x2048 it uses only 16gb of vram and when i make a 2048x1504 it uses 20gb ???

this makes no sense

[raw prawn]

Wait how do I get my prize for the giveaway

[elder hazel]

how to use this bot

?

[unreal vigil]

How can I make img2img here?

[carmine oracle]

Hello does anyone have any idea why is always get black screen?

i installed the v2-1_768-nonema-pruned and worked fine but when i try to generate a image i get black screen

[lyric lance]

what do you mean it worked fine

[carmine oracle]

i used different module now i got this one and this doesn't work for some reason

i followed this video

https://www.youtube.com/watch?v=e3vcYVwEkW0

[carmine oracle]

Do you have any idea?

[lyric lance]

well it depends what gpu you are using and it depends on a lot of things, i never used the model you are using so i can't say really sorry man

[carmine oracle]

this is the 2.1 sd.

i just installed that.

[lyric lance]

whenever I had any problems i just reinstalled the gpu driver and stable diffusion dependencies and it worked fine

clone another version of stable diff and see how it works

[brazen fox]

add --xformers to your webui_user.bat launch options

set COMMANDLINE_ARGS= --xformers

then launch SD from there

[lyric lance]

interesting thing what does xformers actually do, i run without it and its okay, what is that for?

[brazen fox]

it's a form of optimisation, for some weird ass reason it fixes the black image thing that occurs on some GPUs

[lyric lance]

ahhh okay i see makes sense then

[carmine oracle]

So i just edit the webui and add this there ? --xformers "set COMMANDLINE_ARGS= --xformers"

[lyric lance]

I'm using a tesla p40 so i never had any problems with that

[brazen fox]

webui_user.bat, not webui.bat

[carmine oracle]

yes i got it thanks

[brazen fox]

nice, hope it works out

[carmine oracle]

let's see if it works

[brazen fox]

never had problems either on my puny 3060 but some GPUs, even top of the line ones like 4090s, have the weird black image thing

[lyric lance]

must be some gpu architecture thing

[carmine oracle]

Which module you thing it's the best one for anime or i'm searching for a particular style Ukiyo-e style

[brazen fox]

brother I am not the right person to ask that, lol, I don't like anime at all- but I would try looking on civitai.com, and see if there's anything that fits your taste

[lyric lance]

novelai lol

there was a leak of it a month or so ago

[carmine oracle]

Thanks guys

Also @brazen fox it works now.

After hours or installing shit my brain is almost melting down

[brazen fox]

cool, have fun with 2.1!

[old marten]

There is no good reason to use EMA for training models anymore or is there? I still can't understand how it works

[lament osprey]

hello

[fervent thunder]

How do I create those AI video things

[halcyon granite]

when do you think we will start seeing AI that looks legit close up? As in the same quality as a photo or Photoshop

[topaz charm]

the "start here" on this server has a SD version without hypernetworks...is this one inferior to the rentry one?

[full lark]

Maybe 3-5 years in my opinion

I´ve seen some AI generated images that look so realistic now. but mostly because of the subject is far away.

[halcyon granite]

yeah although I find it's closeup objects that look more realistic. With my cityscapes the cars are just random shapes up close

[full lark]

Yeah, it depends on the images

[brazen fox]

I imagine you're mostly talking about consistency here, but there's actual photo level stuff already with the right methods and subjects

[old marten]

Uh... today?

2.1 already has some really good results, I saw a bunch in my regularization images and it was literally a prompt of "person in high resolution" lol

[topaz charm]

how are you using 2.1? followed rentry?

[old marten]

Huh? I just downloaded the model and the yaml

I already had automatic1111

[topaz charm]

ah, I recently reformatted and I downloaded the one on this server, seems to be pretty old

will reget automatics

[old marten]

If ur gonna train on 2.1+ get the newer webui since it has more options

[halcyon granite]

I'll try 2.1 but like I said close-up, for example cars in a city from far away

[lusty mason]

Is there a good method to create 3D models using SD

[chrome relic]

Can someone point me to the site that houses image and their models?

[dawn yarrow]

i know theres a youtube video on that

[lusty mason]

Remember name of it?

[dawn yarrow]

"how to make 3d game assets with ai " albert bosain

bozesan

[brave wolf]

I am looking for a recommendation for an image file manager software for windows. Specifically I just need the feature of easily searching through dirs of images metadata

Sometimes I need a section of a prompt from another image and I can only remember a couple words from the prompt

[lusty mason]

Ah thanks

He does the modeling in the video

I was curious if there's a method to not do modelling

Just type in like normal SD but 3D version of it

[dawn yarrow]

better luck nezxt year 😉

[lusty mason]

Lol

[dawn yarrow]

yeah currently i think it takes a little groundwork to model aroundit

[lusty mason]

I wonder if there's software that turns texture to 3D using depth information

Like materialize but better

I mean that would turn SD image to 3D model

I saw a website doing that but it's not open source

[dawn yarrow]

that is the case

[lusty mason]

Ah I want to create a game that has all its assets generated by AI lol

[dawn yarrow]

cool i suggest image to image with masking in gimp

bosezan has a tutorial on that as well

[lusty mason]

I really like how this is easily possible nowadays, last year people wouldn't have imagined this to happen lol

[dawn yarrow]

my pc is a tad underpowered unforntuantely

[lusty mason]

Use colab

[dawn yarrow]

does it have image to image?

[lusty mason]

Yea

[dawn yarrow]

oh boy

[lusty mason]

It's fast too my pc runs SD but I just use colab instead

Lemme get you the link

[dawn yarrow]

if you could dm me a link

ah thx!

[lusty mason]

https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast_stable_diffusion_AUTOMATIC1111.ipynb

[compact linden]

How do I put in a prompt and where?

[dawn yarrow]

thx bee

[dawn yarrow]

dreambot rooms slash dream prompt: insert prompt here

[pulsar pollen]

Can someone link me to a stable diffusion colab that is currently working? the one I have been using isn't working and I saw there are a bunch of other ones so I would like to use one that is more used

[fervent thunder]

Can someone please answer some questions

1. How do I create those AI videos that almost look like stop motion. I have SD installed already.
2. Can I download SD on decent computers with no graphics card and just integrated graphics

[lusty mason]

welcome mastermike

[golden citrus]

Is anyone familiar with a method for diffusion which moves from 1 image to another, basically using diffusion to fill in the space between ? something like an init image and a target image ?

[still remnant]

Hello everyone, may your diffusion be more stable than I am!

[fluid condor]

Hello. Where can I go to learn about setting up my own training model? Like using a custom data set to train SD? Any help is appreciated!

[cobalt cape]

hello do you know where to get the free the nipple model

[still remnant]

Re: number 2, I have run SD on CPU, which will likely be your option for integrated graphics. It can be done, but it is slow. I was running around 4 minutes per 512x512 image on my i7, compared to around 40 seconds on my still-unimpressive Quadro P1000 with 4GB of VRAM.

This was my laptop vs workstation,btw.

[fervent thunder]

hi, i'm wondering if anybody knows of any web apps or APIs where one can upload an image or images, and a prompt, and receive as output an image variation. e.g. upload a picture of a house and the prompt "black and white", and the result would be the prompted b&w style applied to the original image. edit: looks like dreambooth can do this

[dense dune]

guys I have a question about models

stablediffusion is a model?

if it is what is my image ckpt file?

[fervent thunder]

I see, runpod has a system where you can use SD by renting a GPU. So technically would you be able to run it on a phone

[still remnant]

Thats true, rental services open up a whole world of possibilities, some of them are even free.

[fervent thunder]

Free? where

[still remnant]

I was under the impression google colab had a free tier, I know I've read about others on message boards, but as always, free will be significantly slower.

[lilac reef]

https://discord.com/channels/1002292111942635562/1058572276930465853

[fervent thunder]

Still very cool tho, let me know if you find one with SD webUI for free

[proper crescent]

Hey hi and howdy… is there a version for iPad?

I only saw info for windows

[tender saddle]

Does anyone know how to install another notebook. I have 1.5 in my google drive and I’d like to install 2.1….

[tender saddle]

You could use google notebook and do it on the cloud

[proper crescent]

Okay. Is there instructions for that here. I love instructions

[tender saddle]

I sent some to you

[proper crescent]

Appreciate it

[oblique scaffold]

is anyone in here willing to call me and teach me how to use sd?

[ashen jasper]

If you want visual help, try youtube first

[raven edge]

question is there any talk of increasing the parameter's of the model to like 2billion or 5bill?

[uneven wasp]

what value would that bring?

[forest apex]

@oblique scaffold best I can suggest is try the automatic installer (https://www.reddit.com/r/StableDiffusion/comments/zpansd/automatic1111s_stable_diffusion_webui_easy/), watch a ton of youtube videos, and visit #🤝｜tech-support

[oblique scaffold]

okay

i just gave no clue how to use it

[forest apex]

what do you want to make?

[fervent thunder]

Oh wow. I haven't been on this server in a while. Anyone wanna give me the run down of any news regarding the community since the Unstable Diffusion anti AI activists keep trying to bring down AI art?

[oblique scaffold]

idk like character mashups or cool scenery

[forest apex]

news from the future: they failed miserably, and looked silly trying

[fervent thunder]

I mean they successfully removed the Kickstarter and patron. Now they're going after stripe their new payment processor.

[forest apex]

yea but UD still got major donations directly to their web site, from what I saw

[fervent thunder]

Which uses Stripe to process the credit card payments.

[forest apex]

on one of the anti ai discords they were trying to rally everyone to go and report random small github projects, petty stuff

[fervent thunder]

Yeah I saw them trying to get models removed from certain sites. A bunch of petty stuff

[forest apex]

the power of propaganda

[fervent thunder]

should I download v1.5-pruned.ckpt

or v1.5-pruned-emaonly.ckpt

the latter is smaller in size

my internet is kinda slow so it makes a difference to me

7.7gb down to 4.27

[fervent thunder]

You could always use sneaker net and then download the bigger model

sneaker net?

[forest apex]

the larger model is mostly useful for training/dreambooth/textual inversion

[fervent thunder]

Yeah. Download it somewhere with a faster internet onto a thumb drive then use your sneakers to walk it to your PC.

I’m interested in checkpoint merging other than just normal generation

does that count as training?

[forest apex]

yes

@fervent thunder it seems stripe is still accepting payments from their website, I haven't been active the last few days so I haven't seen any other news, the UD discord has some murmurs of them going after stripe though

[fervent thunder]

Yes they're still accepting payment for now. I've seen Twitter posts tweeting at stripe that they're allowing a company to use their service that allows child pornography.

[forest apex]

the childrens.... what about them

I don't think it'll matter. There are very wealthy technologists with large amounts of crypto, still, and if they can't take fiat they will find another way.

[cedar kayak]

Anyone here can share experience of txt2img with an RX 7900 XTX 24GB ? (I’m doubting between that card because of 24GB vs the upcoming RTX 4070TI 12GB) because of the same price of $800-999 range

If I want to render to 1600*900px, which one would be faster? AMD because of 24GB? Or NVIDIA because of higher FPS in games 15% between the 2 cards and better onboard AI tech ? 🤷🏻‍♂️🤷🏻‍♂️🤷🏻‍♂️🤷🏻‍♂️

Thabks 🙏🏼

Hmmm no AMD fanboys here apparently 😂

[forest apex]

@cedar kayak if you don't get answers here you can try #1011228477954998273

[cedar kayak]

Nvidia it is 🙃

Ah cool ok thanks 🙏🏼

[mortal narwhal]

Is here any hope more images will be added in stable diffusion training as quality is going down

[sleek osprey]

Lol looked up last messages of emad to see if there's anything interesting and his last message is a reply to me I didn't see!

[steep wing]

Lots of hope

More data sets are being collected if you look at their job postings

They are reaching out of Laion

Just look at 2.0 to 2.1, while the noticeable quality was only slightly increased, the number of images went up a lot

[sleek osprey]

Is there a ETA on distilled sd?

[steep wing]

Not a source, but I keep hearing sometime before January 10th-14th

Just from various random people on Discord though

so speculation at best

[sleek osprey]

awesome

What changes apart from lower step count requirement will there be in distilled?

Will it be faster to do textual inversion and stuff too?

[mortal narwhal]

Is it possible to submit our own images. I can create different styles of images and would like to submit

[steep wing]

No idea, they don't even release proper documentation with new models so very little is confirmed

[steep wing]

If you could find what OpenCLIP is pulling from, that would probably be the best way

I know they added more than Laion, but not sure what

[mortal narwhal]

thanks for info. Seems I should just add images on website

[steep wing]

If you can add to Laion then yeah that would increase chances of it being pulled for training

[mortal narwhal]

I tried searching how they do. They scrap websites for images

[short tapir]

Is there a specific way to do a prompt in order for it to generate art of a specific character from specific show?

Iv tried a bunch of different ways but it never seems to work me

Like is there a specific way I have to do a prompt to get it to do that?

[forest apex]

I think it'd depend on if that show/character are in the model, if not you would have to train your own model

[bleak wolf]

You would rather train your own model

[short tapir]

I just wanna make cool fan art

[bleak wolf]

whats your gpu

[short tapir]

How do I check?

[bleak wolf]

hm

many ways

easiest is to right click to windows icon and press device manager

then click on display adapters

then say what it says there

[short tapir]

And Radeon(TM) graphics and Nvidia GeForce GTX 1660 super

[bleak wolf]

you make it, barely, but make it

you are able to train your own model locally

it isnt a laptop right?

if it is then rip

[short tapir]

No. It's a desktop

Also. Unless it's how to train a dragon I have no idea how it works

[bleak wolf]

there are some tutorials online, but it involves technical stuff (no scripting tho)

do you have webui by automatic1111 installed on your pc?

[short tapir]

I don't know what that is

[bleak wolf]

ah well

[short tapir]

Is it a model?

[bleak wolf]

no

it is

a UI

that is very handy

[short tapir]

I only have the stable Diffusion ui

🤔

At least I think it's a UI

[bleak wolf]

like

do u use a website when u generate

[short tapir]

My computer runs on orc logic. Painted red for speed. I have no idea how it works but as long as I believe it works it does

[short tapir]

Sorta? Like I downloaded a think but to gen stuff it opens an internet tab

[olive portal]

any good free online stable diffusion site?

[bleak wolf]

Stable UI

https://aqualxx.github.io/stable-ui/

[olive portal]

whats the link

thanks

[bleak wolf]

does it say 127.0.0. something like that on the tab?

[short tapir]

Yes

[olive portal]

this download to web4? haha no jpeg/png?

[bleak wolf]

then run it locally lmao

[olive portal]

my laptop sucks lol

[brazen fox]

Have you tried the dreambot channels in this server?

[lime relic]

is there an AI to upscale images?

[olive portal]

what is that? i tried MJ thou

[brazen fox]

This server has channels like #1047759808834584606 where you can generate images for free- it's limited in its use compared to a local installation, but it's still free and very useable, a good way to practice prompting as well

[olive portal]

oh yeah i tried that, not really a good result. I think im addicted to MJ result now if i compare with SD, abit heaven and earth comparisson : (

[brazen fox]

That's because MJ basically does all the work for you- it improves what you type behind the scenes with all kinds of different addons and extensions, but then again, it's a commercial business as opposed to open source software

although MJ still uses SD as a base I believe

[olive portal]

im using this https://aqualxx.github.io/stable-ui/ << and Vintendious Diffusion model seems quite okay, not to shabby compared to MJ

"Vintedois Model

[brazen fox]

ah right, I forgot about stable horde- it's pretty allright I guess

[sleek osprey]

Has anyone tried or succeeded in making a proper comic book (good, consistent, coherent art) using sd Or any other art model

[humble dust]

Hi SD experts. Have anyone got xformers to work on the old GTX 970 graphic card?

[bleak matrix]

Good morning, everyone! Happy New Year!

How are we all this beautiful morning?

[young ivy]

This afternoon Imma take a nap, then get ready for this eve

[bleak matrix]

That's a very good plan

I'm just waking up myself, so, you know lol

[young ivy]

^^

what time is it?

[lofty parcel]

Adventure time

[bleak matrix]

7:43AM!

[bleak matrix]

This is a p e r f e c t response

(More like time to play pip install/uninstall)

[frail jewel]

Hey everyone! What's up?

[bleak matrix]

Just messing around with Anaconda

You?

[uncut plinth]

when you are going to release the new model that is 60x faster ???

[winter pike]

i use gigapixel. SD 2.x came with an upscaler but i dont know how to use it

[leaden mural]

So... This came out just now:
https://www.youtube.com/watch?v=7PszF9Upan8

[bleak wolf]

which side are you on fr

[leaden mural]

balanced, mostly for AI art, but with some ethical conditions.

[bleak wolf]

such as?

[leaden mural]

like there must be an option for opt-out, don't monetize AI art from copyrighted work, etc.

[bleak wolf]

respect your opinion

either way

Computer technology is advancing

it might even get to the point where you will need like only 1000 images to train an almost perfect AI model

even if artists/photographers restirct their content

AI will probably get there by brute force anyway

its just a matter of time

[leaden mural]

true

And I respect that

[bleak wolf]

Yeah not really an opinion more of a prediction but ty anyway 😄

[leaden mural]

heh, yw

I know that's gonna happen anyways, with the exponential growth of AI and tech in general.

We can use AI art however we want, but it must have some guidelines to avoid some of the more sketchy stuff from happening.

And most of which have already been implemented.

[north stirrup]

We could go a lot further on the culture side of things

[astral goblet]

Same guidelines that have always applied. If you're a digital painter who is copying someone else's style, you'd credit them. You don't have to but that's the etiquette. An interesting case here is CPG Grey youtuber, his style has been completely lifted many times by other youtubers. Hardly anyone that rips it off credits him as an inspiration, but no one is outraged because people understand why the style is lifted. It's very easy to replicate and very easy to animate. Technically speaking, it's a huge improvement to the production process and allows content to be more expediently created. There's even a ton of discussion on the ethics of this one situation through out the lexicon

[hollow cobalt]

how do i use it

hellloooo

[steep wing]

Head to one of the #1047759875914076160 channels and look in #1014939219904450590 for instructions

[leaden mural]

yep

[sharp pendant]

looking to pay an specialist! need a handful of amazing image models for my project that automates small business social medias
add me if interested

[agile mica]

if someone please help me in #🏞｜general-with-images

[sharp pendant]

are there mods?

[peak orchid]

Artists who use others' art as references shouldn't be to monetize their work then either

[brazen fox]

Where do you draw the line? What if you combine styles? Would that be okay or still infringing? How do you prove that?

[brazen fox]

Moodboards are most definitely not infringing while the general principle is the same- why can companies say 'we want art like this' but individuals without the same resources can't?

[leaden mural]

Not the point.

[peak orchid]

"sketchy" art is a matter of opinion though as well, with widely differing opinions across different cultures and groups

[leaden mural]

A hard limit is with things that are just straight up copies. Otherwise, it's free game with copyrighted work.

[peak orchid]

Its absolutely the point, considering we are using AI to make art

[leaden mural]

And we can still do so, just don't bother monetizing the works that aren't really transformative to begin with and are just there to rip people off their money.

[brazen fox]

Pretty much nobody straight up copies, though- why would you copy something exactly? People are salty about styles and using material to train, not about people duplicating their art because that is just useless

[peak orchid]

Its not ripping people off their money if they are paying for "x in the style of y"

[brazen fox]

I'm not going to create a 1:1 copy of a van Gogh painting and sell that, so where would you draw that line? When is something transformative enough?

[leaden mural]

I have seen it, especially with weirdos taking art from other artists to make a dreambooth of it, then taking the work of said artists, use said model to transform it a little with really minor adjustments, then sells it to the highest bidder, passing it off as his own.

[brazen fox]

People can do that with photoshop as well, that's not something ai is instrumental in

[peak orchid]

We will also soon be at a point where we will be able to determine what sources an artist uses to make things by hand. Such a future could destroy art if we go down the path of making references illegal

[leaden mural]

Yes, because it is a tool. I only seek to stop whatever those douchebags are doing, and some regulation is needed. It's not there to stop AI art. It's just there for those cases.

[peak orchid]

Artists should be careful what they wish for

[brazen fox]

Look, I'm all for artists having autonomy over their work, but most of the concerns stem, in my opinion, from being misinformed and grossly overestimating the personal impact AI will have on an artist's work

[steep wing]

Unfortunately laws have a lot of side things that come with them, putting up unnecessary regulations hurts both sides

[leaden mural]

That's if the regulations are way too over the top in favor of those looney artists lobbyists. It will be alright, if we do something about it.

[leaden mural]

But reasonable ones can. Or are you expecting some sort of a formal agreement with both sides regarding this?

[brazen fox]

Anti ai groups teaming up with Disney in the copyright alliance should be enough of a warning where this all is headed

[steep wing]

Not an agreement, people just need to get over themselves and accept the new tech. This happens literally anytime a new technology comes out.

[leaden mural]

And that's what I'm afraid of. Like I said, looney artists lobbyist. Especially when those groups are with a larger conglomerate designed to exert itself as the only force to exist PERIOD.

We really need to do something about that.

[steep wing]

We can't do anything about changing people's minds that don't want to change

[brazen fox]

You don't fix that by basically extending copyright though

[leaden mural]

And just abandon the artists in the dust? Without even a slightest consideration?

[steep wing]

Bro... The technology needs to be embraced by the artists as well. I am a digital artist and I use this tech in my workflow daily

[leaden mural]

I am an artist too, and I too do use this tech in my workflow. Just needed to be reasonable with regards to how this is used and not go willy-milly like for criminal or sketchy activities.

[steep wing]

There will always be criminals and sketchy shit, we haven't been able to stop that since humans could walk

[brazen fox]

Camera's, photoshop, digital art didn't kill painting, why would AI suddenly end art as we know it-the answer is not in regulating but in educating

[leaden mural]

Not everyone is going to accept it man, not everyone is going to embrace this tech. But by the very least, we need to be considerable with our regulations and our actions at least.

[steep wing]

Yes I agree with that

Some regulations will be put in place I'm sure, just hopefully they are helpful and not restrictive, while still keeping everyone in consideration

[leaden mural]

yep. that's what I was hoping for.

[brazen fox]

There is very little that regulating can actually fix imo, without massively overdoing it

[leaden mural]

If it's not done right, and if those weird anti-ai art people had their way.

[brazen fox]

But that's what I'm saying- I don't think it can be done right

[steep wing]

Yeah that is my worry to

Someone will be unhappy with the outcome, and not just the sketchy people

[leaden mural]

So... We stop getting the government involved?

[fervent thunder]

software must remain free

[steep wing]

I mean ideally yes... They should focus on other shit

[fervent thunder]

they cant prohibit anyone from just making a database of public images

[leaden mural]

Ya do know that AI is going to be a force to be reckoned with. Letting it roam free is a recipe for disaster.

[fervent thunder]

and also cant prohibit anyone from just coding AI

[steep wing]

Yes. I also know that the government fucks up everything they touch

[fervent thunder]

the images that can be posted is the only thing they can change effectively

and how are they gonna restrict AI art without restricting human art

[brazen fox]

Laws will either have an impact on the entirety of copyright and traditional art, or they won't be effective at all-is what I'm worrying about

[leaden mural]

American government, to say the least. But not agrees with the US with regards to some laws.

[steep wing]

Yes, also the government that will have the biggest impact on the way this goes.

[fervent thunder]

to me it seems like its the whole pirating discussion and discussion about copyright basics all over again. its already been proven that pirating doesnt have an impact on original sales, just as using someones artwork to learn from it does not impact the original artists comissions

[leaden mural]

So, what do we need to do then? How the hell do you deal with a corporate amalgamation seeking to acquire more control in everything, without dealing with the government in some manner?

[fervent thunder]

these laws only serve those who sue for copyright infringement

[steep wing]

Bingo

I don't have an answer to that question, no one seems to have a GOOD one

[leaden mural]

I'm asking on the "How". Because there's no reasonable and realistic way to go about this.

[fervent thunder]

i mean what are they gonna do ? outlaw open source?

[steep wing]

And that is what @brazen fox has been trying to say

[fervent thunder]

if they do they really dont understand internet and computers

[leaden mural]

Sooner or later, they will deal with it, and they have done so already with the case of Google vs Author's Guild. They have some form of precedent. What's gonna stop them to do the same thing with AI, with or without Anti-AI lobbyists interfering.

[fervent thunder]

local laws are kind of useless in a global net

even in china people tunnel out and just use the whole net

unless they change how the global net works these laws are ineffective

[leaden mural]

Then tell me @brazen fox, what do you want us to do, with all this in mind?

[glass grotto]

is there a big difference between stable diffusion 1.5 and 2.1

[leaden mural]

A lot, both in favor and against it.

[steep wing]

I asked a custom AI what we should do, it said this
"One possible compromise that could potentially satisfy both sides of the argument about the use of AI in art would be to recognize that AI can be a useful tool for artists, but that it should not be the sole source of creativity or the dominant force behind the creation of art. In other words, AI could be used to generate ideas or to assist in the creation of art, but the final product should still be shaped and influenced by the artist's own vision and perspective. This approach could allow artists to take advantage of the benefits of AI while still maintaining their own personal voice and style. However, it is important to recognize that this is just one potential compromise, and ultimately the best approach will depend on the specific circumstances and the preferences of the individuals involved."

[bleak wolf]

The end is inevitable. AI will brute force take over, what way we choose.

Its like the farm workers when the combine harvester was first invented

[steep wing]

It will shape everything we do, absolutely

[bleak wolf]

Even if it cannot gain sentience

It uses, phsyical occurences that cannot be

predicted,

for random number generation

that alone is its sentience

imo

like

measuring the time between 2 releases of neutrons from a uranium radioisotope

[steep wing]

Well the same AI I asked about the compromise said to do this if it gets too powerful and goes rogue
"The process for deactivating an artificial intelligence that has gone rogue, or has otherwise malfunctioned in a way that poses a threat to people or systems, would depend on the specific design and capabilities of the AI in question. In general, there are a few different approaches that could potentially be used to deactivate a malfunctioning AI:

1. Disconnecting the AI from its power source: This would involve shutting off the electricity or other power supply that is being used to run the AI, effectively "unplugging" it and rendering it inoperable.

2. Disabling the AI's access to data or resources: Depending on the specific implementation of the AI, it may be possible to limit its access to data or other resources that it needs to function, effectively "cutting off" its ability to operate.

3. Deleting or altering the AI's code: If the AI is running on a computer or other electronic device, it may be possible to delete or alter its code in order to disable or modify its behavior.

It is important to note that these are just a few potential approaches, and the best course of action for deactivating a malfunctioning AI will depend on the specific circumstances and the capabilities of the AI in question. In any case, it is important to approach the situation with caution and to carefully consider the potential consequences of any actions taken."

[bleak wolf]

The AI Art stuff is one of the first examples of "sentient" AI in my opinion

[bleak wolf]

You can just hard code a kill switch

that overrides everything

on a hardware level

[steep wing]

If it's coded, can't the AI access it?

[bleak wolf]

.

hard etch*

hard engineer*

you get it

[steep wing]

Got it

thought you meant literal code lol

[bleak wolf]

and we never make, phsyically capable, AI machines. Ever.

because they can disable hardware level kill switches

and then its.. gg

[steep wing]

Yeah I like AI but we don't need Terminator type things

[still remnant]

That would make it prudent to limit computer integration into any kind of physically interacting hardware, especially when it can connect to networks

[steep wing]

Yeah I guess forklifts and other machinery that has computers would be susceptible

[still remnant]

makes them hackable by an AI, and possibly capable of running a subset of it, for example Tesla cars...

[bleak wolf]

and now we know AI has that "sentience" because of the...

wait

if phsyical randomness is the sentience

[still remnant]

at this point its still pseudo-random, real random is kind of hard.

[bleak wolf]

we can just make sure that we never use that "sentience" aside from specific things

[bleak wolf]

someone could make a device that measures the time between neutron releases of an uranium radioisotope

the only thing the ai sentience needs to survive is to replicate that

[still remnant]

I thought an easier model would be to make a microprocessor with gates down around the diameter of a hydrogen atom, at that point, the gate only has a probability of being off, giving you true random results to logic operations.

[bleak wolf]

oh thats easier for sure

and doable today

so yeah

gg

[leaden mural]

Yeah, but like I said, why the heck should we just abandon people like this when AI rolls around?

Also, this is just arguments. I ask for something to do with regards to resolving the matter with both the art community and the government.

[still remnant]

I got the idea from reading about the challenges with making faster modern processors, we've kind of hit the floor of small

[bleak wolf]

ik, pc builder here

[still remnant]

I hate the whole "is AI art art" question or debate, it reminds me of my college freshman art class. It seems like nothing I ever drew, painted, or sculpted was "real art" to most of my classmates.

[bleak wolf]

For real

[still remnant]

If you do art because you enjoy it and use it to express yourself, AI has nothing to do with your art. If it's about "elevating the human experience" AI may have a role, if it's about selling a product to make a living, AI is another tool in your kit, but I don't care to quibble about what art is.

[bleak wolf]

"ART" is entirely up to the person

i consider engineering as "art"

just because we use math and metal instead of paint and brushes does not mean we lack any sort of free thinking or imagination

[untold moss]

Hi. Anyone familiar with how to use the api and inpainting?

[bleak wolf]

Ask away

[still remnant]

Ironically one might consider outrage over AI art as validation of it as an artform, it's eliciting an emotional response...

[bleak wolf]

I hope you dont mind me using this quote, will give credit if needed

[steep wing]

I was about to ask the same thing

[still remnant]

go for it! with or without credit is fine

[untold moss]

well...its not working. I am following the guide here https://platform.stability.ai/docs/features/inpainting and im getting an image that isnt inpainted. I make a mask where the region I want to keep is all white and the region I want to diffuse is black ... but what I get back is a white image and where the black box was I get some low quality approximation of what I asked for.

[bleak wolf]

Thank you

[still remnant]

might try cranking up the steps? Sometimes inpainting/outpainting benefits from way more steps than standard generation in my (very limited:1.8) experience.

I know that's no fun cuz it really slows stuff down, especially when it still doesn't work

nevermind, I reread your explanation, and that doesn't just sound like steps... It almost sounds like it's using the mask as part of your image to inpaint?

[median jolt]

How do you generate AI anime images

[bleak wolf]

Many ways

First step is to use something like Anything Diffusion V3

doesn't matter where you run it

[median jolt]

I have trouble installing Stable Diffusion on my computer

[bleak wolf]

The checkpoint or the webui?

[median jolt]

WebUI

It crashes when installing Torch vision

[bleak wolf]

uh

right

u using correct python version?

[median jolt]

Yes, 3.10.6

[bleak wolf]

did you add it to the envrionment variables?

PATH

[median jolt]

Yes

[bleak wolf]

i see

any errors shown?

[median jolt]

No, it just hangs and freezes

[bleak wolf]

hm

whats your PC/Laptop specs?

[median jolt]

Oh never mind it just unfroze

Uhhh let me pull them up

Well it just gave me an error

Which I don't know what to do

[still remnant]

I use healySAnimeBlend checkpoint and specify "Anime" in the prompt

[bleak wolf]

show

[median jolt]

I can't read it.

[median jolt]

Processor AMD A4-3420 APU with Radeon(tm) HD Graphics 2.80 GHz
Installed RAM 6.00 GB (5.49 GB usable)
System type 64-bit operating system, x64-based processor
Pen and touch No pen or touch input is available for this display

[bleak wolf]

there we go

[brazen fox]

Sorry, I was making dinner and eating, so just missed your comment- but what I personally think is the only option: fight the outrage, educate people. It's a tool. It's not an art generator, it helps in the process. The outrage that is currently here is not about copyright, I think. It's people that are scared of being replaced while they thought that would be impossible. It's fear. You don't fix fear with copyright laws.

Sure, it will make it easier to replicate artists' styles, but at the end of the day, all it will really do is democratize the creation of art and visual imagery. People will still pay for something that is viewed as genuine. People will still want things that a human creative has thought about, has a vision of, etcetera. You can ask a concept artist to draw something cool, and they will draw something they think is cool. They will take in mind the customers' exact needs. An AI won't fullfill that role (yet), because people do not know what they want.

As I said before, people hugely overestimate the negative impacts on the art world and art consumption in general, in my opinion. Art and visually based stuff will just become a lot more prevalent as it is easier and cheaper to make, especially for artists themselves who, paired with it as a tool, will supercharge their workflow.

[fervent thunder]

Is 8gb vram enougth?

[bleak wolf]

for what? Training? No. Generating images? More than enough

[untold moss]

thanks. I somehow got past that issue. I appear to have something else going on with my code. But now the quality of the inpainted image is really low as in not believeable at all. The people being added have duplication and everything just looks wrong. whats a good sampler value to use to ensure the most realistic delivery?

[median jolt]

I want to generate images but have problems looking for something with Stable Diffusion to even work

[bleak wolf]

https://aqualxx.github.io/stable-ui/

Powered by stable horde

a horde of volunteer GPU's

which at some point included mine

[still remnant]

as far as I can tell it's different for each image. sometimes my stuff just winds up looking 'overbaked' no matter what I try, but I've seen sampling steps up in the 80s

[median jolt]

Alright thank you

What are the preset settings?

[median jolt]

Oh god it looks really bad

[untold moss]

ahh interesting.

thanks

[bleak wolf]

why, shouldnt do that

what are your settings

[untold moss]

Hey - whats a good negative prompt to keep things as realistic as possible?

[median jolt]

K_euler and karras enabled

[bleak wolf]

disable karras, what model, what cfg, what resolution, how many steps?

[median jolt]

704x704 and 30 steps

[steady arrow]

Does anyone know how I can use a python script to gen images with auto1111?

[median jolt]

Stable diffusion, the normal one

[bleak wolf]

, cfg, resolution, steps?

[median jolt]

It doesn't show CFG anywhere but it shows resolution and steps

[bleak wolf]

just send screenshot of the page

[untold moss]

actually - how do you provide negative prompts to the api?

[still remnant]

I haven't messed with it, but the ones it uses are in the .\stable-diffusion-webui\scripts folder. You could take a peek at how those are written as a starting point? 🤷🏻‍♂️

[untold moss]

what exactly is auto1111?

[still remnant]

@steady arrow that's a good idea though, I could use more scripting to more efficiently use my overnight batch runs.

[still remnant]

a warm and gui wrapper for operating stable diffusion for those of us who aren't as into the cold hard bash environment.

[untold moss]

https://tenor.com/view/cat-chilling-fireplace-warm-toasty-gif-15674280

Gotcha.

[tepid pulsar]

Is there any way to scale the img2img better than just zooming the windows, getting frustrated on a image that i just got right and hand decent and suddenly she is holding peas and i need to pain them out.

[still remnant]

are you talking about for drawing an inpainting mask?

[tepid pulsar]

Sorry, i ment inpaint mask, getting words still mixed up

[untold moss]

how do you specify a negative prompt when calling the api? I have the following (which is sorta what i saw on huggingface) but its telling me that negative_prompt doesnt work.

prompt= "a cute cat face", negative_prompt = "ugly, disproportionate, disjointed, deformed",

[still remnant]

you could try opening a copy of your image in an image editor where you can zoom and paint a black/white mask, then use that image as your mask

[tepid pulsar]

Oh it would be possible to do that, never thought about that, much appreciated, that will make my life easier

[still remnant]

it could be theoretically be possible to do even in microsoft paint, but anything with layers would make your life way easier.

[tepid pulsar]

Yeah gotta give it a try, was easy at first when there was a giant teacup to remove and add something else but then suddenly she was holding peas 😄

[still remnant]

and one more tip, my understanding is the mask is binary, so any non-white pixel will be interpreted as black, but I haven't played with anything but hard brushes to verify

[untold moss]

found the answer

prompt= [generation.Prompt(text="a mountain landscape",parameters=generation.PromptParameters(weight=1)), generation.Prompt(text="in the style of thomas kinkade",parameters=generation.PromptParameters(weight=1)), generation.Prompt(text="tree",parameters=generation.PromptParameters(weight=-1.3))],

[leaden mural]

Wait, lol. Not done.

[brazen fox]

🤐

[leaden mural]

Unfortunately, this is the real world we're dealing with. Even if we backtrack people by educating them, they have already caught the eye of the government and a media conglomerate, and at that point, there's no turning back.

I have already stated the fact that people still need artists to an extent, even with AI art proliferating, and that there is a place for such a tool like this.

However, information alone won't save AI art. I was hoping for something a bit more concrete that would break this stalemate. It needed some action, some way to provide the government with another view point, and even before that, a means to bring to an agreement with the community of artists in some capacity so that we can avoid going to the law for this.

I am not really sure how we go about this, but some action at least that would help both sides would have been a bit better than just a information campaign.

[brazen fox]

It's exactly due to this being the real world that we can't give in to all of this. My point is that we can have regulation, and there will still be tons of people fighting to end this; a compromise is not a solution because it doesn't solve what most artists seem to be afraid of. It's not like giving a baby a piece of candy to stop it from crying; I wish. The people are inherently against AI and use arguments to protest it, they aren't against AI because of those arguments. Those, or some of those, that spearhead the movement have already said that their goal is stopping AI, not making AI 'ethical' (in their view).

You can see this already with AI being branded as a tool to produce child pornography. Sure, it can do that, but that is a bad faith argument. So can cameras, so can photoshop, so can artists. And they do. It's not something that can be fixed, other than it to be completely banned.

I know education won't fix everything, and it's not a perfect solution, but there is no real solution other than hoping regulators will see AI for what it actually is.

Maybe we need something like actual televised debates, or something high profile like that. Something that has a national news level of reach, idunno- as generally the media (very broad, I know) has already spreaded extremely biased information (from what I have seen so far)

[little plaza]

I really don't see why either the government or corporations would care about regulating the production of any kind of images outside of deepfakes. Which they're already working on. At most they'll make it so that it's only legal to distribute models without children or without nsfw. Disney wont be anti-ai, they'll be anti open source ai.

I don't think the artists lobbying is going to do anything at all tbh.

[brazen fox]

I hope you're right- I'm mostly afraid of stricter laws regarding AI due to the mentioned CP and copyright fearmongering- we'll see where this all goes

[little plaza]

Last time the US government along with massive corporations tried to enact wide spread censorship laws on the internet it went down like a lead balloon (SOPA), they might try to outlaw open source AI i just don't think they will be successful. But yeah none of us know how it will really play out.

other countries will be more successful who have more control over their country's internet already (like China)

[brazen fox]

Yeah true but if it is framed as a net benefit (artists won't starve, no CP etc) it can really hinder the companies like stability

But we'll see, lets just hope for the best- I'm sure we're all on the same page here, just with a different take on the issue haha

[little plaza]

Honestly i don't think the general public cares that much about artists (i say this as a professional artist) 💀

they do care about CP but ... i think the popularity of stuff like Lensa and Midjourney speaks to how well the general public will receive AI as something positive

[brazen fox]

Well yes that is true, but people are easy to influence and especially with the prevalence of mad artists on platforms like tiktok it's a relative breeze to mobilize an army of outraged people who don't know why they're mad, just the fact that they're mad and that people are being taken advantage of- hear something like that enough, you'll start to see it as a fact

Which is why I think educating at some level is so important, show why the arguments are bs, show why it is bad faith

[little plaza]

Oh yeah, i think there is currently a moral panic happening nd there will continue be one, but i think (hope i guess) that the positives for people will eventually outweigh the negatives and they'll not want to give up the things these tools allow them to do. I pretty much agree with you on everything i just am slightly more optimistic i guess hah.

Peoples attention shifts very quickly from manufactured outrage when something new pops up to be mad at, too.

But more popular influencers with big social media followings educating people in a measured/rational way will be very helpful

I think a lot of people once they start playing with the AI apps will be actively looking for reasons to justify it for themselves. (not that they need a reason, but they will want to feel like it's ok)

[sand lily]

is there any sort of gpu benchmarking test using SD that any one has seen?

[brazen fox]

I hope so- we will see how all of this ends

At some point the anger will have to die down as you said, I think, if we will make it that far

[potent spire]

I really dont know if people here and on other servers and platform intentionally or unintentionally mislead with their messages about AI and artists who protest and everything around that topic

I assume some do this intentionally because of propaganda behind it and others do it because they dont know it better

[broken pollen]

I went from generating images at 0.2it/s to 2it/s on my pc but apparently I am still 10x slower than Nvidia.

[steady arrow]

it apparently has a rest api as of recently, https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/API

[raven edge]

Im using Vodlys repo and now all of a sudden im getting black images using anything

[gaunt oak]

Anyone here know if a 3070 would be any good for SD?

[still remnant]

I'm running on a Quadro P1000 (4GB VRAM) and it's usable for generating images, but not for training, if that helps you.

[tepid spindle]

good to be back here

[bleak matrix]

I have a 3070 and it works great

I do plan on upgrading, though

[fervent thunder]

upgrading from a 3070? :O
Isn't that a little early? :3

[bleak matrix]

For training, no lol

For the things I plan on doing, no

[brazen fox]

I don't mind an early christmas gift

[steep wing]

Some artists I'm working on a game with want to understand AI art better, as they currently lean towards the other side of the argument. Does anyone have some good reading material that explains stable diffusion in a non biased way?

[cursive gate]

Has anyone found any great tools for cropping images to use for training?

[steep wing]

https://www.birme.net/?target_width=768&target_height=768&rename=3image-xxx&rename_start=831

[cursive gate]

You are a legend:)

@steep wing if I drop 20 images of a person and want to just capture shoulders up and resize at 1024x1024 will it do this?

Think we need something more powerful that can use image recognition to detect what we are looking for in the image and then apply that across a set of images then resize.

Does Photoshop have this capability?

[steep wing]

This uses auto focus and then you can drag the box to adjust before saving the resized and cropped images, and yes to your resizing question

[brazen fox]

https://www.reddit.com/r/StableDiffusion/comments/zp2zq6/ive_made_an_eli5_in_case_youre_tired_of_arguing/
https://www.reddit.com/r/StableDiffusion/comments/zbi8zl/my_attempt_to_explain_how_stable_diffusion_works/

I don't think this is exactly what you mean, but maybe one of the two can be helpful as a means of showing how ridiculous some arguments might be just by explaining the tech

[bleak matrix]

I'm working on a friendly guide. But it is a work in progress.

I will be linking to the scientific papers

For those interested in reading

You can also watch the recent Stream

[fervent thunder]

friendly guide? like a normal one, but less swearing? ;P

[bleak matrix]

Frankly, it's probably one of the easiest ways to learn

[little plaza]

Computerphile on youtube has a couple of good videos .. not reading material but, seems pretty unbiased to me.

[bleak matrix]

One that's easy to understand/easy to translate (hopefully)

[steep wing]

I'll check it out, thanks!

[fervent thunder]

that's what they all say before meeting me!

[steep wing]

Thank you I will also give these a look

[steep wing]

The one from Stability?

[bleak matrix]

Yes

I have a transcript of it

And link here:

https://docs.google.com/document/d/1FgGOpTiHU7Lythj-1W2wGootrh2AkL4vFv_IqcFtXRA/edit?usp=drivesdk

But really, you can actually hear from the people working directly on the technology

Learn about it

And what they are doing with it

Because it's far more than just art

[rough spire]

Hallo friends. I wish all of you happy new year. So i have a question because SD. I test the Programm but i dont find a Solution, how can i do beatiful picture ?

Tutorial on YouTube was good.

[bleak matrix]

Hello! Happy New Year yourself!

https://docs.google.com/document/d/1aHJ9RBt_vlCwJQBVUUsb7VghKB-wynv7WGTxm9ozL1k/edit?usp=drivesdk

This guide might help you but you can check out #📝｜prompting-help

As well as #1045349359044280360

For ideas 🙂

[brazen fox]

I was very confused as I was expecting an actual mint

But NFTs are still living I guess

[rough spire]

Thank you so much Guys ✌️✌️✌️

[native sundial]

Happy New Year all

[full lark]

Happy new year!! 🤗 😊 🥺

[hardy sluice]

All the best 💗

[brazen fox]

Happy new year!

[potent spire]

Happy new year everyone ^^

[thorny sorrel]

guys, how does token count work?

[north stirrup]

It’s basically how many words you can use / are using in your prompt; how much deeper do you want to go?

[thorny sorrel]

im thinking about weight, like how many tokens would (red eyes):1 use?

[brazen fox]

the weight shouldn't really change it much, it's just the amount of words/tokens that gets counted

[thorny sorrel]

is that syntax correct though?

(what i want):weight?
and in this case, would the token count be "3" because there are three words?

[brazen fox]

you can just select what you want to weigh and ctrl-up or ctrl-down, but if you want to type it it is (red eyes:1)

[brazen fox]

well it's not exactly 1:1 with the amount of words, commas seem to count sometimes as well I believe (not sure though but I think I noticed it some time)

[thorny sorrel]

not using the webUI, so i dunno if that works

[thorny sorrel]

it seems that my parenthesis are also being counted for some reason

[brazen fox]

oh in that case I'm talking out of my ass, I possess only webui knowledge

[thorny sorrel]

pretty sure the webUI is based on what im using, so dont think its wasted info tbh

cant find a good colab that works with Automatic's UI :/

[brazen fox]

Might be, might not be, but it's hard to know for sure without knowing what it is you're using- and even then I'd have to google

[thorny sorrel]

i can post my code, if u're willing to look through it

its fairly organized

[brazen fox]

oh no I want to help, don't get me wrong, but concerning technical stuff I'm quite literally a caveman

might be more of a #🤝｜tech-support kind of thing

[tender saddle]

Noob here

does anyone know how to use dreambooth with automoatic1111 on google collab. I have 2.1 installed and I went to extensions in downloaded dream booth and restarted the UI and I don’t see any change

[thorny sorrel]

will ask them over there, ty for help though

[tender saddle]

Has anyone used dreambooth on collab with automatic1111

[thorny sorrel]

have you been able to get good results with Automatic on colab in general?

[wise stratus]

Happy 2023 all, may the brrr be with you

[astral goblet]

Happy New Years Emad. It's still 430 here but glad to see you're bringing it in!

[fervent thunder]

Easiest way to use SD is through runpod by renting a gpu 20 cents an hour way faster too

[jolly fog]

hi guys i hope u can help me

i can't use deforum on 9:16 aspect ratio
(768x432)
any fix ?
on automatic111

[robust coral]

A Happy Stable New Year To You!

[remote zephyr]

how can i wipe meta data?

[cold warren]

Happy 2023 may the rizz be with you

[strong jacinth]

HAHAHHA

[astral goblet]

ooooo those cuDNN and xformers manual updates helped my 4080 on the automatic1111 so much. I was coming from a vega 64 so it was still very much faster than before, so i felt something was off but wasn't sure

[wild steppe]

Happy New Year everyone!

[main sparrow]

Which fix did you use? was thinking about trying it

there was one on reddit that didn't look too difficult

[astral goblet]

copied the cudnn newest dll's into the venv directory and then did a pip command to install the newest xformers

https://www.reddit.com/r/StableDiffusion/comments/y71q5k/4090_cudnn_performancespeed_fix_automatic1111/ this is what i followed

[main sparrow]

yeah, I think that's the one I saw earlier. might try it in the morning. something seemed off with the 4090 in SD. will give it a shot

[astral goblet]

for my 4080 it was a boost and a half

[desert whale]

Anyone know of a dreambooth repo that works with SD2.1?

[velvet garnet]

hey guys what is "clip Guidance" button i dont see anything on the website?

[peak orchid]

Oh fuck, someone compromised Pytorch's nightly build

https://pytorch.org/blog/compromised-nightly-dependency/

[slate crescent]

yup, on linux

[astral goblet]

thats crazy. of course it happens on the holiday. just like diehard

[versed trench]

https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast-DreamBooth.ipynb

[desert whale]

ty

[glad jolt]

2022 was probably the biggest year ever for ai thus far. Hopefully 2023 can continue this momentum !

[wide juniper]

How does embedding work with non-faces? Can I embed a particular costume? What about? What about ice skates? It won’t turn my characters’ faces into ice skates, right?

[still remnant]

if I had a good enough rig for training, I'd love to make an embed based on the Violet Evergarden anime. Every scene in that show is gorgeous!

[lilac reef]

https://discord.com/channels/1002292111942635562/1058985063661969498

[outer saddle]

question will it also have some similar to ChatGPT?

[unkempt hatch]

If I want to train facial expressions, does it make sense to train the exact same images but with crops?

closeup of a man's laughing face [cropped]
a man laughing [uncropped]

closeup of a woman's laughing face[cropped]
a woman standing on a boat laughing[uncropped]

[surreal frigate]

Who is here the master on model training from a person face. i kind of did all the steps followed ustube instructions, but they dont have LOARA in them, did a check point of my wifes face and she looks like nothing when i generate the pix.

or if you know a channel/person who i can ask about it that would be amazing. i'm just new and still trying to understand this whole thing lol i'm not gonna bore you 😄 just need some directions in a right direction.

[proper osprey]

hello
does anyone have a tip which negative prompt is best to use to avoid 2 people or objects in the image?

[dry trellis]

Especially August to December

[rain basin]

So someone released protogen ckpt based on 1.5 but I can’t get it to work after upgrading automatic111 with yaml files for 2.1 768 . Anyone still use 1.5 models after upgrading ?

[wide juniper]

Help me understand this: what’s the stable diffusion version have to do with anything? People say 2.0 is worse because celebrities are no longer recognized, but doesn’t all this have to do with the models, and not the stable diffusion version?

[cold siren]

models AND training data

in 1.5 you can add magic words "greg rutkowski" to your prompt and everything will look better

[forest apex]

and like everything else, hehe

[wide juniper]

So, is Anything v3 for 1.5 or 2.0?

Is there a place where I can download embedded models that are trained for specific objects/celebrities?

Instead of training my own

[unkempt hatch]

just to clarify, it's not beneficial, right? I should not check that box?

[cold siren]

not "for", it is based on 1.x

[wide juniper]

I see, so it won't for for 2.x then

I want to train SD so I can apply different costumes and shoes for my characters. What's the best approach to do that? Dreambooth or textual inversion?

[rain basin]

ok figured it out, just don't need a yaml for non 2.0 models

[twilit yarrow]

Hi everyone 🙂

Could you suggest me how can I make cartoon portraits of pets? There are freelancers on Fiverr who do this, not sure they are using AI or hand draw it. Appreciate for any input.

and I'm completely new to AI btw

[cold siren]

use anime themed model? 🤷

also, if on sd 2.x use "furry art" in the prompt 😐 not even joking, this "ai" apparently thinks furry animals and furries are the same thing

[gaunt oak]

I take it AMD cards are a no go? Like a 6800xt?

[forest apex]

can anyone recommend a free wiki (with markdown/source access) that I can catalog all of the stuff I'm looking into to use SD?

[bleak matrix]

You can use AMD.

most cards that support ROCm 5.0, i.e. their 6xxxx line

Are supported

[bleak matrix]

A wiki? No. Guides, yes.

[gaunt oak]

Do you think it would come close in terms of performance to the 3070, or is that just farfetched?

[bleak matrix]

You can prompt them here in the server. If you want to learn how to use the bot, you can check out my wip new user guide here: https://docs.google.com/document/d/1aHJ9RBt_vlCwJQBVUUsb7VghKB-wynv7WGTxm9ozL1k/edit?usp=drivesdk

[forest apex]

I mean, I'm looking at fandom.com right now, I tried slimwiki, just somewhere other than a bunch of text files to save various settings and workflows

[bleak matrix]

Or check out #1019350374487961752

And #📝｜prompting-help

For help

[bleak matrix]

Unfortunately, I have not personally had one, so I cannot vouch for anything. Someone in #1011228477954998273 might know, but honestly, it comes down to specs. While SD does require quite a bit (hence I'll be upgrading for training, as I mentioned previously), msny requirements for running it have been lowered, too. Since the 6800xt has 16GB of VRAM? I think? I mean, that's getting up there, and you can train with less, depending on settings, etc.

I can't speak for the future because I just don't know, but I'd do as much research as possible and consider what's best for your budget, etc.

[bleak matrix]

I save my stuff in my documents:

https://atypicalconsortium.carrd.co/

Be warned--they're like textbooks.

[sturdy bridge]

Tree

[gaunt oak]

I see, thank you. I'm not familiar with any AI tech at the moment. I had been led to believe Nvidia was the only way to go, but seeing as I can only afford a maximum of a 3070, I thought the VRAM might me limiting due to the nature of this all

[forest apex]

@bleak matrix well that looks pretty cool, thanks

[bleak matrix]

T r e e

[bleak matrix]

VRAM is certainly a big component but it all depends on what you want to do.

Yes, at base, SD requires a LOT of VRAM.

(Although they continue to work on lowering that, as does the community.)

The lowest rn is 4GB of VRAM

But that's with 1.5, not 2, I believe

Training takes much, much more

But you CAN use Colab and other services for this

So even if you don't have a super massive computer, you can still run it. It's just more difficult

Cloud-based computing really helps mediate for those that can't afford it

[gaunt oak]

Yeah, the 6800xt has far more VRAM than the 3070 which is why I'm kind of puzzled. It seems like it's rather complicated to get it all working with AMD, and a video I saw online of some dude benching cards, everything AMD did super badly in SD

[bleak matrix]

And we are continuing to lower costs, etc

[gaunt oak]

Right, I mean there's midjourney and all that as well. I'm kinda just rooting for SD

[bleak matrix]

Like I said, 2.0, 2.1, anything in the outlined series should work. (I'm not tech support, though I try sometimes lol.)

I haven't tried it, though, so, like I said, I'd ask someone who has

[bleak matrix]

Np!

[gaunt oak]

I'm trying to yeah, pretty difficult to get answers 😅

[lyric lance]

nvidia teslas are absolute beasts for SD tho. very cheap too

[forest apex]

@lyric lance how beast? comparable to what?

[bleak matrix]

Your CUDA cores are a factor, foo, of course.

[lyric lance]

compared to general purpose gpus cost/performance wise

[bleak matrix]

They're used a lot tbh for training

[forest apex]

I've heard they use a lot less power, I'm wondering iterations/second, I get around 6/s on a 3060. memory is the problem though lol

[lyric lance]

they are designed to be power efficient since they are used in thousands in datacenters.

[forest apex]

according to a comment on https://www.reddit.com/r/sdforall/comments/zy6hcq/the_cheap_option_to_big_vram_upgrade/ they are fairly slow, buut if it were just for training models that might not be so bad

[lyric lance]

k80 is extremely old. Also a k80 is not full 24gb of vram it's split into two 12gbs. something like an m40 or m60 would be better and probably the same price too

i mean not EXTREMELY old.. just old i guess

i think m40's are about 150$ now, that's like nothing for a 24gb vram at 250w.

performance is probably comparable to a 1070

[bleak matrix]

Colab+ uses the Tesla T4, I think, most of the time

[lyric lance]

T4 are very nice low power at 75w, tiny little cute thing

[forest apex]

sharp price increase from m40 to m60

[bleak matrix]

It's just kind of finding that nice price point

Plus, you want to have the right kind of space, too

Like, do you want a PC that you use all the time, or is it just for work/SD?

[lyric lance]

yea true, if someone doesn't have a server chassis and put a tesla in there you're gonna be in a world of hurt 😂

[forest apex]

all 3 hehe

[lyric lance]

fun fact they use and EPS power connector rather than a standard 8pin pcie power. for who knows what reason nvidia thought to use a different power connection i have no idea

oh and both connectors fit so if someone doesn't know they're basically shorting the 12v power to ground and causing a fire 🙃

[vagrant blade]

Do any of you know if anything is being done to generate proper hands?

[bleak matrix]

Hahahaha yeaaaaaa

Exactly

[bleak matrix]

You need a LOT of space. The amount of models I have made. Omg. The space. Having to pick and choose is a bad, yet good thing.

Not to mention cooling. While we are more efficient than ever, cooling is also very important to consider, because, unless you plan on being a toaster lol

[vagrant blade]

You solve house heating, though.

[bleak matrix]

Or your own heater, which, I do tbh

[bleak matrix]

Nvidia. Changing things for...one reason or another!

[bleak matrix]

Also, a lot has been done to generate better hands and continual improvements and releases are being made.