#💬｜general-chat

1 messages · Page 18 of 1

gaunt tide
#

Oh do I have to do the like, git clone link thing or whatever?

rich sand
#

i mean, i know my CPU is slow but i really don't wanna pay for these online processing stuff

empty terrace
#

Yeah that will work, I have a browser widget for git.

#

...but I have no idea where I got it.

gaunt tide
#

Not working :/

#

Dunno how to do it

noble basin
#

@rich sand yes i know, but sometimes it cost less money .. if you wait one week than your laptop finish the models and imagine you doing a mistake 🤣

#

anyway do what you want

empty terrace
#

holdon I'll do a quick test.

umbral kraken
#

im absolutely fascinated by what can be done, ive been like a kid in a sweet shop the past 2 days, just cant stop playing around

rich sand
rich sand
umbral kraken
#

yes briefly. saw a video earlier about how to get more out of it, and even use that to help in generating prompt ideas.

#

Just feels like this is only the beginning and the possibilities are going to be endless

rich sand
#

today we can generate an unique customized image that has never seen before

#

things are scaling up fast lol

astral goblet
rich sand
#

but interesting story also

astral goblet
#

he's a sort of influential guy that asimov

umbral kraken
#

does feel like the start of a matrix type scenario, where if you can think it, u can create it. Like in VR, you could create any object and have it there in front of you

#

scribblenauts did it first lol

astral goblet
#

more like the world described in ready player one i think. where massive scale content is crowd sourced

#

the book rather than the movie described it better. the sequel not so much

#

Daemon and Freedom™️ touch on it too

rich sand
#

I think stable-diffusion should be re-written on an low-level programming language

#

Python is really slow when doing complicated math

signal nest
astral goblet
#

thats where the heavy lifting is

rich sand
signal nest
astral goblet
#

pytorch is the library that would bottlneck things if you're concerned about that

rich sand
rich sand
signal nest
#

having said all that I would actually like the pipelines in a low level language for more convenience launching it from custom tools (eg i want to feed rendered images from my engine into img2img) but thats still all possible through other means... it just so happens that organizing all the data and pipelines for ML has been easier for that community in python

signal nest
#

it is what it is.. the convenience of python for organizing workflows has won in that domain

astral goblet
#

python is easy to iterate on. good for rapid development

#

low level code when things move so fast, the timing for that optimization just isnt right yet

rich sand
signal nest
#

definitely

astral goblet
#

i'm sure stability has a lot more in the labs then just what we see

signal nest
#

rust is my favourite language these days.. but i must admit the pytorch bindings in it look horrible :/

rich sand
signal nest
#

but implementing a Rust native ML library and replicating a diffusion model could be nice, but you'd really want to replicate StableDiffusion exactly so that you can just load the same models

astral goblet
#

Rust is primed to dominate all fields, but a big part of it is there's no culture of development. You realease code in rust and few people use it.

signal nest
rich sand
astral goblet
#

Rust right now is like HDTV when tv's were HD Ready and not real hd

#

it's obviously the right move but there's just not enough tool infrastructure yet

rich sand
signal nest
#

I've been using rust on and off since 2014 - i'm actually pleasantly surprised that it's got a lot further with acceptance than I thought it would

astral goblet
#

linux kernel supports it now

#

likely will face total conversion

signal nest
astral goblet
#

c++ doesn't have perpetual motion. it's very much winding down

rich sand
signal nest
# rich sand maybe we can have that in the future

it would be interesting to ask around in the community and see if you could get collaborators for such a project - set out with the explicit goal of replicating StableDiffussion - i'm sure there'd be some interest . Just dont expect it to actually speed up the process. Rust is nice from a correctness perspective aswell as speed. Its interfaces are more solid.

#

heh maybe ChatGPT will help people rewrite their code in rust ..

rich sand
astral goblet
#

The memory safety features of rust are insane. it's absolutely going to convert everything. Anything remaining on old models will become legacy software

#

It's like when fuel injection engines were invented. Who uses carbs anymore?

rich sand
river maple
#

i'd never prototype a ML model in Rust

#

Sure, that's partially because I don't know rust, but also because I don't want to waste time fighting with a low-level language when I'll spend most of my time calling some ML library anyway

astral goblet
#

a lot of it is dependent on projects like pytorch. that's what you'd want to replicate if i'm not mistaken

rich sand
river maple
#

it's the same reason I don't write my webapps in C

signal nest
signal nest
rich sand
signal nest
#

anyway I would definitely have liked the ability to run SD from rust directly..but there's far easier ways to acheive what I was after

river maple
#

Strict typing for ML?

signal nest
#

(i jst want to render things from my game engine and feed them into 'img2img' , similarly i wanted to try alternating diffusion passes with projecting results back onto 3d meshes .. all that would be a PITA in python. but some kind of rust / python interop is far easier than re-implemnting pytorch in rust lmao)

astral goblet
#

wrappers are a thing

river maple
#

metacall

astral goblet
#

first result on google for pytorch rust, i have no idea what it is

signal nest
#

yeah those are the bindings iv;'e seen. they probably work fine. i was just commenting that sadly bindings can look ugly because an API tends to be designed around the features of a particular language. Rust has its own tools, and doesn't have things like default parameters.. hence bindings end up just looking ugly , sadly. still easier than rewriting pytorch i know

river maple
#

Honestly, C/C++ and now Rust aren't popular in the ML space because they're pains in the ass.

As it stands, most ML stuff tends to be integrated with some kind of network service or just done for research

signal nest
#

yes you're right - one solution I did consider was making a server in python and just throwing jobs at it from rust

river maple
#

I'd rather a subset of Python ("Python for ML") that can be compiled to a binary

uneven wasp
#

Well for ml, sop is to run it on the gpu so whatever language you're using is essentially a cuda wrapper

river maple
#

just built-in pytorch and numpy

uneven wasp
#

aka the speed of python doesn't matter that much. throw in quick dev time and great libraries and anything else isn't really worth it for researching models

river maple
#

you don't need most of the language anyway

astral goblet
#

when people can just download this and run it, like it's discord or winzip, it'll get nuts

river maple
#

winzip
damn am I back in the prehistoric age

signal nest
#

i conceed python is indeed very good at "setting up workflows".. just throwing data around

#

its a good "utility language" as i'd call it

river maple
#

oh yes

#

I want machine learning in Go

#

Server-grade machine learning

empty terrace
river maple
empty terrace
# river maple >three colors two?

No, my first personal computer had cga.
We had two color palletes Cyan/magenta/white or Swamp green, Turd brown, and bright red... but you could only display one pallet at a time.

river maple
empty terrace
#

most of the programs I saw signalled it as off.

astral goblet
#

i remember those days. what ugly colors

signal nest
#

the best 8bit machines had nicer graphics than CGA (c64)

river maple
#

PC sucked at graphics for a long while, in my opinion

astral goblet
#

commodore and amiga machines used their own graphic standards. cga was from ibm pc

signal nest
#

yeah 16bit days I had an Amiga, was very smug until Doom came out..

astral goblet
#

amigas kicked ass and had hardware palette cycling and shit

#

heh doom came out "WHAT NOW?!"

#

my first was commodore. i think i was 4 or 5 years old and knew how to type LOAD GAME or whatever the command would be

kind flint
#

hello
Do you think it will be possible to run stable diffusion properly on amd cards in the future? I'm stuck between 3060 for $250 or 6700xt for $300 and want to use stable diffusion.

astral goblet
#

its possible now with ROCm on linux

river maple
signal nest
#

strangely Stable Diffusion has spiked my retro nostalgia - the prospect that retro-games can be upscaled & re-imagined. I would like to make something in a middle resolution that could be downscaled to real retro hardware, or upscaled & detailed further with a diffusion model.

astral goblet
#

first time i saw that was someone used an old anime upscaler to redo all of the backdrops in final fantasy 7. i was thinking "thats smart af"

#

SD can add detail instead of just upscale though. but then you run into the potential issue of too detailed textures on low quality shapes, can look noisy

signal nest
#

right I imagine it'll take a lot of tweaking. Personally i've been drawing some "retro style" things (not quite pixel art but not contemporary game art quality either) and then throwing them into img2img with a prompt randomizer and I find the results interesting. i'm sure with some work on the prompts (and a finetuned model ?) you could take almost any retrogame and give it a new lease of life

empty terrace
river maple
#

AMD flounders in ML

#

I have higher hopes for Intel than AMD

lilac reef
empty terrace
#

I think partly, It's just not a priority for them.

kind flint
#

Since the power difference between the 6700xt and the 3060 is not small, it is difficult for me to choose, what would you buy?

signal nest
river maple
empty terrace
#

I mean, I'd look for a used 3080

lilac reef
empty terrace
#

but if you can't find a good used deal the 3060

signal nest
#

I used C++ most of my career. I classify rust as the best language, however its not the magic bullet its proponents think, you do pay with dev time for the safety. But you need that if you want perfromance and internet-facing

astral goblet
#

rust is disruptive. people have to relearn everything. this angers them.

river maple
#

rust is too explicit for me

kind flint
river maple
#

i'd be pleased with C with a stricter type system and a built-in slice type

empty terrace
river maple
empty terrace
#

IF you want to do ML stuff, then I can't recommend AMD right now. I wish I could...but I can't.

signal nest
#

maybe zig or jai will catch some people resisting rust. i also think herb-sutters "cppfront" is an interesting idea. But I've put the time into rust now - and the underlying language design is nice IMO. I like how it organizes programs, tagged-union pattern matching is awesome, its great for concurrency

river maple
#

I use Zig already

#

It is basically what I want out of a low-level language

signal nest
river maple
#

Zig is basically extended C

#

As a C programmer, I picked it up in a couple days

kind flint
river maple
#

and don't switch languages. just use whatever's easiest/best for whatever task you want.

young spruce
#

Im learning C#. wish me luck

river maple
#

OH YES

#

Stable Diffusion in C#

#

I'd totally do ML in C#

young spruce
#

not for stable diffusion. im learning as hobby

river maple
#

I figured, but I just had to say it anyway

signal nest
#

(never touched c# but everyone i know who uses it likes it. I am a "no-GC" snob however lol. C++/Rust and dabble with python as a utillty language and for the ML frameworks)

river maple
#

Unless you need the bother and performance that comes with manual memory management, I suggest just accepting GC

signal nest
#

maybe pytorch bindings look less horrible in thigns like C#

river maple
#

Trying to do high level stuff like pytorch in low level languages will always be a horrible experience

signal nest
river maple
#

for game engines, I suggest reference counting

signal nest
fervent thunder
#

@kind flint Just shell out for a new 3060. Best value you're going to find right now for a decent GPU for SD.

kind flint
#

yea i think i will

empty terrace
#

Avoid the 40 series like the plague.

fervent thunder
#

Anything in the 4000 series is going to be better than 3000 series for gaming and CUDAs and probably memory.

empty terrace
#

that's not really true.

fervent thunder
#

I stand corrected?

#

Explain

empty terrace
#

The pricepoints relative to the performance are comically bad.

rocky rock
#

What is the prompt for aspect ratio? 16:9 and the like pease.

fervent thunder
#

Oh, yeah. Of course they price thing out of the common price-rance with new stuff

signal nest
empty terrace
#

even for new product the 40 series is baaaaaaaaaaaaaaaaaaad.

#

I mean if you are determined to buy a 40 series, save up for a 4090

signal nest
#

16gb in the 4080 = better for finetuning i'd assume (or is 12gb enough?)

empty terrace
#

vram is king

fervent thunder
#

@empty terrace I haven't really looked up the specs, I just assumed they've only got the 4090 going atm

#

Definitely VRAM is what you want, but CUDA cores help too

empty terrace
#

if you are just looking at it for ML applications, there are 3090's with 24gb of vram

signal nest
#

essential for finetuniung, otherwise 8 is enough

fervent thunder
#

It's ram heavy

empty terrace
#

Anything where you are training a new thing (finetuning,embeddings, inversions,new models) needs lots of vram

signal nest
#

if you just want to generate images using existing models - 8gb is enough. but its possibel to re-train the models to customize them to be bettrer for certain themes, or teach them new objects - if you have 16-24gb(? min is?)

fervent thunder
#

But, doesn't mean an old card with lots of ram will be as good.

empty terrace
#

cheapest way to get bulk vram is to buy refurb NVidia tesla Mseries cards.

#

you can get a 24 gb card for 400USD that way

#

but it's not good for anything else

fervent thunder
#

I heard it renders quite slowly, despite that

empty terrace
#

just machine learning or big math science

#

yeah

#

it's a dedicated training card at that point

#

and you need to figure out cooling, because it's built for server airflow

#

anything 30 or 40 series with 8+gb of vram is good for generation

fervent thunder
#

If you can afford a 4070 ti, get it. I don't see why not

empty terrace
#

the 4080 and 4090 are both better deals.... still not good deals, but at least less actively insulting than the 4070

fervent thunder
#

It has 7,680 Cuda cores on it. Looks sweet

#

AMDs capability layers aren't cutting it and I'm sad

signal nest
#

i would have been happy with 4070ti (perf/watt matters to me) . for light use used 3000 is probably better I guess.

fervent thunder
#

I'm not in Europe, but I expect the power agenda will come here too

empty terrace
#

They don't talk stable diffusion

fervent thunder
#

Gamers Nexus is pretty good

empty terrace
#

It's probably because they are brand new, and almost no one is buying them.

fervent thunder
#

It's just how it works -- Brand-new=Expensive

empty terrace
#

There's a difference between 'expensive' and 'wtf nvidia?'

fervent thunder
#

Yeah, I'm not a huge fan of the company. They're like Intel

river maple
#

unless you get at least 1 SDITs, the GPU is no good

signal nest
#

is there a revised chart of that with the latest releases

fervent thunder
#

@empty terrace, You must hate me for what I said earlier

#

But, you do need to broaden your horizons

#

I'm not looking to make enemies

empty terrace
fervent thunder
#

lol, good. Same, I'm getting too old for petty hate

#

We all have bigger fish to fry

#

I hope Gamers Nexus's never stops growing

#

hair*

sonic carbon
#

is there a subnautica TI or DB or HN?

empty terrace
fervent thunder
#

Hello everyone

empty terrace
#

Hello!

outer crescent
empty terrace
#

The smut models aren't my bag, but I think they have a right to exist.

#

It's also functionally a mirror for a lot of the stuff on hugging face.

signal nest
#

regarding copyright: Would there be any problems publishing a list of timestamps and text descriptions of the images of a film - without publishing the images (.. why? - to take stills from the film and use the text to finetrain a model)

thorn hare
#

what exactly is the meaning when someone talks about "several generations" in SD ?

astral goblet
#

ianal but i've been experienced with IP law, as any creative should

karmic finch
narrow sluice
fervent thunder
#

I can't imagine that any of them could have any kind of copyright claim.

astral goblet
#

if you're creating a marketable product with ai generation and calling it "art from bladerunner" then yeah they could potentially have a claim. even if it is called "swordrunner" and looks entirely like bladerunner, a copyright claim coudl be made. Transformative is a big part of it. Intent is a big part of it. there's all sorts of caveats and is why copyright is usually a case by case basis. Indexs to a dvd though? no question. no problem.

fervent thunder
#

Yeah, but people aren't doing that. It's mostly inspiration from film and other things. There's nothing they can do otherwise.

astral goblet
#

people are doing that. you can't sell fan art technically. people do.

fervent thunder
#

Who are these people? And I'm sure they're taken down swiftly.

fervent thunder
astral goblet
#

you can take inspiration from stuff yes. but if you're publishing it for profit, with the intent of it looking like a copyrighted work, a court might find you in infringement. there are lots of ways to protect against that. some people think the "everyone else is doing it" defense will work. it won't. one day fan art comission artists will have lawyers coming at them, as soon as how much money is made gets noticed

dull stratus
#

Hi! what is the best way to train a self model for free?

fervent thunder
#

I mean all Anime/Manga looks the same, anyways. What difference does it make?

river maple
river maple
karmic finch
fervent thunder
#

Cultured? lol

#

I've appreciated some manga artists, but it all looks the same now

#

Also, why why did Civit send a trolling email to samdoesart?

karmic finch
#

i think its on his video channel?

river maple
#

i'm not touching anti-AI youtube

karmic finch
#

don't then, this ain't an airport

fervent thunder
#

All these artists took their inspiration from someone else. Look at Blizzard, their whole gamedev has been stealing other ideas and making it their own and have made billions.

karmic finch
#

inspiration is fine

fervent thunder
#

Inspiration can't be copyrighted

karmic finch
#

i kinda just don't understand why people defend AI so much overall though specially on those practices as you mentioned

#

if anything none of you will find jobs thanks to AI alone though

fervent thunder
#

Good

karmic finch
#

its mostly corps that will probably pay 1 guy to write the prompts for sketchy ads

fervent thunder
#

The internet and tech changes constantly

#

I've lost my job to the third world many times

brisk bloom
fervent thunder
#

Nice pad he has going.

astral goblet
#

"man of culture" is a 4chan meme

#

it means you're someone who enjoys the anime boobies

river maple
#

like 70% of memes are 4chan memes

fervent thunder
#

lol

river maple
#

the other 30% are a mixture of reddit, twitter, and 9gag

astral goblet
#

i dont give 4chan that much credit. most of their memes are something awful memes

fervent thunder
#

4chan used to be the furnace for a lot of internet culture, but hasn't been in over a decade

#

But, I'm happy to see this guy lose his grift

river maple
astral goblet
#

corporations are going to leverage ai to flood markets with content and devalue them, so that only corporate branded content can be trusted. brandpower! that's my guess anyways

#

they'll likely fund huge ai mills

river maple
#

BRANDPOWER™️ commercials?

#

anyway, I don't care, because I won't need corporate content then

#

I'll make my own AI content

#

infinite AI-generated anime episodes

astral goblet
#

yeah just like i don't have a cable subscription. but a lot of people still do. a whole generation will have to die off before centralized content goes away

river maple
#

inb4 someone says "interacting with people online will be difficult because you won't know who's real or an AI"

  • I can interact with people physically, in person
fervent thunder
#

Cats out of the bag.

astral goblet
#

it'll be easy. "Is you or is you isn't an ai imposter??"

fervent thunder
#

No turning back

astral goblet
#

same way you catch chat bots now a days. Throw it a curveball question

#

only gullible ass people will fall for the fake ai humans, but.. we know how many fo those there are. lots

fervent thunder
#

That's another thing -- How much of social media is real?

river maple
river maple
#

people literally don't exist until they connect with me over an encrypted tor chat server

astral goblet
#

social media is a fad 🧓

supple zephyr
#

Anyone have a preference for stable diffusion for generating coherent text?

river maple
#

sorry, you'll have to wait for deepfloyd to get that

astral goblet
#

i generally roll the dice and if i see any amount of text i'm like "ooo "

#

"it made text that time soo neat"

empty terrace
#

It is currently of questionable utility, because I included too many 1800's medical textbooks, now it just generates pure nightmare fuel.

#

Like it's vivisecting HR Gigers worst nightmares and taking very meticulous notes

supple zephyr
#

that sounds fun lol.
I've on rare occasion get it to generate coherent text with large amounts of guidance & effort. Prompt editing for like 10-15 minutes with small tweaks just to get the tiniest changes. Eventually I can get what I want, but man is it tedious sweat

supple zephyr
empty terrace
supple zephyr
#

let's see them lol

empty terrace
astral goblet
supple zephyr
#

Reminds me of attempting to make blueprints of humans lol

empty terrace
#

I'm thrilled with how the text is turning out, but it slaps gruesome imaginary organic spare parts on every output in graphic detail. and no negative prompt can correct it.

#

I'm going to spend some time this weekend adjusting my dataset.

#

then retrain it

queen pawn
#

What GPU is better for render, RTX 3060 12gb or RTX 3060 8gb TI

north snow
#

does anyone have the server banner picture ?

empty terrace
#

VRAM is king, and the 3060 8 gb is kinda weird in terms of memory bandwidth

#

especially if you want to do any training.

outer crescent
#

has anyone had a look at Karlo?

fervent thunder
#

3060 12gb coming in soon

#

will that run dreambooth?

queen pawn
#

i dont will traning is only for render

empty terrace
#

you can run pretty much anything on 8gb of vram, but if you want to do any training, 8gb is limiting.

#

Embeddings/inversions/etc are all training tasks

outer crescent
#

theres optimization guides and git repositories

queen pawn
#

so for render images on stabble diffusion 3060 8 GB TI > 3060 12 GB?

astral goblet
# queen pawn What GPU is better for render, RTX 3060 12gb or RTX 3060 8gb TI

put it this way, if it were vram vs speed, vram always. you can do larger generations or bigger batches. better for training. ML's domain is certainly vram capacity for the time being. i'd rather have slower (but not too slow) higher capacity vram than faster 8gb. simply put, get the 12gb, or 16 or 24 even better

tawdry ferry
#

I'm using a 3060 with 12GB right now to generate 2048x1472 output and it's currently using ~10GB of VRAM according to the InvokeAI console debug messages. So for me, at 3060ti with only 8GB is not going to cut it. harrisonford

vapid tangle
empty terrace
#

There is no such thing as 'enough' vram.
I've got a refurbed 24 gb server card just for training, and I can still hit CUDA OOM errors.

placid hatch
#

Anyone notice that their outputs are suddenly way worse lately regardless of what model they are using?

astral goblet
placid hatch
#

persists across resets and seeds

#

and clean install

astral goblet
#

define "way worse"

#

for me, sometimes i get memory corruption and it looks like images are half generated with bad cfg. so i just reboot

placid hatch
#

does not generate the same images off the same seeds as before an update, and it is results generally do not end up as expected for the models that I typically use.

fervent thunder
#

how can i use a1111 to create multiple images at once is there anyway

outer crescent
astral goblet
#

if you don't have enough vram, you can do number of batches and it'll do 4 one after another, instead of all at once

outer crescent
#

^ this

#

yeah that's how i do runs of jobs.

fervent thunder
#

worked awsome

broken holly
#

guys is it possible to make a ai based off of a picture?

outer crescent
broken holly
#

how does it work?

outer crescent
#

assuming by a ai you mean an ai generated image

broken holly
#

yeah

broken holly
outer crescent
#

if you type img2img into any search engine it should bring you up a good tutorial with images and examples, much easier than explaining it here

solemn herald
#

anyone knows if theres a way to get text from the prompt actually written in the result?

vapid tangle
#

How can I use the "more variations" function when I run automatic1111 locally? I've been using seeds to get images similar to one image I like, but it doesn't make images that are hardly similar

solemn herald
#

that's only possible if you learn what it should keep as a reference i believe, seeds always give new variations etc.

brisk bloom
radiant linden
placid hatch
drifting oak
alpine bear
drifting oak
junior badger
#

Hello i was suggested this server from unstable diffusion, but that was probably a joke.

#

i highly doubt what I create is allowed here

unkempt hatch
#

in the auto1111 webUI there's the option for "warmup steps"

they mean literally steps, not epochs, right?

e.g. I'm training 100 images at 100 epochs. 100 warmup steps means it'll warmup each image only 1 step

junior badger
#

would anyone be willing to point me to a community that's more adult in nature. That would be greatly helpfull for me, and ty.

unkempt hatch
junior badger
#

i'v been banned

#

i need a place more ... unstable

unkempt hatch
#

for what?

brazen fox
#

you're too nsfw for unstable diffusion? kek

junior badger
#

yes, and I don't bite my tongue

unkempt hatch
#

you making gay stuff?

junior badger
#

no

#

i'm straight

unkempt hatch
#

I would ban you for that

junior badger
#

for what?

#

being myself?

unkempt hatch
#

straight

#

it goes against my god

junior badger
#

i'm used to it 😒

#

my personal best is 48hrs in a community

#

ppl can't handle me

#

idk y

#

but whatever

#

any suggestions

#

I already know, i wont be welcomed here, which is fine

#

i just don't know where to go

unkempt hatch
junior badger
#

😔 maybe I am too much for this world to handle. Maybe it'll be better in the next life. Sorry to waste your time.

#

if it is alright, i'll stay to myself and just lurk for inspiration from time to time

hexed bramble
#

I'm using GTX 1070 with 8GB and the only model that actually works is 512-base-ema. Tried to use 768-v-ema and 512-depth-ema but with no success...

modern oak
#

are there any news on getting xformers working with amd? since 2.1 more or less requires the use of xformers

junior badger
#

I have it working

#

im using a ryzen5 with 6700XT

modern oak
#

do you use full precision?

junior badger
#

also made some of my own ckpt's

#

yes

modern oak
#

when i use full precision i just easily run out of vram

junior badger
#

w/ half

#

without cuda

#

obv

modern oak
#

and it's very slow too

junior badger
#

but i cannot share what I have created or coded here.

modern oak
#

there's a new optimization in automatic1111, maybe that makes it a bit better with no_half

junior badger
#

i see no diff in speed from using SD on my old RX570 to my new 6700XT

#

My post in unstable SD server b4 i just got banned:
After seeing what my RX570 can do running SD compaired to my new 6700XT. I can conclude: I see no difference, in speed or time, quality, ect... I have had the time to learn SD and add extensions, weights, hypernetworks and even created 1 of my own ckpt's. I'v also added styles and more stuff I can't remember at the moment. The webui is neat
and I do like it. So don't get me wrong.

modern oak
#

completely unrelated, but is anyone getting gpu coil whine through their speakers when generating stuff with stable diffusion?

dull mesa
#

I have a question. I have been training model checkpoints from huggingface onto my own face, so for example the mo-di (modern disney) model, but further trained on my pictures. This has some good results, but it also means having to retrain for each style etc. I was wondering, say I merge a SD model that was only trained on my face with such a checkpoint, would that work? So say I have my model, with my face as instance "skwaks", and merge it with mo-di, which has "modern disney" as instance name, can I than generate "modern disney skwaks"?

junior badger
#

personal thought "I'm surprised Disney isn't trying to sue anyone"

modern oak
#

i have no idea what the situation is on windows though but people report that it works

junior badger
#

all i had to do was pip install xformers, and they work

modern oak
#

xformers say they don't support amd

junior badger
#

so

#

ppl say lots of things

#

doesn't mean they are corrcet

modern oak
#

i get a lot of errors

junior badger
#

try a diff python version

modern oak
junior badger
#

or use a venv

#

look you can show me all the articles you want, it doesn't change the fact they work for me

modern oak
#

and the developers of xformers say you'd have to rewrite a lot of internal code for it to work which they don't seem to want to do

junior badger
#

correct, that's your job

#

they dont have to do anything for anyone

#

nor do u n i

#

do for yourself

modern oak
#

so it's true that xformers don't work? i'm confused lol

junior badger
#

they work for me

modern oak
#

you said it works fine but now you're saying it's our job to fix it

junior badger
#

i dont ask anyone to fix things for me, nor do i wait for an upstream fix, i code and edit on my own

modern oak
#

so how did you fix xformers apart from pip install xformers?

junior badger
#

that's all I had to do

#

and the error went away

#

¯_(ツ)_/¯

modern oak
#

i think you're not actually using xformers then

junior badger
#

srry i dont have a better or more tech answer for you

#

how can we test?

modern oak
#

i mean you can install it, but it won't enable if it knows it can't run under some conditions

junior badger
#

i'm willing to execute whatever test you can come up with

#

i dont mind

modern oak
#

what's your startup log? it should tell you if it's being used or not

junior badger
#

i'll look

#

log/images/log.csv this log, correct?

#

it's the only log i have, so

#

and it has no useful info

#

just what images i have created

#

and the prompts i used

modern oak
#

terminal output when you launch

junior badger
#

u said log

modern oak
#

yeah sorry i meant the terminal output

junior badger
#

but sure, i can start it up

#

1sec

#

command i use to launch is:
python3.8 launch.py --precision full --no-half --skip-torch-cuda-test

#

give me a min, to let it start up pls

#

py3.8 seems to give me the best results, so if anyone was gonna ask

modern oak
#

you're not using xformers though, there's a flag called --xformers

junior badger
# 
Python 3.8.16 (default, Dec 29 2022, 02:26:33) 
[GCC 12.2.0]
Commit hash: 310b71f669e4f2cea11b023c47f7ffedd82ab464
Installing requirements for Web UI
Installing imageio-ffmpeg requirement for depthmap script
Installing pyqt5 requirement for depthmap script

Installing Requirment of Prompt-Maker


#######################################################################################################
Initializing Dreambooth
If submitting an issue on github, please provide the below text for debugging purposes:

Python revision: 3.8.16 (default, Dec 29 2022, 02:26:33) 
[GCC 12.2.0]
Dreambooth revision: 50a9c18f29b841ee919e84a3e0cf28517abec726
SD-WebUI revision: 310b71f669e4f2cea11b023c47f7ffedd82ab464

Checking Dreambooth requirements...
[+] bitsandbytes version 0.35.0 installed.
[+] diffusers version 0.10.2 installed.
[+] transformers version 4.25.1 installed.
[+] xformers version 0.0.13 installed.
[+] torch version 1.13.1 installed.
[+] torchvision version 0.14.1 installed.

#######################################################################################################

Launching Web UI with arguments: --precision full --no-half
#

i kno what it is

modern oak
#

you can have it installed but it won't be used unless you add --xformers i think

junior badger
#

i will try

modern oak
#

at least according to the code if shared.cmd_opts.xformers or shared.cmd_opts.force_enable_xformers:

junior badger
#

shutting it down and restarting

modern oak
#

when i try with xformers I just get OSError: libc10_cuda.so: cannot open shared object file: No such file or directory

junior badger
#

Launching Web UI with arguments: --precision full --no-half --xformers
Warning: caught exception 'No CUDA GPUs are available', memory monitor disabled
2023-01-07 02:31:02.121430: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.

#

nothing changes, but maybe this warning helps you

modern oak
#

it kinda looks like you're running stable diffusion on your cpu

junior badger
#

well, i'm team amd, and it's looking for nvidia

#

sooooo

#

prob

modern oak
#

did you setup rocm?

junior badger
#

i did but when i installed anaconda, it Fkd up python systemwide

#

i hate python and all its variants

#

i use amd's aocl and aocc stuff

#

but i was using that stuff before i ever installed SD

modern oak
#

i use distrobox to run stable diffusion to avoid the mess

#

also because i have no idea how to get stable diffusion working on nixos lol

junior badger
#

i have, aocc, aocl-aocc, aocl-gcc, amdgpu, amdapp, amd-securng, and amd-rng installed

#

b4 i ever started using SD

modern oak
junior badger
#

i also use Arco OS

#

btw

#

distrobox?

#

like gnome box's?

#

no ty

#

I use qemu, docker, libvirtd, libsecomp for containerization and virtual stuff

modern oak
#

it's like a convenience wrapper for docker or podman

junior badger
#

sometimes i'll use proot to install a sub distro to a folder even

#

mount it, then loginto it from a new terminal, sharing all my system's resources running a distro without ever having to init

modern oak
#

yeah, it just does it for you conveniently through a cli

junior badger
#

proot can do some cool stuff for certain situations

#

just set the proot distro to use a diff tty

#

then you can have gui on like tty4 if you want

#

or use spice or vnc, ect ect...

#

well, at least i'm getting along with someone, just by talking tech, so that's nice

#

too bad i can't share my art in a server for sharing art

#

😔

modern oak
#

if you got banned from unstable diffusion for sharing your art it's probably best to keep it to yourself lol

junior badger
#

im banned from there for speaking

#

not my art

#

ppl there really liked what i was able to do

#

my problem, is I dont bend my knee for anyone

cold crater
#

installing stable diffusion on my pc is a nightmare

#

something with torch

#

wont work

junior badger
#

and i dont bite my tongue

#

if someone makes me feel a certain way or asks me, i'm more than willing to express myself

#

but, it sure makes my life lonely

modern oak
#

do you often end up in arguments?

junior badger
#

not that i start, i just finish them

#

i also have really bad PTSD

#

to where i'm not allowed to work

modern oak
#

you mean people often start arguing with you?

junior badger
#

yup

#

and because of my ptsd

#

i only have two modes

#

fight

#

or run

#

😔 i hate it, and worst part is, it's not my fault

#

someone or something did this to me

#

and i had no controll over it

#

and what it did to me

modern oak
#

yeah it doesn't sound nice to be in constant fight or flight mode

junior badger
#

afk, brb

modern oak
junior badger
#

linux and programing are therapy for me, it keeps my mind occupied. and yes it sucks, bad. life becomes even more lonely when you live on edge, my family has to know on doors before entering a room im in just so i dont jump from getting scared

#

and you would think it wouldn't spill over into my virtual life, but it does, and maybe even more

#

since we dont see body lang we only see text, ya kno

modern oak
#

yeah

#

i don't know what to say other than it sounds like you need to see a therapist, or if you can't maybe healthygamergg on youtube is a good channel

junior badger
#

and i usually dont have this opertunity to let someone kno, so i'm seen purly as some aggro, or scared person

#

depending on the situation, ofc

#

no, some who needs a therapist wouldn't have the insight i have

#

i'm fully aware of my situation

#

and have had all the therapy one could get

#

i'm just looking for a place i can share my art, and wont get mad at me, if I speak my truth when asked

#

from what I gather unstable diffusion was the most hardcore art server

#

and i was too hardcore for them

#

so, i'm fkd

modern oak
#

well as far as I understand you're telling me you have some deep insecurity problem causing you to often be in fight or flight mode

junior badger
#

no

#

i'm secure with who i am

#

ppl just cant handle what they put out and it gets back to them

#

ppl come at me crooked and have no idea, what box they opened

#

and it gets me into a situation where i get banned

#

i dont go looking for trouble, but when it finds me, its just no good for anyone

modern oak
#

eariler when we were talking about xformers it felt like we could easily start arguing

junior badger
#

and i can't find a community, that can handle that kind of raw energy, compiled with art

modern oak
#

i think i'm good at not biting, avoiding or just playing dumb to not start arguing

junior badger
#

ye, i felt that too

unkempt hatch
junior badger
#

not much more to tell

#

just looking for a place to share art and where ppl don't punish others for being themselves

cold crater
junior badger
#

specially when its artists

modern oak
#

so either you have to change the way you behave to avoid starting arguments with people or you have to find some sort of utopia of people who it's difficult to start arguments with lol

modern oak
cold crater
#

nvidia

modern oak
#

ah, i don't have nvidia so i'm not sure if i can help

junior badger
#

hmmmmm, ye lol, like an Anarcho society of artists lol

#

that would be nice

modern oak
junior badger
#

and yes, i kno u r in unstable diff server @unkempt hatch

#

u just tried to bait me, good try tho

cold crater
junior badger
#

and i have to deal with ppl like that

#

they'll follow me

#

just to continue the bs i have to deal with

#

i think they get off on it

#

but idk

#

too bad idk how to test this xformers thing for you or I would

lofty jungle
#

Hi everyone, Stable Diffusion disrupts the masked area when I use inpainting. What am I supposed to do in order the keep the masked area as it is?

junior badger
#

well, i'm going to continue my search for an Anarcho like society of AI artists, though I already have a good suspicion it does not exist

#

instead of following me around, in the shadows, you could just come with me on this journey @unkempt hatch

#

or not, its whatever, there will always be more like you, waiting for me, anywhere I go

#

i'll see you all around, if not in this life, the next one

#

✌️

unkempt hatch
lofty jungle
#

I've masked it in a program which I coded.

earnest shadow
#

Guys

lofty jungle
#

Watch is colored black and rest of it colored white

lofty jungle
solemn herald
dull mesa
#

I have a question. I have been training model checkpoints from huggingface onto my own face, so for example the mo-di (modern disney) model, but further trained on my pictures. This has some good results, but it also means having to retrain for each style etc. I was wondering, say I merge a SD model that was only trained on my face with such a checkpoint, would that work? So say I have my model, with my face as instance "skwaks", and merge it with mo-di, which has "modern disney" as instance name, can I than generate "modern disney skwaks"?

bleak matrix
#

Good morning, everyone!

#

How are we all today?

#

You mean Stable Diffusion?

#

Through 25

#

You can also download it locally

forest aspen
#

hey guys funny question

#

if i use 1.5 model that is 512x512 trained (the original model) to train 768x768 images, will it work correctly?

#

I'm curious has anybody tried

dull mesa
#

There are several model checkpoints for 1.5 trained on 768x768 images, so I guess it's doable

trim nymph
#

how to inpaint, cant find the option anymore

#

in dreamstudios

trim nymph
bleak matrix
#

once you generate your image, you need to go to editing

wide juniper
#

Just downloaded 2.1 ckpt and safetensor. My automatic1111 webui gives me an error whenever I switch to them. Help please?

outer crescent
#

do you have the yaml file in the same directory?

#

should be a yaml in the repo you got the 2.1 ckpt, save it in same directory with the checkpoint's filename but .yaml

wide juniper
#

Thanks!

outer crescent
#

np

wide juniper
#

I have an old photo that I wish to restore. Should I use stable diffusion’s of upscalers or should I use one of those AI photo restoration websites out there?

cold siren
#

hey people, i used to be able to go above 1600x1024 on my old ass gpu, but after the latest auto1111 update i can't even render 1536x1024

#

wtf is going on with memory requirements there 🙄

muted temple
#

I would like to generate an image that shows a cat with an eagle beak in the style of dali.
For txt2img I use the prompt "cat,eagle beak,in the style of dali,dali", but it just doesn't do that.
I have already tried "cat,eagle beak, in the style of dali,dali,beak,eagle".
Does anybody see what I'm doing wrong?

wide juniper
outer crescent
#

where are you getting 2.1? huggingface?

wide juniper
#

Yup

outer crescent
#

768 version?

fervent thunder
#

hey is there any difference really between inferring using fp16 and bf16

#

specifically on a 3060 Ti

trim nymph
bleak matrix
#

as it does work (and is access through the image itself)

wide juniper
#

I'm not a programmer btw. How do normal people access 2.1?

outer crescent
#

do you have automatic1111 installed?

compact vigil
#

can I use hugging face link to automatic1111 colab version?

frail sparrow
#

can anyone help me, I've installed on my desktop python, stable diffusion and a model as well, but when I try to open "webui-user" it doesn't do anything

fervent thunder
#

Hello everyone

cold siren
hexed bramble
#

Hi, is there a way to run model 2.x on GTX 1070 with 8GB?

cold siren
#

2.0 works on 4 gb

bleak matrix
hexed bramble
cold siren
#

i run webui-user.bat 🪟

#

with --medvram

#

inside the .bat, that is

hexed bramble
cold siren
#

you might also need --no-half, especially for 2.1

fervent thunder
bleak matrix
fervent thunder
supple valley
#

How many images is recommended for training a person with textual inversion embedding method?

hearty schooner
#

Any actual machine learners in here?

thorn hare
#

what exactly does merge models mean ?

#

or what does it do exactly ?

sage sky
#

does anyone have txt2mask

#

aka unpromptu

grand wharf
#

I've tried: dark shadows, increased black level, strong blacks, strong contrast, etc... but nothing changes it.

astral goblet
# thorn hare what exactly does merge models mean ?

It's aprocess of training models off each other. It produces a new model that is roughly the same size as each of the original ones, but with a blend of each model's knowledge. It's basically model breeding

signal nest
#

if its the same base model finetuned, i think its possible to literally just blend the weights to produce a mix, but let me double check that. I speculated that it would be possible to do a 'mixture of experts' approach by producing weighting factors for a bunch of finetunes (dynamic blend) .. feedback in the developers section seems to confirm that this isn't a dumb idea.. but I haven't even delved into basic finetunes myself yet.

floral zealot
astral goblet
ocean mango
astral goblet
#

I find that most of them are 1.5 models with hardcore pornography models blended into it, for some kind of misunderstood approach of getting better anatomy knowledge into the model. As if the AI is learning form and natural drawing, instead of denoising towards a 2d plane showing hardcore smut

#

I like the idea of model breeding, but the gene pool right now is, well, there's gotta be a word here that doesn't sound mean. ... but yeah its one of those words

astral goblet
signal nest
#

(what I have in mind is coming up with say a set of say 8 contrasting finetunes from which you'd be able to mix more variety. Those could be stored in RAM and mixed on demand (eg if you have 64gb RAM + 8gb GPU ). a second much smaller net would take the text prompt and figure out the model-blending weights)

astral goblet
#

that's not how merging works. Models need to be stored in vram to be used too. It's a training process so to create a merge, it has to think for quite some time about it.

#

Your process might be better suited for embeds. they're amazing in 2.x models. you can use many at one time and blend on the fly, and they're less than 1mb each

#

Textual Inversion make better experts imo

signal nest
#

failing that maybe it would be possible to store a common base model and swap out some critical layers. anyway that idea came from a claim that it was just possible to 'add, subtract etc' which I haven't had verified, i read it in passing. (maybe someone was infact confusing embeds with finetunes in that claim)

#

to be clear in my idea the blend would be uploaded to VRAM once done to actually generate an image. (but if blending is non trivial.. ok)

astral goblet
#

a lot of people use fine tuning as a generic term for all model refining, i've found

outer crescent
#

but yeah embeddings are amazing

signal nest
#

once i organize my kit lol i'll try it. "finetune amodel on star wars imagery .. finetune a model on The Expanse imagery.. blend them.. and ask it what a fighter, a cruiser, a spacestation, or a futuristic soldier looks like.."

astral goblet
#

few other good science fiction domains that are not so cliche, startrek styles, are starship troopers (sequels even), foundation (appletv), dark matter and firefly

#

maybe a little event horizon

signal nest
#

maybe you'd have to constrain the finetuning process to keep the models compatible enough for blending (i've heard that it loses some of its base information as you narrow it, and you can combat that by keeping some generic images in the finetune training set)

ocean mango
river maple
#

automatic1111 finally added licenses

outer crescent
#

give proper links

astral goblet
#

now we'll see if all the people who pull from his repo will correspond to it

river maple
astral goblet
#

i think he picked mit for the webui architecture specifically and added licenses for all the other projects he pulls

jagged lintel
#

sry new, and there are so many chat rooms I dont know if im right here or not...

If I want to train a Hypernetwork (only got 8gb vram so cant use embeddings)

How many Images should I use? and can I use multiple things.

For example have multiple types of clothings or such.

narrow sluice
alpine bear
jagged lintel
alpine bear
#

Embeds are trained on images. I could train an embedding called 'sky' with the initialization text 'sky_style', and train it on images of a tree. When I then use the word sky_style in my prompting, it will influence the output in the direction of the trees the embed was trained on.

#

sky_style might have had a different influence in the model before the embed. The init text of the embed will replace the original model's understanding of said text when used in the prompt.

jagged lintel
#

am I limited to 1 prompt?

For example if I would want to train on Castle and Mystical Forest
and a picture has both in it

Do I just use it in both embeddings that I will create, or should an embedding only include exactly what's needed

astral goblet
#

worth noting that --xformers requires cuda/nvidia

alpine bear
#

Although, both of those are very well known by many models. So embeds shouldn't be necessary, unless you have very specific styles in mind.

astral goblet
#

some overlap of training images in focused embeds wouldn't be such a bad thing. if one image has a castle in a mythical forest, use it in both sets

#

since this is all so new, its worth it just for the learning experience

dull stratus
#

does anyone have a good tutorial on how to use the dreambooth extension?

#

i'm lost

river maple
#

I used to know how to use it, and now... hehahahahaha

lilac reef
#

Anyone using automatic1111 on Linux as I tried to run it but it said some error about bash/r?

umbral kraken
#

'CUDA out of memory', anyone help on how i can help with this issue? running an RTX 3080, SD 2.1

ocean mango
#

i got 3060 use --no half in 2.0

#

not in 1.5 tho

umbral kraken
#

ive generally been okl the past few days, but i suppose ive been trying higher res images, and its struggling

#

currently got it set to --no-half

floral zealot
#

somone can get me the links for novel ai model and zeipher f111? or the berryMix?

nimble barn
#

hey need a bit of help I dont know if this is the right place to ask, Also if you could reply to me via DM, Thanks. I've been looking to attempt to download Stable Diffusion, I followed the Youtube guide BUT right at the end I got Error code:9009 in phython when attempting to open Webui.user (in the youtube video this was webui.bat?) I read on Reddit that by Editing webui.user and typing the loction of phthon beside 'set python =' this would fix the problem, but instead of error 9009 I got a different error that says 'stderr: Error: could not find a version that satisfies the requirement torch ==1.12.1+cus113 (from versions: none) Error: no matching distribution found for torch==1.12.1+cu113

ocean mango
floral zealot
#

thank you

ocean mango
nimble barn
#

can you help me do that lmao

ocean mango
#

try yt

#

supposed to check it when you install python

thorn hare
#

what is the point of doing a model merge with 0,0 strength ?

ocean mango
nimble barn
#

@ocean mango all youve done is confuse me

ocean mango
#

do a windows search for environmental variables

#

see if python is there under system variables

nimble barn
#

its not

ocean mango
#

so you have to add it

nimble barn
#

ok let me try that the enitre python folder or just the .exe ?

ocean mango
#

C:\Users\xxx\AppData\Local\Programs\Python\Python310

nimble barn
#

I sdded it to both system varbiles and also user variables

#

error 9009 ( deleted the floder path in the webui.user)

#

also its python 3.11

floral zealot
#

Does anyone know how to fix it, the berrymix model is putting two women on top of each other and I put them in the prompt 1girl

river maple
ornate tartan
#

How do i upscale an image (making it bigger, without making it blurry?)

#

Is there anyway to do that in the Web UI?

ocean mango
ornate tartan
#

Damn alright fair enough

ocean mango
ornate tartan
#

I wish there was a tool to do it in stable diffusion itself but thats fine

ocean mango
#

there is a script if you use 1111

forest apex
ornate tartan
#

i do use 1111

sonic carbon
sonic carbon
#

or in img2img you can use the SD upscale script

#

prompt "detailed, intricate etc" cfg:max denoising:0.1 steps:100-150

ocean mango
empty terrace
ocean mango
#

hmm dobt see it

#

that site is amazing dont know they do it

umbral kraken
#

can i use the basujindal fork with SD 2.1? im trying to reduce VRAM usage, and struggling to find an answer

floral zealot
sonic carbon
floral zealot
#

512x512

placid egret
#

Newbie question but is Thera a large language model being worked on as part of the stability projects, if so where is it being coordinated and what is it called?

floral zealot
shrewd egret
#

Also a newbie questions. Is it possible to set up textural inversions as a Style setting in Automatic1111? What is the best use case for using styles?

sonic carbon
#

@thorny depot

thorny depot
forest apex
#

oh that's how its done ...

sonic carbon
forest apex
#

summoning a mod :0

sonic carbon
#

i see 😅

thorny depot
#

Membranous wings unfurl through billowing fumes of ashen smoke and mist. I AM BECOME MOD, DESTROYER OF WORLDS.

narrow sluice
thorny depot
#

In all seriousness, feel free to ping me if you see nonsense like that. Happy to help.

forest apex
#

thanks

junior badger
#

For anyone unsure if SD works with AMD, it does. I'v gotten SD to work with my old RX570 GFX card and my new 6700XT GFX card. From my experience, I see no real gain in time or quality.

#

So for all the talk about SD only good for NVidia GFX cards. I call it out as BS.

#

It may work "better" with NVidia. But it is not limited to just NVidia.

forest apex
#

@junior badger anything in particular you had to do or an up-to-date guide you followed?

junior badger
#

When it comes to code, I just wing it, though I should mention I'v been a purely Linux only user for over 6 years. and am confident in my "wing it" abilities

forest apex
#

Ok so this was on linux not windows

junior badger
#

correct.

#

I havn't touched a windows machine since win3.1

forest apex
#

what kind of speeds do you get with the 6700XT?

junior badger
#

slow ¯_(ツ)_/¯

#

ish

#

idk what are good or bad speeds

#

I'm new to this.

#

and I also havn't seen any charts or tables with data to show what are acceptable completion times

#

I guess we'll have to wait for Linus-Tech_tips to do that, lol

forest apex
brisk bloom
#

on my rtx3080 I can gen a 768 image with 30 samples in about 12 seconds

#

average

junior badger
#

my 6700XT is nowhere near those times

#

not even close

brisk bloom
#

yup thats the big difference with nvidia

junior badger
#

I understand, but for the sake of this convo, I'm a linux pure-ist

forest apex
#

on a 3060, 20 steps DPM++ takes about 10 seconds

junior badger
#

and prefer to use open source code as much as possible

#

dispite its downfalls

forest apex
#

yup. I try to use them all for the sake of support

junior badger
#

I should also mention i'm a fully (paid dues) FSF member since 2017

#

Free Software Foundation

forest apex
#

my local LUG has been canceled since covid hit 😦

junior badger
#

dang

#

srry to hear that

#

but regardless of my SD times moving from an RX570 to a 6700XT has been amazing

forest apex
#

do you do any gaming on that machine?

junior badger
#

specially for my Steam library 🙃

#

World of Tanks is like a 1000x's better

#

(exaggeration, but you get it)

forest apex
#

its funny, I upgraded to a 1060 6GB, was playing games a little bit before that, but then after I found SD haven't really played at all

#

rather watch videos and read tutorials,l ol

junior badger
#

well, ofc this is new and exciting in so many ways.

#

I totally get you.

forest apex
#

this is the game now

junior badger
#

mhm 🙃

#

if you click my pfp and follow the ART link you can see what I'v made by using SD sites (before I locally hosted SD)

#

I'm in the process of building my own personal gallery website to showcase my art.

forest apex
#

nice. evil snow white is trippy 🙂

junior badger
#

Thnx 😁

forest apex
#

I have to pick up a tripod so I can take some decent video for img2img conversion

junior badger
#

(I have no idea what they are saying, I don't speak german, lol)

#

I just like the visual aspect and the thought of Snow White being bad

forest apex
#

I read the english lyrics of one of their songs the other day, very poetic

junior badger
#

She's my new Wiafu now lol

forest apex
#

I wish playground would do prompt blending properly

junior badger
#

playgroundai?

#

I just solved there pedo problem a few days ago

#

they were, oblivious too, (sus)

forest apex
#

yea, I get totally different results in automatic than I do on their site

junior badger
#

if you searched the word "topless" it was quit evident they were neglagent unknowingly or willingly, allowing img2img of young girls.

#

I don't think I need to go into more detail

#

and I called it out on there discord

nimble barn
#

can i add dreambot to a private channel on discord ?

junior badger
#

hopefully they made some adjustments

forest apex
#

I remember when I first heard about the site, yea it should be a lot better now, I'd see prompts that were obviously people trying to generate CP

junior badger
#

mhm

nimble barn
#

can i add dreambot to a private channel on discord ?

junior badger
#

can you? ¯_(ツ)_/¯

#

have you tried?

#

I would first try, then if you fail, then ask.

#

see what I mean?

nimble barn
#

can i add dreambot to a private channel on discord ?

ocean mango
nimble barn
#

I mean i forgot how to, been a while since I picked up a bot

ocean mango
#

oh no

#

can with midjourney

junior badger
#

I don't see why ppl reward an answer to someone that just repaets a question by being rude.

ocean mango
#

not sd

junior badger
#

but that's just me.

ocean mango
#

to shut him up?

nimble barn
#

dude your tatts are bk

junior badger
#

it doesn't matter, and not really my place to continue my personal thoughts on the matter.

nimble barn
#

you look like a down syndrome with or without them

junior badger
#

I'm not trying to stir up the pot

ocean mango
#

ok enought help back to generating animations

junior badger
#

see what I mean, How he instantly went to using name calling ect...

#

that's what I was trying to avoid

#

Low IQ, moral standards, and ethics.

nimble barn
#

can you? ¯_(ツ)_/¯ i mean @ocean mango
have you tried? i mean @ocean mango
I would first try, then if you fail, then ask. I mean @ocean mango
see what I mean? I mean @ocean mango

junior badger
#

I do hope he gets the bot working though, I'm not holding his low IQ against him.

forest apex
#

stable horde works for me when I'm not at the desktop ^_^

nimble barn
#

@junior badger check out dreambot 1

junior badger
#

>'')>|===NICE==>

vale spindle
#

Anyone know of a Google Colab that can be used for "Deepfakes". I hate using the term because it sounds nefarious, but I just am looking for a Colab that can train on a face and then overlay that on a video. If anyone knows a legitimately good/dependable one hit me up and ill buy you a coffee or beer via BTC (100% serious).

Also, sorry if this isn't allowed to be asked in here. I just joined.

junior badger
#

@nimble barn no ty, i'm not going to engage w/ you, and enable your name calling

river maple
#

what even is this?

forest apex
#

@vale spindle you'd just have to do an img2img of the frames of the video you want to alter. that, or using the ebsynth program

vale spindle
#

holy shit...

#

never. thought. of that....

#

or heard anyone talk about that?

#

is that a thing terbo? ive also never done a deepfake but i have been messing with SD for the past 2 weeks

forest apex
#

enigmatic_e on youtube has some videos that cover similar techniques

vale spindle
#

Is that a thing terbo? Do people do it that way? Or is that something you thought of ont he spot

#

ahhhh

#

shit eh. thanks man.

#

DM me dude, just so i have you in my inbox, i like to have people like yourself as contacts to pick your brain, and ill make it worth your while by sending you some coin every now and then

#

no pressure though just throwing that out there

astral goblet
#

i never had capable enough hardware for a deepfake until a couple years ago, and i just never got into it. This image generation running locally really caught my attention though

forest apex
#

same, it wasn't cool enough sounding to invest in a video card at the time

vale spindle
#

i wish i knew about things like colab/runpod years ago

#

but mind you, maybe nothing was usable for colab/runpod back then

junior badger
#

well there is an ethical reason to not endulge into deepfakes vs image generation

vale spindle
#

for sure

forest apex
#

@junior badger I think Sassy Justice resolved all of the ethical issues 😄

junior badger
#

i'm unaware of what you mean.

forest apex
#

the south park creators created a show that is purely deep fake. it's hilarious.

#

I saw somewhere recently they just got funded for more episodes

junior badger
#

I havn't watch TV in 15+ years

forest apex
#

me either 🙂 its on youtube

junior badger
#

and it's called "sassy justice"?

forest apex
#

TBH they spent a lot of cash on it. then covid hit and they stopped. But I bet using SD it would be a whole lot easier.

astral goblet
#

Disney is investing into it too. The name deepfake comes out of the goon porn addict community, the same driving force behind unstable diffusion. they compiled together a bunch of tool sets and hacked together most of their own, for the specific purpose of putting people into pornography. What a bunch of horny geniuses.

It's unfortunate though because the tech will never not have that legacy now. That's where it was born and raised.

junior badger
#

yeah, I was just banned from there discord server

astral goblet
#

they're a touchy bunch

junior badger
#

they all liked my art, even made it on there Starboard

short tapir
#

Is stable Diffusion down currently?

junior badger
#

But i also don't bite my tongue and let ppl know how I feel about whatever I'm brought into

#

and I don't bend my knee to anyone, ever

#

I keep it 100

feral kelp
#

is hirexfix broken, or is it changed and I'm using it wrong? appears to just be doubling my image size, and taking forever as a result

astral goblet
#

a1111's ui? it doesn't auto calculate things anymore and there's no indication that it doesn't. you have to calculate the final size and multiplier yourself

#

it'll scale what you generate by the multiplier everytime. so the initial w&h won't be your final product. if you want 1000x1000 you have to manually put 500x500 and a mutliplier of 2

junior badger
#

from my experience in programming, everything is a suggestion, and the best way to get any software to work the way you want it to, is to edit the source to fit your own needs and wants

#

(how very linux of me to say this, lol)

feral kelp
#

hmm ok thanks, I'll do some experimenting

forest apex
#

@junior badger some of us don't have the source, maaaan 😦

junior badger
#

😉 That's the best way, and the best part is, it's just software if you bork it, delete it, and reinstall it

#

you mean, you didn't install it from github?

#

where you can read all of the source code

#

why?

#

oooohh windows... right

#

yuck

astral goblet
#

a1111's ui is all open

forest apex
#

is surrounded by raspberry pi's

astral goblet
#

dude i got a dell

junior badger
#

lol

#

😂

#

🤣

astral goblet
#

aww dude

junior badger
#

90's joke, noice

astral goblet
#

it's actually a franken dell. i loved the alienware case so i rebuilt a pc inside it. not a dell component left

forest apex
#

I'm on a dell too, it takes like 5 minutes to boot up, I never reboot

junior badger
#

the TV comercial

#

from the 90's

astral goblet
#

that's the one! kid was so rich by the end of that commercial campaign

junior badger
#

shoooot i bet

#

that was fkn funny tho bruh, good on ya

astral goblet
#

ohhhh he got caught with pot and fired!

#

i forgot that part

#

OT sorry... back to gens

junior badger
#

also, the BS of the 90's

floral zealot
#

what is the best Sampling method to use for anime on berry mix?

silk coral
#

If so where can i check how much i can create?

feral kelp
#

but it also displays it right there in the channel too

silk coral
#

Great thank you!

keen lichen
#

Is there a discord for a1111?

#

or a new way to download?

junior badger
#

The editing is so well done, this 41 mins must have taken months

forest apex
#

4 billion doses, that should get us through the night

junior badger
#

lol

#

it's a 🍿 movie, I swear it's that good, And I don't even like Star Trek. But I'll watch this.

forest apex
#

the 60's cartoons were pretty good

junior badger
#

1940's Looney Toons

#

The Edgy Years

steep mortar
#

anyone know how to get auto1111 to save images with the full txt of the prompt and not truncated?

forest apex
#

@steep mortar the prompt is saved inside of the PNG

steep mortar
forest apex
#

as well as all of the generating info. as far as the filename, no idea,

#

you can also save a text file besides the PNG containing that info.

junior badger
#

Just edit the filename ...

#

click to highlight file, press F2 to rename it.

#

add all the info you want to the filename.

#

Am I missing something?

steep mortar
junior badger
#

highlight them all and mass edit filenames

#

oh, I should mention I use Linux, and that's a thing we can do easily

#

idk what OS you use, but I'm guessing Windows, and for that, idk.

steep mortar
#

windows

junior badger
steep mortar
#

its wierd that auto1111 just truncates file names to 128 chars

junior badger
#

then don't rely on it and do it yourself.

forest apex
#

@steep mortar inside of the modules\images.py it is a hard limit. You could probably change it.

steep mortar
#

hmmm

forest apex
#

But for those 100 files, a python script could do it

junior badger
#

what is windows limit for filenames?

forest apex
#

256, but I guess you can make it longer. sounds like a headache, lol

junior badger
#

there ya go

#

edit the py file to 256 vs the 128

steep mortar
#

awesome, found it thnx

junior badger
#

team work makes the dream work

steep mortar
#

thnx guys

junior badger
#

yw, hi fives @forest apex

forest apex
#

well. I've run into that problem too, I'll look at seeing how it can be extended or maybe have the 128 character limited accessible through settings

junior badger
#

lol, you guys name files that long?!? I mean, it is your file, but jesus.

#

What I do is make a Folder with the subject name and inside make a README.md file with my notes for everything inside the folder

#

it's a good practice to get yourself in the habit of doing.

forest apex
#

Well. If I was generating for a particular reason, I wouldn't have a problem with that, but when testing, it might be easier

junior badger
#

specially if you plan on sharing the folder, it helps the next person by having 1 file that they should read with the info they should know.

#

and by making the file named Read Me, in all caps, should make it obv that maybe they should read that file

#

lol, but you'd be surprised how many ppl wont read it 😂

forest apex
#

unless it jumps out at them huh

junior badger
#

like go into your SD folders

#

you'll notice in the folders you'll find README files

#

😉

#

you should read them

#

(that's why they are there, lol)

#

surprising how many people are clueless sometimes

forest apex
#

or clued in to something totally else

junior badger
#

yeah, then when they ask the developer and wonder why they never get an answer

#

because they already gave you the answers, in those very files

keen lichen
#

Best extensions for a1111?

junior badger
#

all of them, if you're not asking about anything specific

forest apex
#

Image Browser, Deforum, Dreambooth, SD Latent Mirroring

robust coral
junior badger
#

sorry, I can't answer without you giving context.

#

i could tell you all the wrong ones depending on what it is you want.

forest apex
#

@robust coral what is Unprompting?

junior badger
#

prob means negative prompts

#

but idk

keen lichen
#

img2txt?

forest apex
#

that's CLIP

junior badger
#

I'v never even tried img2txt

#

what's that about anyway? Does it try to explain the image

keen lichen
#

I guess it makes a prompt used to recreate the input image.

forest apex
#

it does a fairly mediocre analysis of the image and gives you a prompt

#

its not horrible

junior badger
#

oh, I use the prompt generator extension

#

you give it like 3 words and it'll make you a bunch of different prompts

swift snow
#

Do safetensor files work the same as ckpt files?

forest apex
#

there are a lot of extensions on github too. not all of them are in the extensions tab, and there are also the custom scripts that weren't made into extensions, from the automatic1111 wiki

junior badger
#

i think they are more for hypernetworks

#

but idrk

keen lichen
#

hypernetworks?

silk coral
#

Heyy guys🙏🏻😹

forest apex
#

@swift snow they are non-pickle files that often load faster, but you can't train off of them or merge them in the same way as ckpt files

junior badger
#

yea, i also dont know which web ui i'm using vs yours

swift snow
#

Saved me a google search 😂

forest apex
#

gotta think about the lurkers too

junior badger
forest apex
#

one of my favorite automatic1111 scripts was the Animator script. I don't think it has been updated in a while, it ran into some problems but it was like a seed stepping, pre-deforum script

junior badger
#

also the way my terminal is setup (vcs aware) it auto updates the software from github 🤓

#

so, like just now,

#

24 files changed, 823 insertions(+), 275 deletions(-)