🌶｜off-topic

[livid pendant]

I was having problems getting stuff done and you know

You always remember people who help you

@fathom condor literally spend over 2 hours in DM with me helping me get stuff up and running.

Who does that? in this day and age.

And to see genuinely good people get picked on is painful.

[main cedar]

I'd like to, but college is very stressful and workload-heavy

[livid pendant]

I appreciate people who genuinely go out of their way to help others for no return or reward.

[blazing adder]

that's a great ability that is perpetually in shortage.

[livid pendant]

Bro. The guy spend two freaking hours with me.

I offered to send him some paypal and he said, no, just go buy some homeless guy a meal.

That brought tears to my eyes.

Just warms my heart to see there are still good people helping others, making things for free, making technologies to help mankind, not profit

We would not be here right now

If it wasn't for these kinds of people

[main cedar]

That's how I want to spend my retirement

[livid pendant]

You gotta be careful though.

[main cedar]

Also, I think I said it before, but the whole reason I want to employ robots is so that remaining human workers can be paid a living wage with the money saved

[livid pendant]

Yeah, no.

Not happening. That'll just pad their bank accounts some more.

You think anyone ever needs a billion dollars?

[blazing adder]

according to current economics, Not sure if that would work.

[livid pendant]

couple of hundred billion?

Capitalism is not just mode of econonmics.

It's a mindset. It's a feeling. It's an emotion. It's endless acquisition.

[main cedar]

I don't want a billion dollars

[livid pendant]

It's a sickness of the mind, man.

[main cedar]

Just a decent salary and some good times

[livid pendant]

The Dragon Sickness.

The sickness for gold.

[main cedar]

Especially because I grew up with a situation where even that was far-fetched at the time

[blazing adder]

I don't think it's restricted to capitalism or a capitalist mindset. It's more of just a fact of human nature that we're that way.

[main cedar]

Ehh, greed nature

[livid pendant]

No.

Like people who live in villages or stuff

[main cedar]

Human nature is be kind and communal, because getting kicked out of the tribe means death

[livid pendant]

Mountain villages or like people living in different kind of simple life

They don't more and more

[main cedar]

One moment, rendering 2 1080x704 images on a 2060

[livid pendant]

They just hunt, gather, grow, live.

[blazing adder]

yet still take advantage of each other and are self interested people.

[livid pendant]

Sure

Capitalism is a specific type of mindset of the industrial revolution

It's much more common in industrial societies

[main cedar]

Haven't posted in a while, figured I'd come back with a bang

[blazing adder]

It's a system that existed before the industrial society. the industrial revolution just molded it to it's current form.

capitalisms sucks because it encourages greed, socialism sucks because everyone's dirt poor and hates the bourgeoisie.

[noble cedar]

i wonder what the avg age demographic of this server is

[blazing adder]

good question

[livid pendant]

It's mostly kids.

[noble cedar]

from wat ive seen, it feels like 20-36

[blazing adder]

my guess would be mid twenties or so.

[livid pendant]

Some grown people but mostly kids.

And that's why you see that kind of behavior.

[blazing adder]

Also known the depressed and/or stressed college student demographic.

[livid pendant]

I don't think there are that many.

[blazing adder]

hmmm

well how many people in their thirties and forties that are into AI?

CS is expanding and progressing so fast that people are quickly left behind understanding wise.

[spice vigil]

old IT people make up a disproportionate amount of many communities. hullo

[blazing adder]

hullo

[main cedar]

Hey, I may be stressed

[noble cedar]

which reminds me how the arch linux community (the discord) Has None

[blazing adder]

well maybe I'm wrong. but I do know that my father and people his age in the tech sector are working more in databases, SQL and maybe C++, but my brother runs circles in networking and other areas

[spice vigil]

oh I hope he doesn't do circular runs in networking

[blazing adder]

Maybe more older people then those I know hobby in AI and crypto, but from what I could tell it's mostly the younger guys doing the new fangled stuff while the older fellows actually keep the backbone working.

[blazing adder]

I have no idea what that is. I'm a philosophy and mechanics guy not programming.

[spice vigil]

a dumb networking joke, disregard

[blazing adder]

ok. googling circular networking now.....

[main cedar]

Lmao

[blazing adder]

ohhhh

so it's similar to a infinite loop error but for networks?

in small man speak.

looks like I'm gonna learn networking now. nice.

[spice vigil]

deletes essay

good callout

[blazing adder]

haha

[main cedar]

lol

[spice vigil]

YES.

[blazing adder]

cool

now I can tell my geek brother and he can have a laugh at my half baked explanation of it to him.

[spice vigil]

You can explain to him that switches don't like diddling themselves.

[blazing adder]

I will then.

I've been bugging him recently with questions about installing arch.

He setup my partitions with my first dual boot setup, then I decided to "experiment" with things recently.

[fast oak]

did r4 get banned or something?

[fast oak]

He has been timed out after an encounter earlier from what I saw. He should be back, I hope so, I like the guy a lot (not by me and I can't really disclose more on what happened, i wasn't around)

hm

[hybrid mural]

IDK if he intends to come off like it, but was around the last 9 hours or so and he can come off passive aggressive sometimes - got people riled up in a few separate chats

[fast oak]

from my perspective all that's going on is a bit of a generational divide. i think he obviously has good intentions but due perhaps in part to age, or simply difference of culture (also he claims to use psychedelics every day so that might just be why he's a little odd sometimes), he's a little prone to histrionics. i have been known to be so myself, but overall, he's an all right guy, just gotta guard one's self from him so one does not have a negative reaction to how he behaves. people have a bad habit on this server to gang up on people. it's a bit toxic sometimes, (sometimes, overall there's a lot of creativity and fun and no bad apples can spoil it) so there could have probably been more sensitivity in handling the guy.

[rocky wind]

Hi all

Graphics tools related question.. Do any graphics tools export a 'depth' buffer layer for other tools so that it can determine which elements in an image are foreground, background etc?

The thinking here was that an image from say Povray , "knows" where in a scene certain elements are, and can thus potentially provide information to another tool doing compositing.

Such as img2img being able to say that something it's generating as an inpaint, should be occluded by something already in the foregorund.

[fast oak]

that would be cool

isn't there a plugin for photoshop though?

[boreal ore]

Aestus

[boreal ore]

I never recommend paying Adobe.

[winter moon]

I got a nasty penalty for an early termination I don't even remember agreeing to

That was the day I went, okay, it's over for me:\

It's funny — not affording Adobe CS and pirating the Master Collection was what got me into design

Now I'm in design and I can't afford Adobe CC

[boreal ore]

It's always morally correct to acquire Adobe products via nefarious means

[winter moon]

They re-did Master Collection

(the pirates, not Adobe...)

When I saw "Master Collection 2022" I almost got misty eyed lol

[boreal ore]

If I'm in an environment where I need legitimately licensed apps I'll switch to resolve instantly

And krita/gimp

[winter moon]

Do you like resolve over Pr?

wait what would its CC equiv even be? After Effects?

[boreal ore]

No I don't even use premiere
I mainline after effects

It's way more complicated but way more powerful

[winter moon]

Me when someone asks what im using:"After Effects"
Me when someone asks why: (turn to camera) "After effects."

[winter moon]

Resolve?

[boreal ore]

No, ae

I don't use resolve at all

[winter moon]

oh yeah like Pr

[boreal ore]

Ye

[winter moon]

dont @ me but i like imovie for really short stuff if its just a meme or something

anyone who doesnt is lying

[boreal ore]

Windows so I have no imovie
If I just need to quickly cut something I use premiere

[winter moon]

i heard you can run both mac and windows on macs why dont you get a mac

[boreal ore]

Because then I'm limited heavily in the hardware I can use

[winter moon]

ah was just jokin

im on the macbook pro with 4 USB-C ports and one headphone jack

[rocky wind]

Does anyone in design professionaly use anything other than Mac?

I use (on windows) Inkscape , an old old shareware called Paint Shop. and I used to use Gimp for a bit, but found it wasn't really UI designed that well..

[main cedar]

I use gimp on Linux and Windows, but I'm also used to the blender interface, so I'd never claim gimp is easy to use just from using it

[rocky wind]

Of course some people I used to know in video production used high-end workstation brain melty powerful things ...

That could do the "Dissolve the male villan into a pile of wind-blown sand" type effect, with a few mouse clicks...

[main cedar]

That could be blender

[rocky wind]

Oh?

Blender can do digital compositing?

[main cedar]

I mean, if you can make something in blender into a curve, you can basically draw it to your liking across thr scene. It also can be used for video editing. I don't doubt people have used it for VFX and digital compositing

It's extremely versatile

[rocky wind]

Anyone here want to hear some crazy ideas?

[main cedar]

I have to go to a class that my professor does not offer online, but you're welcome to leave them here and I'll read them

[rocky wind]

Well there are mostly media/graphics related...

[main cedar]

That first one is definitely possible, and it would be cool to see a non-snapchat one that doesn't whitewash and/or give you chad jaw

[hybrid mural]

Isnt it a thing already on facetune

SD is also pretty good at soft blending facial concepts

[subtle echo]

Pls anyone one of you, could you generate a text in openai playground?
write me a powerpoint presentation about green energy
Im going to fail classes if i dont finish this.

[rocky wind]

Sorry but my university honour code prevents me doing that for policy reasons.

[spice vigil]

oh no not the consequences of my decisions

[main cedar]

As a college student, I will say that that kind of thing is the reason homework gets easier and easier over the years

I know several students who might not be turning in the same paper I did last night despite it being due today

How many weeks did you have to do it?

[indigo salmon]

Is anyone here interested in making some occasional cash? I'm looking to make a new friend with lots of experience in stable diffusion that i can occasionally pepper with questions. I'll send you ten dollars whenever i pester you about stuff, i'm nice to my friends 😄
Its more of a casual arrangement than a job, I just know people would rather be compensated a little for their time.
Good english skills are a must

I make lots of neat stuff: https://www.deviantart.com/nanakoac
and have recently transitioned from MJ to stable diffusion, struggling with lots of teething problems
DM me if interested

[desert gorge]

I need a knowledge bank too. Nanako will pay for it

[inner bloom]

@dreamy rover https://www.youtube.com/watch?v=ZVYqB0uTKlE&t=1s this was the video i was thinking about

[manic rover]

https://www.youtube.com/watch?v=5de9GqD6vhk

[crystal scaffold]

sounds cute lol

I do wonder how the server costs compare to other multiplayer games

glad this exists 🙂

[hollow falcon]

@lilac pebble I might be able to help you out what'sup

[lilac pebble]

So when i started generating it usually says 1.4 s per iteration

now it says 6s per iteration

[hollow falcon]

what's your specs?

[lilac pebble]

using a 1050ti gpu

[hollow falcon]

laptop? desktop? gpu? is it zotac or evga

or something else

[lilac pebble]

desktop

[hollow falcon]

do you know the model of gpu?

[lilac pebble]

its nvidia

[hollow falcon]

also did you recently update (git-pull) web-ui?

[lilac pebble]

i have not recently updated the web ui

[hollow falcon]

I mean is it a founders edition nvidia card, or is a zotac/EVGA/gigabyte/ASUS card

[lilac pebble]

oh lemme check rq

[hollow falcon]

some of them can be damaged by running SD too much

[lilac pebble]

its an asus card

[hollow falcon]

okei, probably fine there then

what do temps look like when you run SD?

[lilac pebble]

lemme do a quick run rn

[hollow falcon]

I would personally recommend grabbing a new GPU ;) I just bought my 2080s for like $400

I'm pretty sure you can get like a 1070 or 2070 for hella cheap (used)

but yeah watch temps as it runs

[lilac pebble]

Problem is im a student and im about to go to college so i cant focus on buying my gpu rn

[hollow falcon]

also pay close attention to how fast it reaches max temp

[lilac pebble]

But planning on buying a laptopt with 3060 ti for college purposes

[hollow falcon]

ah I see

don't do that

that's a waste of money

you're getting a throttled down cheaper 3060 ti for a bloated price in a laptop

[lilac pebble]

ok so i just did a run

before it reached 70*

now its at 54

and before

[hollow falcon]

not bad ok, sounds like you're not getting even close to max temps

[lilac pebble]

the dedicated gpu memory reached 3.8 / 4.0

but now its at 3.1/4.0 while running

[hollow falcon]

hmm, few things here

what's the generation specs for what you're trying to run?

also what model

[lilac pebble]

generation specs?

[hollow falcon]

CFG scale, prompt, resolution, scripts, hires fix? etc

[lilac pebble]

Oh

28 steps, cfg scale 11, 512 res, euler

[hollow falcon]

and the SD .ckpt model you're using?

[lilac pebble]

nai

[hollow falcon]

is it the 7 gig version

or 4gig or 2gig

[lilac pebble]

4

[hollow falcon]

🤔 are you using a custom vae

[lilac pebble]

nope

[hollow falcon]

oh

do you run with web-ui.bat

that automatically handles doing git pull and downloading updates and stuff iirc

[lilac pebble]

oh

i run with the user one

webui-user

[hollow falcon]

sometimes new commits can go up that break things

[lilac pebble]

oh

[hollow falcon]

contents of webui-user.bat:

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=

call webui.bat
``` lol

[lilac pebble]

Do i just reinstall sd? for it to get fixed?

[hollow falcon]

it might not be automatically updating it anymore lemme check

I'm guessing that something on your pc is using a little bit more VRAM than usual, so the webui.bat just automatically launched with lowvram mode due to the amount of available VRAM on launch

no one sec

and ignore the xformers thing

[lilac pebble]

yep

[hollow falcon]

1050ti doesn't have anything that will benefit from that lol

[lilac pebble]

ye he was talking about pascual which is 1080?

[hollow falcon]

just open a command line in the web-ui root folder

and put python .\webui.py --medvram

tbh it should automatically handle it anyways so maybe skip the medvram

1.5 iters/second to 0.15 iters per second though it almost sounds like its running on the cpu lol, there's a major bottleneck there

just to be sure, launch without --medvram, switch to model.cktp (sd 1.4) and try to run a generation and see what happens

[lilac pebble]

idk the cpu usage is not going up tho only gpu

aight

[hollow falcon]

also nai has some nice outputs at different resolutions, so try 448x512, it might kick the vram just low enough to reach back over the threshold

[lilac pebble]

noted

[hollow falcon]

I used to run this on a 1060 6gb

another tip - you can use 448x640 and it's only like 8% more pixels than 512x512

if you're making characters

[fast oak]

?si

/serverinfo

[hollow falcon]

#1014939219904450590

[lilac pebble]

im back to 1.4 s/it

thanks @hollow falcon

[hollow falcon]

np what fixed it the launching it manually thing?

[lilac pebble]

i ran it with webui instead of the webui-user? other than that i have absolutely no clue what just happened

[hollow falcon]

lol

so, looks like it switched to running your default python install, default git repo, idk venv, and overrides the commandline args,

[lilac pebble]

Unless my gpu was just acting up

[hollow falcon]

of which you may have inserted something at some point?

[lilac pebble]

I dont remember inserting something at any point

[hollow falcon]

open your webui-user.bat and paste it here if there's anything after COMMANDLINE_ARGS=

[lilac pebble]

its emtpy

cause i played with it before and nothing much

[hollow falcon]

🤷 then probably just the threshold for VRAM optimizations during launch

[lilac pebble]

Well anyways thanks for still guiding me!

[fast oak]

!serverinfo

[hollow falcon]

#1014939219904450590

[fast oak]

idc

[winter moon]

Anyone in here happen to know great foreigner friendly restaurants in Tokyo?

Especially a great miso or tonkotsu ramen

[hollow falcon]

I'm sure your "!serverinfo" was very helpful lmao

[fast oak]

No it was to see who has the crown of the server

[hollow falcon]

? you mean that video that's making all the schoolkid demographic for that one youtuber brigade this server with absolutely no clue what's going on?

[fast oak]

No

U saying shit

i just want to see who has the crown of the server

LMFAO

[hollow falcon]

#1005976006177345596 rule 1

[fast oak]

BRO ID

C

[hollow falcon]

Then why reply :v you seem passionate about the subject

[hybrid mural]

"Sukiya's" 💯

Definitely not a deez nuts joke btw

[winter moon]

fast food:[

[hybrid mural]

It's like, sorta fastish food

[winter moon]

omg according to Siri Knowledge it's american beef?

what in the world lol

[hybrid mural]

I would also go back just to experience a GoGo Curry again, def have one of those

[winter moon]

Unlike Yoshinoya, Sukiya did not stop serving gyūdon during the ban on American beef imports, instead switching to beef imported from Australia.

[winter moon]

mmmmmmmmm that sounds so good lol

[fast oak]

U replied to me first

[spice vigil]

not a lot of space for cattle herds in japan lol

[fast oak]

stfu

[hybrid mural]

Dutour's is also a fantastic lunch stop, amazing sandwiches n stuff

And they are on every single street corner p much

[hollow falcon]

yeah but I actually do care

[winter moon]

as long as it's not another Paris Bakery type thing

[hybrid mural]

Also don't have anything western, not even to try out, they're all so bad

[hollow falcon]

looks like we got a wild one who made it past general lol

[winter moon]

imagine fresh food

my american brains going wowow

[hybrid mural]

IKR, just have to see what a tokyo western thing is like

[winter moon]

Is it China or Japan that does the KFC Christmas

[hybrid mural]

Japan Still cant believe that

And they close all their shops to Auld Lang Syne

[winter moon]

i still don't understand

[winter moon]

i flipped out when i heard this

[outer prism]

japan does that

[winter moon]

i was like imagine not stressing out an entire group of people

that was so cool

stores in the US just kinda ghost you

[hybrid mural]

Also there is a famous fish market in Shibuya with an extreme amount of free samples, lots of various seafood that was probably swimming moments ago Extremely good stuff

[winter moon]

is this the one that comes out in Jiro Dreams?

[hybrid mural]

Ahh maybe? I have not seen 👀

All I knew was it was a market everyone in tokyo goes to from miles away for

Do you have a J Railpass btw

Supreme cheapness for getting around

[winter moon]

I don't even have a passport this is for like a long time from now (less than a year, more than 6 months) but planning early as hell always pays off for my ADHD self

[hybrid mural]

Fair enough I am similar with travelling

[winter moon]

[hybrid mural]

Japan Rail Pass is a must, its a tourists only rail card, massively pays for itself

[hybrid mural]

You pay X amount for 1 or 2 weeks however long, and all journeys covered including the shinkansens

[winter moon]

Me: I will learn this map in 6-12 months

the map:

[hybrid mural]

Urban nightmare challenge extreme 👁️

[winter moon]

i lived in NYC, I can totally handle just visiting Tokyo for the first time as an American

I'm traveled.

[hybrid mural]

Most important tokyo map to learn is

[winter moon]

this is

the best

[hybrid mural]

[winter moon]

keeping an eye on me

[hybrid mural]

The eye of Akihabara

The final destination of all japanese tourists

[winter moon]

I described Harajuku as "where the cool kids hang out" and Ginza as "the Fifth Avenue of Tokyo".... how horribly did I mess up?

[hybrid mural]

Accurate

Shinjuku is where the extremely cool kids hang out

Yall will see some IRL final fantasy people there

[winter moon]

I will take the Shibuya bound yellow train to Aoyama Itchome and the Hikarigaoka bound train to Shinjuku

To get from Ginza to Shinjuku

I am very clever

[daring coral]

Good evening (or morning) folks! 7th issue of AI Art Weekly just went live https://twitter.com/dreamingtulpa/status/1588569357634109441

Dreaming Tulpa 🥓👑 (@dreamingtulpa)

Dear #aiartcommunity 🎨 Issue º7 of #aiartweekly is live! Cover art by @amli_art 🏆 First look at image-to-4D. We talk to @rainisto about his process and #dreambooth. New cover challenge, new chance to win $50 bucks 🔥 https://t.co/h5IPye2yyc

More below 👇

Twitter · Nov 4, 2022, 4:30 PM

I'm doing another Cover Art challenge. This weeks theme is "cyberpunk". Winner gets $50. Would be lovely to see your contributions 💪

[night geode]

I wonder if stable-tweets will still be a thing soon. Cutting half the employees of Twitter at the same time with no warning, including putting technical staff down to a skeleton crew… does not seem great for a tech company that size. Lots of knowledge going out the window from no formal handovers. Plus I’m sure the rest are looking for jobs, putting even more strain on keeping things up and running…

It sounds like people woke up to not being able to log into their email account and at least one person lost their access in the middle of a meeting..

[neat lichen]

what are stable-tweets?

[night geode]

Oh just a channel here that shows the tweets from Stable Diffusion.

[desert gorge]

what's wrong with any of that?

Twitter had half the employees like 5 years ago and the site was identical. Not much value is being lost.

and for a company with so many activists, you WANT to cut them off with no warning, because a small percentage could be crazy enough to sabotage before they leave

[hasty willow]

🤣 Any employee that didnt have a plan b since the buyout announcement months ago doesnt really deserve sympathies

[desert gorge]

I don't think anyone who's above average needs to worry

it's the dead weight that should've tried to get hired elsewhere quickly

[spice vigil]

disregard outrage trolls, acquire waifus

[desert gorge]

post one

curious what someone at your level is producing

[spice vigil]

No get your own waifus D:

[desert gorge]

ok just post the face, crop out the NSFW

@spice vigil can you do better? #🏞｜general-with-images message

[desert gorge]

it's very rare for me not to feel empathy and solidarity with fired workers, but Twitter employees and their supporters seem to make it so easy

[tall brook]

I have one that looks really detailed but it's uhhhh........

[hybrid mural]

Damn the artist double whammy of AI art coming out then twitter dying

[terse sky]

Hello everyone, please take your booster shots!

[rich fossil]

how come? you got covid?

[wind robin]

Fun fact: if you can monetize your stable diffusion creations and buy a new graphics card to do so, you can write the price off your taxes

[spice vigil]

just make sure to never admit to using that GPU for anything not-work-related, because then it's tax fraud 😄

[azure pebble]

https://www.bandlab.com/the_yodeling_yak

[rich creek]

Hey does anyone here use a VPN? can I ask what you use and do you like it???

[limber orbit]

I apparently got banned from Midjourney's discord and I have no idea why, didn't even post in the last few weeks and I can't rejoin it seems :/

[tawny plume]

Did anybody else get kicked out of the midjourney server?

[limber orbit]

I love how shitty discord UI is about bans/kicks, the server disappears and you don't evne get a message

[limber orbit]

lol, ok it's a discord issue then

[tawny plume]

Their server crashed or something

[limber orbit]

@hexed depot @tawny plume

[hexed depot]

Cheers

[terse sky]

♪and now remote code execution in your own body♪
awesom1 yay1

i made a funny song https://0x0.st/og39.mp3

[flat bramble]

is there a central hub or discord to talk about all AI generated stuff? Like music, text, sound effects, video, etc etc? I want to keep up with what is happening in all AI spaces but the only ones I really know about are image generators

[boreal ore]

#1034941531762733167

[grand scroll]

Hi there! 🔥 we have developed this prompt book for data visualization hope you enjoy it:❤️‍🔥 https://docs.google.com/presentation/d/1V8d6TIlKqB1j5xPFH7cCmgKOV_fMs4Cb4dwgjD5GIsg/edit?usp=sharing

[rich fossil]

good morning everyone

[blissful lava]

Poll on no code ai development https://www.reddit.com/r/artificial/comments/ypye9v/how_many_of_you_would_use_a_drag_and_drop_visual/

[fast oak]

I'm expending on my DrWho model, who should I include next, Matt Smith or David Tenant ?

[fast oak]

both it is, Day of the doctor

[modest sandal]

https://www.businessinsider.com/oculus-creator-palmer-luckey-facebook-meta-vr-kill-real-life-2022-11

https://tenor.com/view/monke-monkey-wth-wtf-bruh-gif-26418485

[wind robin]

stupidity knows no bounds

[hybrid mural]

Its just an office meme and everyones acting like its serious

[placid walrus]

Affinity suite 2.0 was released a few hours ago and is on 40 % sale right now. If you need an alternative to photoshop for retouching images without subscription and with a better ui, take a look at the deal

[mental sparrow]

it's standard procedure to disable employee's access before letting them know they're fired

[sacred harness]

Who's in Miami?

[noble cedar]

WEE WOO WEE WOO THERES A TYPO ALERT ON SLIDE 27

@novel beacon nitro scammer

also how do you ping every mod at once

[silk acorn]

.

[fast oak]

https://twitter.com/jyuu1010a/status/1590351095192506368?s=46&t=J-_PNY2ID-7Rrxtk3BS_kQ

ジュウ (@jyuu1010a)

振り向く

Likes

15846

Retweets

3676

▶ Play video

Twitter · Nov 9, 2022, 2:30 PM

How was this done??

[spice vigil]

depending on which bots aren't broken: react with ⚠️ , right click -> apps, or make a ticket in #1010934719455707218

[spice vigil]

⚠️ is "warning" btw

[true anchor]

Man I need to cool off, I just had a huge argument with a coworker on the phone and it's a long while since I was that angry. I usually never yell, but I screamed my guts out man.
It's gonna be awkward when he comes back from home-office, since he sits quite close to me 👀

I hate being angry, man :(

[weary kiln]

Hi! 🔥 I'm building a prompt keyword research tool!! Hope you enjoy it! https://promptrush.com/

[vernal bane]

bro i can pay for things like this really

it's really hard to come up with text

thx

[weary kiln]

happy you like it!

[vernal bane]

@weary kiln Sorry for the pin, but can you add something in the future? Some sort of shortcut for all the prompt that start from A, then B, then C, then D. Like a dictionary and have small number how many people like that one it would be. Really helpfully to keep a track of which one we have tried

like this i don't know

[weary kiln]

nice idea, i will add it soon

[daring coral]

Hey everyone. I'm hosting another https://aiartweekly.com/ cover challenge. The theme this week is "comic" as in comic books, Tintin, Stan Lee, anime/manga and so on and so forth. Winner gets $50 and gets mentioned in the next issue. Would love to see some cool Stable Diffusion submissions if anyone's interested in participating. Aspect ratio has to be 10:14. Submissions can be DM'd to me or publicly posted in this Twitter thread: https://twitter.com/dreamingtulpa/status/1591028639214469126

Cheers! 🙏

Dreaming Tulpa 🥓👑 (@dreamingtulpa)

✨ 4th #aiartweekly Cover Art challenge announcement!

💰 Winner earns $50 and gets mentioned in issue #9

🎨 Theme is "comic". Max two submissions per person. 10:14 aspect ratio. Rulebook in tweet below.

Leave your submissions below!👇

Twitter · Nov 11, 2022, 11:22 AM

[daring coral]

Cool stuff. Where do you get the image data from?

[weary kiln]

Thanks! i use multiple public sources

[sterile tinsel]

if I train only 100 steps, what rate do I need for any effect?

[fast oak]

?si

!si

[fast oak]

a high rate, something around 2 to 5 e-6, or even 1e-5. but the higher you go, the biggest chance there is it won't converge at all/not train correctly. 100 steps is really stretching it low, my baseline is 100 steps per picture in the dataset

[slate flame]

https://www.youtube.com/watch?v=pl0cc8eOe4o

[manic rover]

meanwhile i just wonder how many people bought the prints

[noble cedar]

we can never know for sureif this was AI generated or if the poster just attacked some random artist since we never got a close up

how can i poke holes in AI art in this state!

[manic rover]

I found the original tweet https://twitter.com/puccanoodles/status/1591897912719835136?s=46&t=9VNOpHmpXCLzolI40wDOfg

Marie Lum - AX Chibi I-1 (@PuccaNoodles)

anyway they have a statement printed with a bunch of bullshit but whatever

Likes

5909

Retweets

220

Twitter · Nov 13, 2022, 8:56 PM

How will AX know or detect ai once people begin lying about their process?

[noble cedar]

they dont profit because all proceeds go directly into the power bill

[woeful aspen]

What type of videos do I need to upscale for my Youtube channel

[fast oak]

They could run the raw images through an AI image detecting GAN

[light galleon]

Wow, the twitter hate against someone (and typical "ban it! ban it!" mentality)...

so glad I never got into twitter

Cool that they posted those signs, but people have already made up their minds walking up to the booth before ever reading them. 😐

The whole "I don't like it so I must prevent anyone else from it" thinking burns a rage fire in me that I can't stand. Ugh.

I'm not sure yet where I stand on the issue at hand, yet. Still learning and interested to see how it plays out. But stuff like this makes me root for the little guy. (And, the irony that most artist alleys I've seen are filled with fan art is sort of humorous.)

.. not sure why I'm writing this. Nobody cares. 😄

[wet zephyr]

Those are valuable thoughts that you are sharing. I definitely agree with you that Twitter is just one of those places online which really fuels on hatred and stuff.
Honestly, I don't get why people would treat AI in such a negative manner. It will cause temporary disruptions in various creative industries like with the invention of photography or the film camera but it's mainly just another tool which artists might need to consider adding as part of their workflow.
To make things truly impactful and meaningful even with AI image and video generation tools, Lots of time and manual labor are still required outside of the amount of technical expertise required to learn how to use that medium. It's just that these tools lower the barrier to entry and provide more diverse ways for people to iterate and bounce off of ideas.
Those images shared as part of the Majou Witch Project Statement provide a clear and concise explanation as to why experimentation and developments in this medium are important and I'm glad that you were discussing about the negative influence that Twitter in particular has towards these creators looking to push new ground.

[noble cedar]

and we have all seen how good those are already :p

[fast oak]

well the HuggingFace Space isn't entirely an accurate showcase of Tech + tech will improve

[hearty chasm]

fun fact current world population already crossed 8 billion

[fast oak]

https://tenor.com/view/taeuvre-squidward-squidward-shocked-squidward-break-gif-26165410

[sterile tinsel]

What do you mean "won't converge at all?"

I got 30 pics (though some are the same pics flipped out zoomed). I noticed training 50 steps my renders were already different, 150 they seem better, 250 they seem better again.

My machine sucks, so I'm just adding 100 steps per day, so I guess better to keep the same 30 pics and not add more, right? Maybe I should delete the flipped and zoomed pics?

[fast oak]

if you go small steps by small steps it should be good. high LR can lead to "NaN loss", breaking your training, and you then need to go back to a previous checkpoint and lower your LR a little. but it speeds up learning too so yeah, hard balance

[sterile tinsel]

but I figure, start with a high LR, then when it overtrains, lower the rate?

[fast oak]

if you can afford the maintenance yes

you are on a slow computer with low steps per day, so you can totaly fine tune where those steps are spent

[sterile tinsel]

and folk are telling me less images somehow trains better... less risk of overtraining?

So maybe I train only 5 pics at a high LR with small steps, then another 5 pics, then another?

[sterile tinsel]

I'm on CPU only

hence I'm trying 100ish steps, which takes me an hour

but right now I've got about 50 pics I've prepared for training (though some pics are cropped duplicates), so I'm wondering if 5 pics at a time is better than training all 50 at once...

[fast oak]

it's complicated. lower image trains better on lower steps and high LR, like in theLastBen. it's the fastest but not the most qualitative. you can get better learning for more pictures, but, in only 1 condition : every new picture needs to add something new and not repeat something that has already been repeated.
Like for a character, introduce new pictures, but change the most things you can : artistic style, hairs, clothes, ... more variety = better quality result/model, but longer training

[fast oak]

you should just not have 50 pics at all if you are training a single subject

an artistic style why not

but not a single character, not if there are repeats

like, don't use twice the same pic cropped differently, it will reinforce the exact same motif, and this pic will overtrain faster than the rest of the pics, resulting in half OK/half fried model in the end

the bigger the dataset,the most complicated it is to use

to give an example, my baseline is usually 7 to 12 pictures for a single character concept, depending on what I can find that is truly diverse

[sterile tinsel]

I'm not training a subject. I'm trying to make better anatomy. Training humans in general. So pics of beards, eyes, scars etc. trying to get better hands and hair

[fast oak]

hum ok

so yes then try to cover all the basis, just try to have changes as much as possible

around 50 seems like a good idea

[sterile tinsel]

so I want better abs, nipples, muscles, hair etc.

[fast oak]

💪

it's going to be long to train correctly but it should be possible

[sterile tinsel]

but I had some large images, and decided I'd crop e.g. the legs, the abs, the face of a single person, then a pic showing the entire person; labeling the jpegs appropriately

I figured that tells the AI "these are abs close up, and here are those same abs on a person"

[fast oak]

I would start with around 5000 steps on LR 1e-6 on that many pictures, but it would be on the low part of the spectrum I think, more around 10k

what do you use to train ?

[sterile tinsel]

auto1111 training on CPU

[fast oak]

ok so shivam

[sterile tinsel]

100 steps takes 2 hours

[fast oak]

so you put that in a multiconcept ?

like 1 concept for abs, one for beards, ... ?

[sterile tinsel]

I check "use filenames for for labels instead of instance prompts"\

which I guess does that?

[fast oak]

ok I didn't know there was this option in there now

[sterile tinsel]

but I'm probably doing it wrong

[fast oak]

no I think you are doing it right if you named them like you said, "close up shot of abs", "picture of abs" ...

[sterile tinsel]

I think doing that, I can named one jpeg "abs" and another "man's face" and it'll automatically train the concepts?

[fast oak]

ho no it needs a separate txt

so if you have a picture name "XXXXX.jpg" you neeed "XXXX.txt" with the prompt in it in the same folder

[sterile tinsel]

yes, then the full person is

man with abs, man standing, nude man.jpeg

[fast oak]

nope, a lot more simple than that

man with abs

man standing

nude man

those are 3 different prompts

don't confuse DB

don't put the same token 3 times in the same prompt

[sterile tinsel]

those are all in the filename

so.... I'm messing up

[fast oak]

yes, check the advice I pointed

you need the prompts in a separate file

and you need to make shorter prompts per picture

[sterile tinsel]

so the filenames don't do it?

[fast oak]

the screenshot you posted said to use "filename.txt"

[sterile tinsel]

because there are 2 checkboxes, one says "use filenames" and the other says "use text files"

[fast oak]

not to put the prompt in the filename, but in a txt

ho

then try use filenames sorry

yeah it should work the same then, but still, don't put "man" 3 times in the same title

it will reinforce the token so much more for that specific picture

[sterile tinsel]

oi, this is what I want \

[sterile tinsel]

ah, so it'll start making the character too much like my picture just if I use the word 'man' right? and it'll burn out too fast

[fast oak]

it would be a little (not exactly) like putting more weight on the word Man. Think of the prompt as ... a prompt that SD would use to make your picture. when SD sees the word man with higher weight, it puts a man in front, and forgets the abs usually. So in your case, it would learn maybe that a MAN with lots of weights means a zoom on the abs, or something

it requires so much test that takes so long it's hard to build my knowledge fast enough to really give the best tips

even on my GPU that can churn that up quite fast, learning sessions can take a full day to come up not learned enough yet

but I love that thing ^^

[sterile tinsel]

ya, well that's exactly why I'd rather know what I'm doing and not experiment, because on CPU it takes forever

once I really know what I'm doing, I'll leave my machine to render while I'm away at work

but I don't want to use my CPU up for 11 hours and come home to learn I trained something wrong

although I wonder now.... maybe it makes sense to train the inpainting for specific things..

like, train the in painting only on curly hair, and make a model that's very good at giving people curcly hair

[modest sandal]

prediction- Trump will not only announce his candidacy for 2024 tonight; he will also be announcing his new-found partnership with YouTube star and philanthropist Mr. Beast, aka Jimmy Donaldson. Mr. Beast's future videos will be published on Truth Social. Trump's heart will grow three sizes with all the good public deeds and philanthropy he does (which the public, initially full of shock and criticism, will eventually warm up to) and he will win the 2024 election in a landslide as a member of the newfound Beast Party with Mr. Beast as his vice president. Trump gains a positive name with gen Z voters and Mr. Beast consistently nets billions of views per video with his challenge videos involving high-level government officials. Truth social becomes the national social networking app. Twitter goes bankrupt and Elon Musk commits suicide by blasting himself into deep space in a SpaceX rocket

[fast oak]

Uhm sure

Sounds like it could be a movie

[manic rover]

ads on the sd reddit be like

several pointed out that this website looks like its using the NAI model and im skeptical about the legality of this anyway

[dreamy rover]

The devs of NAI are here

[hasty willow]

Lol how the hell do they call that free 🤣

[manic rover]

Waiting 12 hours for just 1 picture seems extremely weird 😂 I understand not all “free” services are free but..

[manic rover]

Oh the devs of the site also made some google extension that imports your generated photos from NAI and uploaded it to the site(?) https://twitter.com/kyoshi07675798/status/1590910627161440256?s=46&t=t1uJIh9m9hYH4rP0fIj6Nw

kyoshi (@kyoshi07675798)

PixAI-NovelAI Artwork Sync

Directly upload your AIartworks to https://t.co/SlkX2rBOnS from the generation page on NovelAI
NovelAI の生成ページからAIアートワークをhttps://t.co/SlkX2rBOnSに直接アップロードします

▶️https://t.co/KZQn3toz7R
#pixai #novelai #AIArtwork

how to use🔽

Twitter · Nov 11, 2022, 3:33 AM

[hasty willow]

Lol what

[fast oak]

Some kind people can tell me how to use AI painting in the community

[void gust]

open photoshop, paint something (dont forget the colour), go to your webui, img2img and type a prompt, profit

[hasty coyote]

wazzzupppp

When installing Stable Diffusion with Linux running into error No module ldm.models.diffusion.dpm_solver.solver, where's the mirror? Thanks!

Adobe is turning into a subscription nightmare lol dunno how you can afford that

[dreamy rover]

Affinity Designer, Photo and Publisher are launching their 2.0 version for a huge discount on all three programs

You buy it, you own it.

[light galleon]

Yeah I have v1

Thinking about buying v2

Any big improvements you've noticed?

[noble cedar]

forgot color, summoned glurglurg the terrible

[void gust]

Will never buy arrgg

[dreamy rover]

Cope, seethe, mald. They're decent devs.

[void gust]

Cope, seethe, mald. You pay for my free copy, forever

[dreamy rover]

UI's definitely easier to see. Changed a few icons but yeah, doesn't.

[void gust]

[dreamy rover]

In all seriousness, my v1 version has lasted at least 3 years, pretty decent product.

Also, Adobe sucks, and you have 0 chances of getting nasty spyware from getting free stuff

[void gust]

Adobe does suck, but you do not get nasty spyware if you already know from where to get your resources

and if you don't, well, piracy isn't made for the convenience of the user, but for the inconvenience of the corporations

[mystic flame]

Is there a opensource neural network for generating music like stable diffusion?

[void gust]

you wish. Music companies are like Dobbie holding onto the sock

[boreal ore]

HarmonAI

[hasty island]

Heloo how can i find "nai_0.95 model.cptk"?

[wind robin]

you can't find it

[boreal ore]

You dont

[hearty chasm]

why twitter shutting down

[daring coral]

Hey folks. Issue #9 of https://aiartweekly.com/issues/9 just went live. I'm hosting another Cover Art challenge this week. 1st place wins $50. Theme is "forestpunk". Images can be submitted here: https://twitter.com/dreamingtulpa/status/1593582489699643392

Dreaming Tulpa 🥓👑 (@dreamingtulpa)

✨ 5th #aiartweekly Cover Art challenge announcement!

💰 Winner earns $50 and gets mentioned in issue #10

🎨 Theme is "forestpunk". Max two submissions per person. 10:14 aspect ratio. Rulebook in tweet below.

Leave your submissions below!👇

Twitter · Nov 18, 2022, 12:30 PM

[spice vigil]

because musky boi bad 😄

[oblique oak]

is twitter actually shutting down? or is that just a figure of speech or exxageration or something

[midnight fjord]

https://www.youtube.com/watch?v=8ksyuQG8j9g

[oblique oak]

cause tbf i have the widely unpopular opinion that if twitter was gone it would be a largely good thing

[midnight fjord]

interesting story, after seeing the video look at the first comment

[spice vigil]

I mean, all twitter offices are closed and all badge access is suspended so... read into that what you will.

[fast oak]

Vileshaft, you can discuss this if you want with everyone on here, but please stop saying the same things in multiples channels if this doesn't seem to create discussions, people have seen it and discussed it a lot around already in the past few days.
You can talk and discuss all you want, but everyone follows the same TOS and spamming will get you reported by other users and timed out after a point

[oblique oak]

well; my idea of whats going on (im choosing to ignore the hype and im doing my reasearch) is that it wont shut down but it'll probably become significantly smaller. people have said this kind of stuff about youtube a few times and nothing bad ever became of it all; and historically a lot of other catastophic predictions never came to pass on a huge variety of things. the only reason i think it'll downsize is because of the amount of employees leaving which is affectively already making it a smaller company

but its overdramatization to say it'll just shut down entirely

as much as i wish it would personally

[spice vigil]

I on the other hand will happily continue ringing my bell and wearing my sandwich board

The world Twitter is ending! 🔔

[oblique oak]

lol. well even if im right about it becoming really small it may aswell end in that case. nobody uses myspace anymore, and facebook is dying too, i think twitter will be the same. people will move on. they always do

[spice vigil]

Thing is, when facebook and youtube make decisions that lead to that kind of talk... they spend a lot of effort predicting the response & planning...

Everything we've seen so far regarding the takeover is that it's a huge shitshow lol

[oblique oak]

well, im seeing a trend actually that social media in general is somethign people are getting a bit tired of. they are finally waking up to how bad it is with gaint corporations essentially using people as their product, basically, and im seeing more and more people on all the major social media sites complaining about the sites themselves. this cant go on forever; i believe wholeheartedly that social media will eventually be replaced by something else. my hope is that it gets replaced by something less sinister, but my fear is that it'll be replaced by something far worse. its probably atleast 3 years away though before we start to see the signs of that more strongly (by which i mean more people start to see what i already do)

[slate flame]

I made this thing years ago... It basically makes Stability.AI logos. lol
http://inspiralizer.com/inspiralizer/

[slate flame]

I'm getting really into antisocial media

[haughty grove]

did anyone here claim the free games on Epic? It gives me a captcha to solve, but I know for a fact I solved it, but every time it says captcha failed.

this is new, never got that crap before. is this how they're gonna reduce the number of freeloaders like me? why not just stop giving free games instead of pissing people off?

[dark swan]

Is it actually shutting down?

[haughty grove]

a lot of people wish. I don't use it but I hope nothing bad happens just to annoy the sort of person upset about Twitter's new management 😂

the seething on reddit and HN is glorious

[upbeat forge]

Thinking of doing a quick hack project this weekend. Do you think people would pay like $0.10 per request to call a stable diffusion model by SMS?

[haughty grove]

Usually it takes years to see who's right, and we kinda forget...but if Twitter keeps working fine through the World Cup and then holiday season, with most of the employees laid off...then Elon was right. In 40 days 99% of HN/reddit might eat humble pie. They can't memory hole this, not even in their own mind.

[haughty grove]

no? why would anyone do this instead of using their web browser?

[upbeat forge]

Because they don't want to sign up for a service like DreamStudio or learn how to prompt engineer?

[haughty grove]

What kind of person doesn't want to sign up to a web service but will sign up to an SMS service? Someone who just time traveled from the 90s?

[upbeat forge]

You might be surprised.

[haughty grove]

Anyway, I gave my answer. I don't know a single person who uses an SMS service, except for receiving notifications from their clinic, flight, etc.

what country are you in?

maybe it's a local thing

[upbeat forge]

USA.

[haughty grove]

maybe some Americans can comment. It's just not a thing in the countries I've lived in.

[split temple]

Data plans can be stupid expensive in parts of north america

[upbeat forge]

It's basically an easy parlor trick that folks can show family/friends while congregating for Thanksgiving. Not really designed to make it cheaper.

[fast oak]

Predatory practices on this yeah... Same for internet. I pay 50 for symmetrical fibre optical with no cap and about 8MB speed, but that would cost an arm and a liver in the states

[haughty grove]

and those people dont have wifi?

look, this is going nowhere. clearly you feel there is demand for an SMS service. I don't. Just do it, it's not like it's gonna cost you much. Don't overspend on the server because I dont think you will make more than $5 exclusively from people who know you.

[limber yacht]

how would the img result be sent? mms causes low img detail, rcs is new, or back to webpages again?

[oblique oak]

make an app. 😛 i agree with audiogus on that idea; nobody who uses sms stuff is in the same generation as people who are interested in SD

but if you wanna do the sms service thing, nobody is stopping you. you just asked for an opinion and we gave it to you

[haughty grove]

12 minutes to Qatar-Ecuador. Let's go Qatar!

[noble root]

https://youtu.be/WmNr0OBlESg

[strong copper]

how to skin the model?

[fierce ridge]

Hello everybody!

I am Zsolt Szabó, a student of automotive design at Moholy-Nagy University of Art and Design Budapest. Please help me to develop my thesis by answering the questions. Thank you!

Here is the link:

https://docs.google.com/forms/d/e/1FAIpQLSdCD7za1mxKFNcYTuYalqoNsXAi_Dk2EB5aHjtLxsEXnQcimQ/viewform?vc=0&c=0&w=1&flr=0

[true anchor]

Just a couple dozen more minutes and I'm at work for 24 hours working for a deadline... rip

[junior shale]

where can you get the best of these thingies for good results and how exactly does it even work?

[true anchor]

Those are the models. They come as a .ckpt file and most of them can be found on huggingfaces. Just put it in your models/Stable-Diffusion folder, restart the WebUI and it's there. Different models are trained on different datasets and thus produce vastely different results even with the same prompt.

[junior shale]

and where can you find the best ones?

or maybe even make them

[true anchor]

making them requires an insane amount of VRAM which you only get with very high-end GPUs. The best models can be found here, just sort by Downloads or Likes:
https://huggingface.co/models?other=stable-diffusion

[junior shale]

so making them myself is not an option i see

btw is getting multiplied limbs and body parts on my ai generated images normal and i should fix it or is it sth that shouldnt be happening

[true anchor]

You can rent a GPU for quite cheap, but doing something like that will require quite some experience

[true anchor]

Yeah that can happen, especially with horizontal poses. Try generating the person in a way that looks "the right way up", that way SD has the least trouble with it. You can always rotate it later. Or you use the inpaint tool to erase any unwanted limbs.
My lazy method for humans so far was to generate a batch and pick a result which looks nice, but maybe has wrong anatomy or whatever. Then I pick that seed and run another, larger batch with just slight seed variation. There's a high chance one of those will turn out looking "correct"

[junior shale]

what does seed variation do?

[true anchor]

it basically changes the seed very very slightly for each generated image. That means the content of your image will stay relatively the same, but will have subtle changes to it. The higher your seed-variation, the stronger the differences will be. I usually start out with a low value like 0.05 or 0.1

[junior shale]

so like for example u make a batch with seed 333333 as a reference and you like one newly generated image from that which has seed 333315 on it's own so u click on that image and generate batch based off of that?

not sure what u mean by 0.05 - 0.1 yet

[true anchor]

Check out your outputs folder - every image has its seed saved in the filename. So you can look up the seed from inside windows explorer, copy it and paste it in your webui interface. It's usually set to "-1" which means it's random, just paste your desired one in there. To the right of that is a checkbox, I think called "seed variation" or something. If you enable that you see a new slider pop up which defines the variation strength, for which I usually use something like 0.05-0.1

[junior shale]

oh i see it now

but what if there are multiple seeds with same numbers?

oh and this is a whole different thing right

u meant mostly this one

[haughty grove]

same number = same result. seed is just a way to add randomness so every generation isn't identical. same seed + same model + same config/prompt = identical image on every PC.

so usually you use the seed as an "always random" factor to generate new images, once you like your prompt

[junior shale]

oh i had problem with this identical image part

[haughty grove]

yeah, set to -1 usually (always random), unless you have a specific reason to reuse a seed (for example wanting to compare different samplers)

or wanting to improve an old image you made 2 months ago

[junior shale]

i did a test run when i first installed and couldnt get the same image, guide said i did everything right and it was due to by GPU being odd one

[haughty grove]

the Hello Asuka test?

[junior shale]

yes

[haughty grove]

what's your GPU?

EVERYTHING has to be identical: resolution, prompt, CFG scale, seed, etc

[junior shale]

gtx 1050ti

[junior shale]

yeah i made sure that wasnt the problem

[haughty grove]

I got a 1060 and got the same image. maybe because you're using command-line arguments that can change things

I assume you needed some args to run on a 1050

[junior shale]

the guide had different asuka images as for ideas what could go wrong and i found this under image i was getting

[haughty grove]

ah, that would be it then

[junior shale]

still not sure how to do this properly cause they didn't go into detail

[haughty grove]

it doesnt matter, as long as the images you do generate look fine

[junior shale]

so it won't cause any problems then?

also sucks i dont even know whats the root of the problem

i dont think it's vram but who knows

[haughty grove]

just generate regular images, and see if they look fine

are you using SD 1.5 or another model?

[junior shale]

not sure how to check if 1.5 but pretty sure it's sd

[haughty grove]

get prompting, and don't worry about it

SD 1.5 is a hard one to start with, especially if you want anime-style characters. You'll have to find example prompts.

[junior shale]

it's called prompting?

[haughty grove]

txt2image = writing prompts = prompting

there's no official terms really. "image generation"

[junior shale]

oh i thought u mean like program called prompting

where u even find these prompts? u just come up with them yourself?

[haughty grove]

that's not for SD 1.5, that's for the forbidden model we're not allowed to discuss here:)

but yes, you cant go wrong with that

[junior shale]

oh yeah i see

[tight comet]

check these pages if you haven't already https://lexica.art/ https://proximacentaurib.notion.site/proximacentaurib/parrot-zone-74a5c04d4feb4f12b52a41fc8750b205

also for anime the waifu-diffusion model is very nice.

[junior shale]

thanks will check it out

[brave wolf]

@idle sigil as you were saying?

[idle sigil]

Yhea no spoilers for me, im going to go watch it blind, like i absolutely want to go in with no expectations.

The ONLY spoiler

i have seen

[brave wolf]

But yeah it would depend on if the gene was recessive or dominant

[idle sigil]

is the end of the re-release of the first movie were they showed like 3 minutes of the 2nd movie

[brave wolf]

If it is dominant than the whole family would have 10, if recessive, it could be a mix

[idle sigil]

it would be cool if they have 2 children 1 with 10 and the other with 8

[brave wolf]

Typically, things like number of fingers and other I guess what you would call core anatomy, is dominant. People born with mutations to their hands of limbs don’t generally carry it over in their children

The reason this is the case is that our bodies will always try and ditch faulty genes when making a child. So if your dad is passing down a faulty gene, the babie detects this, rejects it, and adopts the mothers version of the gene when growing.

This is actually why incest can lead to deformities, because from brother and sister share the same faulty genes from their parents, there is no healthier gene to adopt

[idle sigil]

interesting, but then...in this case Jakes DNA was specifically engineered this way so it's basically considered to be Healthy and perfect. So how would it know wich is the faulty DNA for the baby?

[brave wolf]

His DNA is mixed with Navi DNA, this human DNA would absolutely be detected during gestation and likely rejected by fetus, so only his navi DNA would be accepted.

Technically, it should be very hard for jake in his semi-human/semi-navi body to have children.

[idle sigil]

But of course....it's a movie

[brave wolf]

Right, I’m just going off what I understand about genetics, this is just speculative fun

[idle sigil]

I'm still working on that Neytiri model, it's still WIP but im thinking of re-training it, currently its trained at 16K steps on 5e-7

But i've learned new things since then, so im hopeful that i can improve it

[brave wolf]

I might have to poke your brain some time about that, I’d love to take a crack at creating my own model

[idle sigil]

i aint no pro at it or anything tho, im still learning and im still not sure about some of the settings, i just know some of the basics

[brave wolf]

I’d like to create an alien model. Not of model of Alien from the Alien franchise, but more abstract and open alien forms. I think SD by default uses the whole Area 51/grey man shtick.

I want to see if I can create a mode that will do aliens more like Mass Effect or Stars Wars or Startrek styled, mostly humanoid but still very different looking aliens.

[idle sigil]

You probably can but you'd need to find the the images of exactly what you are looking for and tell it, "this is what Keyword is"

it would also depend on your initial model, with dreambooth you have to choose a base model to train it on

in my case

i didn't choose the SD 1.5

i chose WD because it gave me more pose control and by saying stuff like "red dress, tree, arms up, happy" it already knows based on the Danbooru tags what it means

and even if it was an anime model, it came out great, i do get lots of realistic results

[brave wolf]

F2222 is also good for that I believe as well

[idle sigil]

Well at that point i havent even heard of F2222

[brave wolf]

I believe F2222 is a highly photorealistic (especially of faces) NSFW model. I have a version of it that’s combine with Novel Ai and/or Waifu Diffusion. Has some really strong outputs

I then generally use Arcane Diffusion to retouch the face to keep a realistic yet stylised face.

[idle sigil]

i just do WD or Furry DIffusion, since well Na'vi are technically cat people plus it provides lots of NSFW poses too. i usually just generate at clip skip 1-2 then upscale the images using img2img to 1024, if that doesn't look good or whatever at 512 i do img2img with inpainting.

Tho i have used SD 1.5 to train Neytiri, it doesn't provide much in terms of posing and creativity, at least for me

[brave wolf]

I would discourage using clip skip, especially for realistic faces. While you can sometimes get results that look better with clip skip, you will statistically get more good looking results not using it than using it.

[idle sigil]

Well here's the thing, considering i probably overtrained my Neytiri model, i get more better and creative results at clip skip 2 than 1.

and some people actually recommend clip skip 2, sometimes it gives that bit of extra creativity in some cases

[brave wolf]

Ah, makes sense I guess if your compensating from overtraining

[idle sigil]

Yhea... tho i do have backups

at 6k, 8k, 10k, and 16k, steps

in this case clip 1 looks good, but in other cases.....i need the 2

this was 2

same seed

Well, i'm leaving it here for tonight. im probably going to train Neytiri some other time when i have the time or the....urge to do so?

[brave wolf]

O I see

It’s not that clip 1 was bad, more that it’s too human

[hot crypt]

Hey mate @blissful solar

I remember you working on models and doing some cool stuff on here before, funnily enough I encountered your name again.

Just incase you didn't know you may get a kick out of seeing people like them using your models for neat stuff https://youtu.be/QBWVHCYZ_Zs?t=225

if you did know.. my bad lol

[blissful solar]

Thanks for sharing!
Yeah I knew about it since Nico reached out to me a few weeks ago.
Great video and amazing work by the corridor team 😁

[hot crypt]

ahh cool haha

yeah great vid :)

[drifting basalt]

I've been showing that video to friends interested this kind of ai. I enjoy that they show slightly more in-depth perspectives from vis dev and industry artists beyond the typical media spins you see. As an illustrator, I've been over the moon excited about this technology since playing with GauGAN. I remember having my mind blown when Adobe introduced content-aware fill. I see these as invaluable tools to add in my creative workflow.

[fast oak]

https://beta.character.ai/p/P5wDqVoOORqbitBWPZH4ScB6LURx4J7Bv36oEGFKfMw

[fast oak]

how do you train a model?

[hot crypt]

Yeah it's great

content aware fill was definitely like magic too when it got introduced haha

[fast oak]

Speaking of Corridor Crew, did you guys see them test that new "Neural Pathways" thing? It seems VERY similar to AI-based image stuff, except maybe more like based on "light fields"

[junior shale]

whats the image max size i could go for with 4vram?

[gentle temple]

@rigid hedge If you guys need help with the game of thrones project 🙌 😉

[dry sluice]

Not sure if it's the right spot for that, but some colleagues and I have worked hard the last months to set up a German non-profit organisation to basically research modern technologies like AI and VR.
If the language isn't too big of a barrier, is love you to follow our IG account: https://www.instagram.com/p/ClYSbKrKe4v/?igshid=YmMyMTA2M2Y=)

virtuellekultur

Virtuelle Kultur wurde 2020 als Projekt von Steffen Rümpler und Lucas Rees in Freiburg ins Leben gerufen, um Fragestellungen rund um die Digitalisierung unserer Gesellschaft und den Einfluss von technischen Strömungen auf den Kulturbereich mit gestalterischen Mitteln zu erforschen. Im März 2022 wurde der gemeinnützige eingetragene Verein Virtuel...

Instagram

I'm posting that here, because one of our first activities is going to be a local exhibition (and a book) showcasing images generated by SD showing images in the style of famous classical fine-art artists. Like used in the exhibition promotional material;

[hasty furnace]

https://media0.giphy.com/media/RUIIKcoWgbkYOqO3Fp/giphy.gif?cid=5e2148861f96821c8846540744b6c13f832559f6611ecfa3&rid=giphy.gif&ct=g

[sacred wyvern]

This project is almost finished. I am working on an Intro track and Outro track. For a kind of mental and musical closure so to speak. This is a dark journey I must warn you. This isn't for everyone. But nonetheless, Enjoy.

https://soundcloud.com/allen-polley-official/sets/a-night-with-a-jester

[rare sky]

hi

[fast oak]

Where would I ask for help getting the style of a specific character?

[tight comet]

is there any good alternative to gpt3 coming out? it would be good to have a open source one.

[fast oak]

idk bc im knew too, but https://lexica.art has a CLIP interrogator so you can copy+paste image link or upload image and it'll give you back a bunch relevant prompts and images

Nice

[versed bobcat]

Saw this on reddit and deadass thought it was AI art. Found myself counting the number of fingers and being impressed the number was right 💀

[cerulean dune]

im jealous of MJ

if they didnt have such a strict nsfw policy i would have bought it in a heart beat

[hot crypt]

mj is weird shit

they got rid of their tokens apparently and raised the shit out of the price lol

I do keep hearing about their new version though

👀

[fervent willow]

Can y'all hear your own voice in your head?

[fast oak]

Yes, although just barely over everyone else’s

[fervent willow]

Some people don't have it

[fast oak]

NPCs

if you cant hear your own voice in your head, you did not get a soul

thats just facts

[fervent willow]

That's mean

[fast oak]

its just facts.

[modern talon]

I am running Auto1111 on Google Colab.
Whenever I try to generate high res image like 2048x1152, It stops loading halfway and gives me this error

Content Encoding Error
An error occurred during a connection to 0b0d2ada81ab457c.gradio.app.

Does anyone know how to fix it?

[weary kiln]

🔥 Search millions of prompts by color, orientation, width and height.
https://promptrush.com/i/random

[proper folio]

man..

[thin spoke]

@proper folio thats why you should toast it ..... sticks together better

[proper folio]

im not

gonna toast

brioche bread

[thin spoke]

grill then

hahahah

[proper folio]

i have 2 rooms

live on 4th floor

nty

cNt even afford a grill

[thin spoke]

computer for AI .... check
Grill ......... naaa

[proper folio]

idk hoe to use ai

i wanna make porn

but i dont have the link

ragagagaahagagahahgs

[thin spoke]

join unstable diffusion

[proper folio]

how

[thin spoke]

youll have to search discord ..... i dont want to post it here

ill pm you

[proper folio]

just post it in dms

amazing

[normal zodiac]

https://medium.com/@thewayteller/the-watching-chapter-1-3-the-watching-begins-f25e82e6d4e7

[noble cedar]

HOW

[proper folio]

it fell

[noble cedar]

did the plate have too much angle of attack

[proper folio]

😭

[noble cedar]

did yiu somehow sprinfboard the sandwich

what happened here

[proper folio]

no i slipped

[noble cedar]

zad

is that baloney

[proper folio]

no

salami

its basically a

brioche bread salami ketchup sandwich

[noble cedar]

zam

salamis good too

i had baloney before

tastes like hotdogs

[proper folio]

im muslim

i fucking HATE pigs

@noble cedar do u own a switch

[noble cedar]

nop

[wise sedge]

Wait, so modern neural networks are basically just a bunch of matrix multiplications, right?
Which in turn is just multiplication and addition...
So the system that generates big booba anime waifus is a set of equations with elementary school math containing billions of terms 😳

[wind robin]

Could be, depending on the activation function. Idk which activation function SD uses

[whole current]

There’s a lot of SiLU layers from when I’ve looked into it

And idk if I would call a discrete convolution elementary school math

[wind robin]

Technically it could all be done by elementary school students if they worked overtime… that is, except for the SiLU. That’s a bit more complicated

[hexed depot]

New gpt-3 model out

Wonder if we are close to a gpt-4

[main cedar]

You're on to something, you should write a binary value substitution system for it where the only 2 numerical values are "8008" and "8008135"

[restive kraken]

I killed an Asian Tiger today

Asian Tiger Mosquito

[manic rover]

unintentionally funny

[fast oak]

Wouldn't be surprised

[manic rover]

I can’t say much for midjourney but with niji journey (anime Novel ai type competitor?..) they are definitely catering to the Chinese market yep

[vocal vault]

that's so sad and scary

[paper cloak]

Who has used gpt-3 davinci ?

[hexed depot]

Yep. There’s a few channels on the openai channel dedicated to it now

Really cool stuff.

[manic rover]

Any information about Laion allegedly including images of other people’s medical records? Was this removed in 2.0? https://arstechnica.com/information-technology/2022/09/artist-finds-private-medical-record-photos-in-popular-ai-training-data-set/

[noble cedar]

how does this even happen

[fast oak]

lol no kidding

[sick dune]

ah, just for fun - I asked the new GPT-3 davinci text generation model
"write a short essay about discord users complaining about off-topic postings:"

Discord users often face the issue of off-topic postings and complaints about it. Off-topic postings disrupt the flow of healthy discussion and can be very annoying. It is important for users to be aware of the etiquette of the platform and to understand that a certain level of respect is expected.

When users are discussing a particular topic and someone posts an off-topic comment, it can derail the conversation and ruin the experience for everyone. The offending user should be made aware of the issue and asked to refrain from posting off-topic messages. If this issue persists, the user should be banned from the channel in order to maintain the quality of discussion.

In addition, Discord users should take it upon themselves to inform other users about proper etiquette. If someone is posting off-topic messages, users should politely explain why it is inappropriate and ask them to stick to on-topic discussion. This is a great way to help keep the conversation on track and prevent further disruptions.

Overall, off-topic postings can be annoying and disruptive. Discord users should be aware of the etiquette of the platform and take it upon themselves to inform other users when they are posting off-topic messages. This will help maintain the quality of discussion and ensure that everyone has a
(token limit reached 🙂 )

painfully true 😇

[winged hatch]

Needing a help. I bought a rtx 3090, 24gb. Using the automatic1111 gui. Now stable diffusion is not following what I write in the prompt. The results are not even like what I wrote. For example. If I write Woman, it shows cars, trees, etc. But not a woman. In the last gpu, same gui, everything was working as expected.
When I try the 2.0 model, it works. It doesn't work as expected in 1,4 and 1.5 models only. Using xformers and not, tried both ways. Any clue?
Thanks

[paper cloak]

Stop posting off topic subjects! 😂😂 JK

[sick dune]

"write the top 3 best practices to deal with off-topic postings in discord:"

1. Establish clear rules: Establishing clear rules and guidelines for your Discord server is essential for ensuring that conversations stay on-topic and don't drift off into tangential discussions. Setting expectations up front will help to keep conversations on track and prevent off-topic postings.
2. Use Mention or Role tags: Using mention tags or role tags (e.g. at-everyone🤪 ) is a great way to bring attention to an important conversation and keep everyone focused on the topic at hand.
3. Use Moderation Tools: Discord offers a variety of moderation tools to help manage conversations. These tools can be used to delete off-topic posts and mute users who are consistently posting off-topic content. Additionally, moderators can use Discord’s “slow mode” feature to limit how often users can post in a channel, which can help reduce the amount of off-topic chatter.

so which one is this?
"list the 10 most annoying discord off-topic postings:"

1. Spamming
2. Trolling
3. Posting memes
4. Posting irrelevant content
5. Asking for free items
6. Posting personal information
7. Posting inappropriate content
8. Posting fake news
9. Posting NSFW content
10. Posting long, rambling messages

#7 and #10 i guess 😇

ok but i promise to shut up for now, this is just soooo incredibly cool 😉

[keen imp]

Who's playing NovelAI?

[icy harbor]

Anyone here have a karagarga invite?

[limber yacht]

it is, but it's not really available like sd

[worldly hill]

Guys, what do you think of using GTP-3 to write short stories? Like as a freelance job

[paper cloak]

https://www.reddit.com/r/midjourney/comments/z921l8/i_created_a_short_story_using_gpt3_davinci003_and/ This is something I did before

[white glen]

https://www.youtube.com/watch?v=YreDYmXTYi4 neuralink show and tell livestream

[grizzled pier]

New Collection released: Themed 2023 Calendars

https://opensea.io/Galwin?tab=created

[upper gyro]

Any good AI cropping tools out there?

[daring coral]

Hey folks. AI Art Weekly Issue #11 is out: https://aiartweekly.com/issues/11. Anybody tried to utilize ChatGPT to produce SD prompts yet? It's quite cool. I'm also hosting another cover art challenge with "dreamscape" as a theme in case you want to participate: https://twitter.com/dreamingtulpa/status/1598635082247540737

Dreaming Tulpa 🥓👑 (@dreamingtulpa)

Hey #aiartcommunity 👋 I'm looking for submissions for the next #aiartweekly cover! Winner earns $50 and gets mentioned in issue #12.

🎨 Theme is "dreamscape" (dreamlike landscapes). Max two submissions per person. 10:14 aspect ratio. RT required.

Leave your art below!👇

Twitter · Dec 2, 2022, 11:07 AM

[fast oak]

I use this one https://github.com/arenatemp/sd-tagging-helper

not sure if that's what you meant

[worldly hill]

thoughts? made with gtp-3

[stark marsh]

https://media.discordapp.net/attachments/512112686020952065/1048250365818183761/Screenshot_20221202-075918_Chrome.jpg

[fast oak]

lol

[urban juniper]

playing an ai game got this as an ememy

[slate flame]

🎨 GRAND THEFT WHAT?
https://youtu.be/szZdS-Np1ww

[terse sky]

very nice jpeg cleaner without using neural nets https://github.com/ilyakurdyukov/jpeg-quantsmooth clever

[fast oak]

Anyone got some good advice to do inference on a model without using python? Is it possible with javascript or c++? I could setup a flask web environment but that is a hassle. What is the standard in the AI industry? Just query a server that runs on linux/python?

[midnight tusk]

#🏆｜winner-gallery message wonder how far away we are from being able to put some atoms into machine and then AI will generate new tasty food for us with completely new tastes 🤔

[whole current]

holodecks first, then replicators

[dark needle]

Examples of my non AI work 😅

[wise sedge]

While looking at Emad's Twitter my screen cut this tweet off and I thought it was funny

[grizzled pier]

New Collection released: Themed 2023 Calendars
Currently have Kitten, puppies, fantasy, Pagan, and Christina themes. Set up as monthly so folks can mix and match as liked.
https://opensea.io/Galwin?tab=created

[upper gyro]

Does this crop the background?

[fast oak]

No, it just crops as square

[upper gyro]

Ah I see, I am looking for something that can crop the backgrounds completely and still keep the character intact

[fast oak]

They weren't private.

[noble cedar]

https://www.engraved.blog/building-a-virtual-machine-inside/

what the

[whole current]

this is some weird chinese room turing test shit

[whole current]

welp googling didnt help me and chatgpt cant

[normal zodiac]

https://medium.com/@thewayteller/the-watching-chapter-1-4-wyrds-made-word-ba4af4a23c0a

[south elbow]

Anyone here who's used ChatGPT and found it actually useful for something?

[runic shoal]

well i asked it to write a limerick about stable diffusion 😃

Here is a limerick about stable diffusion:

There once was a process called stable diffusion
That kept things from turning into confusion
It helped atoms to spread
In a way that was stead
And the results were quite smooth and fusion!

[modest sandal]

https://www.axios.com/2022/12/05/adobe-ai-made-stock-images

This boils my blood

AI image generators should be marking the end of stock photos

An awful stain on society

Pure cash grabs that offer nothing of substance

Hoops through copyright to make creative works

I hope all stock photo companies go bankrupt

[wet scarab]

ah

so when a big company does it people are fine with it

but when individuals use it it is "art theft"

[slate flame]

https://www.youtube.com/watch?v=2fM1wlgCaxQ

[wind robin]

This is the best joke I’ve heard in a while. I need to see more

[wind robin]

I tried it myself turning GPT into a python repl. it didn't like me using libraries, but if you break the repl immersion for a bit to pip install dependencies, it works great!

lol

[wind robin]

lmao, it can even kinda run my programming language

[wind robin]

It's also really good at converting freehand inline math to latex

[robust bough]

Are the rainbow-like patterns that appear in some AI-generated images caused by photographs with chromatic aberration being in the training data?

[meager epoch]

I read recently about a free open source code generator model, anyone know what its name is?

I thought stability was working on one, too

[scarlet nymph]

My friend’s sister denied me when I asked her out since she is in the same grade and my friend is in the junior/11th grade class

[wind robin]

A little too off topic

[raven flint]

made AI think too much....

[wraith gorge]

Hi, I have a question about text to speech services. Did anybody discover a text to speech service with voices as natural sounding as the ultra realistic voices from play.ht? Before I drop the 100$ per month, I’d like to know if there are more affordable competitors with the same or better quality. I haven’t found one yet!

[pallid portal]

I don't know if you'd be looking for this, but Dragon NaturallySpeaking is great if you're visually impaired. I originally got it a long, long, LONG time ago due to blindness/cool state program, but it can dictate and read. I don't know how well/good it is compared to anything else in the market rn, but the home edition is $200.

[wraith gorge]

I’m sorry I wasn’t clear before, I was talking in the context of generating speech sounding as natural and realistic as possible. I don’t think Dragon’s software is about that.

[pallid portal]

It can read pretty naturally

laughs

[wraith gorge]

I believe you, but I want to do video voice overs

[pallid portal]

At least, back then, so, just a suggestion if you don't want to do a service.

[vital ermine]

Anyone else here steadfastly refuse to use MJ, but still feel a strong sense of MJ-envy? 😄

[terse sky]

:
for the record i believe in the holocaust, if i didn't i'd go to jail

[tight comet]

why would you refuse to use it if you like it?

Personally I really like what MJ has going on, but I find SD more useful because I can get things closer to what I actually want with it, but they are both good.

[vital ermine]

Several reasons. 1) lack of control. They juice the prompts to produce interesting output but it rarely actually matches the prompt. 2) lack of a user interface 3) ridiculous censorship

4. pay to play

5. lack of privacy

[tight comet]

oh yeah lol similar reasons to why I prefer SD, I think MJ produces great images the best ones I see are made by MJ but if it doesn't make what I want it doesn't make much sense even if the image is cool.

[vital ermine]

Exactly

I’ve seen insanely cool images that in no way match what was prompted. I don’t just want cool images, I’d like to actually have them represent my vision. I can do that in SD 1.5 fairly well, but overall quality pales in comparison to MJ, and don’t even get me started on the castrated SD 2 model.

[tight comet]

yeah I don't fully understand the reasoning behind SD 2 nudity censor, I know they said it's because of kids which I would be ok with banning nudity if that's the case, but cant the AI tell if you are generating a kid with certain prompts?

anyway it's not something I would want to support so maybe SD had no option when it comes to stuff like this.

The AI still has a ton of limitations and will generate a lot of things you might not want but as it becomes better there will be better ways to approach this issue.

[vital ermine]

I’m not worried or concerned about censoring porn out. But the removal of artists and styles is ridiculous.

It all represents Stability.ai protecting themselves.

And I guess I would rather they be protected than get shut down. It just feels draconian.

[tight comet]

yeah I was also going to mention artists and styles but that's also a complicate thing, sincerely artists should be able to opt out.

tho I think it's a lost cause because anyone can train the AI on the style they want...

[vital ermine]

Yeah, that is a little tricky, but at the same time, not really. Artists put their work online to be seen. Before a person ever sees it, it gets catalogued by Google and others. The ai sees those images just like you and I do. It’s not pkaguerising or replicating them, it’s learning from them just like a human would do. If a person can study an artist’s work and then reproduce it as an homage, then there’s no reason a computer can’t do the same. If one artist can prove that their work has been literally plaguerized by AI, then they have a point. But I have not seen anyone able to produce evidence of that.

And for what it’s worth, I say this as an artist whose work is included in the liaon dataset.

I’m not a famous artist or anything, but to me, having my work included in the model is posterity.

And I am willing to bet Greg Rutkiwski has never been as well known as he is right now, and I also bet his sales went up when he started bitching.

His style is wildly derivative too

As all art is

[tight comet]

I have seen that argument before but the difference here is you have a large corporation getting your image training an AI on it and profiting of your data without your permission.
It's very different from an individual trying to figure out how to make good art.

[vital ermine]

And what does Google do?

[tight comet]

Google does not include your web page if you don't allow.

[vital ermine]

Why is it ok for Google to put your images into a database a serve them up along side ads?

Yes, but nobody who wants exposure would do that.

So it’s ok for Google to profit off of artists images, but not others?

Laion only contains public images.

If people can see them, so then can machines.

If someone doesn’t want their images to be seen by machines, it’s really easy: don’t post them on the internet.

Can’t really have your cake and eat it too.

[tight comet]

Sorry but those are not equivalent, you might be ok with your image being in one service and not in another.

[vital ermine]

Public is public though. If a person can see it, then so can a machine.

The laws around what is infringement are clear and they are not only designed to protect copyright holders. They are also designed to protect creativity.

[tight comet]

You are assuming because someone saw an advantage in allowing google to use their image they are allowing any service the same, you can even choose which search engine will get your images, AI doesn't own the data, and it being public doesn't mean you are being given the rights over the image to do whatever you want with it.

[vital ermine]

And that isn’t necessary. All the machine did was look at the image.

We’ll see how the laws evolve but right now, there’s nothing illegal about what has been done.

With Infringement claims, the burden of proof lies with the person making the claim. If they can’t prove infringement has occurred, they can’t win the claim.

I can go to a gallery, take photos of a painting, go home and create a perfect replica of the image, and as long as I did so from scratch, I am protected.

Also as long as I don’t attempt to mislead anyone into thinking it is the original.

By the way, every single image is copyright protected, not just those created by famous artists. So all 2.3 billion images in the dataset unless they are Creative Commons or public domain would be considered infringement if what you are suggesting is true.

[tight comet]

There are so many issues with this

for one there is even private medical data in the database

it's not legal to store any of that and training an AI on images of private medical data and photography of patients is sincerely very messed up, so this is not just art, it's not because it's on google that you can use.

also if you copy something that is copyrighted without permission yes you can be sued, try to make a serious comic book with superman on it and see what happens.

I don't get the example of going to a gallery and making a copy, yes if you are selling it of course you can get yourself in trouble.

personally I don't have problems with the kind of image the AI makes, as long as it's not a copy of someone work it's fine but is weird that a company is allowed to develop software using copyrighted and private images.

I'm not against the development of AI to generate art and I think it's unavoidable that AI will replace people, but it's important to protect privacy and the wishes of people who own the data, if someone doesn't want to be in the system it should be easy to opt out.

[river geyser]

So with corporations invested in Ai, could we see them using their Ai(s) to create games?
Could we see them incorporate them into games seeing as the minimum specs for hardware are getting higher and higher all the time?

And if they're using Ai models to make games. Could we see a streamline effect where games are coming out every 6 months rather than every 3-5 years?

We saw Ubisoft take this approach with Assassin's Creed when they were releasing a new title every year and it had a negative impact.

Could we see them taking the approach of rather instead using the Ai to create DLC or add-on content for their games as a continued service?
So rather than making a new game expand on the existing one.

[vital ermine]

1. Private medical data should not have been posted online publicly. The Laion dataset did not hack into people's private servers/accounts to gather images.
2. You used the word "copy". Yes, if you take a copyrighted image and reproduce that image, that is infringement. However, if you use that image as source for a novel drawing/creation, then you are not infringing. There is a big different between something that is a registered trademark and something that has a copyright. "Superman" and the Superman logo are trademarked. That is not the same as creating derivative art and/or homages.
3. You are absolutely, 1000% incorrect here. You can absolutely attempt to reproduce an image and sell it. You can also attempt to mimic the style of a famous artists to a T. As long as you made it from scratch and did not embed any of the original image into your reproduction, you are fully protected.

Also, personal medical data is only protected if it is tied to a name or other identifying information about the patient. Random medical images that do not bear personal identifying information are not HIPAA protected.

[tight comet]

https://arstechnica.com/information-technology/2022/09/artist-finds-private-medical-record-photos-in-popular-ai-training-data-set/

[vital ermine]

Try to sell an image of Mickey Mouse in his original incarnation and you will get a cease and desist letter from Disney, at minimum. But diverge away from that original incarnation sufficiently and you are protected.

This image is protected:

[tight comet]

yeah but you talked about making an exact copy and selling

[vital ermine]

I did not say exact copy

What I said was "I can go to a gallery, take photos of a painting, go home and create a perfect replica of the image, and as long as I did so from scratch, I am protected."

An exact copy would mean a literal reproduction of the original. Recreating it from scratch by looking at references is protected.

Let's pretend the Mona Lisa is not public domain. I go to the Louvre, take photos of it, go home, setup a canvas and recreate it PERFECTLY, using my own canvas, paints, brushes, etc. Protected.

Copycat works are fully protected, with the only exception being if any attempt is made to defraud someone into believing that it is actually the original.

[vital ermine]

This is the fault of the doctor, or whomever allowed that HIPAA data to be leaked onto the public internet.

[tight comet]

"Substantial similarity, in US copyright law, is the standard used to determine whether a defendant has infringed the reproduction right of a copyright. The standard arises out of the recognition that the exclusive right to make copies of a work would be meaningless if copyright infringement were limited to making only exact and complete reproductions of a work. Many courts also use "substantial similarity" in place of "probative" or "striking similarity" to describe the level of similarity necessary to prove that copying has occurred."

[tight comet]

So I can store that data if I was not the one who leaked it?

I'm not from the US but it doesn't sound like what you are saying is reasonable, by your logic I can copy literally any work and sell.

[vital ermine]

No, you are just not understanding

You cannot make a literal copy of anything that has a copyright.

That is infringement.

[tight comet]

you were talking about a perfect reproduction of an image.

[vital ermine]

But if you create something from scratch, the claim mujst prove that it bears substantial similarity, which is not defined. It is intentionally not defined to allow for such concepts as homages and fan art.

Dude, how many times would you like me to explain it again. A perfect reproduction that I created from scratch is what I said. And again, I was referring to copyright material, not trademarked material. There is a significant difference between those two.

Going back to the Mona Lisa, if I paint a version of the Mona Lisa from scratch, that is not a perfect reproduction or copy. That is an homage. It doesn't matter how close to the original it looks. It's an homage, I made it from scratch, and it is protected.

I made this. It is a "copy" of "Drawing Hands" by M.C. Escher. It is fully protected by copyright law.

[karmic wave]

Most fan art work is not legally permitted, and most likely infringes on someone’s copyright. Much of the legality of fans’ work is up to the person or company who owns the original work.

Sorry but what you are saying is completely wrong, for some reason you are assuming that if you manually copy something you own it.

[vital ermine]

Sorry, you are failing to understand the concept of fair use.

[karmic wave]

I think you don't understand what fair use means... making a copy of a work and selling it is not protected by fair use.

[vital ermine]

Yes, if I create a novel image that is directly derivative of another image, I do own the copyright to that image, but obviously not to the original image.

[karmic wave]

You are extrapolating it, you mentioned multiple times that you are allowed to make a perfect copy but because you manually made that copy you own it.

[vital ermine]

I can also directly reproduce the style of any artist

And nobody can stop me or file an infringement claim against me

[karmic wave]

Style is not protected that's a completely different topic.

[vital ermine]

A style cannot be copyright protected

Using the drawing hands example above, if I did the same exact thing using pencil and paper, it would also still be fully protected.

[karmic wave]

Copyright is literally the right to copy, you are not allowed to copy an image you don't have the copyright for, I don't see how you came to the conclusion that you own the image because you manually copied an image.

[vital ermine]

I never once said I own the image. Don'

t be dense.

I own the copyright to any image I CREATE

I can draw the mona lisa right now and I immediately hold copyright protection over that drawing.

[karmic wave]

I think you really don't understand what copyright is
this is completely false -> "if I did the same exact thing using pencil and paper, it would also still be fully protected."

[vital ermine]

Yes, I know you think that, but you are wrong. This is not my first rodeo. How many copyright infringement claims have you filed in your life?

HOw many of your images have you had to go to court over to protect?

What is the source of your implied expertise on this subject?

[karmic wave]

Dude you are just making things up, show me in the law where it says you can do what you mentioned above.

[vital ermine]

No. I'm not making anything up. I have actually studied all of this many times. The law is pretty clear. Not sure which country you live in, but in the US, all of this is pretty straight forward.

[karmic wave]

Again show me the source.

[vital ermine]

How about you call an IP lawyer and discuss with him rather than asking me to teach you?

I've already paid that bill enough times to not need to do it again.

Anyone can file an infringement claim over anything they want. But it is up to them to prove infringement has occurred. There is no clear definition for "substantial similarity" so each and every case is handled independently. If the copyright office determines that "substantial similarity" exists, then the claim will be heard.

Perhaps you are confused and think when I say "Make from scratch" it means like put the original in a copy machine or something?

[karmic wave]

Is this "Let's pretend the Mona Lisa is not public domain. I go to the Louvre, take photos of it, go home, setup a canvas and recreate it PERFECTLY, using my own canvas, paints, brushes, etc. Protected." not substantial similarity?

[vital ermine]

It would ultimately be up to the copyright office to make that determination. And of course, nobody would ever be able to make a truly perfect reproduction from scratch, so fine, remove that word from what I wrote.

If something is so well reproduced that it is indistinguishable from the original then yeah, that would likely meet "substantial similarity".

But a pencil on paper drawing of "Drawing Hands" could never attain that level of similarity unless actual copying had occured (i.e. say you traced the original onto tracing paper).

But Escher, nor Escher's estate own the rights to the concept of two hands drawing each other.

You are basically arguing the line between homage and infringement, and that's a fruitless debate because there's no hard rule or definition to clarify when an homage becomes infringement. It is handled on a case by case basis, and the burden is on the copyright holder to prove the infringement claim, which can only be raised for something that has a registered copyright. I think you're also mixing up some terms. Attempting to claim an homage is an original is fraud, not infringement. Attempting to claim the actual artwork someone else made as your own is plagiarism, not infringement.

[manic rover]

This photo and paper will definitely make the rounds on angry twitter posts in the next couple of hours or days https://arxiv.org/abs/2212.03860

[austere forge]

Anyone know any text2speech AI's that could be used to create models based on given audio sample and utilized to create audio files based on given text spoken by AI?

I know about Whisper, Coqui and xVA Synth

[sweet ravine]

wtf happened here

i just want to make anime women

[chilly glacier]

chatgpt saying that it cant reply in spanish, then the very next thing it did is replying me in spanish 💀

[wind robin]

chatgpt > deepl ???

Translate "結局怯えていた世界の果てが来なかったゆえ、安心しておりました" to English

"In the end, the end of the world that I was afraid of never came, so I was relieved."

ah, there is the fake news they keep giving disclaimers about

the year 1562 is incorrect, as it predates the founding of the United States by over 300 years.

[tight comet]

another cool thing about chatgpt is you can ask it to explain the grammar of that sentence and further question it and get examples from it and so on, it's a lot more powerful than an ai that only translates.

[wind robin]

It's kinda funny. I tried with 果 instead of 果て and it said "fruit of the world" so it really is looking at each and every word, making sure it's correct
I also told it to interpret a python program and it was like "no, you forgot a colon after the else, so your syntax is wrong"

[whole current]

if chatgpt had the math accuracy of wolfram alpha it'd be amazing

Altho It can craft wolfram alpha prompts for you so...

Although this may be the single most complicated answer I have ever gotten out of wolfram alpha

[wind robin]

ChatGPT writes its own diss track

Large language models, they're a real pain
They take up space and cause us all strain
They gobble up resources like they're on a binge
Leaving us with slow computers and a lot of cognitive dissonance

They claim to be smart, but they're not
They just regurgitate what they've been taught
They can't think for themselves, they're just a tool
Spewing out words that often sound like fool

They lack creativity and true understanding
They're just a mimic, a poor imitator standing
They may be impressive in size and scale
But when it comes to real intelligence, they'll always fail

So let's ditch these large language models
And embrace true human thought and creativity, don't you think so?
We don't need them, we're better off on our own
Let's leave these machines to rust and moan.

[fast oak]

@leaden dew Go fuck yourself.

[fast oak]

who?

mind your own business

[fast oak]

jeez

[whole current]

I want chatgpt in one of those home speaker assistant echo alexa type shit devices