#💬｜general-chat

1 messages · Page 57 of 1

vapid tangle
#

maybe im not good with prompts but i dont really see anything new or amazing

stuck furnace
#

Seems like the first time i ever used stable diffusion to me personally

sly eagle
#

Hi! I think I asked this yersterday, if there's a way to paint lineart images with SD in google colab?

brisk bloom
sly eagle
#

Yes the one i'm using have controlnet, I saw the inpaint options, but usually i only get the same image (not painted)

#

for example, I want to paint a character with a certain coloring style

#

but it doesn't work

brisk bloom
#

use your line drawing for the contronet section. Above in the normal img2img section insert an image that has the colours you want to use

#

the style will mostly come from your prompt

sly eagle
#

But how the ai will know if i want a certain color in the hair

brisk bloom
#

from your prompt

sly eagle
sly eagle
#

How many max steps do i need to train properly a model?

pine comet
#

whats the most realistic ai model

#

ping me

rigid kayak
#

hola

lunar vigil
#

s

sage wasp
#

How to create image?

normal arrow
#

Is there are reason for all these deleted channels? I am new here

young musk
#

I don't see any deleted channels and I joined today

normal arrow
#

Can you see all the channels in art-movements?

#

This post in particular just shows a lot of deleted channels for me

young musk
#

Yeah I see all of em

normal arrow
#

Weird, for me all of them are deleted except for the anime one and the modernism channel

#

Here, you'll find a plethora of threads awaiting your art submissions. Unleash your inner artist within our new forums and share your creative endeavors with our stable community!
Welcome to Dreamer Communities!

Anime
anime - dreams of a world where animation transcends its limits and becomes an immersive experience that offers an escape from reality!

Art Movements
art-movements - dreams of vibrant and dynamic art movements, a celebration of individuality and artistic expression!

#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel
modernism
#deleted-channel
#deleted-channel

Humans & Animals
humans-animals - dreams of character creations, be it the complexities of humanity to the wildest creatures of your dreams!

#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel

Landscapes & Buildings
landscapes-buildings - dreams of the harmonious balance between the beauty of nature and man-made structures!

#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel

Objects & Vehicles
objects-vehicles - dreams of three dimensional objects that evoke feelings of nostalgia, comfort or joy, and can serve as reminders of the people, places, and memories!

#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel

Genre & Styles
genre-styles - dreams of the whimsical world of cartoons, from the mystical realm of fantasy to the spine-chilling genre of horror, from the stunning art of photography to the trippy realm of psychedelia!

#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel
#deleted-channel
Send exceptional posts into Starboard!
Have you heard of our Starboard? It's where we collect all of the best posts in our server, and it's curated by the community!

#

I cant share images so I copy pasted the text

young musk
#

skill issue

young ivy
#

good morning

copper crystal
#

better discord often just screws up discord. there are others and they're often malicious and send your login tokens to other users.

Don't run scripts on your discord client. it's a really bad idea.

normal arrow
#

I dont use a special discord client though

copper crystal
#

those channels are long gone. They're threads now

#

if your client is remembering them, it's modified

normal arrow
#

I can not see the threads either though

copper crystal
#

yeah. modifying discord often breaks it. i said that

normal arrow
#

if you can not see this message then I guess there is something wrong with my client

#

I did not modify my discord though?

copper crystal
#

i see that message. all those link to threadded topics. not channels. The channels are long long gone

normal arrow
#

Weird

copper crystal
#

discord is definately broken in a weird way where it's remembering shit it shouldnt'. that sounds like it was modified to me. friend did it for you. you did it months aog and forgot. Or you just don't want ot admit it because its against tos. i don't see how discord could believe they were still channels other than it was modified to see stuff that moderators had changed and other hacky discord stuff.

#

as a fix i would uninstall discord 100%, delete all registry keys and folders and anything associated with it, then start over

normal arrow
#

I'll look into it later, I feel like there might be some other solutions since neither of your described scenarios are applicable to me

lost swift
#

I have a small question to ask
why this generating speed seems very slow. the generating speed of my computer is 3.59it/s,(4080 laptop)

wild steppe
normal arrow
wild steppe
#

Glad to hear it! heartpop

marble plaza
#

is stable diffusion xl free from copyright issue? (like gettyimages watermark)

copper crystal
copper crystal
alpine bear
half finch
#

civitai is down?

marsh hare
half finch
#

Do you know When can it be restored?

marsh hare
#

I have no idea, but it keeps getting shut down every now and then so next time it goes up just download as much as you can

split crescent
#

It's there a Collab for this?

half finch
#

@marsh hareok ,thank you

vast ingot
# half finch Do you know When can it be restored?

this is their latest news, about 4 hours ago, an announcement on their discord :

We've identified the issue: Our database replicas are failing every 6 hours or so, and we're not 100% sure why. We're doing our best with our limited knowledge of Postgres but its slow going. To that end we're currently looking for a Postgres expert to consult with us to help get some longer term fixes in place. If you or someone you know is interested, apply here:

We have what we think is a more stable solution going out now, (Thanks 💬︱general for the feedback!) no eta on that yet but hopefully not more than 4 hours.
As always we appreciate your patience and understanding as we work to get your waifu's available once more ❤️

glacial oar
#

any update on Stable LLM?

#

just excited abt what they gonna come up with

pine comet
#

What’s the most realistic ai model

#

Is it chilloutmix

#

?

vast ingot
#

(well, civit is back down already)

fallen garden
#

where can i create my images

marsh hare
#

Anyone here versed in python/pytorch that can help a brutha out?

ebon beacon
#

why do people use civitai? its so hard to search for anything and it mostly just shows images of the same style

#

and every category is just flooded with pictures of girls

lapis nova
#

venv "C:\Users\Mooi Kin Heng\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Commit hash: 22bcc7be428c94e9408f589966c2040187245d81
Traceback (most recent call last):
File "C:\Users\Mooi Kin Heng\stable-diffusion-webui\launch.py", line 355, in <module>
prepare_environment()
File "C:\Users\Mooi Kin Heng\stable-diffusion-webui\launch.py", line 260, in prepare_environment
run_python("import torch; assert torch.cuda.is_available(), 'Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check'")
File "C:\Users\Mooi Kin Heng\stable-diffusion-webui\launch.py", line 121, in run_python
return run(f'"{python}" -c "{code}"', desc, errdesc)
File "C:\Users\Mooi Kin Heng\stable-diffusion-webui\launch.py", line 97, in run
raise RuntimeError(message)
RuntimeError: Error running command.
Command: "C:\Users\Mooi Kin Heng\stable-diffusion-webui\venv\Scripts\python.exe" -c "import torch; assert torch.cuda.is_available(), 'Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check'"
Error code: 1
stdout: <empty>
stderr: Traceback (most recent call last):
File "<string>", line 1, in <module>
AssertionError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

Press any key to continue . . .

hi anyone know this issue? need help

young ivy
#

takes awhile to prepare a character generated from nothing but noise

#

hmmmm

weak kelp
young ivy
#

you have an nvidia card right? @lapis nova

#

it's not easy to create on your own and there's plenty on civit, but the big names have it all handled @weak kelp

weak kelp
young ivy
weak kelp
#

oh. didn't realize that

lapis nova
young ivy
lapis nova
bitter lintel
#

Hi guys,

Is anyone experiencing this when trying to login dreamstudio?

#

Hmm, looks like logging in didn't work. Please try again.

young ivy
next haven
warm junco
next haven
#

Thank you.

#

@warm juncoI downloaded and saw that one model was added. But where is it? How to use it?

warm junco
#

You can click on the third button below the Generate button. There you find embeddings, loras etc, then select the embedding

next haven
#

Found in textual inversion.

hasty bane
#

Guys, I post pictures of anime/game characters on twitter. I don't do nsfw and I don't promote any products. In short, is it possible to file a lawsuit against me over this? I don't have many followers right now, but who will know in the future...

warm junco
hasty bane
#

I've been researching a lot about copyright, but this is a headache

real zodiac
#

I'd still be careful if, say, you trained a lora on some very recognizable style and are posting images that some reverse image search files as that artist... thinking

verbal tapir
#

can i write prompts in google sheets and generate images there instead?

#

is it possible to link google sheet and stable diffusion tgher?

real zodiac
#

wow that's not a sentence I thought I'd ever see

marsh hare
#

I guess the correct questions is.... why?

#

also l0l at discord's sad attempts to get people to buy nitro.

#

literally effects that we've had since the days of msn messenger

verbal tapir
#

i saw those habit tracker thing template, random thought of generating images for visuals maybe characters to "lvl up"

#

just a random thought

drowsy garnet
#

Am I the only one who can't access civitai now?

humble orbit
#

I'm getting in. Quite slow tho

regal snow
torn charm
#

could someone pls share a bit of info about models with me? i am currently using v1-5-pruned-emaonly and wondering how to find new models and how to choose the best models.

bleak matrix
#

Good morning, everyone sunsmile

#

How are we all today?

warm junco
balmy jungle
#

yes there is a problem

#

friends of mine are dying

proud locust
#

does anyone know how to fix the hires fix that is in Automatic1111? I have the latest version and for some reason, the Hires fix is adding geometry. It's horrible.

ember rover
#

hi all

#

been messing around with AI art casually since nightcafe early days, but I'm finally using it offline

bleak matrix
#

Nice! How are you enjoying it?

ember rover
analog mango
#

how can i get help about stable diffusion

ember rover
#

what do you need help with?\

analog mango
#

check ur dm pls

#

ugh dm closed 💀

ember rover
#

why cant you ask in here?

analog mango
#

im trying to send image and i cant send it here

bleak matrix
#

I'm glad you're enjoying it!

bleak matrix
#

If you need help with prompting, there's things like #📝｜prompting-help etc. I recommend taking a look around the server, as there are a lot of great places to get help

#

@analog mango

ember rover
#

do more people use 1.5 or 2.1 here?

bleak matrix
#

I wouldn't say I have any particular numbers, but there are many people using 2.1

wooden sphinx
#

I put "too many hands" in the negative"

#

and it literally just created a person to fill in the hand

#

🧠

peak orchid
#

This will mean the megacorps like Getty Images will have monopolies, and won't help small time artists

narrow cairn
#

It has been a while since i been active in this subcategory in the internet. Has Stable foundation figured out the issue with fingers/hands yet?

verbal vale
#

all this preemptive legislation has to end. No one knows for sure how AI will effect us at large. Greed is pushing all of this.

floral kestrel
#

hey guys , noob there, what make that a lora work with some models but not with others? even when sometimes they look similar style like anime for exemple.

ember rover
#

shouldnt it also be illegal for artists to learn from copyrighted works too?

peak orchid
solar hollow
obsidian meadow
#

Hi all!

#

Is this the latest version of Automatic1111: Commit hash: 22bcc7be428c94e9408f589966c2040187245d81

#

Git pull does tell that I already have the latest update

pine comet
#

whats the most realilstic ai model

broken gull
#

firefly

obsidian geode
#

How to disable showing the image in automatic1111? I want to reduce the downloading of pictures and only stock them

tall gorge
vast ingot
obsidian geode
#

thank y

vast ingot
#

no prob 🙂

forest lily
dawn viper
#

anyone recommend a discord where i can generate some AI Pictures?

bleak matrix
olive jacinth
#

hi frens. is there a good api or protocol for creating videos on social media? seen a bunch of reels in the past week that are morphing images with movement, character based. I created a good one with an app, but I'm looking to use the actual beta instead of relying on a third party app.

mossy bone
#

Lebon james

wooden sphinx
#

I assume I should just slowly increase CFG scale the more tags I put in

#

Especially if I am getting my tags ignored randomly

pine nest
#

how do I make a picture?

fervent thunder
copper crystal
wooden sphinx
#

Where do people share their art?

waxen pasture
severe sonnet
#

Hello friends
When I want to generate a photo in the automatic 1111, I receive this message: NansException: A tensor with all NaNs was produced in VAE. This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
Time taken: 31.30sTorch active/reserved: 3324/3910 MiB, Sys VRAM: 5565/6144 MiB (90.58%)
It used to work fine, now it's like this, what should I do to solve the problem?

warm junco
severe sonnet
#

Can this error be fixed for high quality image processing and weak hardware?
error : OutOfMemoryError: CUDA out of memory. Tried to allocate 768.00 MiB (GPU 0; 6.00 GiB total capacity; 4.45 GiB already allocated; 0 bytes free; 5.22 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
Time taken: 2.86sTorch active/reserved: 4953/5368 MiB, Sys VRAM: 6144/6144 MiB (100.0%)

pine comet
#

whats an extention

#

for stable diffusuion

#

i am new to all of this

fervent thunder
fervent thunder
#

I'd assume to increase speed thinking

severe sonnet
marsh hare
forest lily
#

i still enjoyed using it tho i liked all the faces it made with no interaction

marsh hare
#

You can literally create realistic looking faces in two seconds flat with what we have

forest lily
#

Yeah but i didn’t want to make them i wanted them made for me so i could just. reload the page

tough torrent
#

When are the XL safetensors coming out?

full carbon
#

How do I make NSFW images?

cerulean merlin
#

Has anyone installed sd cn animation? I cannot figure it out

alpine bear
young musk
#

people really ask anything

fervent thunder
#

1.5 doesn't exclude afaik, but yea...it's better to take custom model anyway from https://civitai.com/

waxen pasture
#

ok I keep forgetting but I also see it everywhere... what does afaik mean?

brisk bloom
#

as far as I know

waxen pasture
#

ohhhh

#

that makes sense

vestal dew
#

anyone used controlnet 1.1? How is it?

lost swift
#

@alpine bear ,I have small problem in SD,might need your help

#

my notebook computer use rtx 4080laptop version as GPU. but its generating speed of SD is 1.5-2it/s

alpine bear
#

Hehe, aye

lost swift
#

which is very slow

alpine bear
#

Ya

lost swift
#

🤔how to fix it?

alpine bear
#

Cudnn.dll files, and python 3.10.9

#

Use xformers too

lost swift
#

OK I will read them carefully

snow mason
#

@long wind I'm using AI for my graduate thesis project and since I'm using your models for most of my output, I wanted to credit you properly with creating them. However, as I was working on my paper, it occurred to me that saying "models created by Sir Veggie and uploaded on HuggingFace" might raise some eyebrows, amusing though it is.

Is there a better way I can credit you for making and releasing models like wlop_100?

alpine bear
#

Veggie's got their name on HF. Could just credit 'Veikka Pulsa' but I won't speak for em catlurk

snow mason
#

I wasn't sure if that was their real name or another internet handle heavysweat

alpine bear
#

But I think its real

long wind
snow mason
#

Thanks!

long wind
zenith thunder
#

Not sure about where to ask but : Does anyone here have a program they use to make gifs? I want something that makes high quality gifs that also tells me what the file size will be.

#

is there some ai powered program that can output gif vids?

#

ezgif is okay but I want something better

copper crystal
#

ffmpeg is best.. its all command line options. might have a way to only output the potential filesize

viscid musk
#

hi

zenith thunder
#

ah i see

zenith thunder
willow coyote
fervent thunder
#

Is there a bot that i can use for txtto image in this server?

vale steeple
#

POV: feeling __ having MX130 stressing itself for 8 min for 500x800 pic.

silk lodge
#

so the platform is down?

vast ingot
silk lodge
#

caught SyntaxError: Invalid or unexpected token (at auth0.f9e84e44.js:1:8532)

vast ingot
#

same for me yeah

#

same exact position for error too

silk lodge
#

not loading on my ios phone either

#

or safari desktop - so I think broken from a deployment I guess 🙂

vast ingot
#

I'll try to make sure the devs can take a look

silk lodge
#

was just about to start integrating lol

#

my mind probably broke it

remote iron
#

NansException: A tensor with all NaNs was produced in VAE. This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

#

how can i fix this

ember rover
#

oh yeah I found a bug in A1111 btw

#

who do I tell?

vast ingot
ember rover
#

I'll just leave it then. it's a pretty niche issue that has a relatively easy workaround

knotty blade
#

im trying to install atomatic 1111 where i can ask my questions ?

ember rover
bleak matrix
#

Good morning, everyone

#

How are we all today?

vast ingot
#

Hey Sunny 🙂 as Tony the tiger would say, greeeeat !

#

how are you ?

bleak matrix
#

I think I'm doing pretty good this morning, so far, all things considered! And yourself, Guiz?

#

What are you up to?

bleak matrix
vast ingot
#

I'm trying to finish captioning but...

#

it's so long ^^

#

I'm at 80% of the dataset done now

bleak matrix
#

Yeah, captioning takes such a long time to do right

vast ingot
#

I did the test by auto captioning the rest of the dataset

#

it works, but I'm doing it by hand for real

bleak matrix
#

To me, it seems that it's like.......it gets it half right, or almost right, or really good.

#

Depends.

alpine bear
#

"Mowed the garden"
Okay guiz smugshroom

bleak matrix
#

Sometimes, I see what happens with it, and then I play with it.

#

OMG

alpine bear
bleak matrix
#

He runs, but what will happen next?

#

On this episode of general chat

#

I just realized that my characters in Unreal are GIANTS compared to the character height

vast ingot
alpine bear
#

Perhaps my joke has flown over your head sfomegalulcat

fathom imp
#

Does anyone have any suggestions for a music ai, maybe one that takes samples for reference?

wild steppe
fervent thunder
#

Someone , share me a good prompt

#

I wanna test one txttoimg ai

fathom imp
clever venture
#

guys

#

did anybody here try the kohya_ss gui?

#

and is willing to help?

fervent thunder
#

what is "baked" in a model's name suppose to mean?
already include vae or lora inside the model file?

fervent thunder
warm junco
shadow birch
#

Hey guys, does anyone here use LORA models with Google Colab?

woven wagon
#

Hi, If I wanted to create an image based on both tex prompt and image, what method in python should I use ? does such a method exist? Does Stability offer sucha functionality? @vast ingot Last Question I swear xD

fervent thunder
#

ummm controlnet?

hazy ermine
vast ingot
hazy ermine
#

Look what I found

woven wagon
fervent thunder
#

how do i change which browser webui1111 opens in?

warm junco
#

You can change it in windows at standard apps

fervent thunder
vast ingot
dense agate
#

Hello Morning. Is there a Moderator here that i can possibly reach out to.
Thanks 🙏🏻

vast ingot
#

sure

#

what's happening ?

dense agate
#

@vast ingot hi 👋🏻 are you by any chance the moderator 👨‍⚖️

vast ingot
#

I am one ^^

#

I wouldn't say "the" :p

#

tell me mate

dense agate
vast ingot
dense agate
#

May I DM you I wanted to ask something in private thanks

vast ingot
#

I can't DM directly, nope, sorry, I get too many

#

but I'll see you in the ticket in a second

dense agate
fervent thunder
#

"Guizmus , The Moderator" thinking

vast ingot
#

we'll do a movie soon

alpine bear
vast ingot
#

I want to see what words I may have overused in a dataset captioning, what will be trained enough and what could need a change before I start.
I would need to dump my file names in some way and analyze the words, but somebody may have encountered this problematic already ?

hushed lichen
#

Wassup edited, because I don´t want to get banned

#

Bruh

#

The auto mod is kinda bad here

maiden bough
#

What's the relationship between deepfloyd and SDXL?

mild plinth
#

is there a way to force SD to render subject only? - with 100% black background. I did try "character sheet", "empty black background" etc. Nothing seems to work well so far. It will do characters in this manner, but i'm trying to get some vegetation.

#

(transparent BG would be even awesomer)

open garden
#

but thats just my opinion

mild plinth
#

makes sense

maiden bough
#

try (minimalist), (simplified character design), (line art)

ember flax
#

Regarding #dreambooth I wanna ask if it is a sensible idea to train my first set of photos based on , say, chilloutmix first. Then I use another set of photos to train based on the checkpoint that I have already trained by the first set of photos ?

mild plinth
open garden
#

lets try asking an ai

#

be right back

#

im using bard

#

the one that's supposed to be bad?

#

i pinged you

#

in general with images

#

i signed up when it came out

#

there's a waitlist

#

like dall-e

#

somewhat i guess

#

not really reliable for coding though

#

it gave

#

so u can look at that for reference

#

can it code acheong?

#

no

#

probably not

#

Google has just revealed in an interview they're rolling it out slowly

#

so it can do much more, but they're letting everyone get used to the tiny features

#

so it can do more than this

runic igloo
#

Hi all 🙂 I have a question. I am using Deforum in Stable Diffusion. I have an automag 1111 installed. Version 1.5 (on the screenshot).
I heard that a new model is out. Please tell me - does anyone have a tutorial on how to install a newer version?
P.S. I am a newbie, sorry for the stupid questions.

#

v1-5-pruned-emaonly.ckpt [cc6cb27103]

vast ingot
#

hey there

mild plinth
#

you just drop models into "stable-diffusion-webui/models/Stable-Diffusion/"

#

@runic igloo

vast ingot
# runic igloo v1-5-pruned-emaonly.ckpt [cc6cb27103]

well, there is a new model since 1.5, there has been lots.
"base" official models go up to 2.1 currently, XL is in beta on their website but you can't download it yet.
But civitAI has thousands of models for you to choose from

#

string some models together ? WHat do you mean ? merge existing models or train a new one ?

#

I'm no python dev though. I do a little but chatGPT is mostly there for me.
I think there is an extension for automatic1111 and civitAI

#

nope I'm not going voice, lot too tired

#

you'll need to explain more clearly what you want to do, I don't understand you there. I thought you wanted to download models : civitAI or Hugging face are the place to go

#

even through code

#

HF mostly because you can clone

#

you'll need to build the library by hands, or web scrappers for each type. there isn"t a common place for each, it's all over the web

copper crystal
#

downloading and experimenting with the models is really the only way to learn them. there's no standardized way of defining what a model can do and what it is meant for

#

hope that the author's description is reliable

#

often times it's "This is the greatest model that fixes hands!" though

vast ingot
#

you should ask the #1011228667659178055. I can't be your help on this there. I have too much on my plate to do, you're asking me to think on things I have no idea about there, and I'm far from the best person.

copper crystal
#

Start with a spread sheet to build your notes out. in time you could convert that into a DB

#

I would love to talk about modellinga DB for these purposes but again, there's no standardized way of defining a model yet. i don't know how we'd get there though. (DB Model in this context, not an AI model)

#

i'm not sure what string models together means still

#

but why call it "stringing models together" what does that mean?

#

a database of notes doesn't really mesh with that phrasing imo

#

i think you need to work on defining your goal better.

#

before asking for help

runic igloo
copper crystal
#

i contextualized notes here yes

wild yew
#

hii

#

I am new to all stabble diffusion things is there someone that can help me on DM?

copper crystal
#

You really need to define your goal a lot better sorry

#

i obviously don't understand you

#

that sounds liek notes, but you're insisting now that i made that up

wild yew
#

I want to learn where from can I download models, because civitai only gives me hashes

copper crystal
#

I did but you're saying you're not making a db of notes now

vast ingot
copper crystal
#

so i don't understand, obviously

#

now i'ts suddenly an argument

#

you don't need to bash me for not understanding. i've told you a few times we're not getting there

vast ingot
#

Let's stop there

#

This is not a place for personal attacks, nor responding with more

#

I'm not gonna sort who is right or wrong, you have the right to differ opinion. block yourself, move this to PMs or I don't care

#

those aren't the talks that are OK on this server

strong kindle
#

Hello all. I'm new to SD and have a couple of hardware questions. I have older system with an Intel Core i5 4440 and a NVIDIA Quadro K2200 with 4GB VRAM. I have A1111 installed and have been running SD locally. It's been working to generate images, but it can be a bit slow depending on what I'm attempting. I can't go larger that 512x768 at 20 steps, and I can't upscale. It's fine as I learn, but I know I'll need better resources to do more and grow. I've been looking at systems to upgrade to, and from what I've read something with a Nvidia 3060 with 12GB of VRAM is what I'm aiming for. Outside of the overall implications of have a faster processor for all computing use, is there any other key factor I should look out besides getting a Nvidia card with 12GB or more VRAM for using SD? Is there any consideration for AMD vs Intel, or if I go with AMD consideration between a Ryzen R5 or R7?

vast ingot
#

I think I have said my part on this : if you want to continue discussion on this level, take it to PMs with them. I warn them too on exactly the same thing, this ways explicited for both of you, not just one

fervent thunder
#

I think generally there most popular upscalers are

  • ultimate SD upscaler
  • 4x-UltraSharp

Can't help with link on 1st, never used it.
Your link has link to ultrasharp, but it's...uh...not for "everything", let's call it that

I don't do upscaling , probably someone else can provide more info

zealous maple
#

Is there a comparison of different fine-tuning methods? I have a custom domain-specific data-set of ~10,000 images of items that I want the model to be good at generating variations of.
How can I find out if I should train using lora or regular continued training of the SD model?

abstract quarry
#

I would try Lora with very low learning rate 🤷‍♂️
in theory it should be comparable to full fine-tuning, although some people report they had problems with it

ruby gull
#

Hi all, new to the community!

#

Just wondering if there was a YT resource for prompt tips and fine tuning basics for auto-photoshop SD plugin specifically?

#

at the moment it feels like I'm just throwing paint at the wall and hoping it makes what I ask for lol

#

I should say that I looked through the 'start here'thread' too but couldn't see any video tutorials or anything?

fervent thunder
#

How to use the bot

#

Usually authors tell in which folder you should put it
(if it's not in description, I have no idea )

#

I have only 1 upscaler downloaded manually and it's .pth

fervent thunder
shadow wyvern
#

greetings, how do i manage to create a seamless pattern?

zealous maple
#

How do LAION, Google, Stability AI, etc create labels for billions of images for their diffusion models?

They aren't human labeled right? That's too many images..

copper crystal
wraith magnet
#

What is the best way for me to learn how to make photorealistic images? My people keep coming out looking like monsters with lazy eyes.

zealous maple
#

Interesting.. does that really translate well into prompts?

#

One thing that doesnt' explain is how Stability AI and Midjourney have solved the problem of creating prompts for billions of images

echo bane
#

I've just started using stable diffusion with automatic1111 yesterday and I was looking into inpainting. Using version 2.1 I generate some image and then send it to inpainting but results are really poor and I was wondering If I must install or configure something or maybe I need to use another model for inpainting. Any help or tips ?

zealous maple
#

🤔 How does that help you with descrbing images though. As far as I know, the input to training diffusion models is an image and a description, not a classification.

I.e "a grey cat in a yellow rain jacket" How does classification help turn that into a training sample prompt for a diffusion model?

copper crystal
#

theres pretraining too. clip in stable diffusion 1.5's case.

lunar raft
#

ahoy

rancid tangle
#

Can you "paste" an object previously generated inside another image?

ocean geyser
#

There used to be a channel where people would generate textures

#

I thought there might be a thread for it, but I cant find it if someone can link it?

clear pasture
#

dont know were to ask this but, i've been searching for the embed:bad_quality": "c1c5471862. Can't find that embedding, please help to were to download it.

wild steppe
#

You'll need the community archive role from #👥｜roles to view the 2nd link

gleaming furnace
#

Does discord delete metadata on an image when uploaded? If I download some AI art will it fit into PNG Info?

brisk bloom
ocean geyser
thorn spire
#

Im looking to upgrade my GPU currently Nvidia 750 with 2GB vram, stable diffusion 1.5 is pooping when I push past 1024x768. But leaping into the RTX 4090 with 24GB is a bit painful. What benefits (perhaps resolution?) besides speed that I can only unlock with the RTX 4090 that I can’t with the RTX 4070 with 12GB?

fervent thunder
#

for stable I dont think anything

#

quite a bit for games I suppose

#

hard to top a 4090

#

you guys seen any place to make ai music thats on the level of stable for images?

marsh hare
#

4090 is overkill unless u got money to burn

#

at least if ur serious about diffusing, cuz XL is probably gonna be a vram hog, at least during the first stretch

kindred wraith
#

If you have other models you like better for SD is it safe to delete v1-5-pruned?

#

I'm trying to save space where I can.

marsh hare
#

Yes, and no

#

base 1.5 is way too valuable to delete IMO especially if 1) it gets blown off the internet and 2) it's a good idea to use it as part of a blended model if u ever make ur own merges (which you should)

fervent thunder
#

(including me in most cases)

warm junco
fervent thunder
#

hm...let's see

#

yea, you're right, it doesn't have info anymore

narrow ice
fervent thunder
narrow ice
#

it's don't work

#

aw

#

It's work

#

😄

#

thanks

warped sedge
#

is there a way to pause rendering?

solid bough
#

hello everyone! first post here so sorry if it is in the wrong place....
You guys know is ConfyUI is on discord? looks like an interesting project but I cannot seem to find any community for it.

vast ingot
#

hey there Niconico

#

well, I do love comfyUI, and it gets talked a little more rencently around here

#

but I don't know of their dedicated community for it

#

you can find quite some workflows around though

warm junco
solid bough
#

I come from vfx background and the idea of a procedural production of ai art is very tempting the node workflow could really be a big step onthat direction.

abstract quarry
wheat stump
#

привет

bleak matrix
#

Good morning, everyone!

fervent thunder
#

why there suddenly come out a downloading in shell when I generating a image and it stuck on 98%?

#

I am confusing, why would this downloading happen?

warm junco
fervent thunder
# warm junco You activated restore faces it seems

sorry, I don't quite understand.
this is my positive prompt:"a 19 years old girl, best quality, masterpiece, (realistic:1.2), 1 girl, brown hair, brown eyes,Front, detailed face and breast, beautiful eyes, small breast lora:koreanDollLikeness_v15:0.1, white jacket, jeans, fashion, smaller head, fashion girl, wrinkle t-shirt, hair covers ears, hands in pocket lora:japaneseDollLikeness_v10:0.2, real face, real skin, realistic face, realistic skin, rough skin"

#

oh it's an option

#

I see.

#

I will learn more from google. thx you so much

bleak matrix
shadow birch
#

hey guys, what are your favorite LORA models?

upbeat leaf
#

Hello everyone, how’s it going?

robust lodge
#

The previous versions of sd run perfectly on 3090s

clever venture
#

guys

#

what prompt should I use to make the skin of the character like literally black?

echo bane
#

can someone help me understand how to find appropriate inpainting models for sd 2.1 ?

rugged zinc
hoary plover
hoary plover
#

wget works tho

tough jungle
#

damn fail

rugged zinc
#

Videos don't load in browser, they just download

#

But the file was too big to embed in the message

tough jungle
rugged zinc
#

Sweet

tough jungle
#

this one is 35mb

oak holly
#

Can anyone give me a link how to make a photo od multiple characters in it?Im trying to make my dnd party.

zealous maple
#

Are there any benchmarks comparing Nvidia GPUs vs AWS Inferentia for StableDiffusion-like inference tasks?

karmic brook
rugged zinc
#

It's the thought that counts suffoPrayge

karmic brook
abstract quarry
placid sage
#

trying to find SD 4.5
what channel would it be in? or just link pls

lost latch
#

Hey folks. My name is Vatsal. I am a Computer Science student. I have received a research fellowship in deep learning and a data science internship at a top vc backed startup. I’m currently doing ML research at Stanford.

I am looking for people with experience fine tuning stable diffusion for Text2Image with excellent results. I have a custom dataset with image-caption pairs that I built readily available so a decent amount of the work is done already. Didn’t have time to do proper fine-tuning cause I was so busy.

I am really looking for anyone with significant experience fine tuning stable diffusion that can help me wrap up this project with good results.

Ideally we would also package the product and ship it as a web app.

slim raptor
#

guys, is 378 words too long a prompt?

young ivy
#

damn I forgot. Does anyone remember what the name of the extension for the one that checks your model for CLIP deviations, numbering etc?

daring cypress
#

is the local url down rn?

foggy atlas
#

guys. what happens if you use a Lora as the checkpoint?

young ivy
#

you can't

fervent thunder
#

Are there any photoshop plugins for comfyui? I'm having a hard time finding it with search engines if there is

oak holly
gentle blaze
#

Would any one be willing to help make a picture for my Kick Starter. I want good art and this does take practice and skill for good results.

ivory merlin
#

I'm currently running A1111 off of Colab because my current mac is from 2018 with a intel chip. I'm considering upgrading to a new Macbook pro with the following specs so I can run it locally. Would this be enough? Any limitations with running it locally on a newer MB Pro?

Apple M2 Max with 12‑core CPU, 38‑core GPU, 16‑core Neural Engine
64GB unified memory
2TB SSD storage

shadow birch
#

Guys, is there any colab link that loads the lora models from google drive?

fervent thunder
#

It creates folders with Embeddings,models, loras

shadow birch
fervent thunder
near silo
#

I am interested in that 2 week challenge

#

checking out the songs to see if I can do anything cool with them

#

I need to resonate with these songs in order to try to do this challenge, but the signs are not good 😅

thin widget
#

I created it for the 4090 Grotto community on Discord. If you wish to join feel free to DM me 🙂

shadow birch
near silo
#

I just listened to all of the songs in the challenge and unfortunately I am not feeling these songs at all. As much as I would love a shot at $1500, I just can't sink the proper amount of passion into songs I don't get anything from, personally

Best wishes to all of those that compete. Would love to see what you all do!

#

I get the artistry of the songs, and they are very well made from one music producer to another, just not my cup of tea

fervent thunder
near silo
#

Yeah, that seems like an extremely oversaturated thing to document IMO, but who knows

charred iris
#

hey guys

#

I had a question concerning SD

#

is it in any way possible to generate images using negative prompts only?

coral hound
#

@drifting lagoon

https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence

All you have to do is prove/show that your prompting as a human determined the expressive elements/details/style in a prompt. Attention/emphasis, break padding, negative prompts, etc. You can refine your prompting in SD to exercise control over what's generated and how, not letting the AI do it for you.

And really, even if prompting alone wasn't enough, and in most cases won't be due to people just typing in a single line and pressing generate... people that actually work on and refine their AI art beyond just prompting, such as... inpainting/outpainting details, or any other kind of human controlled influence, they'd be able to copyright it.

drifting lagoon
coral hound
#

So does legislation in that document not read that as long as you have sufficient human influence on the overall work and not let the ai handle it, you can copyright the work?

#

Because the reasoning why simply prompting alone wouldn't work is very clearly defined.

#

Which means all you have to do is make it fit the criteria. And you can do that with prompting.

#

To be more specific. The documentation reads that you prompting doesn't qualify in the case of the AI deciding on the elements of composition for you.

drifting lagoon
#

I don't really see how they can make it any more clear than #27 here: https://www.federalregister.gov/d/2023-05321/p-46
The prompt could be protective if it is creative enough, but not the outputs themselves. They view writing the prompt in a similar way to giving instructions to an artist for a commission. It doesn't matter whether you tell the artist you want them to "draw a girl with pink hair and brown eyes in anime style" or you write an entire book describing every detail of the image. In both cases the artist created the image and so they have the rights to the image, not you. You could copyright the book that you wrote to describe the image, but the artist would still have the rights to the actual image. The only difference here is that the "artist" in this case is an AI and therefore can't hold rights to the image so nobody does.

inner wave
#

Is there a way to put an image sequence (from a video) and have an sort of animation as a result?

broken smelt
#

yea thats how animations work even without AI

inner wave
#

i mean with stable diffusion xD

copper crystal
#

"While some prompts may be sufficiently creative to be protected by copyright, that does not mean that material generated from a copyrightable prompt is itself copyrightable." The actual words. It doesn't exclude new copyright either

inner wave
#

like i have an image sequence, how can I have it exported with some prompts, for example? Like you would do with a still image, but in this case several images that would result in a sort of rotoscope

copper crystal
#

imagine reading the cited link

meager eagle
#

How do I install loras in google colab

coral hound
# drifting lagoon I don't really see how they can make it any more clear than #27 here: https://ww...

And nowhere does anything say that any/all material generated from a prompt can't be copyrightable.

The office gave an example on prompting not being enough because of how their understanding of how prompting works, specifically stating that due to a lack of control of how the AI understands and generates prompts, it's not sufficient enough to say that a human had influence and creative control on the output.

drifting lagoon
coral hound
#

It says it doesn't mean that it is. It doesn't say that it can't be.

thin widget
coral hound
#

Go on? What does this have to do with what we're talking about?

#

Because my whole point in this conversation/the overall topic is that the more human input/influence is involved, the higher your chance of being granted a copyright. And the more the AI does, the less your chances. Because it's case by case right now.

#

It's due to their current understanding of how prompting works that has them use prompting as an example.

#

You can be granted a registered, documented legal copyright. That is necessary in order to go to court for your copyrighted work.

#

You have a general copyright for the things you make. Yes. But from a legal standpoint, the copyright has to be registered on a legal record with documentation.

#

This still doesn't have to do with what we're talking about. I knew what you're saying already.

#

You can't even go to the court and sue if you don't have a registered copyright.

#

This conversation isn't about the prompts

#

It's about the AI generated material from the prompts

#

Being registered for a copyright or not

#

And nubby said that the work cannot be, when nowhere in the documentation does it say that.

#

Correction, edited

#

I understand where you're coming from, I'm not disagreeing with what you're generally saying right now

#

But purely in this documentation

#

It's saying that just because you prompted something, it doesn't mean you can register it for a copyright

#

But what I'm saying, is that it's not saying you can't register it for a copyright

drifting lagoon
#

I'd be happy to be proven wrong. For now, the documentation put forth by the US Copyright Office states that the product of a generative AI requires additional human authorship in order for a copyright to be registered. It's quite clear as far as I can tell, and I have not seen any specific line showing otherwise.

coral hound
#

If I had seen a single line saying that AI generated work from prompting alone cannot be registered for copyright, it would've been case closed.

drifting lagoon
coral hound
#

But, one. The offices understanding of the technology is new, as is almost everyone's.

#

"Based on these developments, the Office concludes that public guidance is needed on the registration of works containing AI-generated content. This statement of policy describes how the Office applies copyright law's human authorship requirement to applications to register such works and provides guidance to applicants.

The Office recognizes that AI-generated works implicate other copyright issues not addressed in this statement. It has launched an agency-wide initiative to delve into a wide range of these issues. Among other things, the Office intends to publish a notice of inquiry later this year seeking public input on additional legal and policy topics, including how the law should apply to the use of copyrighted works in AI training and the resulting treatment of outputs."

#

So things could be subject to change, and likely will be. For better or worse regardless of what side anyone is on.

#

And two, again. Nowhere does it say that AI generated work itself cannot be registered for copyright.

drifting lagoon
#

Yes, hopefully things will change in some ways. I'm not talking about how things should be, or how they may be in the future. I am only talking about what the document we have says right now.

coral hound
#

It just said that it doesn't automatically mean just because the prompts can the generated work is.

junior sluice
#

I have 0.6 credits left and in dreamstudio I do something which says below takes less than that, but it still tells me "not enough credits". What could be going on?

coral hound
drifting lagoon
#

The issue is that they do not consider you to be the creator if you prompt a generative AI for an output. You created the prompt and so the prompt could potentially be copyrightable, but the AI created the product and so it is not copyrightable. Modify the output enough and you can claim to be the author. How much you have to modify it is the only real question left.

coral hound
#

I mean. It seems they've been "generous" with allowing some AI works to be registered at all. So I think they just need further specific guidance and understand on exactly how the process works. And even then, most things will have to be a case by case basis and defended/proven/"argued" in court.

coral hound
#

Yeah, fair use has pretty much always been a thing, which is why we can use models trained on other people's work legally to begin with.

near silo
#

What did I miss-

#

I came to chat generally, but stayed for the rage

drifting lagoon
near silo
#

When did this happen? I remember the ruling of the US Supreme Court and US Department of Copyright

drifting lagoon
near silo
#

The SC and DOC ruled that AI works cannot be copyrighted right out of the Ai itself, as the process can be streamlined to be pure Ai to AI.

They said that all things AI generates are public domain, as their data sets were public domain. However, if you can prove transformation of the original content, you are still eligible for copyright.

Ie, a drawing, a hand edit, an input of your own creation, really anything that would lead to a normal public domain image being copyrightable

#

So an AI image cannot be copyrighted unless you had some provably human input. For example, that includes using a photograph you took as an input.

coral hound
near silo
#

but its also a great way of saying that the average person on some sketchy AI generation app on their phone won't be able to copyright low effort crap they just happened to spam the keyboard to make.

Its saying that the general masses of people who put in no effort cannot copyright

near silo
#

We had a whole breakdown of it less than 24 hours after it was ruled in this very chat

coral hound
#

Nowhere in that document did it say the generations can't be.

near silo
#

i actually have the document as a PDF on my PC

coral hound
#

It used prompting as an example, and the example isn't even completely accurate/always applicable to every program.

near silo
#

Damn it, can't send images here

hallow plaza
#

Whats the best way to train a model on realistic people? My friends want to do ourselves as fantasy character portraits. I tried using Lora, but I keep getting stuff that sort of looks like us or is cartoony. Is dreambooth better, or Everydream2? Or are there some really important lora settings to get percises results?

#

So far in my test Lora < Dreambooth

#

But I dont know between Everydream or Stabletuner which is best.

shadow birch
#

does anyone know if in camenduru's colab, we can change the code to make the lora models be loaded from our google drive folder?

#

exactly, I shared it on google driver access for those who have the link. so, I put the link in place of the higginface url, but when running the code, it gave fatal: repository not found

#

hmm... will only git repository work?

#

ok

#

I replaced the line of code:
!git clone !git clone https://huggingface.co/embed/lora /content/stable-diffusion-webui/models/Lora/positive

for this:
!git clone "the drive folder where the models are" /content/stable-diffusion-webui/models/Lora/positive

#

I understand, is it the beginning there? I need to change !git, is that it?

#

and if I would change it to: "!wget"...

#

ok, thanks bro!

#

Thank you for your help

fervent thunder
modern crater
#

Guys, is there a website that can generate SD with chilliout model and korean likeness lora?

delicate mural
#

ty

modern crater
#

Thanks, will try it

thin widget
#

The installation instructions there are good enough, but I've provided some improvements in my guide that will lead to a performance increase and better maintainability. Hopefully people will find it useful. If it's not useful to you that's totally fine.

sinful otter
#

hi

bleak matrix
#

Thanks for letting me know. 🙂

#

<3<3<3

#

Whelp, tis almost midnight! 🌙 Off to sleep I must go. Have a wonderful evening, everyone!

visual gale
#

Is there an official Stable Diffusion API from the Stable Foundation, like the way other gen-AI companies have commercial APIs (e.g. - OpenAI/Dalle-2 API? I did some web searches, and it seems like the top results are just ML service providers (HuggingFace, etc.) that have Stable Diffusion instances available publicly.

ocean geyser
#

Are there a few websites where people upload the pics with metadata?

glad pond
#

Hey I'm new to Stable Diffusion and I'm not sure how this works where do I learn about that and how I can use this program. Also I heard Stable Diffusion is powerful enough to make ultra realistic looking humans is that true as well. Thank you in advance for answering my questions.

broken smelt
#

i learned all by myself just by installing automatic1111 and trying it out myself tho, theres really not that much to it

#

theres instructions on how to install it if you scroll down

thick otter
#

@broken smeltmy output pics are less saturated than the preview, what do you think is causing this

broken smelt
#

i wonder if olivio ever made that VAE explaination video

#

SD first makes images in a 64x64x4 square, the VAE is whats used at that point in the process. the base model installed with a1111 comes with one inside. you can change it though, so look for fun ones to try out

warm junco
#

Its needed for Color and Detail correction.
Most models have it already included.
But most anime models and merges with anime models not

broken smelt
#

its just a trained model, your question is the same thing as asking how a phone works which isnt really necessary to use the phone tbh

#

unless you really want to get into the ML field of research lol

warm junco
broken smelt
#

put another way, if u were to inpaint an image and chose the "latent image" option only, the picture would be a big colorful blob, since that's the latent image.
a vae decodes that and turns it into what we know of as an actual image and not those blobs

#

like i said interesting stuff but in terms of actually making art, not really necessary to know

glad pond
#

Okay so this doesn't work like Midjourney on a discord but on a computer program instead? VRAM is what exactly. So I'll also have to get the latest python and something caked Automatic1111? This sounds so different from Midjourney...

#

Okay. So in comparison Midjourney is crappy compared to Stable Diffusion?

broken smelt
#

ya the very first iterations of ai art gen was all done in google collabs i believe, midjourney does something similar where you dont run it on your computer. you dont have to pay fees though if you have a decent computer and run it yourself, which is why automatic1111 is so revolutionary imo

glad pond
broken smelt
#

yep you can just follow the instructions there, i think youll need python and git

shut edge
#

yes

glad pond
broken smelt
#

yep thats correct

#

git is just how developers pass their code around pretty much

#

the instructions go over how to install and use it

#

its fairly simple but if you need, i believe there are some step-by-step videos too that you can google for

warm junco
broken smelt
#

heheh, thats the same guide that i followed that made me download 3.10.6

#

we come full circle

warm junco
#

xD yea because on Automatic its still the "recommended" version, he should update that to 3.10.9

glad pond
glad pond
#

How is Stable Diffusion with Hands and Feet?

#

I get tentacles pretty much with Midjourney.

warm junco
glad pond
# warm junco Its difficult at first but not impossible. It depends on the model you use, the ...

I've seen people make almost flawless characters with this program and I want to make allot from this. My goal is to make 3 Anime Novels turn 1 into an app game and all 3 into Animes with a program that can give me free range. Midjourney has soft bans on words and when I tried to create a character with a Bloodred Dress on it let me create the character then erased it once the program discovered the word BloodRed...

broken smelt
#

yeah censors are lame

glad pond
# broken smelt yeah censors are lame

I was trying to show the evolution progress of a Slime to a Slime Humanoid male and female. And even though they had no private body parts showing, just because they had Humanoid forms and the body was pretty much the same color the creation was erased.

warm junco
glad pond
#

I felt that way.

broken smelt
glad pond
#

True

warm junco
glad pond
glad pond
warm junco
#

It will not teach you how to use it xD

#

Just the basic get ready stuff

glad pond
#

Okay that's what I need to... Lol...

warm junco
#

You learn more by Experimenting with it. For Questions feel free to ask in this dc

broken smelt
#

yeah imo the best way to learn is to just install it, throw a few runs and see what different stuff do

#

kinda similar to learning photoshop for the first time

#

you could watch a whole video on everything but thats boring, id rather just click on everything and see what they do first for a few mins on my own

#

in my experience, that first learning session helps you learn from the videos better

#

but mabey thats just how i learn

glad pond
fervent thunder
#

Double the efficiency.

Generate while you watch videos on how to generate ✅

glad pond
glad pond
fervent thunder
#

I mean...youtube doesn't take alot of resources

warm junco
fervent thunder
#

✅ watch videos on your phone then

#

Idk , I always have browser on, if what I have after that - not enough for SD - That's SD problem, I'm not turning youtube off for it 😄

random ice
#

evening...

fading parrot
#

any suggestion on which model best use for horror, lovecraftian image generation ?

random ice
#

or morning, rather. Anyway I had a question - is there some way to exit A1111 webui on Linux, without hitting Ctrl-Z? I always end up having to terminate the dead process, after that

random ice
fervent thunder
#

Probably something that can do dark images...?

random ice
#

yeah, there are horror themed loras and checkpoints

fading parrot
topaz elk
#

for anyone curious, I may have just discovered that you can use walmart eyes in 1.5 and give that extra weight to give people blue eyes without making their outfit and everything blue. may put people inside walmart if you aren't specific. do with that what you will

topaz elk
glad pond
split fossil
#

hello which chat can i ask questions about the installation of stable diffusion

vast ingot
outer parcel
vast ingot
umbral tulip
#

sry for late follow up; but got any specific models in mind?

quick eagle
#

hi

ember rover
quick eagle
#

okay, gtg

ember rover
#

lol

#

cya

quick eagle
#

back

#

i created this account in school as a joke

#

now i am with my main

fervent coral
#

hi

bleak matrix
#

Hello everyone 🙂

little rampart
#

Noob here in deforum 2d how do I control the heads and objects from getting cut off the top screen wich axis do I use ?

marsh rock
#

Has anyone used gif to gif in auto 11 11 - I only get 1 image from a GIF

#

Help!!!

muted jungle
#

the github doesnt work for StableLM

steel frigate
#

repo gives 404 lol

eager phoenix
#

link ded

steep vector
#

yoo they just dropped a language model

carmine coyote
#

nice github 😎

gray fiber
#

lol did GitHub rug?

quiet ice
#

404 moment

broken gull
#

404 yes

gentle ledge
karmic plinth
#

404 pog

broken cave
#

lol

muted jungle
#

stable diffusion but text

merry pendant
#

404 lol

karmic brook
merry pendant
broken cave
#

i think they didn't realize it's still a private repo

past flax
#

1st of April yet again

karmic brook
#

coming up shortlyyy

buoyant jetty
merry pendant
#

interesting to have a new LLM to play with. lots of LLaMA stuff going wild lately

broken gull
#

yes guys it's working!!!!

ember rover
#

are you guys able to run llama on your personal rigs?

muted jungle
#

been waiting for open source language model

#

thanks stable team

ember rover
#

or are yous using cloud?

broken gull
cosmic kettle
#

Did they forget to make the repo public lmao

karmic brook
cosmic kettle
#

The one in the announcement doesn't exist. Seem to be either private or typo

#

Huh

muted jungle
#

how much vram thinking

cosmic kettle
#

I'm guessing a bit over 16GB for the 7B

#

At FP16

maiden bough
#

github link is dead

carmine wedge
#

yup

buoyant jetty
#

Ugh. So many channels here. I don't remember where I spotted that announcement ...

lean terrace
#

yep

carmine wedge
steel frigate
#

it alive now

maiden bough
#

awesome

tough jungle
#

the huggingface spaces link on the repo is a 404

steel frigate
#

huggingface link in repo still private though lol

storm shard
void egret
#

so like

steady arrow
#

Why did they make an LLM based on the terrible pile

void egret
#

precisely WHO is going to be able to run 65B?

dusk prism
void egret
#

wouldn't this take like an A100.

karmic brook
#

live nowwww

void egret
#

i don't think even a 3090/4090 can run it.

cosmic kettle
#

HF repos are private too trollskull

dense stratus
#

hugging face down?

steady arrow
void egret
#

unless ya'll did some wizardry with vram costs.

cosmic kettle
#

Maybe the announcement was made in haste?

broken gull
#

so how to download and use it?

merry pendant
#

now to see if it runs in oobabooga

cosmic kettle
steady arrow
#

And all the other models are equivalently as bad as gptj and gpt neox

maiden bough
#

4096 context window!

cosmic kettle
#

What architecture does it use?

maiden bough
#

excellent

cosmic kettle
#

NeoX?

steady arrow
maiden bough
#

NeoX was pretty alright

storm shard
halcyon granite
#

is the language model thing like chatGPT and can I run it locally? maybe use it in my game?

wild steppe
steady arrow
maiden bough
chilly crow
#

Is this coding focused?

cosmic kettle
zealous solstice
#

is that one still a buzzkill? as in it wont respond to controversial things

cosmic kettle
#

But a 3B model on 800B tokens sounds promising

nocturne gull
maiden bough
cosmic kettle
#

Excited to see how it turns out

broken gull
chilly crow
#

Sound really cool

#

Can't wait to try it out

maiden bough
charred mesa
#

huggingface stuff still not public yet

#

very sad

#

NVM

void egret
#

i am strongly doubting anyone can like

charred mesa
#

ITS UP

void egret
#

actually run the bigger models.

#

at all.

chilly crow
#

Yes up

marble minnow
#

How do I run LM? 👀

smoky bolt
#

just heads-up HF link is broken

tough jungle
#

in the notebook

smoky bolt
#

But gz exciting all!

vast ingot
broken gull
#

yeay I saw

maiden bough
halcyon granite
#

more competition is better

chilly crow
#

Where can we submit feedback for StableLM?

little sleet
void egret
broken gull
maiden bough
steady arrow
void egret
#

ahkay well

#

yeah

#

basically no one is running that.

broken gull
charred mesa
void egret
#

not many people have dual 3090s

maiden bough
void egret
#

or an A100.

steady arrow
#

Why go smaller than llama ugh

void egret
#

like for all intents and purposes just about no one is running 65B

#

i'm not about to drop 5,000$ for it.

steady arrow
#

Redpajamas is doing 1.2t on 7/13b

charred mesa
#

3B and 7B sound good, especially if we can quantize them to int4

broken gull
nocturne gull
dense stratus
#

is the chat on hf using my gpu?

maiden bough
# 
StableLM-Alpha models are trained on the new dataset that build on The Pile, which contains 1.5 trillion tokens, roughly 3x the size of The Pile. These models will be trained on up to 1.5 trillion tokens. The context length for these models is 4096 tokens.
charred mesa
#

int4 quantization will truly democratise it

steady arrow
charred mesa
#

and also CPU CPP ports which use RAM instead of VRAM

void egret
#

because right now they are totally bonkers

maiden bough
chilly crow
#

4096 token context length
Double than llama

#

Amazing

vast ingot
charred mesa
#

niceee

steady arrow
#

Emad come clarify this ':c

broken gull
dusk prism
broken gull
#

it's taken so long to load 🥲

dusk prism
#

starship launch as well

halcyon granite
#

hopefully Stable model will be less woke

broken gull
charred mesa
#

175B planned too ?????

I guess if DreamStudio or something offers it as an API I guess 🤷‍♂️

vast ingot
maiden bough
#

dev says "1.5 T. Wait no let me double check"

charred mesa
#

"We would love to port llama.cpp to work with StableLMs"

That would be awesome! 👍

untold ocean
#

The Tuned model seems to be referring to instruct finetuned from various open source datasets

red dove
#

how do i use Stable LM im not very intimate with coding so can somebody give me a guide?

steady arrow
#

Sadly it's with alpaca dataset as fine tune instead of gpt4-llm which is alpaca but with gpt4 outputs instead -_-

vast ingot
charred mesa
#

Still no answer from the AI yet XD

steady arrow
#

Everything they've done is like two steps behind llama 😭

charred mesa
#

if only llama was truly open ffs 😫

#

it's research only and non-commercial

untold ocean
steady arrow
charred mesa
#

it would have almost single-handedly fixed the open source LLM issue, as I personally think Llama is one of the BEST models

steady arrow
#

So that same applies regardless lol

charred mesa
#

because ClosedAI 🤷‍♂️

steady arrow
#

They used alpaca dataset to fine tune, which is gpt3 outputs, while there is a gpt4 version of alpaca by Microsoft called gpt4-llm

#

Stability fine tuned on gpt3 outputs @charred mesa

cosmic kettle
#

I'll try and port the 3B to ggml. With its small size, should be able to run on about 2GB of RAM. Though the context length worries me a bit. Might bump up the requirements to 4GB

steady arrow
#

It's literally part of their fine tuned stablellm model on hf

vast ingot
charred mesa
#

the "experimental dataset" is just a bunch of GP3 outputs??

#

oh ok we'll go there

maiden bough
steady arrow
maiden bough
#

imo that's better than gpt4

steady arrow
#

To fine tune with

charred mesa
#

oh the Tuned one

#

Damn!

steady arrow
#

When they could've used gpt4 version of that dataset that's my sadness

alpine orbit
#

Hey guys, is their a way to invest in this project?

narrow sluice
#

Can someone clarify if they used EAI's V1 or V2 Pile?

steady arrow
bleak matrix
#

Yeah, if you have questions, or would like to discuss, please feel free to head over to #1098025024541167646 there to give us your feedback, etc!

steady arrow
maiden bough
#

Where did redpajamas get their dataset?

steady arrow
#

The same places llama did, they are all opensource copyright free datasets

#

Afaik

narrow sluice
steep vector
#

Oh boy I bought an RTX 2060S 12GB a while ago and I hope I can use the 13B model

void egret
#

6.7b typically already takes around 12GB on most models.

steep vector
#

yeah that's what they said when SD was released and only high-end customer GPUs could run it.

void egret
#

unless they pulled some serious wizardry with this i wouldn't get too excited.

#

but hey, i'd love to be wrong.

charred mesa
#

Guys, hopefully 4-bit GPTQ support comes to StableLM

void egret
#

would be nice to see 65B be possible to run on reasonable hardware.

charred mesa
#

which will drastically decrease VRAM requirements

void egret
#

and not dual-3090s or whatever

charred mesa
#

it is already possible on LLaMa, OPT and BLOOM and it's great

steep vector
#

I bet we are going to see some insane optimizations from the community and medium range consumer PCs will be able to run 13B models in a few months.

charred mesa
#

next best thing would be the CPP ports (like llama.cpp), which have quantization by default which are good enough

#

it's slower, but it uses RAM instead of VRAM, and I suppose a lot of people have more RAM than VRAM 🤷‍♂️

median anvil
#

Man, with all these AIs I hate myself for buying RTX3070 instead of 3060 thinking that despite having less GB it would be better for gaming...
And then SD was released 2 months later

charred mesa
#

oh yeah the 3060 12GB

#

it's so weird that THAT gets 12GB specifically :/

vast ingot
vast ingot
signal dome
#

why is the download so slow ? for sd language model from hugging face

#

i am getting like40 kb

#

8hrs to download bruh 💀

vast ingot
#

well

#

possible that, given the 4 models just got released, HF has a hard time following the huge demand

#

hey Flex 🙂

signal dome
#

yoooo hi there! @vast ingot

#

oh ye possibility the overload

vast ingot
#

I'm cloning it too, I have 8MBytes internet usually, and it goes at 600k right now

#

but I started downloading quite fast after it just got released, so not sure how their "queue"works

#

I'm just happy it hasn't crashed yet

signal dome
#

looks like we gotta wait tomorrow

#

cuz even tho ama is finished peoples gonna be rushed to download again

vast ingot
#

I still need 3 files of 9+GB so... yeah, tomorrow seems right

signal dome
#

would i be able to load this with alpaca electron ?

vast ingot
#

I haven't used alpaca electron. it's a UI for LLMs ?

signal dome
#

yep lemme show u where do i show u

vast ingot
signal dome
#

ohk

vast ingot
#

so I guess it should also work in there

#

still, the guy that said that, also said it was using 18GB VRAM

visual gale
vast ingot
#

dreamstudio

#

but you are also able to download/run an endpoint yourself, if you so choose

sly eagle
#

Sorry if this a repeated question but, can someone explain me why during training my model, it doesn't mimic well my drawing style and just draw a buch of abstract lines? Is like it doesn't recognize well the images or something

daring ledge
#

Hi all.. is this discord just for StableDiffusion, or can I chat about other related things or potential ideas here?

#

I note with interest some new language models, and the page for them linked me to this discord.

vast ingot
daring ledge
#

Unlike some other LLM's (not mentioning any names)

#

I'm delighted at the strength of responses 🙂

#

Now to move on... The StableLM relase is just a model, what clients can it be used with?

#

I was following the development of something called GPT4ALL, and (when I get a new computer) if StableLM can be used with it?

vast ingot
daring ledge
#

Are there plans to enable StableLM to be tied up to StableDiffusion directly?

#

What's needed to run StableLM locally?

vast ingot
fervent thunder
#

Can someone help me in VC

vast ingot
#

I don't see how anything would stop this now

fervent thunder
#

I want to create a LORA model but don't know how to create one.

#

Can someone help?

vast ingot
#

sorry Jack, I have almost not the time to even come here right now, I can't. I'd love to when I get the time, but this is a rush day for me because of StableLM

fervent thunder
#

Can anyone else be able to help?

vast ingot
daring ledge
#

I'm also trying to figure out if a sutibaly trained StableLM like model could be used on Wikipedia (as an assistant) , Using LLM generated content in actual articles though seems to be not allowed.

vast ingot
fervent thunder
#

Can anyone help me please?

fervent thunder
#

I am using Vladmandic/Automatic

#

I am in Diffusers 2 VC

#

@marble cedar

visual gale
sleek osprey
#

its 4/20 where deepfloyd

vast ingot
grave totem
graceful hedge
#

Can stable diffusion be run on an RX 5700?

strange crest
#

Hey everyone
I wanna apply for stability ai program and I've got a great idea I'm just unable to find the link
Could someone just guide/redirect me

copper crystal
#

sometimes i feel like southpark got it right with how we'll interact with AIs. Awesome-o 5000 nailed it

broken smelt
copper crystal
#

i think it might be good for python2 stuff but maybe and also i have no idea

silver vortex
#

i'm trying to fine tune a model on 15 images of my friend, using dreambooth and v1.5, but during inference it doesn't actually preserve what she looks like that much.

using prompts like "zkz pro headshot 8k linkedin", it doesn't produce results which preserve her face

also just prompting "zkz" it sometimes just gives an image of a car?!
any ideas what im doing wrong lol

broken smelt
#

this is a lora right?

silver vortex
broken smelt
#

ah okay

silver vortex
#

i did use LoRAs for this use case and it was ok but i was told i could get better results by fine tuning

broken smelt
#

some people achieve some crazy good finetuning i agree

visual gale
maiden bough
daring ledge
#

Okay what license is it using?

#

(Aside: If Stable Diffusion ever decides to opt for a Creative Commons License... (cough cough) )

empty rivet
#

hello everyone, sorry if my english is not very good, I'm a french director/technician who work in french film industry, I would like to know : Dreambooth doesn't work anymore ? Who replace dreambooth ? With colleagues from the profession we trying to create a movie with stable diffusion for character and gen-1 for landscape

#

we saw "Rock,Paper, Scissors" and corridor's tutorial and they don't use controlnet and new plugins, we are late of the technology that's why I am here tonight

untold ocean
#

test

#

ok cool haha

waxen pasture
#

big

empty rivet
#

if someone can help us it would be great x)

waxen pasture
gloomy iron
#

Today, I was trying to generate some large images using Stable Diffusion. Instead of directly using the hires option, I first created smaller images and then used the img2img feature to enlarge them to more detailed larger images.
My question is, are the sampling steps adjusted automatically? I always chose 150 steps, but earlier the command prompt window displayed 76 iterations, and just now, the same image showed 91 iterations in the command prompt window. Could anyone please explain what's going on?

empty rivet
waxen pasture
waxen pasture
empty rivet
waxen pasture
empty rivet
waxen pasture
#

try that

gloomy iron
empty rivet
waxen pasture
waxen pasture
gloomy iron
#

I am trying up upscale a 5761024 picture to 11522048

empty rivet
gloomy iron
#

576x1024 to 1152x2048

waxen pasture
halcyon granite
#

is there a tiled upscaler available for txt2img (besides highres fix) for auto1111?

abstract quarry
# gloomy iron thank you for answering my question. So that's how it is. For the 76 iterations...

it also depends on the sampler..The stochastic samplers like Euler A might get even worse with high number of steps, while deterministic samplers like Heun are just converging and don't improving anymore with many steps. In general I would rather use less steps (like 20-40 steps is often enough), but that might depend on your images (photographies might need a bit more steps than animes or comics)

warm junco
gloomy iron
gloomy iron