#💬｜general-chat

1 messages · Page 13 of 1

bleak matrix
#

Channels

vagrant blade
#

Doesn't look too good...

lyric lance
#

i mean you gotta give it some credit man, still draws hands better than i do 😂

vagrant blade
lyric lance
#

yea usually it works better if i put some stuff into the negative prompt

bleak matrix
#

Keep in mind that less people may do better

lyric lance
#

generally i look at it this way. AI first struggled making something humans could recognize, then AI succeeds but struggles making faces, then AI succeeds but struggles making hands. and here we are

bleak matrix
#

There are some beautiful hands it can make

lyric lance
#

oh yea, and i have a hunch that Ai is gonna go suuuper big this year especially with text based stuff. like KoboldAI etc, where you literally just ask it to do you job for you

#

🙃 That's why I'm buying out a bunch of V100's and P40's since they'll go up in price 100%

#

and A2's since they're just so small and adorable

bleak matrix
#

I really do think so. I also think ppl don't realize how much AI is used already lol

#

I think the more apps there are, the more ppl use it, the more it will explode

#

Which is great

#

The more people are involved, the better

#

SD really turned things on its head

lyric lance
#

oh yea 100%. i kinda feel sorry for some software devs tho because a lot of them might become completely replaceable

#

or many other jobs in fact might become a bit obsolete

bleak matrix
#

I think that we are still going to be needing plenty of people

#

Technology always shifts

#

As it does, new kinds of jobs open up

#

I think, and this is just my personal opinion

lyric lance
#

personal opinion or not yea you're right in thinking so i agree

bleak matrix
#

You can never take out the human component of people that people need

#

There will always be niche markets

#

Just like people will want to support local communities

#

There are still plenty of countries

#

And people that need help

#

And our help as a community

#

Talent and skill is never wasted

#

The jobs may change

#

People may or may not change

#

But ultimately, it comes down to choosing what you do with your life

#

And if you want or need more jobs? Create them

#

Ask the AI to make up ideas if you can't think of 'em, hahahaha 😂

lyric lance
#

no matter what happens I'm save in a warm datacenter hall 😛 someone has to be there to fix up them servers or else karen won't be able to watch her netflix hehehehe 😂

lyric lance
bleak matrix
#

We all need and rely on people to keep the backbone of the net going

#

All the engineers, the people doing hard work, the farmers

#

The basic needs of life--everyone still needs 'em at the end of the day

lyric lance
#

yea most people if not 99.99% of people haven't even set a foot or seen what is going on in the background, datacenter technicians and engineers are like on the very end of the chain supporting everything but we don't get recognition really

#

not that i need any recognition lol, it's just funny how i try to explain to friends or family what I'm doing and I'm like well.... um.... uhhh... how do i explain lol 😂

bleak matrix
#

ELI5

#

It's hard to explain things to non-tech people

lyric lance
#

hahah yea i usually don't bother i just tell them "just internet n' shit" :)

bleak matrix
#

Hahahahahaha! Those that know already know, and those that don't...do some research?

#

I think dealing with the public/customer service/tech support is probably one of those jobs where it will try your brain in ways you didn't think possible.

lyric lance
#

oh oh oh hahahha get this, i wish i could EXPLAIN the feeling of when you're standing behind a full row of 48U racks at max power just the hot exhaust going through the hot isle containment and the SCREAM of the fans just sends chills down your spine of the power

bleak matrix
#

My dear mother. When I went to college, aeons ago. She put some of my computer parts in a corndog box....it got put in a freezer and rescued five hours later.

lyric lance
bleak matrix
bleak matrix
#

Never trust a non-techie with computer parts. That was my lesson

lyric lance
# bleak matrix Whiiiiiiirrrrrr!

Oh god it's insane I tell you, it's even better when you purge the whole pod and go underneath the floor tiles and you get to feel the wind in your hair 😂

lyric lance
bleak matrix
#

I've been near server racks, though probably not near as many as you.

#

I don't trust myself LOL

lyric lance
#

one time had a guy pull the wrong circuit breaker 😬 it was NOT good

bleak matrix
#

Oh no!!!!

#

There's only so many redundancies

lyric lance
#

yea I KNOW you'd think that with so many redundancies it would be fine but for some DUMB reason they don't connect the BBU's to the TOR switch so you only have power going into the hosts themselves

#

so you had a TOR switch powerdown basically so what is even the point of that BBU?!?!

bleak matrix
#

You might as well just say someone toppled over the racks

bleak matrix
#

That's my question

lyric lance
#

it's designed by someone in the office that literally had no idea what they're doing probably

bleak matrix
#

P r o b a b l y

#

🤣

lyric lance
#

yea it's a mess like you'd think everything is designed to work perfect... nope 😛

bleak matrix
#

Why is it someone who doesn't know what they're doing makes decisions? Hahahaha

#

Well

#

We'd like that to be the case

#

But that's not life

lyric lance
#

OH MY GOD tell me about it.... management doesn't have a CLUE most of the time of what I'm even doing

#

and yet they get the big money

bleak matrix
#

Well

#

That's why we got SD, yo!

#

So we are COLLECTIVELY

#

Brighter

lyric lance
#

yes :)

bleak matrix
#

Cuz

#

My brain is 🤣

#

Off the rails

#

But there are a lot of amazing people in this community

#

Talented, well-educated and passionate people

#

And that's what I love about it

lyric lance
#

hey look I'm a bit "off" here and there a bit so i get ya 🫠

bleak matrix
#

My brain is made of melted ice cream

#

laughs

#

The stream was a really great opportunity to show how people are working together, imho

#

For the greater good

#

As a community

#

(As opposed to people who aren't genuinely invested, etc)

lyric lance
#

oh i didn't know there was some sort of stream that sounds interesting

bleak matrix
#

Frankly, I think AI is going to make people smarter

#

Yeah!

#

I wrote down highlights/transcript

lyric lance
bleak matrix
#

And then

#

There's a tranacript someone gave me, too, also copy/pasted, but the link is

#

Also there

lyric lance
#

ah cheers for that

bleak matrix
#

It's a GREAT stream

#

Very technical

#

But a lot of GREAT, passionate people who are working on this technology and it's really enjoyable tbh

bleak matrix
lyric lance
#

I'd be really nice to be more involved in that but i have like so so little time but It's nice to see people working at something as a collective for no profit it's a very heartwarming community

#

generally not used to that personally, me and my friends would bite and fight each other 24/7 constantly 😂 in a playful way of course, i got some really bad habits from that

bleak matrix
#

Oh, for sure

#

You know, everything moves so fast, and it can be so hard to keep up

#

It's so easy to get burnt out

#

I myself don't keep up with everything

#

Just because it's too much for one person

lyric lance
#

i remember back when i was OBSESSED.. and i mean obsessed.. with linux and I'd be in these chats and getting into arguments all the time like literally drawing blood with people over dumb stuff it was great 😛 ............ but i got better....

#

have to have some peaceful time now ay

bleak matrix
#

And this community is special, I think

#

Very creative

#

Linux! It's been a long time since I used it

#

I used to dual boot but

#

Eh

lyric lance
#

creative it is true. but actually what do you think about people like in the art community

#

they're like oh Ai is strealing stuff

#

that's just like....... lol

#

i think it comes down to the fact that they just don't understand how it operates

bleak matrix
#

But this is not the same kind of feeling

lyric lance
#

no the same feeling? do tell

bleak matrix
#

As it's blended with a wide variety of computer science, engineering, and various types of art--writing, music, drawing--but also has far reaching hands into applications like education and therapy

#

Medical application, etc

#

People from all walks of life

#

But people who have often been studying these things for years, or their whole lives

#

Who are putting their knowledge from their careers into AI, or switching their careers to AI

lyric lance
#

oh yea i can imagine. i used to paint a lot with like oil pains, but kinda lost interest in it over time. or maybe just didn't have time for it or something. I remember chatting with a few people who were like really engaged in art and even in college way back I think it takes a special type of character to pursue this career

#

same as any career to be honest

bleak matrix
#

I'm an artist, both traditional and digital, and game developer, but I am disabled

#

A lot of ppl are also here to help disabled people be able to create, help with chronic conditions, etc, too

#

I try to help people understand art and technology to make it easier to understand and use SD

#

Artists have worked hard. I went to school, too

#

I've been on DA for 20? Years now

#

I think

lyric lance
lyric lance
bleak matrix
#

I chose game development (Besides loving games) because I love writing, I am a pianist, and an artist. I do some programming. (Done web development, but mostly old school.)

#

I couldn't pick because I can do all of the things and want to do all of the things.

#

My body may not be able to keep up, and my mind is on the fritz most of the time

#

But

lyric lance
bleak matrix
#

I've spent my life studying

bleak matrix
lyric lance
#

no problem friend :) it doesn't cost me anything to be nice

bleak matrix
#

Because I've dedicated my time to learning, I figure that, at the very least, I can do something good with the knowledge I have.

#

This is what is so great about SD

#

Because no matter where you are in life

#

You can bring that knowledge to the table

#

Maybe you're not an artist, but you're a farmer

#

Everyone absorbs information a the time

#

Art, in a lot of ways, is simply an interpretation of that information

#

That knowledge about being a farmer can be translated to SD

#

You know about manufacturing parts, what things look at sunrise, how animals act, etc

frank moon
#

The rendered image is not remaining on screen on Stable Diffusion. This happening to anyone else?

bleak matrix
#

Key words people may never think of you may know, which gives you a unique perspective

lyric lance
#

i hold a lot of respect for people who continue to learn, when i was out of school i YEETED the hell out of there and was like nope let me be on my own terms and just get to work on stuff and get wisdom from people more skilled than i

bleak matrix
#

And you can also take photos, learn how to make datasets, and have access to things others may not, etc

bleak matrix
#

I. Lol

#

I taught myself HS

sleek osprey
#

Are there any sneak peaks or anything of someone using distilled sd

leaden mural
leaden mural
bleak matrix
#

(Not that old)

#

Let's just say around 40)

bleak matrix
#

I'm surprised

#

But also pleasantly so

#

That a lot of people within these spaces seem to be multi-talented

#

With multi-directional skills

#

Which is great, because, regardless of AI, that means that there's always something to learn/excel at

wide juniper
#

Is there a place where I can download textual inversion embeddings for specific known people so that I don’t have to train everyone by myself?

unkempt hatch
#

I wonder if I could train a concept using 3 identical images with only 1 change, and appropriately captioned, e.g.

Man sitting on a chair with green eyes Man sitting on a chair with brown eyes Man sitting on a chair with blue eyes

would it learn what the eye color means? Or would it overtrain the chair?

bleak matrix
#

Depends on your images

undone ibex
#

hey guys

glass grotto
#

guys is S.D 2.1 is better than S.D 1.4 ?

dawn yarrow
#

shoutout to commodore 128

old tinsel
#

is there name for "anti variant?

bleak matrix
old tinsel
#

as in, thing that does something that is not expected, as a variable?

storm osprey
#

Hello is there a place where I can learn how to use the danbooru tags?

#

I am new to this and I want to learn

cerulean ether
#

we can't add this bot to our own server?

#

@tight quartz

tight quartz
#

@cerulean ether You can add your own bot to your own server - but not take this bot and run it in your server.

cerulean ether
#

please guide me through

tight quartz
cerulean ether
tight quartz
#

I meant building bots inside of your own discord. You'd be communicating across in such that your own API key is used to reference the engine that will allow sending and receiving of prompts and images.

wide juniper
#

How can I connect my laptop’s automatic1111 webui to my PC with a decent vram?

gleaming cosmos
#

@wide juniper You can run the webui on your powerful PC then use port forwarding to connect to it over your vlan. By default it only listens to localhost, but you can setup a tunnel

#

What do people use for photorealistic images of people? I am found the dreamlike-photoreal does very good but mangles the anatomy a bit. I found the moDi model to get the anatomy just right but very cartoonish. Mixing them 50/50 is giving me very good photorealistic human results. Does anyone have a model or model combo that they think does well for this?

lilac reef
gleaming cosmos
#

ahh even easier.

silk coral
#

Is there a limit on creating?

north stirrup
dapper island
#

is anyone having any trouble using stable diffusion atm ?

north stirrup
dapper island
#

Automatic1111

#

but after the update i cant generate any images

north stirrup
#

OK, what does the log say?

silk coral
dapper island
#

@north stirrup send it to you prvt

north stirrup
north stirrup
dapper island
#

File "C:\Users\Name\Documents\AIArt\stable-diffusion-webui\venv\lib\site-packages\gradio\routes.py", line 284, in run_predict
output = await app.blocks.process_api(
File "C:\Users\Name\Documents\AIArt\stable-diffusion-webui\venv\lib\site-packages\gradio\blocks.py", line 983, in process_api
data = self.postprocess_data(fn_index, result["prediction"], state)
File "C:\Users\Name\Documents\AIArt\stable-diffusion-webui\venv\lib\site-packages\gradio\blocks.py", line 930, in postprocess_data
prediction_value = block.postprocess(prediction_value)
File "C:\Users\Name\Documents\AIArt\stable-diffusion-webui\venv\lib\site-packages\gradio\components.py", line 3308, in postprocess
file = processing_utils.save_pil_to_file(img, dir=self.temp_dir)
File "C:\Users\Name\Documents\AIArt\stable-diffusion-webui\modules\ui_tempdir.py", line 18, in save_pil_to_file
shared.demo.temp_file_sets[0] = shared.demo.temp_file_sets[0] | {os.path.abspath(already_saved_as)}
AttributeError: 'Blocks' object has no attribute 'temp_file_sets'

north stirrup
dapper island
#

thanks

wide juniper
#

Is more VRAM going to make things faster?

astral goblet
astral goblet
wide juniper
#

Well I’m looking at rtx3090 not ti. That should be the cheapest 24 GB card right? Does this mean it probably offers the most value right now?

astral goblet
#

no. its still very much a premium product and is priced as a luxery item not a value item. i've seen 3090 at higher price tags than 3090ti's, only because they were one of the only 24gb cards available

#

might've been an oc edition of a 3090 too

wide juniper
#

What’s oc?

astral goblet
#

overclocked. some manufacturers do it at the factory

lilac reef
#

Yep, in AI/ML vram is the number one thing and the more you can get the better it is even if it means sacrificing some speed/cuda cores.

wide juniper
#

What gpu has good value for sd then?

lilac reef
#

You tell me because I think they are all 50% over priced.

wide juniper
#

I see

astral goblet
#

"good value"

lilac reef
#

Nvidia, and AMD too, got that sweet crypto money and crypto mining is dead but they refuse to let it go back to normal.

astral goblet
#

I think they realize theres a ML revolution going on that they can bank on too

#

the day someone gets a library going that lets multiple gpu's share a memory pool for training, it'll be like mining all over

lilac reef
astral goblet
#

as is tradition

lilac reef
#

I watched a video and they let the 4090 have a feature (hardware wise) that you only see on A100s while they kneecapped everything else. Due to that one feature it is still, overall, 56% faster than the 3090 in AI/ML.

astral goblet
#

yeah usually one instruction set changes everything lol

lilac reef
#

If they had not kneecapped it well over 100% faster

astral goblet
#

yeah they still got the pro market products that come off the same dies. they give the pro end the best bins and then the ones that don't bin as well have features disabled until they test well enough. then they develop a product around that. It's not always nefarious. a lot of the time it's just how the dies bake out in the lithography machines

#

"not always"

lilac reef
#

I am due for an upgrade as my 1060 is near death (sad) but when a 4060 is looking to have 8-10gb of vram that is shit considering it is 3 gens later now so six years.

astral goblet
#

my vega 64 was something of a 1070 equivalent. i was feeling the belt on that one for the past year

forest apex
#

@lilac reef how is it dying?

lilac reef
#

besides the 500-650 price tag.

astral goblet
lilac reef
# forest apex <@315547788211388417> how is it dying?

Must be ram heat or something I can't measure but I had to redo the tim, and the pads but to try and train SD it throttles itself the longer it goes. Has to be ram related because the card stays cool but this gen it didn't have a way to measure ram temps.

lilac reef
astral goblet
#

prosumer is one of the most lucrative hardware markets haha

forest apex
#

@lilac reef tim?

astral goblet
#

it's why sony sells a pro line of the playstation hahaha

forest apex
#

looks at his 2nd hand, usually 64F running card

lilac reef
forest apex
#

ahh

astral goblet
#

shouldn't you not put insulation on a gpu?

silk coral
lilac reef
# forest apex ahh

Mine had all but dried out. The pads aren't perfect as this required 3mm pads but evga was no help, and everyone said 1.5 or 2mm bought both for 30 bucks and said to hell with it I am not buying again.

astral goblet
silk coral
astral goblet
#

or a macbook with one of their fancy neural chips

silk coral
astral goblet
#

look into automatic1111 webui. there's instructions on their githb for installation

silk coral
#

sure

astral goblet
silk coral
#

oh i thought you misstyped xd

astral goblet
#

yeah it gets tedious with all the initials haha

silk coral
#

i only have 128 mb vram on display 1

astral goblet
#

yeah your laptop gpu is a classic "gaming" laptop.. meaning it doesn't have a dedicated gpu. it has an integrated gpu on the cpu die and shares memory with the system

silk coral
#

so i need like a pc

astral goblet
#

or a laptop with a dGPU yeah

silk coral
#

damn thats unfortinute

astral goblet
#

or a mac that has a dedicated neural engine chip thingy

midnight crane
#

I'm not sure at this point, I'm still a bit confused about EMA weights, maybe try both and compare the results? I think that's what I will do.

unkempt hatch
midnight crane
unkempt hatch
true dawn
solemn hill
#

Is there some sort of extension which allows you to upload models from interface? Like, the uploader from Prolapse's colab, but within the UI itself?

vivid portal
#

Anyone good at img2img and know how to outpaint a tv around a picture?

#

Have tried so many times

#

I just don’t know how to do that 😅

compact prism
#

Hi! I would like to ask if you would be interested in taking part in our exhibition of AI outputs, on the topic "communication / dialogue". I would like as many nationalities as possible to participate in the exhibition. I currently have 25 author outputs ready. The exhibition will be in Zlín in the Czech Republic. Under my auspices and the Tomas Bata University in Zlín. The output can use any text to image model. Any aspect ratio. The exhibition will also have its own website. Each author will receive a catalog from the exhibition. If interested please let me know in DM! Thank you.

bleak adder
#

Can I run Stable Diffusion locally on my computer?

brazen fox
#

6GB or more of vram, although preferably more

bleak adder
#

Sounds like a lot for a laptop

#

I'll check later

still remnant
grave barn
#

hey guys i am new here

#

couldn't find a guide how do i use stable difffusion ai ,

#

?

still remnant
#

👋🏻

#

@grave barn what's your goal? Would you prefer command line or to use a graphical interface?

grave barn
#

command line

#

but what is a graphical interface?

still remnant
#

I use Automatic1111 so I have text boxes and sliders to enter things like positive and negative prompts, steps, sampling method, resolution, etc. on the command line, I'd be formatting a long string of text to do the same thing.

#

I guess the other question is do you want to run it locally, or like with google collab or another cloud service.

#

once you know your environment finding guides online will be a lot easier.

grave barn
#

i see

#

thanks alot

brisk bloom
#

Is there a way in A1111 to run img2img, send the output image back and run img2img again, like 10 times or something?

#

Im doing it manually now but maybe there is an automated method?

still remnant
#

Loopback script should do that

brisk bloom
#

so the script knows to select the last image I guess?

still remnant
#

yeah it keeps feeding the output into the next iteration

brisk bloom
#

Oh wow! I will try that. Thanks!

still remnant
#

only downside is I think it maxes out at 30 iterations

#

nvmind, the slider only goes to thirty, you can type any value you want.

brisk bloom
#

works perfectly. thanks again

fervent thunder
#

Hey is there a set of docs somewhere? Having trouble finding out how to do the aspect ratio

still remnant
fervent thunder
#

like midjourney has the --ar 2:3 for a 2 to three aspect ratio output

still remnant
#

I've been crunching it manually.

true dawn
#

manual all the way baby

still remnant
#

I made a spreadsheet.... I should convert it to a google sheet to share.

#

You enter the square pixel space, and it gives you resolution at different aspect ratios with similar pixel count

true dawn
#

(short side/ 2) x 3 = long side

fervent thunder
#

Thanks! Would you mind dropping a quick example for me of what that prompt would look like?

still remnant
#

for a 768 square, a 2:3 would be 627 x 940

#

at 512, it's 418x627

fervent thunder
#

Thanks, I checked out some prompts and figured it out! Appreciate the help 🙂

bleak wolf
#

Well

#

Caspian's Tank Diffusion is in the works

#

It will take me 98 hours of CUDA

#

But inshAllah it will be worth it

#

I will make sure to share it with the community inshAllah

#

So if you want to see the progress of the tank diffuser boi

#

Hmu

bleak adder
#

Is this AI capable of outputting images as detailed and specific as mid journey?

signal nest
bleak wolf
#

If you know how to use it

still remnant
winter pike
#

the thing about midjourney is that they add extra prompts without you knowing, you can even type "fksdhfkjdhsf" and it will give you a very detailed portrait

#

you need to learn how SD thinks and it will become a very useful tool

bleak adder
#

Where do you download it to run it locally?

#

I'm very dumb with programming

#

Midjourney creates very clean images. I haven't had the same success here yet but I just got here.

dawn yarrow
#

theres github repos

signal nest
# bleak adder I'm very dumb with programming

heh i can't remember off hand but yeah .. i'm using the "diffusers" python library. i've been able to make my own prompt randomizer front-end , assists for running img2img on batches of tiles.. so much fun. Hell in the past 2 months that i've paused from running it there's probably a whole load of new things to try out

bleak adder
#

Can you feed it images to base it's output off of like in Midjourney?

signal nest
signal nest
signal nest
bleak adder
#

I recently gave it a "gates of hell" prompt and it feels watered down

#

Midjourney I mean

#

Seemed less detailed than I expected too

#

And it does weird deepdream stuff like putting a door inside the door

signal nest
#

what i've been doing is stringing together lots of thematic and descriptive words, and spitting out lots of random permutations of those. 5 seconds per image on rtx3080 - its fascinating watching what it'll make

bleak adder
#

Are you using the cloud or are you doing that locally?

#

It seems I'll need to have a strong understanding of python to get this thing to work right

#

Maybe I could pay someone on Fiverr to build a UI for me

signal nest
#

yeah i'm sure someone will have done this by now

#

i must admit i suck at UI programming.. most people running locally are probably just using python - but there's no reason a local UI can't be done

bleak adder
#

Do you know of a tutorial on YouTube that may be able to at least walk me through how to do the python stuff myself?

signal nest
winter pike
latent flower
#

Just wanna inform the following, because like me, someone might be using this version cause of the latest fixes: PyTorch-nightly has been compromised. If you installed this version during December 25-30 UNINSTALL IMMEDIATELY and consider formatting your OS. Source: https://pytorch.org/blog/compromised-nightly-dependency/

signal nest
bleak wolf
bleak adder
#

We're like the old school computer game programmers my dudes

latent flower
bleak wolf
bleak adder
#

Can you get good outputs here in the dreambot channels or is that thing a bit nerfed?

latent flower
bleak wolf
signal nest
bleak adder
latent flower
bleak wolf
signal nest
bleak adder
#

Business ideas...

#

I have them.

#

Too many

bleak adder
#

Is it okay to poke around here for people who offer services?

#

I'm looking for someone who could help me build something

#

Something with stable diffusion

#

I need to get it to work and then I need to slap a UI onto it

bleak wolf
bleak adder
#
  1. Pretrained models for different but reliable outputs (e.g. poses, backgrounds or settings)
  2. Take images and then train on them
  3. Apply 1 to 2
  4. Give output (bonus: upscaling)
  5. Save the output to private individual profiles
  6. Other stuff: stuff that UI into a sleek mobile app
tender scroll
#

I prefer SD to MJ in so many ways but can’t deny there’s a crispness / resolution to mj images that has me envious. Is there anything on the horizon that will put SD over the top?

trim hemlock
#

Will we ever see an open source alternative to ChatGPT, and will it be possible to run it offline on a personal computer? Like what Stable Diffusion is to Midjourney and Dall-e? Arte there any projects like this in the making?

signal nest
still remnant
signal nest
#

i've just been reading a big explanation of why someone expects the gap will widen (between what can run locally and in the cloud) r.e. that kind of AI

trim hemlock
trim hemlock
signal nest
still remnant
#

At work I've pretty well got my group sold on paying for it when it switches over, we use it so much to start scripting commenting old scripts/programs, etc. lots of business use.

#

saves dozens of hours a week.

frail isle
#

some1 wanna change tha world ?

#

we could do a social network where people could say if they ok or not with decision that they gov take in theyre name.. respecting democracy for the first time ...

trim hemlock
drifting mauve
#

heeyeyyy, does anybody knows if i can add the bot to my own server?? Or I cant do that

inner nacelle
#

can sd v2+ be used in Automatic1111's webui?

winter pike
bleak adder
inner nacelle
#

what is the difference between ema and no ema by the way?

lilac reef
#

one is for training the other isn't.

lofty monolith
#

Hello everyone.

proper pollen
#

hello guys, if someone is good at drawing digitally and editing. please DM me, i need your help

swift snow
#

What are embeddings. Aren't they just checkpoints?

grizzled harness
floral zealot
#

hello guys somone can tell me the diference when (()) are used in one prompt?

leaden mural
# bleak matrix Linux! It's been a long time since I used it

Linux is getting better recently. A guy I know knew of a Linux distro that can basically run any windows game and apps + customizable and works pretty much like windows. Only problem is with anti cheat, but I just hope it would get fixed soon.

leaden mural
fervent thunder
#

Wat

leaden mural
leaden mural
# fervent thunder Wat

Sorry, this is what happens if you try to continue a conversation from the opposite side of the time zone.

fervent thunder
#

Lol its alright

proper pollen
wooden raven
#

Hi guys im new to SD, I was wondering if there are any resources on learning how to create our own anime model

bleak wolf
#

Search up:
Dreambooth
Textual Inversion
Hypernetworks

Etc.

What are your system specs?

wooden raven
floral zealot
#

sorry about my eng

bleak wolf
bleak wolf
floral zealot
bleak wolf
floral zealot
#

thx

meager nest
#

To use models that use 1.5, do' I need 1,5 in my models folder?

#

I just have standard 1.4

#

;w;

bleak wolf
bleak adder
#

anyone know the input for feeding it an image?

lilac reef
wooden raven
#

anyone know whats a safe place to dl f222 model?

marsh geyser
#

am i able to use stable ai to make fan art of characters from my favourite marvel and disney movies?

bleak wolf
mortal narwhal
#

Is there any stable diffusion model which is trained on large number of images

marsh geyser
#

isnt any ai art model trained on images?

unkempt hatch
#

this should maybe be an obvious answer but

higher LR has higher risk of forgetting cooncepts, right?

but the tradeoff isn't linear.... learning longer and a lower LR won't make it 'forget' the way training a short time at a high LR will?

for context, I'm using individual custom BLIP captions with detailed natural language and no class images (because there are no tokens or classes etc...)

bleak wolf
#

5.85 billion images, not enough?

mortal narwhal
bleak wolf
elder cradle
reef epoch
#

@wise stratus whens the next soontm tease 👀

uneven heath
bleak wolf
uneven heath
#

Ikrr SD has used a subset of it tho

#

not like whole 5bn images

bleak wolf
#

bruh

#

time to make 5 billion diffusion

uneven heath
#

we wont be able to run it 🤣

#

unless we rent

bleak wolf
#

nah

#

we can optimize it like fuck

#

by poruing millions of $ into its development

#

heck

uneven heath
#

damn

#

lol

bleak wolf
#

we can make an ai who would optimize the AI

uneven heath
#

i need to look into how it works behind the scenes

#

but 2.1 is very good with custom training tho

#

and embeds, loved it

velvet oasis
#

silly question, but If I'm doing some dreambooth, should I resize all the images to be the same like 256x256 or 512x512? and so I can learn for the future, how do you know?

tepid cobalt
#

in stable diffusion 2.1 they remove ability to copy artist style right? or im missing something here

signal nest
tepid cobalt
signal nest
# tepid cobalt too bad, in 2.1 i can't achieve any style beside photo realistic

I'm still interested in a model that is purely trained on photos to bypass these anti-AI arguments altogther - however it would be good to have both models around of course. One thing I'm wondering is if other filters could emulate hand-painted art styles (there's non-neural algorithms for that, i.e. procedurally trying to rebuild a photo with brushstrokes) - and then you could perhaps finetune (or train a model in the first place) on that

signal nest
tepid cobalt
brazen fox
signal nest
brazen fox
tepid cobalt
brazen fox
#

It's not adding something to the model, rather it guides the prompt where you want to go, using vectorized tokens- it's basically a prompt shortcut

brazen fox
signal nest
#

(thanks for th explanation . "nudge a prompt" .. sounds like the vector output from the text encoder, probably just added to whatever prompt you give, ok. maybe its possible to create that vector directly e.g. working back from images, i.e. to give more control than you'd get from text)

brazen fox
#

you don't need to download an entire new model to get the results you want this way

brazen fox
#

But training an embedding is pretty much what you said

brazen fox
signal nest
# brazen fox It's hard to explain not knowing exactly _how much_ you know about all of this h...

I couldn't implement S.D. from scratch or anything but I've got an overview in my head.. 3 components - (i)text->embedding vector (text encoder), (ii)VAE ( goes to and from a compressed version of image space to work in), (iii) the U-Net (does the main work of diffusion iterations, uses the embedding to guide working on an image in the latent space of the VAE). "custom model" would be a new version of the whole thing

brazen fox
#

I just woke up so I still have a bit of brain fog, but that sounds pretty correct

tepid cobalt
signal nest
#

i gather that going from image -> embedding for diffusiuno models is much harder than earlier generative models (VAE's and GANs) because there isn't a simple latent space from which the whole image is generated.. but there is supposedly as process that can actually do this.

brazen fox
signal nest
brazen fox
#

or train embeddings on their own style

near falcon
#

Any1 knows how do you refine a generated image. Like creating different poses of ur char with face and outift remaining same ??

tepid cobalt
signal nest
#

(and yeah this is of interest to me aswell, i do some handpainted stuff aswell and it would actually be pretty awesome to have a net that can replicate that , fill in from sketches and turn photos into it..)

brazen fox
#

The plan is for me to write a quick start guide on training embeds soon- that might come in handy for stuff like this

#

I just need to find the time 😄

tepid cobalt
signal nest
# tepid cobalt that sounds like a great option

i've definitely seen S.D (still using 1.4 here) do very interesting things on my fully shaded handdrawn image, i.e. re-interpreting the shapes with a sense of 3d but substituting elements based on a prompt. I've never actually tried to jjust make it a straight photoreal re-interpretation though.. but from what i see everyone else doing it seems like that should be possible)

brazen fox
#

If you really want to take the time you can make a composite of inpainted parts, then do a low denoise img2img pass to make it fit together

signal nest
brazen fox
#

So if you have a composite image with clear cutout parts etc, you can add a little noise and denoise it again- to make it all more coherent, while not changing everything too much

signal nest
#

class StableDiffusionImg2ImgPipeline(DiffusionPipeline): .. def call( self, prompt: Union[str, List[str]], image: Union[torch.FloatTensor, PIL.Image.Image] = None, strength: float = 0.8, num_inference_steps: Optional[int] = 50, guidance_scale: Optional[float] = 7.5, ... ): ... I thought it was that 'guidance_scale' parameter that controls how strongly to use the prompt - I also thought the caller to this pipeline would add the noise - i can't see a parameter here , but the example or UI probably does it for you. Now i'm guessing quintusdiast is using webui etc rather than python so there would indeed be a noise parameter somewhere)

brazen fox
#

Guidance scale is exactly that, but I meant that that wasn't what my earlier denoise img2img comment was about 🙂

signal nest
#

(i definitely remember seeing the effect of varying noise .. sorry my experiments are on another machine , i'll dig it up to see what i was doing ..)

brazen fox
#

I'm not a programmer or technologically gifted though, so my knowledge on that stuff is limited

cold siren
#

i saw a reddit post where op used negative CFG scale (for science!), but apparently (now?) AUTO1111 does not allow this 🤔 does this even make sense? is this be the same as swapping positive & negative prompts?

signal nest
#

(so what I did was started with the 'diffusers lib' img2img example - I think the actual noise addition you are talking about was implemented there . i made something to generate random prompts, and do tiling textures via feedback hacks, and run batches on a grid of tiles cut out from an input image . i think there's a better way to do repeating tiling tesxtures out of the box now . I use a mac mini M1 as my daily driver and was running StableDiffusion on a desktop PC which is off at the moment. I've got sidetracked doing actual art again lmao .. my intention is to use SD to upscale and vary it, but i'm hedging bets on the copyright/legal debates by focussing on art thats complete without it first..)

leaden mural
dry flower
#

Does it make a big difference what checkpoint i use in SD?

sage abyss
dry flower
#

Oh really? Is there anywhere i can get new ones?

elder cradle
brazen fox
signal nest
#

i see them conflating ai art and NFT/crypto people aswell. i hated NFT/crypto because I wanted GPUs for AI

brazen fox
# signal nest i see them conflating ai art and NFT/crypto people aswell. i hated NFT/crypto be...

I’ve been playing around with several generators, and so far none have mimicked my style in a way that can directly threaten my career, a fact that will almost certainly change as A.I. continues to improve. It’s undeniable; the A.I.s know me. Most have captured the outlines and signatures of my comics — black hair, bangs, striped T-shirts. To others, it may look like a drawing taking shape.

So all of this fearmongering in the title and the rest of the article has the conclusion that it's pretty much a childs' impression of her comic when you prompt her name. Yet she felt the need to say

The LAION data sets have also been found to include photos of extreme violence, medical records and nonconsensual pornography. There’s a chance that somewhere in there lurks a photo of you. There are some guardrails for the more well-known A.I. generators, such as limiting certain search terms, but that doesn’t change the fact that the data set is still rife with disturbing material, and that users can find ways around the term limitations.

Which is not at all related to her argument, and does not have anything to do with the image generation process- it just means they are available on the web somewhere that has been scraped

God I hate this haha

night stirrup
#

Hi, i have a question, how can i set an image size (like for a wallpaper) ?

bleak matrix
# leaden mural Wait, how are you disabled? Sorry for the question. I'm just curious.

Oh, many ways lol. I'm like a crypt. I have many rare autoimmune conditions (Lupus, Reynauds, Sjogrens, fibromylagia, etc), deal with paralysis of my limbs, loss of feeling, have weakness, carpal/cubital tunnel, gotta have my knee replaced, starbismus/lazy eye, almost legally blind in my right eye.....along with other rare conditions. I'm a circus with ADHD but I keep on going. I usually use a cane to walk.

main sparrow
rough spire
#

Hallo guys before 1 week chatgpt knows What stable diffusion is but now they dont what this is. Why?

bleak matrix
main sparrow
bleak matrix
#

The d r e a m

#

hasn't read this far

bleak wolf
bleak matrix
#

Also, you can do traditional styles with 2.1, etc

bleak matrix
bleak matrix
#

Can't say I always succeed 😂

#

But I try

bleak wolf
#

Could you help me with Dreambooth Lora? Im trying to make a model for the community

bleak matrix
#

It's 2023; the slate's clean. Right!?

bleak wolf
#

I have a question rather

bleak matrix
#

A s k away

bleak wolf
# bleak matrix A s k away

So i saved the parameteres right after my ckpt save(set it on like, 1k steps?), so when i continue training with the loaded parameters, does it continue from the image it left off or despite the steps, does it go back to the first image in my dataset?

bleak matrix
#

I'm not 100% sure, as I haven't had time to work with Lora. But I would imagine that it works the same as TI or DB locally.

bleak wolf
#

resumes at the image it left off?

bleak matrix
#

I would probably just go with no

bleak wolf
#

I see

bleak matrix
#

I'll double check and if I hear differently, I'll let you know

bleak wolf
#

from my dataset

bleak matrix
#

nods

#

I just haven't used it myself, so, unfortunately, I just can't say either way confirmed

bleak wolf
#

so either way

bleak matrix
#

Yup

#

Sorry. Was reading the math

#

Basically

#

Lora, from what I understand

jolly fog
#

anyone having an issue on google colab, that won't appear the public url

bleak wolf
bleak matrix
#

Guided training, or what you might call textual inversion/mini fine tuning 2.0

lavish current
#

The cloud model I was using is in error state 😩

bleak matrix
#

But who knows

lavish current
#

Is there another Finetuned Diffusion that's not the Anzorq one

bleak matrix
#

The Github does a pretty good job of explaining how it works

#

I will probably try it, at some point, but I'm currently busy making models and tutorials for 2.1

#

Esp for Dreambooth, etc

#

1.5 included

lavish current
#

Guess I can't generate anything for a while until this is fixed

bleak matrix
#

What exactly is the error?

#

Now I'm curious

lavish current
#

It says "Preparing Space" and never moves past that point

#

I'm on Chrome OS so I can't use a local model

#

Figures the moment I actually have time to make them this happens

bleak matrix
#

The owner will prolly need to fix it

forest apex
#

@pastel beacon that is a specific model, you'd have to look for something similar someone else has compiled, either as a checkpoint or embedding

main quail
#

so what is the most popular distro these days for local stable diffusion?

forest apex
#

@main quail I'd think Automatic 1111 followed by InvokeAI

main quail
#

ive been trying to get invoke working, automatic drives me nuts due to no spacebar support

rough spire
#

Hello guys. I have a question about SD. I have Nvidia GeForce 3060 Ti 16gb, but SD Told me. I have Not so much GB. Why?

clever cradle
#

SD is using what's called "Dedicated Video Memory" or "VRAM" for short

#

VRAM is also

A: a fraction of the total memory on the GPU
And B: Can be impossible to allocate manually depending on the graphics card

#

Nvidia is unfortunately one of those graphics card companies that doesn't let you change the memory allocation for VRAM

#

Believe me I have tried

#

I even went into my PCs BIOS and there was no setting for changing VRAM

brazen fox
rough spire
#

Oh yes Sorry.

#

nevertheless it should work

clever cradle
rough spire
#

I mean 8gb

clever cradle
#

Is 8 GB the total memory on the graphics card or is it the VRAM?

#

those two values are not interchangable

#

Speaking of VRAM, is there some way I can get Stable Diffusion UI to access the shared video memory?

brazen fox
#

8GB is 100% all of the memory on such a card- the virtual memory you might see in the task manager is 50% of the ram you have

#

has nothing to do with the actual chips on the gpu

rough spire
#

I send a Screen

#

Oh i cant

brazen fox
bleak matrix
hard basin
#

Hello. What are the default modules of stable diffusion 2.1. Is there a glossary or index of the modules in alphabetical order?

lavish current
#

Ayyy Finetuned finally loaded

crimson veldt
#

hmm

#

i just tested nijijourney and mid journey v4

lavish current
#

Paidjourney

crimson veldt
#

i have got an idea on weather how their ai works and its not what we all think it is

#

yeah paid shit

#

and the results are preety narrow

#

my idea is , its not really just a image gen model

#

its a tag builder ai

#

like give it an apple and it automatically adds high quality tags around it (just so the people dont complain about shit results)

#

its not better than stable diffusion but it is better at pleasing people however

#

well about nijijourney

#

still with narrow stuff man i want stable diffusion anime model

muted skiff
#

when I switch models in a1111, is there a way to autoload a VAE, by specifying it in model.yaml? I've been copy/pasting the VAE file with different names, it adds up

crimson veldt
#

i use colab notebooks so idk

#

i am here for a question

#

when are we getting a good text ai that is better than clip or blip

#

sd is already at its peak but the language models makes harder to generate good images and takes many tries

muted skiff
crimson veldt
#

imma be back after reading that shit

muted skiff
# crimson veldt oh thanks

I tried it (it's just an SD fork), and though it's rough around the edges it makes images SD/NAI/etc is simply not capable of, and far more accurate for the input tags

crimson veldt
#

or is this a model like anything v3?

muted skiff
#

not a model. it replaces the way text prompts work, to be far more accurate for the input

#

I mean it's a model too, but he forked SD for his model to work. Worth trying one evening, it doesnt take long to set up

crimson veldt
muted skiff
#

if you're familiar with danbooru tags, try it, you'll quickly see what I mean

crimson veldt
#

yeah ive been using anything v3 and other models like yohan

#

they require danbooru tags but are not "that" acurate

#

from the char anifusion showed it looks better at understanding tags and prompts

muted skiff
#

this one is accurate, but wont make flawless images like Anything3. (maybe it does at higher steps, I didnt try since I have a crappy GPU)

crimson veldt
#

i dont even have a gpu

#

i use colab lol

muted skiff
#

free or paid?

crimson veldt
#

free

muted skiff
#

how long does it take to make an image?

crimson veldt
#

but i just get 4-5 hours

muted skiff
#

say at 30 steps

#

4-5 hours a day is plenty

crimson veldt
#

euler a

muted skiff
#

nice

#

I'm just not gonna associate my real name and phone number with SD gens though. Nice try, Google.

crimson veldt
#

check out camenduru for colab

muted skiff
#

I'm still holding out for become Prime Minister and booting out that lamer Trudeau

crimson veldt
#

this guy created colabs for most models, you can just run it without writing any code

muted skiff
#

so it's notebooks with preloaded models?

crimson veldt
#

yep

muted skiff
#

nice. Have you uploaded your own model before?

crimson veldt
#

nope

#

its hard since idk about ai stuff and i am new

#

i am not familier with coding too

#

so if i get a model i wanna use i just contact camen and he upload a colab

leaden mural
muted skiff
#

ah, you dotn need to know coding, I just meant that if you download a model from some other website (because it's not available on camenduru), you can load it via Google Drive.

bleak matrix
crimson veldt
#

ah that

#

did it but in early version and things messed up later i found out that novel ai model needed specific settings

fervent thunder
crimson veldt
#

then i never tried because i found preloaded models

leaden mural
# bleak matrix ADHD is fun lol

well, when your attention is all over the place and dopamine is hard to come by for your brain, it tends to go into weird places, really weird places...

But hey, I tend to learn easily and handle well under pressure, so that's a plus. And then I tend to hyperfocus on things that I may or may not want, so there's that.

crimson veldt
#

all these things developed superfast

bleak matrix
#

I'm alive! S o m e h o w. (I had a rare blood disease that almost x'd me multiple times...strokes, etc.) That is what is most important--continuing to persist despite difficulties in life.

#

That is why I believe, or rather, want to continue to work with SD, because it can do a lot of good

#

I know there are many good people in the world and I believe, and can see, many good people working together behind this technology

crimson veldt
#

i too want to make a career in machine learning

bleak matrix
#

There are many good people in this community. Kind people

#

There's so much to learn, too.

crimson veldt
#

ai community is 90% good people unlike artist community

crimson veldt
bleak matrix
crimson veldt
#

ill start after my exams are over

bleak matrix
#

The Lupus does make it worse

#

I'm open about my medical stuff, to a degree, simply because I know there are other people who have struggled like me

crimson veldt
#

hey are you both programmers?

leaden mural
bleak matrix
#

And I want them to know they can have hope, and to not give up

bleak matrix
leaden mural
#

It sucks being in a third world country

bleak matrix
bleak matrix
leaden mural
crimson veldt
bleak matrix
#

Ive spent most of my life without it.

leaden mural
crimson veldt
bleak matrix
#

There are a lot of challenges with ADHD.

crimson veldt
#

but i am talking about human only art community

bleak matrix
#

You know, I didn't expect it, but calm.

#

It was super weird

#

Lol

crimson veldt
#

sam does art apparently lost all respect after misinforming thousands of people

bleak matrix
#

It's supposed to help with narcolepsy

leaden mural
bleak matrix
#

But did it help with mine? No

leaden mural
bleak matrix
#

It slows you down

#

Makes it easier to organize your thoughts

crimson veldt
#

hey whats maven role stands for?

bleak matrix
#

Over time, it helps those pattern of thoughts emerge, and make them easier to access

#

It's for moderation; I'm a new mod

#

I just help when I can

crimson veldt
#

any idea how big stability ai team is?

bleak matrix
#

Cuz health lol

leaden mural
bleak matrix
#

I'm sorry but I can only answer general questions

crimson veldt
#

nvm its around 250 people big

bleak matrix
#

It's helped improve my life

#

Ofc, everyone is going to be different

#

But it's certainly improved my memory, for sure

leaden mural
bleak matrix
#

So, what brought both of you into the AI community?

crimson veldt
#

dicover dall e mini for meme

#

discover midjourney

#

dicover some other ai

#

discover unstable diffuion through some blog and reach stable diffusion

#

do all this for ai por-

bleak matrix
#

Sounds like quite the journey!

#

(Of course, I'd love to hear anyone else's story, too)

crimson veldt
#

when i heard about stable diffusion it was able to produce shit but when i reached here there were already big things out

#

i reached here after the releasing of anything v3

#

kind of funny how i knew about stable diffusion models before SD itself

leaden mural
bleak matrix
#

Niiice

leaden mural
#

Found out about SD from "two minute papers", looked around, poked at the parameters in fast stable diffusion colab, and the rest is history.

crimson veldt
#

i am continuously using ai after that

bleak matrix
#

I love Two Minute Papers

#

Great channel

crimson veldt
#

yeah short and covers what we wanted to actually hear

leaden mural
crimson veldt
#

i discovered him after image gen ais

#

airtpreneur was the one for me who introduced stable diffusion stuff

bleak matrix
#

Sometimes I look at that bit there's always so much news on Discord at this point

#

And so many AI-related Discords

crimson veldt
#

thousands of ai discords

#

they all happened to exist after SD

#

now they are contributing in ai, great decision for taking this project open source

dawn yarrow
#

ah two minute papers love that yupyup

crimson veldt
#

finally i have found what i want to be after school ends

dawn yarrow
#

drunk?

#

maybe... full of doritos?

crimson veldt
#

no lol

#

get into machine learning

#

i hope things goes smoothly

bleak matrix
#

Doritos

#

And...Mountain Dew?

signal nest
#

anyone tried finetuning SD on images from googleearth

#

as a map generator

bleak matrix
#

Like, for cartography?

#

Or just in general?

signal nest
#

i'm thinking for creating fictitious layouts, you could then guide it with img2img "i want a coastline here, some buildings over here, now flesh it out.."

#

also you could make repeating worlds

leaden mural
#

how about the EDiff-i?

signal nest
#

(add something to train looking at a point on a map and hallucinating a panoramic view from ground level .. that could be pretty interesting)

bleak matrix
#

With in-painting, or a game?

leaden mural
signal nest
#

as a gamedv tool.. i'd imagine this h appenging offline

leaden mural
#

FRICK

#

is the spammer gone?

bleak matrix
#

I think so

leaden mural
#

that's strange, he's a year long member. probably a hack?

#

wait, it's still a leaf. Either a lurker or just hiding.

dawn yarrow
#

meh

leaden mural
#

back to topic

#

i know of someone who created a EDiff-I plugin for SD, though its still under dev. Not to mention not released as an extendion for the WebUI

bleak matrix
#

There are a lot of developers here

#

Working on cool projects

#

There's a lot of application for game development and AI, I think

signal vale
#

Can anyone please help me to understand if I use SD locally, do I own the generated output, if it's distinguishable from the training data? Or who owns it?

north stirrup
fervent thunder
#

at the moment I believe it's a gray area. But if there was someone who'd own it, then it'd be the one who created that image. So either you, or the AI :P

north stirrup
#

Inpainting, img2img etc. can likely produce copyrightable works

north stirrup
fervent thunder
signal vale
#

In other words, if I have commercial applications that generate logos to print on T-shirts, then anyone can pick the logos from my gallery?

fervent thunder
#

a funny detail is that if someone were to take a spray can, and paint the side of a building, then they own the copyright to it, and if someone removed it, then they who removed it are doing something very naughty. US law is a little wild sometimes :P

north stirrup
bleak matrix
north stirrup
bleak matrix
#

Just a moment

fervent thunder
#

BECAUSE SOME JURISDICTIONS DO NOT ALLOW THE EXCLUSION OF CERTAIN WARRANTIES OR THE LIMITATION OR EXCLUSION OF CERTAIN CATEGORIES OF DAMAGES, SOME OF THE ABOVE LIMITATIONS MAY NOT APPLY TO YOU. gotta love the only important part of it :P

north stirrup
# bleak matrix https://stability.ai/faq

What is the copyright for using Stable Diffusion generated images?
The area of AI-generated images and copyright is complex and will vary from jurisdiction to jurisdiction.

Eh, vague

signal vale
#

What if I get the output of SD, add a couple of details, and put a copyright on that. Is it possible?

north stirrup
#

Copyright requires original, creative work

#

The threshold for “creative” is pretty low but does exist

signal vale
north stirrup
bleak matrix
#

Because not every country

#

Specifically

#

Recognizes public domain

north stirrup
#

Yeah, the jurisdictional inconsistencies will be awful for years

bleak matrix
#

It (DreamStudio) was released under a world-wide license:

#

(Just look at the DreamStudio ToS for more info.)

#

The license for you would run locally on your machine you can read on Git, or by just looking in the folder

signal nest
#

img2img copyright , hmm. "can't copyright generated images" (sounds fair to me,actually). but then if i draw my own images, i can copyright them , right. Then if i feed to img2img.. whats the result.. my copyright? or not? :/ at least I'm possibly edging toward beleiving that I could use img2img generated images in a game i release.

#

also is there such a thing as an imagge license like this: "can only be used to train opensourced AI" - kind of like copyleft.

dry flower
#

how do you know which version of SD you have?

fathom olive
#

Take, for example, something like Gwent. Sure, CDProjektRed has undeniable copyright on all of the art for Gwent cards, but even if every image was in public domain (I personally believe) it wouldnt devalue the final product. They would still have whatever licensing agreement exists for The Witcher game adaptions, and they'd still have the rights to Gwent the Game 🤷‍♂️

signal nest
leaden mural
#

@bleak matrix I still don't know this. Have you tried making a game before?

signal nest
fathom olive
#

Out of curiosity, what kinda game stuff are you making with it? I've been working on card-game-esque use cases since it seemed pretty straightforward 😅

#

Havent found any reliable way to get coherent assets yet, though I've seen some promising proof of concepts

signal nest
#

i also want to try running it on lowpoly models to texture them, not sure if that will work, but even making multiple views (like oldschool doom sprites) could be interesting, with enough view angles, maybe yoiu could approxiamte that in a manner similar to NeRFS or feed back into photogrametry..

#

and yeah i've seen people suggest that the best use of SD is infact just as "concept artist"

fathom olive
#

I guess I haven't paid too much attention to the texturing side of things simHmm I could see that being pretty useful, but I'm a bit too addicted to Substance Designer that I don't think I'd enjoy using it for that kind of stuff.

signal nest
#

right its quite possible that Substance Designer will remain superior to what i'm doing :/ but i am addicted to handrawing textures, so i'll see how far I get

fathom olive
#

If you're going for the ps1/2 demake style I could see the low poly texting working amazingly well, since the whole point of that is creating "too much" detail and then crazily downscaling it

bleak matrix
wide juniper
#

Where can I find Dreambooth training examples besides training faces and styles? I want to see what I can do and what its limitations are before I upgrade my GPU for it

bleak matrix
#

Other stuff I am under NDA

signal nest
fathom olive
#

You'd need to do the standard painful stuff with UV unwrapping and such to finagle the textures into place, but there's your "creative input" anyways. I don't think anyone would argue that creating a game texture out of public domain work means you abandon the game texture's copyright (assuming actual work was done in the conversion, and the texture isn't just a 1-to-1 of the public domain work)

#

I've wanted to mess around with the demake art style for a bit, but I was always put off by needing a more "realistic" style (I prefer hand-painting stylized models in Substance Painter, too 😅 ), but if I could set up a work flow to generate it, that would be a really cool way to work faster.

signal nest
#

i'm not really an artist, i'm a programmer - my main focus is my custom engine , and i'm just improvising art as far as i can . I can draw a little bit. I did a lot of pixelart growing up (16bit days) . it seems like with "img2img" it should be possible to do something thats a lot better than "nothing", or the risk of trying to pay an artist for a game that is unlikely to make any money. I'm building for myself really. (and i'm aware "fair use" -personal use, show friends etc has already been ruled as ok by the courts)

fathom olive
#

I also am not really an artist, and should be a programmer, but I forgot to keep telling myself that over the years and have probably spent hundred to thousands of hours practicing art at this point sim3 It's all supposedly for game dev projects, but then I keep making more art and keep not making games sim3

lavish current
#

I wanna buy Alice and Sparkle

fathom olive
#

But yeah, it sounds like you're gonna be A-OK regardless of any legal decisions. I would be baffled if they somehow managed to stick some kind of license on AI-generated art assets that would immediately de-license any products that use them, and even if one day your game textures end up public domain... 🤷‍♂️ who cares

#

not a lawyer obviously 😄

lavish current
#

It's funny that book made artists get mad lol

#

And I say this as someone pursuing a career in children's book illustration

bleak matrix
#

Also, understanding how it works is fundamental to getting good results

bleak matrix
#

Although I always enjoy seeing new programs, technology, and ways of doing things come to light

bleak matrix
bleak matrix
timid crypt
#

aye does anyone know if there's a way to select the og hires fix for automatic1111, I'm pretty sure it changed

#

from an img2img kind of thing to a straight up upscaler

dark grove
#

can you use stable difussion without CUDA like a boss?

glacial wedge
#

does anyone know who owns "Playgroundai?

high marten
#

I am trying to find a SD model that has the knowledge of Craiyon

#

I want a version of SD that knows a bunch of obscure game characters.

floral zealot
#

where i can see the version from sd?

still remnant
# dark grove can you use stable difussion without CUDA like a boss?

I've been working on this for a while as a local install on a Windows system, as far as I've found, while there are ways to make it work, none of the other methods are as memory efficient as CUDA. In my case, I'm hoping to just gut up and buy an NVIDIA RTX A5000 by March instead of hacking my dual AMD cards into working.

dark grove
still remnant
#

then again maybe MY perception is skewed 🤷

swift sluice
#

Hey guys!
Do you have any articles or git repos you would recommend for v2.1

floral zealot
#

Somone know any good nsfw or anime model?

quick basalt
#

The Anime model I use is Eimis Anime diffusion

#

But it's one of the first anime models, so I think there are new ones that might be better

sonic carbon
fervent thunder
#

when using automatic1111, is it possible to have it generate using different prompts during a batch? So it would generate Prompt A for the first generation in a batch and use Prompt B for the second, Prompt A for the third, Prompt B for the fourth etc? or possibly just do them half and half, like for a batch of 4 do the first 2 as Prompt A and the last 2 as Prompt B? Or all of a batch using Prompt A but then change to Prompt B for the second batch, and back to Prompt A for the third etc? basically just some way to do multiple prompts without having to come back and start a new batch each time

north stirrup
fervent thunder
north stirrup
#

So e.g. with the prompt foo bar you’d put bar, baz, qux in the script to run foo bar, foo baz, and foo qux

fervent thunder
lofty monolith
#

Huh did something change about automatic1111?
I can't seem to find settings for "first pass size"

quick basalt
#

Guys, it seems that Unstable Diffusion released their model

frank oracle
#

Guys

#

Should I update Git?

past vortex
#

is there a good tutorial for how to use embeddings?

sinful mango
#

Just published a video on the subject of AI & Copyright: AI vs. Blood Mouse — Disney, AI Art, and Copyright

The Copyright Association, representing big corporate interests such as Disney, appears to be making moves against open source AI image tech. A lot of artists are hopping on board after the Concept Art Association started a GoFundMe campaign to join the Copyright Association and lobby for copyright law expansion. The video explores some ways they could potentially try to outlaw open source AI models. Enjoy!

https://www.youtube.com/watch?v=5pIVVpoz5zk

past vortex
#

also I really dislike the newest update to Auto1111

#

I have the Upscaler set to none and no detailed faces and yet it's still doing a second pass on every pic

#

for no reason

potent spire
#

and if i was in their shoes id make the same step tbh

#

the reaction within the SD Reddit was absurd and delusional as expected too

#

to this news

past vortex
#

what news

sinful mango
potent spire
#

just watching it 😄

potent spire
past vortex
#

ah, that's probably a good move

#

really AIs should only source from free use pics and whatnot

potent spire
#

i mean i would join too for several reasons

sinful mango
dapper geode
#

hey guys, just tried using stable diffusion on Ubuntu Subystem on Windows. Could anyone help me? The process always gets killed and i rly dont understand why. Thank you guys

potent spire
#

also ppl on SD Reddit share fake news or generally misleading and false claims

#

about Adobe and others making their own AI art generator

#

which is technically true

#

but

#

not how those people are selling it on Reddit

#

so artists wont cry about Adobe Ai Art generator

past vortex
#

but anyway, anyone know how to make Auto1111 stop doing the pointless second pass?

potent spire
#

second pass?

past vortex
#

I tell it to do a batch of 10, it does 20 passes

#

like when you do detailed faces and it does a second pass on the face

#

but I don't ahve restore faces on, nor do I have the Upscaler selected, it's set to None

#

it's only become an issue since he added in the Upscaler direct into txt2img

#

can't post pics, otherwise I'd just show you

#

suffice to say, it doubles your batch times for no reason

jade meadow
#

did you know that there are 22 million people in america that are millionaires or make at least a million a year what's your excuse not being one of them

tawny quail
deft ingot
past vortex
#

I was born upper middle class and am a physician and I'm still not a millionaire 😦

#

(being born wealthy is a huge part of it)

jade meadow
jade meadow
alpine kayak
#

I'd like to ask a simple question. How long does 20k steps for embedding take usually? Is it around 3 hours?

#

(Seems like a good way of knowing whether this is fine, my rtx2070 is nearing 81c as I'm doing this)

velvet oasis
#

how can you tell which models can do depth2img? Like, I know the stable2depth one has to be able to do it, but can others like 2 or 2.1 do it? What about 1.4, or other diffusions like dreamlike-diffusion?

sacred shell
#

Guys, any colab notebook where I can use pre trained concepts/model?

unreal flume
#

Did anyone have any semblance of success training a Textual Inversion with Anything V3 through Automatic Web UI? Because so far it’s failing really badly for me.

steady arrow
#

Guys im trying to do a batch generation of 8

#

But I run out of memory?

#

I dont run out of memory without batch?

quaint yarrow
#

how do you get the 2.1 bot to do a prompt?

#

dream : ?

brisk bloom
steady arrow
brisk bloom
#

Yeah A1111, not sure about the highres setting. I feel it produces better results though

steady arrow
#

seems to take like 6x as long lol

#

But I think I need it because Im generating 512x960 images

#

also my dedicated gpu memory stays relatively high use after doing these gens

past vortex
# quaint yarrow what is a 2nd pass?

on your actual Stable Diff window, you'll see it go from 0-100% once and generate an image. When it's doing a second pass for whatever reason, be it restore faces, hires fix, etc, it will do a second pass instead of spitting out hte image in the folder. So a batch of 10 will get you 20 lines of progress bars filling to 100%

steady arrow
#

It's stuck at 5gb used after the batch of 8 with 1.7x high res fix upscale

past vortex
#

I have everything off and no upscaler running, but it's still doing two passes per image, effectively doubling the render time

steady arrow
past vortex
#

it's started since they added the built in upscaling on txt2img if you want

past vortex
#

I updated Auto1111 after like 3 weeks

steady arrow
#

I just updated today also

#

Last batch gen it got to peak of like 7GB gpu ram used, now its at 9GB

#

It doesnt seem to flush whatever was in the ram 🤔

#

and now I get error: RuntimeError: CUDA out of memory. Tried to allocate 3.75 GiB (GPU 0; 12.00 GiB total capacity; 7.74 GiB already allocated; 1.78 GiB free; 7.77 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

#

grr seems completely random if it will spike the memory into failing and then the memory stays maxed unless I shut down auto webui

#

Been a bug since october..

hot marlin
forest apex
#

Art for Me, not for Thee

astral goblet
#

people are selling ai art? how do i get in on this

#

if theres a new tech wave hustle, i want in

steady arrow
forest apex
#

@astral goblet it's here man, the number of things that have exploded in the past few months will only increase exponentially

astral goblet
#

and then hoping you get out of the ponzi scheme before it pops.

#

i want to get in to new schemes. not the same old scheme

forest apex
#

well crypto is still up if you bought it before the pandemic :>

sinful mango
forest apex
#

I'm talking to some Anti-AI people, one person is rationally asking questions, the others are doubling down on "Art is not theft!"

#

In the end they want to stop "AI" from reaching any popularity, they are convinced "we will loose all our JoobsSSS!"

astral goblet
#

"learn to code" rings true again

forest apex
#

one actually told me "learn to draw!" I'm like, oh the dozen programming languages I've learned isn't enough...

lavish current
astral goblet
#

i have aphantasia, not complete but it's there. i can't visualize images in my mind. instead my primary mode is dialog. i describe everything very rapidly in my mind. like a very fast talker that i manage to keep up with

lavish current
#

We made book dummies using traditional tools, and then the following years while I had free time I made dummies for two other books

astral goblet
#

it has always really prevented me from drawing

#

i can draw but i can't see any end product, so i'm working blind

lavish current
#

The fact that an AI children's book got people heated on Twitter is just funny to me

astral goblet
#

i've lived the "not all people can draw" argument most of my life

carmine fulcrum
#

not all people put the time into learning to draw well

sinful mango
astral goblet
#

i've got probably 100 sketch books of drawing. i been to visual arts college. i've put the time in. but aphantasia means a blank slate is entirely empty for me. I have a lot of ways aroudn that to still be creative, but it usually involves painting over "stolen" ip or other compromises i make in order to actually see what i'm trying to describe in my head. being mind blind is a huge hinderence that most just don't understand

past vortex
#

I wouldn't say all styles are public domain, if you're an artist and developed your own style, a la the Sam Does Arts guy, building models specifically based only on your art to then iterate and generate more stuff that essentially is directly mimicking your work via direct sourcing is pretty messed up, but I'm clearly in the minority on that opinion so I'll leave it at that

forest apex
#

@sinful mango 5:14 in rings so true, artists being given the illusion they are joining some grass roots operation

hot marlin
astral goblet
#

sam does art doesn't have a unique style. he has a style that he sticks too, but it's not his or original

forest apex
#

@past vortex perhaps, but the primary tool of the beginner artist is to mimic advanced artists they adore

past vortex
#

eh, it's still his work, and he shoudl have a say on if AI source his work or not

sinful mango
carmine fulcrum
astral goblet
#

style isn't something you can own, otherwise a corporation would just create a blanket style ownership mat and control it all

past vortex
#

it's no different than if I'm in the park with my kids and some dude starts photographing my kids

#

as a parent I have a right to privacy for my kids

hot marlin
past vortex
#

you can't just use their likenesses in an ad if they happen to both be wearing Nike hats or whatever

sinful mango
past vortex
#

we need to give permission

astral goblet
#

copyright is about as limiting what rights you have as well as giving you rights over your work. one limit is that the style alone isn't yours

past vortex
#

I think Sam's argument of "it should not be you ahve to opt out, you should have to opt in" is pretty fair

#

and reasonable

carmine fulcrum
#

they may not be using machine learning, but they are using AI. But soon they will be using machine learning based AI tools if they want to work at a studio.

past vortex
astral goblet
#

there are 100 other laws like "trademark law" that apply too. Sam Does Art may be a registered trademark, but i don't think it is

hot marlin
past vortex
#

I can't take one of his pics off insta and then put it in my, say, calendar and sell it for money without his permission

astral goblet
#

No, but if you look at that pic on insta, that's fine

#

insta has a license to show it to you

sinful mango
past vortex
#

to show it, not resell it or reprint it for monetary gain

forest apex
#

If Stable Diffusion had trained on Wikipedia images, the artists would have had no say. And the world wouldn't have been any worse off.

astral goblet
#

yes. So looking at it is fine. The ai models don't store the image after it's looked

#

that would be an insane claim. 100s of trillions of images in one dvd? insanity

#

instead what we get is a small 4gb sized neural network configuration that is good at denoising images

past vortex
#

the issue is if AI is sourcing its DB off of copyrighted image, even f the content generated is a new image, it was still based off of a copyrighted source and is taking modeling data directly off those source images, it's not the same as me looking at his work adn saying "hey, I think I'll make a drawing in a similar style"

#

because even though we call it AI, it's not actually intelligent