#💬｜general-chat

1 messages · Page 48 of 1

whole nacelle
#

I'm building the interface for a new Ai-generation software and building some inferences in order to see how the pre-alpha is working. ^_^

bleak matrix
whole nacelle
#

As soon as we'll have the first beta available (it will be a free software for casual users) I'll post the link somewhere here ^_^

bleak matrix
#

As for what people want, it just depends. I haven’t used Invoke's canvas in awhile

outer ravine
#

hey guys why is it i cant change the checkpoints from the drop menu?

bleak matrix
buoyant jacinth
bleak matrix
#

Also, I think if you are looking to do things on a canvas, then really, for artists--it is all about layers, layer properties, brush systems, being able to flip, select, etc.--since there are already plugins for other programs, I would maybe focus on additional tools to make it easier (like cloning/healing, etc), making brushes from what you create, maybe making style sets (kind of like actions, in a way, etc), perhaps @whole nacelle

whole nacelle
bleak matrix
#

Particle brushes--clipping--things people are already used to without having to have a Creative Cloud subscription.

#

The way Corel does some of its brush systems, along with choosing specific color palettes to go along with your image, etc, might be useful.

whole nacelle
# bleak matrix Also, I think if you are looking to do things on a canvas, then really, for arti...

imo, that's a more "Photoshop" approach, for AI-generations I think we can stop thinking about illustration as we did before and we have new tools to build. I like the idea to flip, crop, resize and rotate pictures (of course) and the idea of having custom brushes looks cool, but in my opinion the true heart of these softwares are iterations and how to "Move" a whole picture towards a specific result. I want to offer a lot more control on the whole process.

buoyant jacinth
bleak matrix
whole nacelle
buoyant jacinth
whole nacelle
bleak matrix
#

Most people probably want customization, and for most artists I know, they love brushes, and making brushes. So that has always been pretty popular. And it's often pretty fun, so people might enjoy that

buoyant jacinth
whole nacelle
bleak matrix
#

CSP probably has one of the best brush systems for custom brushes

minor vapor
#

can anyone help me ?

whole nacelle
buoyant jacinth
bleak matrix
#

And being able to do something with what you created--and really transform a piece further. If you could generate basic shapes--i.e, say, make a brush that gives me various maple leaves, and then you could paint maple leaves--or, you can make a big maple leaf on a layer and instantly make that a brush--that's just cool because it would take a lot of the difficulty out of brush making

bleak matrix
buoyant jacinth
# whole nacelle Do you use that?

I haven't yet, but the video has an example. My needs are fairly simple at the moment. I can say comfy works great with a 750TI 2GB vram.

broken gull
bleak matrix
vast ingot
#

(sorry no chill for those)

bleak matrix
#

(Ikr, goodbye)

vast ingot
#

Wtf is wrong with people sometimes

broken gull
#

euh?

vast ingot
#

Nothing

broken gull
#

ah

bleak matrix
#

We got rid of it

vast ingot
#

An offensive person that isnt around anymore

bleak matrix
#

Go t e a m

buoyant jacinth
bleak matrix
#

indeed

broken gull
#

it does exactly that with the images it saves

#

but it has a different format iirc then your average image

buoyant jacinth
#

Yeah, that's one annoyance. I do wish it saved a plain prompt version as well.

broken gull
#

so we can just share it with a friend and he can drop it in a UI he prefers and have everything ready to work further on it

#

that be next stage of AI imaginary

bleak matrix
#

I haven't tried ComfyUI yet but it does look interesting

whole nacelle
broken gull
#

what did you just say? ConfusedCat

buoyant jacinth
#

A reply to something from august. That takes dedication. Or a search.

onyx python
#

why do i only get a .png's up in "start here" when i press a image. it have been like that from the start a few months now. like this https://ibb.co/PzxhHzF

buoyant jacinth
#

those aren't buttons--they're only images. By default, discord shows a full size image if you click one.

#

the channel names in each section are links tho

onyx python
#

no links for me

#

properly just me being a retard, but it still nags me😅

buoyant jacinth
onyx python
#

no

buoyant jacinth
onyx python
buoyant jacinth
#

woah. Yah. That's bonkers.

onyx python
broken gull
buoyant jacinth
#

if the bot were blocked, they should see the blocked messages lines, I think?

onyx python
#

Thanks i will try to look up the bot thing maybe its that

buoyant jacinth
#

is it just that channel or all the channels in the start here category?

onyx python
#

no its all the channels in the start here

buoyant jacinth
#

mobile app, browser app, or desktop app?

onyx python
#

all

onyx python
buoyant jacinth
plucky moss
#

any body here?

#

can help me with generation cob web

buoyant jacinth
peak spoke
#

Anyone have any tips at changing perspective angles not sure how best to search for ?

lament solstice
#

anyone know how to use safetensors instead of ckpt on 1111, i apologize for my ignorance, know nothing about stable diffusion

fervent thunder
#

Asked in #🤝｜tech-support but got a weird response so will ask here:

Say I redownload v1-5-pruned-emaonly.safetensors, do I download v1-inference.yaml and rename it to v1-5-pruned-emaonly.yaml?

lament solstice
warm junco
regal snow
#

Does anyone know if there's some special way to get more controlnet metadata to be saved? Such as, annotator resolution, midas depth, Low/High %, etc?

#

Eh, actually I suppose I'll just enable the Auto-Save detection map feature. Seems like I could reproduce anything using this plus the metadata that it does append to the generated images

novel robin
hollow swallow
peak orchid
#

Have you guys and gals seen this extremely disturbing interview from OpenAI on open source yet?

#

When asked why OpenAI changed its approach to sharing its research, Sutskever replied simply, “We were wrong. Flat out, we were wrong. If you believe, as we do, that at some point, AI — AGI — is going to be extremely, unbelievably potent, then it just does not make sense to open-source. It is a bad idea... I fully expect that in a few years it’s going to be completely obvious to everyone that open-sourcing AI is just not wise.”

Source: https://www.theverge.com/2023/3/15/23640180/openai-gpt-4-launch-closed-research-ilya-sutskever-interview

forest trout
#

So I tinkered around with 2.1 a bit but haven't fully migrated over because I'm just not getting good results. Any suggestions?

broken gull
lament solstice
#

Im curious abt smt, is there a way to register a name with a model so like, lets say you create your own custom creature name it splunkymoboppin, and you want to render that creature in a landscape, could you train the ai to recognize that name so in the prompt you could simply reference "splunkymoboppin" and it'd understand.

fervent thunder
hearty dune
#

Anybody encountered an error where the t2iadapters just don't work?
i downloaded them and re downloaded and nothing. the regular models works fine, but these don't

#

Controlnet's models that is

fervent thunder
#

how do I generate an image

surreal perchBOT
#

@fervent thunder

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

raw narwhal
#

does anyone have a model recommendation for tropical Beach Landscapes? Rendering 960 x 540 and upscling to 1080p, everything comes out nice except the trees and fine details tbh, looks a bit stuck on.

lavish current
#

Trying to upload and I'm getting "file cannot be empty" errors, what does this mean

#

@raw narwhal Deliberate Diffusion is my favorite model for everything, otherwise Dreamshaper is good

#

Analog and Redshift Renderer are also nice for photos, though I'm not sure how well Analog does at higher resolutions

#

Also it depends on the prompts I guess

raw narwhal
oak holly
#

wait is it not better to have over 120 steps than 20?

raw narwhal
#

The anime models i have used seem to die above 20steps

#

but for photo realism higher is usually better

oak holly
#

where do you get your models from?

hearty dune
#

civitai

lavish current
#

Mostly because I got tired of the censorship on Huggingface cloud models :x

#

Man I need to free up storage space

#

SD is too powerful 😭

#

The Chromebook went from: book/magazine reader, to portable Saturn, to image generating machine, lol

vestal dew
#

how do you run SD on a chromebook?

vast ingot
#

Notebooks i think

#

No graphics card on there

#

Well i guess it depends, you can install chromeOS on mostly anything i believe

vestal dew
#

chrome OS won't run SD on a peanut machine

lavish current
#

I use cloud models

vast ingot
#

Its main use is to run things in the cloud usually

#

Minimal local hardware and price

vestal dew
#

SD doesn't know what mohawk hair is

calm acorn
#

Has anyone tried training SD to know about several characters? Basically, I want SD to know about several humans and do img2img. I've tried a few options, but found that the model is not very good at the end.

vestal dew
#

youve tried loras, one with each person>?

vast ingot
#

It requires to balance each concepts, see witch one trains the fastest, and the slowest, and adapt

#

In particular, doing training in multiple periods.

#

First training a little the concepts that will train the slowest

#

Then expend the dataset to the others

calm acorn
#

So that looks like first training on one character, then adding another character to that same dataset and training on the checkpoint

vestal dew
#

you're trying to make it a mix of characters?

calm acorn
#

not a mix, trying to actually teach it about several concepts individually

vestal dew
#

so do each one separately, not all together

calm acorn
#

I've tried that, didn't work great (I used joe pennas repo)

vestal dew
#

what is your end goal? What do you want to generate image wise?

calm acorn
#

basically images of my family in different styles

#

controlnet/non-finetuned SD misses out on some specific features

vestal dew
#

so do a lora with each family member. If it isn't coming out right, adjust your traimning

vast ingot
# calm acorn So that looks like first training on one character, then adding another characte...

More or less. It's more like, if you wanted to train 3 characters :
you do a first training with lots of checkpoints and all the dataset and you try to find at how many repeats/epoch each was train the best
Let's give an example. Concept A takes 10 epochs, B takes 5 and C takes 5
Then you start a new training, for 5 epochs only on A. You continue then this training with the whole dataset for 5 more epochs

calm acorn
#

Got it, what repo have you been using @vast ingot ?

vast ingot
#

And I would recommend dreambooth, not Lora, when hitting so many model weights

calm acorn
#

I've been having more success with dreambooth as well, so I've been sticking to that

vast ingot
calm acorn
#

yeah would be great!

#

Thanks!

vast ingot
#

There is 2 guides in one. One is focussed on the training params and workflow, the other one the training methods out there

#

It's still quite new, I'm open to feedbacks too

fervent thunder
#

I guess I should congrats everyone here, AI won.
Generated images CANNOT be copyrighted, officially

vast ingot
#

Is it a win though ? I thought people wanted to be able to copyright their pictures ?

fervent thunder
#

but seems like I'm wrong and there's already different kind of claims...eh...

#

let's wait on that more

vast ingot
#

It's gonna be a battle of the ages imo.

#

Not sure AI is sentient enough to win yet though

#

Just a loss for now

fervent thunder
#

honestly , I'm okay if quote "generated" won't be copyright-able, but if we'll modify it with out painting, inpainting and whatever other tools we have - that's whole another story

vast ingot
fervent thunder
#

eh...yea, tha'td be dumb too, idk

vestal dew
tall heart
#

What do I do when Hypernetworks refuse to be created?

#

on A1111

#

much help

vestal dew
#

I find it funny that people are claiming AI has replaced human artists - it can't even draw hands or feet

warm tiger
#

anyone else been getting literally dozens of random messages and various group invites from people within this chat?

#

yes. there's an extention you can DL right on the webui extention page to do this

#

now how to actually set it up i wouldn't know

karmic brook
warm tiger
sullen walrus
#

Anyone got gpt 4?

gilded sable
#

Copyright Office Launches New Artificial Intelligence Initiative
Issue No. 1004 - March 16, 2023

uh oh

Today, the U.S. Copyright Office launches a new initiative to examine the copyright law and policy issues raised by artificial intelligence (AI), including the scope of copyright in works generated using AI tools and the use of copyrighted materials in AI training.

This initiative is in direct response to the recent striking advances in generative AI technologies and their rapidly growing use by individuals and businesses. The Copyright Office has received requests from Congress and members of the public, including creators and AI users, to examine the issues raised for copyright, and it is already receiving applications for registration of works including AI-generated content.

sullen walrus
#

Ayo

#

Could you Provider me with it for a tommorow?

#

I have an exam

#

And im desperate

#

Vc?

#

Nah how would that work in an exam

#

I cant talk during an exam

#

Iw ould but i have a sparkasse card and it only takes Credit card from visa and mastercards

#

So i need someone who can Provider me for tmrw

#

You can change the password, i just need it for like 3 hours

#

Tried on previous exams, got3 aint good

#

Nah man

#

I got money though

#

Paypal

#

What would that bidnas look like?

#

Now we in a pickle

#

How do i know u dont run off with the cash

#

Yes but if you run off i loose 10 bucks. If you dont trust me you still wont loose anything

#

Look ill cashapp ypu after the exam is over how does that sound

#

Ok look 5 bucks now

#

5 after exam

#

So you dont have paypal?

#

Aight

lofty zephyr
vestal dew
#

What does it mean?

normal coral
#

you guys reckon prompt bleed through will ever be solved

#

or is there already a way to mitigate it a bit

willow coyote
#

So, after I've trained my own replicate/dreambooth and the pictures are looking kinda nice. I ran the training proceedure three times, once for each cpkt I found (from https://replicate.com/cjwbw/hasdx): https://gist.github.com/basicfeatures/de7e6469d0d928d07118697ca8de4c6e

Now what's the next step to getting the trained subject to do things like go up on a stage in front of people and sing etc.? Is this part of some other solution I need to combine with SD or Dreambooth? It doesn't do anything I tell it to in the prompt other than the standard stuff I copied/pasted from hasdx.

normal coral
#

im trying to force certain body proportions to generate but the moment you input long legs short torso or whatever it just doesnt do anything because AI cant tell whats supposed to be short or long

sullen walrus
#

Man cmon

#

I can do it on Streamer youll se it Asus for multiple

#

Asks

#

Come in any vc ill Show u

#

@misty prawn come pow vc

warm barn
#

hi, which channel is for writing prompt for stable diffusion? I am new

winter pike
sullen walrus
#

Look ill still need help removing your Account from my device

warm barn
winter pike
#

very fast generations + youre helping future models

warm barn
viral topaz
#

hello, can we use SD v1.5 for commercial purpose ?

obsidian geode
#

of course, so many people are already selling stuff. Remember its open source tought

viral topaz
#

because in their TOS they said for research use only

vestal dew
#

then some other inmate in prison makes an embedding out of you

uncut halo
#

Hey guys, first time here but I've been using SD for a little while and am trying to find some interesting ways to use the tools.

I've got a question: Is there any way to use SD or the tools around it to essentially generate a prompt or set of associated words and categories from an image?

What I'm trying to do is see if there is a tool available that I can run locally that can categorize images and make them searchable by context, that doesn't rely on uploading everything I have to someone else's server. I saw Kive.ai which is extremely similar to what I'm hunting for, but obviously is only a cloud service. Has anyone seen anything similar that can be run or built locally?

sullen walrus
#

Man r u even willing to work with me her?

nova lily
#

Anyone else notice that if you want to generate any sort mask you only get typical covid masks.

orchid lance
#

not really... but you can specify it by brand and type of mask

nova lily
orchid lance
#

yeah, just use that and then put the medical mask stuff in the negative prompt

nova lily
#

👍

nova lily
sullen walrus
#

Anyone here got got4?

orchid lance
#

you mean GPT-4? sure.

sullen walrus
#

Not the api

orchid lance
#

...yeah? on the OpenAI's site?

sullen walrus
#

Perfect

#

Could you hook me up withit for an exam tmrw?

orchid lance
#

lol no

sullen walrus
#

Why not

orchid lance
#

just do your work. you don't need AI for that.

#

not gonna help you cheat

sullen walrus
#

Well did what i could but theres no time left

#

Cmon man dont you wanna know how Well itll do

#

Im really desperate at thus point

orchid lance
#

when I did the exam... I had plenty of time to study, plenty of time to write and did all that without a computer.

#

if you say you're desperate and you have little time left... don't you know what that sounds like? you were putting it off.

#

the whole put of the exam was to put in your efforts.

sullen walrus
#

Yeah but for me theres no time lebt, it was oen exam after another i had to let something out

orchid lance
#

if you don't think you're getting enough time, talk to the teacher.

#

but I'm not convinced

sullen walrus
#

The teacher wont change exams because of one Student of 30

#

Out of

orchid lance
#

yeah... and you want to cheat with the AI.

#

come on

sullen walrus
#

Exactly

#

I mean there were to many exams, thats the last option

#

If they were spaced out over the some weeks i would have Managed, but its to harsh here in germany with exams

#

I just dont want my Grades to drop because of one exam that i couldnt have possibly learned for

vestal dew
vestal dew
vestal dew
nova lily
vestal dew
#

with a weight?

orchid lance
nova lily
#

If weight means :X.X no, I was told they have same affect

#

so I usually just use ()

brisk bloom
orchid lance
nova lily
#

oh tanks alot for that page

#

never would have known, will learn alot

vestal dew
nova lily
#

Hi, how to you outpaint? Where is it?

orchid lance
vestal dew
vestal dew
orchid lance
#

oh for crying out loud.

vestal dew
#

(()) means weight 2, but so does (thing:2)

orchid lance
#

why in the world would you jump in if you don't know what you're talking about?

vestal dew
#

I saw people talking about GPT4

orchid lance
#

and?

vestal dew
#

So I mentioned it was for pay...you have to pay to use it

orchid lance
#

false

vestal dew
#

is that accurate?

#

So I can access GPT4 for free?

orchid lance
#

there's a waitlist

vestal dew
#

to access it for free?

orchid lance
#

yes, and there other ways to get it too.

#

Like Bing for one.

vestal dew
#

So if there's a waitlist to use it free, then I can't use it free right now. My statement is accurate.

#

Bing lets me ask questions, get responses? Or it's dumbed down version?

orchid lance
#

🤦‍♂️

#

you're KG_38, aren't you?

vestal dew
#

Ok, where can I go right now to use GPT4 free? I want to input SAT passages, get it to provide answers for SAT questions.

#

ChatGPT only does Reading well, not Writing or Math

orchid lance
#

just what I thought.

vestal dew
#

So let's be clear. RIGHT THE F NOW, this second, I cannot go use GPT4 like I can ChatGPT, right?

#

I thought so

orchid lance
#

not helping you cheat

vestal dew
#

cheat?

#

I am a tutor

#

I use ChatGPT to help write explanations for questions. I create explanations for each test

#

share them with students who are taking the test

orchid lance
#

yeah, I don't think you could make it more obvious right now.

vestal dew
#

It right now only does Reading well

#

see my name? Frank TUTOR

#

It also writes maps for Reading sections pretty well, if I guide it

#

I have tutored SAT and ACT and GMAT and LSAT and other such tests for a little over 40 years

fringe pasture
#

Un saludo desde Colombia

orchid lance
astral goblet
#

Oh neat. Discord has more ads in it now

vestal dew
#

Just as you must prompt SD, know what you're doing to get good images, to use ChatGPT for SAT you have to know what you're doing to use it. Someone like you Nacon wouldn't be able to use it for SAT.

orchid lance
astral goblet
vestal dew
#

yes, but dumbed down

#

You can't input a whole passage from SAT, then have it answer the questions. ChatGPT can do that

orchid lance
vestal dew
#

It does well on Reading, OK on Math, makes a tinker's bollux of Writing

#

I read that GPT4 is much better with Math and English, it scored very high on SAT. So I wanted to see it myself, and if possible use it to help write explanations

astral goblet
#

same guy as what? same ol franktutor really. unable to figure out how to use the tools for his needs so he blanket evaluates all technology as being a giant fail case

vestal dew
orchid lance
#

yea yea.. very convincing.

#

🙄

vestal dew
#

SD still can't do hands and feet well. It absolutely fails in that regard, most of the time.

astral goblet
#

Yeah i know i know. You love to hinge on your fail cases

#

same ol same ol

vestal dew
#

noipe

#

I love to use tools to do what is needed

astral goblet
#

still can't figure out ctrlnet hey? did you stop trying to use unprompted at least?

vestal dew
#

Right now, ChatGPT doesn't work for this use on anything but the Reading sections, which is why I asked about GPT4.

#

I cant run controlnet

#

I have a weak computer

astral goblet
#

"doesn't run" lol i know i know. You find a fail case and fixate. i know same ol same ol

orchid lance
#

I think I'll just block him

vestal dew
#

huh?

#

It won't run on my GTX 970.

astral goblet
#

so SD is very capable of doing what you need, but since you're unable to with the resources at your means, you've decided it unilaterally can't do any hands or feet ever. kay

orchid lance
#

there we go.... half of my screen is filled with "blocked messages"

#

so much better

vestal dew
#

good. It might be all of your screen

#

No flow, but I can't do it right now.

#

And, those claiming it does hands and feet well have yet to show me one example of it.

astral goblet
vestal dew
#

Then generate 10 images....show me 10 images with good hands and feet. Go

#

Even with my 970 crap, I can fix the hands and feet

astral goblet
vestal dew
#

but it still generates over half, maybe over 80% with bad hands and/or feet

astral goblet
#

"GO MAKE ME 10 IMAGES NOW" heh yeh bruh, no. you've wasted enough of my time and now you're getting all demandy as usual haha

vestal dew
#

Well, when I can run controlnet, I will certainly try it

#

if it solves the problem, I will be very pleased

vestal dew
#

Flow: "Claim!"

Me "Prove it"

Flow: "he's making demands"

normal jungle
#

I have a question for anyone able to answer it

#

I recently got SD and I keep getting a “out of memory” error

#

How do I fix that?

vestal dew
#

what is your GPU?

normal jungle
#

3050

vestal dew
#

how much VRAM?

normal jungle
#

I believe 16 gigs

vestal dew
#

wow

normal jungle
#

Too little?

vestal dew
#

I think it's 8

#

no, that should be plenty

#

what are you doing when it gives the error?

#

generating an image, training?

normal jungle
#

One sec let my pull it up

#

Yeah just trying to generate and image

vestal dew
#

what size image?

#

is hires fix checked?

astral goblet
#

3050 is only 8gb. you'll want to run a version that has medium vram requirements, like a1's and the --medvram switch, also use xformers

#

1mp images might be your top out

vestal dew
#

you have to do medvram witgh 8 GB?

astral goblet
#

it can help. slows it down but gives you less oom errors

vestal dew
#

How much vram is needed before you don't need medvram?

astral goblet
#

i got 16gb and get them if i push it hard

vestal dew
#

wow

astral goblet
#

i bet i could get 24gb erroring out

vestal dew
#

sure, for higher resolution

astral goblet
#

16gb error if i'm generating a desktop wallpaper l;ol

vestal dew
#

I have 4GB,. use medvram, make images 512x768

vestal dew
normal jungle
normal jungle
vestal dew
#

Wow, I would think 8 is enough for that size

normal jungle
normal jungle
vestal dew
#

in your stablediffusion dir there is a file called web-user.bat

neon oriole
#

you should try ComfyUI

vestal dew
#

set COMMANDLINE_ARGS= --medvram

astral goblet
vestal dew
#

save it

#

re-run it to launch SD

#

I have to go do pullups...in the rain

normal jungle
#

Uhhhhhh

vestal dew
#

I run medvram with 4GB for 512x768 without errors. I thought 8GB would be enough to do that size without medvram

astral goblet
#

i believe in you

vestal dew
#

guess not

normal jungle
vestal dew
#

wherever you installed

normal jungle
#

Like in my files?

vestal dew
#

or search on your system for web-user.bat

#

yes, in your files

normal jungle
#

I searched it and it’s not popping up

vestal dew
#

Mine is here D:\Automatic1111\stable-diffusion-webui

#

in there is webui-user.bat

#

I forgot the ui part

#

webui-user.bat

normal jungle
#

HOW THE FUCK DO YOU KNOW

#

Uhhh yeah I forgot the ui

vestal dew
#

how the F do I know what?

normal jungle
#

That I forgot to type the ui

#

Ohhh sorry

#

I misread that

#

I’m like half asleep

vestal dew
#

I said I forgot it

#

not that you forgot it.

normal jungle
#

Yeah I know I misread it

vestal dew
#

ah

#

OK i have to go

#

the rain has subsided

normal jungle
#

Ok good bye

vestal dew
#

it's my window

upper wadi
#

I have a question about the krita add on for stable diffusion

normal jungle
#

Well damn

upper wadi
twilit willow
#

Because they're pruned, insignificant weights are removed providing little difference in actual use

thorn hare
#

is there a autofill prompt completer plugin for SD ?

vestal dew
#

??

dreamy igloo
warm junco
winged heron
#

what happen to emad?

sullen walrus
#

Anyone here got chagpt4?

broken gull
winged heron
#

never see @wise stratus come online

orchid lance
#

guess he's very busy?

hearty dune
#

Is there a way to map part of a prompt to a XYZ location in the image? for example, something like a man standing by the window looking at a bird (50x, 250y, 1z) where the bird will always be generated at 50,250,1 location of the image, while the man and window's locations still be randomized?

sullen walrus
#

Man the exam is in an hour, u have Solid proof, cmon help a man out

sullen walrus
#

Doomer please talk to me

#

The exam is in an hour

#

I really need that

#

Lets go vc i can Streamer and you will see it needs 2 Verifikation

#

I wasnt doing nothing shady

#

Where

#

What do you mean?

#

Man i dont even know how to use my computer

#

Who do u think i am

#

We can do that but i would need to have 2 hours of ur undivided attention

#

And you would have to send fast replies

#

Ist 9:00 here

#

If u notice anything shady you can always log me out by changung pw

#

Exam starts in an hour

#

Man we Talkessel do i really come off as the guy that would do it

#

I mean i needed help with logging into chatgpt, do you really think id je capable of it

#

Ok man its ur way im just saying it would be way way way easier

#

I think u blocken me or smth

fervent jacinth
#

Hi

vast ingot
#

welcome

fervent jacinth
#

Thanks!

vast ingot
#

new to SD?

fervent jacinth
#

Yea

vast ingot
#

great magic world to explore

fervent jacinth
#

Im looking for an AI that can turn my drawings into something more realistic/great looking digital art

vast ingot
#

that certainly can be done. can't really show you in this channel though, we don't have images in there

fervent jacinth
#

I only used Midjourney before that worked but it often adds weird things that I dont want

#

Okay 🙂

plush lark
#

hi there!, is there or will there be a big update on stable diffusion soon? other ai platforms seems to have a big update recently.

ashen lion
#

Hey there , is there any model or SD version that get any close to midjourney quality? I have not being around for a few weeks and would like to know how far SD have evolved

vast ingot
#

there starts to be quite a lot of good mix on civitAI for example. What you ask is quite subjective, but using a nice model that fits my personal likings, and a style randomizer on top, I get quite some qualitative results

cedar garden
#

Hey

vast ingot
#

Hey and welcome around 🙂

cedar garden
#

I’ve used this bot for a while now and only just found out u lot have a server 😂😂

vast ingot
#

yep, but no bot that makes pictures here though

#

you use what bot exactly ? MJ?

cedar garden
#

I believe it’s dream diffusion

vast ingot
#

well, I didn't know about those one

cedar garden
#

I have also used MJ aswell

vast ingot
#

here, it's the Stable Foundation discord, with mainly Stable Diffusion and others from the group.
We do have lots of ways to make good pics though :p

#

it's done through websites like dreamstudio, local install on your computer, or even running it in the cloud in a notebook for example

warm junco
#

DreamDiffusion is a discord bot using the resources of StableHorde

plush lark
vast ingot
fervent jacinth
#

How do I tell it which one is which

cedar garden
#

I have a lot of anime images I have made using dream diffusion 😂

vast ingot
#

you will be able to specify the artstyle by a prompt usually, describing it, and also finding a nice model that is trained on things you find aesthetic. check civitAI for lots of those

vast ingot
#

you may love the place

#

it's not their rush hour right now though

cedar garden
#

Holy sht some of there art is AMAZING 🤩

vast ingot
#

they are quite good, I'll give you that

#

and there are really lots of good anime models today

ashen lion
#

What I like about SD is flexibility but at same time it makes it much more time consuming

vast ingot
#

that is true

ashen lion
#

But I still needs to learn more about the injections of styles

#

They removed the random artist thing from latest automatic11

vast ingot
#

it's the brick that makes the art, you have the main tool itself, with all the sliders to deal with.
It can do a lot more because of it, but it also feels more like a very specialized tool.
Like you would need a training to operate a big demolition truck for example, SD will require some real taming

ashen lion
#

Disco diffusion was a monster to tame lol

vast ingot
ashen lion
vast ingot
#

(not mine to share publicly, they were given to me by a friend)

cedar garden
#

I can’t find the anime channel 😂

warm junco
vast ingot
#

Nah sorry

#

I linked to the archived one

#

We need to rename it, i do that all the time

cedar garden
#

Why is it archived 😂

vast ingot
#

We reformated the server, and anime is now a forum

#

So the old channel got archived

cedar garden
#

Ah

vast ingot
#

You can visit #👥｜roles and assign yourself some to access image archive too if you want, but my last link works

#

Damn I'm not good on phone lol

cedar garden
#

😂😂

vivid portal
#

Auto1111 does take folders for model files, right?

vast ingot
#

It will sort them following those in the drop-down menu of the UI

vast ingot
cedar garden
#

Any tips for newbies 😂

#

To create amazing masterpieces

vast ingot
#

masterpieces models for a start

#

highres fix secondly

cedar garden
vivid portal
#

@vast ingot do you know if the hash number that is added after the first load of a model in auto, does that number have to be named when using different ckpts in x/y?

#

Or is the name of the ckpt enough?

vast ingot
#

just the name of the file (and its path if it is in a subfolder), without the .ckpt

vast ingot
#

you start by finding a GOOOOD model for anime, that fits what you like

#

civitAI has tons of those

#

and the anime community has the best tips

vivid portal
vast ingot
#

they will also tell you about lots of little add ons, like embeddings, to get even better results

cedar garden
#

Well I’m on phone sooo 😂

vast ingot
#

you can still browse and bookmark ^^ https://civitai.com/
using phone, you could potentially even run SD in the cloud, but that would be a real hassle. You'd better wait to get back on PC for that

vivid portal
#

Would be cool if x/y worked with batches.

#

To make it quicker

#

But I guess that would be a bit hard to fix since it’s going through lists of things

vast ingot
#

it does, no ? I use it on batch size 4, and put only 1 axis, and it compares 4 seeds on that axis

#

it just applies the batch size to every picture you had on batch size 1, so usually that means you remove the "seed" axis

warm junco
fervent jacinth
#

This is working great

#

Its only making my images a bit to smooth/simple but thats probably bc of the settings

cedar garden
warm junco
#

You can access your local SD webui by phone yes

cedar garden
#

😂 how ? I don’t normally do stuff like this

#

Iz literally noob at this

novel robin
#

mj5 really that special? Saw photorealistic models on civitai.

warm junco
cedar garden
#

??

#

I’m on phone so how do I do that

warm junco
#

Is your pc running a local webui ?

cedar garden
#

I have no idea 😂

novel robin
#

u need a pc to run automatic1111

warm junco
#

There is only an app for iphones that works on the phone localy.

novel robin
#

he could try stable horde i guess

cedar garden
#

Will try stable horde now

novel robin
ashen lion
cedar garden
#

Idk

shy cosmos
static badge
#

how do i use controlnet on SD automatic1111?

uncut plaza
#

<@&1025179534330433656> GOOD MORNING DAVID.G9 @desert plover

vast ingot
vast ingot
warm tiger
#

does a LyCORIS file get tossed into the LoRA folder?

autumn oak
#

does anyone know how to claim a instagram page?
theres a page i want for my business that has 0 followers & followed - any help would be appreciated 👍

vast ingot
plush falcon
#

Hi, I'm New, where is the dreambot server?

#

I don't think it's showing up with me

#

Nevermind

#

Bye

vast ingot
#

that was fast

#

there is no bot currently

surreal perchBOT
#

@plush falcon

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

green brook
#

Hello , new to SD! Quick question about using Controlnet to pose. Under the "Model" drop down in SD I have nothing , how do I add options there

plush falcon
#

Oh ok

vast ingot
#

there are 8 models total to download

green brook
novel robin
#

whats sd xl?

vast ingot
# novel robin whats sd xl?

it's the code name of something that has been teased form some days now. we don't know a lot, just some screenshots of results here and there. seems like a new model

broken gull
novel robin
quick vault
#

someone told me to check out "film on HF" to make a vid from a few frames(like interpolation but I just have 4 still images), but I can't rly find what film on HF means when googling any help?

grand fox
#

Does stable diffusion work on a egpu?

cedar garden
#

Haha I finaly got a good Chibi prompt LETS GO

autumn geode
#

score

stable bramble
#

Hi all!

I’m a PhD student at Stanford working on foundation models, and thought one of my recent projects would be of interest to this community. (Mods—I’m new here, so if there’s a better channel for this, please advise!)

We just released on online game (https://artwhisperer.io) where you collaborate with a text-to-image AI model to create target images and compete with players around the world. Takes only 3 min to play (to get a high score you may need to play longer) and helps us study Human-AI collaboration. New image challenges are released daily. Try out the game and share with your friends!

cedar garden
oak holly
#

where can I find imgs with prompts

#

so I get some ideas?

cedar garden
#

I’m not really sure what your style is so I can only offer you where to find anime styled art 😂 #1072013871730131004

oak holly
#

so If I download a pic from there and put on png checker I will get the prompts?

vast ingot
cedar garden
#

Well I just wanted to make you aware just in case

#

I could also send you some of my prompts if u wanted @oak holly

vast ingot
#

just so you know how to report easily though : right click on a message=> application=> report to staff,
or react to any message with ⚠️ and it goes into our "moderation queue"

cedar garden
#

Ah ok thanks 👍🏻

vast ingot
#

the ⚠️ will disapear but we'll have the report

cedar garden
#

Kk

oak holly
#

@cedar garden can you dm me?

cedar garden
#

Sure thing 2 mins

vast ingot
#

Use open pose mode, not canny. It will copy the pose buy that's all, so new clothes and face.

Or yes, inpaint a new face, that Works too

regal snow
#

well, you could just copy/paste her face in then img2img it

#

in an external editor that is

#

Well if you have an openpose controlnet, it might guide it into the correct orientation during img2img

#

uh huh well that would work if you trained her into your dataset

#

Oh well, you didn't mention that?
inpaint that area and call your embedding and it should work. Guided inpainting always has better results so you could also try pasting her face in

slim hill
#

When can we use Llama with a ui?

regal snow
#

maybe try an X/Y/Z plot to see what settings get closer to expected result, then dial it in

#

If you are using automatic1111 click "Scripts" at the bottom

#

Try something like
CFG Scale 5-7 would use 5 / 6 / 7
Denoising 0.4-0.8 [3] would use 0.4 / 0.6 / 0.8
Steps 20-30 [3] would use 20 / 25 / 30

#

for sake of example

vast ingot
regal snow
green brook
#

Alright friends - could use some more help! Trying to get dreambooth running but I see it needs some dependencies - I'm a complete noob to the CMD line side of things, can anyone guide me through installing dependencies (xformers,diffusers it seems)

bleak parrot
#

bro can someone help me like when i want to download sd localy when i open that user cmd thingy it first downloads a 2 gb file then after 2-3 mins it give me alot of red texts and codes(BTW i have a amd radeon and a 8 gb ram) pls help

regal snow
#

From what I've read, the quality is more important than the quantity

#

Don't use any blurry images at all

warm junco
limpid berry
#

hello there

#

were i can generate images? sorry im new on this

fervent thunder
limpid berry
#

ok

fervent thunder
#

And people are gonna respond over here? You just posted here cuz you think more peeps are gonna see it

#

Which is called spam

#

@misty prawn

split tangle
#

why the poop?

fervent thunder
split tangle
#

a valid questions....

fervent thunder
#

Read the chat

split tangle
#

okay thanks

fervent thunder
#

This isn't tech support 2, gawd

#

Mods here used to care, smh, back when this chat was active

#

Anyway, hope everyone's doing ok today, I mean, if there's anyone out dere

#

I'm excited for what Emad said a couple days ago

#

He really stood up to GPT-4, I hope Stability makes a kickass LLM with their partners

#

That sounds more like privacy violations? Please leave.

#

Please do NOT come here and turn what semblance of community we have left into a parody of 4chan

#

And if you have thoughts like that, keep them to yourself

#

Does anyone know how to ping a mod?

#

Dude...

#

I didn't say anything about politics, what's wrong with you?

#

@thorny depot @vast ingot

#

Could you please get rid of this guy for dragging politics into general for no reason?

#

I don't know what his deal his

#

Uh huh...

#

@wintry stream

#

This has nothing to do with politics, you're being uncivil for no reason

#

You're obviously right-wing, but this isn't the place for your politics

#

How did I enforce politics? Order isn't political

wintry stream
#

Thanks for flagging ❤️

fervent thunder
#

What? I'm not arguing politics man

wild steppe
#

He left after getting timed out 🧹

fervent thunder
#

Haha, ok, thanks, just when I thought I was being a tightass the guy proves me completely wrong

wintry stream
#

No tolerance for that kinda stuff here - appreciate you calling it to attention

wild steppe
vast ingot
#

sorry was a little slow to intervene

thorny depot
#

Aye. I was just looking over everything now.

thorny depot
fervent thunder
#

Who else was here during the beta? I remember back then you needed an army to manage the chat alone.

#

Now most people gen on Dreamstudio or their own riggs, or one of the half dozen online services out there

thorny depot
#

Ahhh, the olden days....

wild steppe
wintry stream
#

It's mostly just that a lot more information is out there and there are lots of places people can go to learn 🙂

Personally I think it's a good thing that more information is out there and people can find it in the ways they want to

fervent thunder
#

Hehe, it's amazing that it's only been 6 months or whatever

fervent thunder
wintry stream
#

It's not that we don't care anymore, quite the opposite. It's just that the community here needs a lot less hands-on and we'd prefer everyone be able to build what they want without us trying to be too controlling of it

fervent thunder
#

I apologize for any perceived sleights

wintry stream
fervent thunder
#

They SAY they want a tech question answered, but half the time they've got a fetish which they don't care to act out in more appropriate spaces

#

And sometimes, it's someone like that

#

And if you admit to being a woman...tch tch

fervent thunder
#

anyone using the unipc sampler? just noticed it not sure its benefits

fervent thunder
#

This is the general

#

yeah and im making general convo

#

so chill the fuck out

#

Asking what the benefits of a sampler are is a technical question

#

People are good about responding there

vast ingot
#

sorry a little my fault on this one. I warned them this morning for similar behaviour, I could have prevented that.

fervent thunder
#

yeah because this place is dead now

#

you used to have all kinds of chats here

fervent thunder
#

Hehe

#

or if you stop killing convos by redirecting everyone to technical chat

vast ingot
#

UniPC is really nice to be honnest

fervent thunder
#

This is the Discord for a $1 billion company, we can give a shit, we can win

vast ingot
#

on low steps and low CFG it does wonders

#

like try to prompt "a cat" with 1 step, and CFG 5

fervent thunder
#

so its benefit is it works good at low steps

fervent thunder
#

At least a lil

vast ingot
vast ingot
# fervent thunder At least a lil

well I can only concur. I personally was out of here for some months, going through some rough stuff, and it's been a lower speed here since I came back.
But it gets up and up each day with people doing what they can. I know I'm happy I found some old friends back here, and see them more than not nowadays. As well as quite some new faces.
And yeah, you're right, it's everyone's job, and thanks for also doing your part, I always loved our convos around here

fervent thunder
#

I think a big part is just lack of a new official model for so long

vast ingot
fervent thunder
#

It's never been better.

vast ingot
#

that's true, it gets quite some energy

fervent thunder
#

And was started by former staff @fervent thunder, the current owner if I recall

#

I don't know why, but people only come here for relatively vanilla stuff now

vast ingot
#

possibly, not sure. I'm always happy to go on the reddit of SD. lots of good developements on there almost daily

fervent thunder
#

And it does annoy me, Emad is based, even during that one incident which I won't mention he always did the right thing

#

I like him and trust him from what I've seen

vast ingot
fervent thunder
#

I'll be candid though, I disagree with the decision not to opensource current and future models

#

And I LOVED Emad's recent tweet, but it smacks me the wrong way for him to criticize OpenAI like that when Stability does the same thing

#

With a MUCH weaker argument for the impact on society

#

That stuff bothers me, but I'll keep it lowkey on that in the man's own house

vast ingot
#

I hope we see things come out. it's been a busy week, AI wise. we've got some new things in #1084896022368624640 but not as big as some other players out there.
but judging by the size of the train that hit us the last time we did a @ every one by error, I'm not that worried on people coming back on the next news.
The thing is, the playfield is moving so fast currently, attendance on discord goes by waves

#

(this is just my personal oppinion, I'm not relaying any official info here ^^)

#

welcome but what language is this x)

fervent thunder
surreal perchBOT
fervent thunder
#

I want OpenAI pressured into loosening restrictions, spending more money on models, etc

#

The closer Stability gets to their tail, the more stuff like that will happen

vast ingot
#

yeah don't start sluring around

#

NSFW rule too

fervent thunder
#

On Unstable they do, this isn't the place for it

vast ingot
#

still, don't go "deepshit" on people.

fervent thunder
#

I actually disagree, but I get why it's not allowed here

#

I remember during the AMA's a lot of people came out in support of artistic nudity

#

I feel it can be used for good or bad, like anything else, personally

#

To each their own of course

#

Haha, anyway, current models eliminate or greatly suppress nudity in outputs

#

Which is the more common criticism as far as I'm aware

#

Within the community

#

So you're barking up the wrong tree I'd say.

#

Ok, that's actually true, you're more knowledgeable than you let on

#

I want people to do that personally though

#

Hmm, I'll reserve my personal desires in regards to this for DMs or a less formal chat

#

But yes, I want people to be allowed to have that

#

Whoa, no, not at all

#

I am not a child pornographer, nor do I sympathize with it

#

Well, nothing stops them from looking up porn now,

#

Except like, child safety locks

#

Which you can install

#

You can keep kids from accessing NSFW-capable models the same ways

vast ingot
#

same as you can put a safety lock on any computer

#

that's the problem of the mother. currently the models aren't geared to prevent NSFW. websites that distribute it have a NSFW filter though

#

I don't see how AI is a new problematic for those

#

I can't understand half of what you say there

fervent thunder
#

Yea, the ultimate problem here is AI generated NSFW is not, at least in that way, more problematic than the internet existing

#

There's something to be said for deepfakes and other fringe cases, but no organizations which allow NSFW allow those either

vast ingot
#

don't you are some kind of doomer here ?

#

same theme

#

strange

#

let's go

pastel fern
#

Now that there's actual competition they stop being open

vestal dew
#

how about go after those who do something wrong rather than cut off the tools ?

#

You may hit someone with a hammer. Therefore, nobody can buy a hammer.

little yoke
#

Newbie question. I’ve made a few LoRas and they are always the same size of around 9,300kb. It doesn’t seem to matter how many images I use and how many epochs etc. I’ve seen some from others that are much bigger in size. Is bigger better or does size not really matter for a Lora to work well?

vestal dew
#

Just got SD to generate an accurate image of the man who kill my father. It's in general with images

small sable
#

Hey all, I'm just getting into img2img. What all do I need to do after the base install?

opaque lintel
#

how do i browse what models are available for SD

alpine bear
opaque lintel
#

how do i even begin to browse that place?

alpine bear
#

To start searching there you'd need to go to the civitai website.

alpine bear
hallow plaza
#

Hey anyone use runpod. I find it odd that I have to load up on dollars and they can't just charge me. Do all the services work like that?

past edge
#

you have to load up $10 at a time. thats pretty much the same everywhere else

neon vector
#

Is there an extension or plugin that allows you to pick or rate images and use the results to influence the subsequent generations for that same prompt?

#

maybe it runs clip interrogator (slow) on them and ranks the top tokens common to bad and good images

#

adds the top good to the pos prompt, adds the top bad to the neg

eager cliff
novel robin
vast ingot
#

don't think so. the video presentation feels like a small scale training, not a controlnet conditionned output

novel robin
#

but controlnet could make variations of the base image?

vast ingot
#

yes, it could, keeping the same contours, ... but it may still change the face enough for the person not to be recognisable anymore.. without training SD (that is possible too) I don't see it happening with a good quality

visual lark
#

Hi guys,
I'm planning on getting a new GPU. Next to gaming I also want to get into Machine learning, doing transformers and Text to image.
I am not to sure of what to expect with each card, My current guess is that i should get 12 GB VRam if I also want to do fine tuning.
Is this reasonable? I mean doing prototyping on my machine and then renting some gpus could also be an option, no?
So what should i get? I was thinking about getting an rtx 4070 ti ( the 900 € already hurt :/ )

vast ingot
#

there is a nice GPU benchmark in the pinned message there too

#

to compare perfomance/$$

wraith fern
visual lark
humble orbit
#

If I want to train TTS on my voice, what would be the best way to go about it? I would prefer something with an easy to use UI but I can use a CLI at a medium level and have a friend who can help me.

bleak matrix
#

Good morning, everyone! How are we today?

sinful vector
#

please help - how to install the full automatic 1111 on AMD Graphic Cards. I saw shark but it is not the full stable diffusion.

#

( on windows )

#

with --skip-torch-cuda-test something seems moving in the installation, let's cross fingers

steady spear
#

How to start

surreal perchBOT
vast ingot
half cove
#

poster with newspaper

vast ingot
surreal perchBOT
#

@half cove

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

half cove
#

thanks!!!

sinful vector
#

I got it running, but when I press generate RuntimeError: "log_vml_cpu" not implemented for 'Half' ( AMD GPU CPU - 32 GB, 8 GB VRAM )

visual lark
vast ingot
sinful vector
#

okey thanks! ... I've tried the --precision full --no half but I still get RuntimeError: "LayerNormKernelImpl" not implemented for 'Half'

elder cradle
#

--no-half actually

#

you forgot a hyphen

sullen peak
#

Hi guys

normal stream
#

Hi, I have a problem with Stable Difussion using Linux, in which channel can I ask my question?

fervent thunder
normal stream
#

TY.

brazen rose
#

一座复古未来主义的空中浮岛

vestal pond
#

Is it possible to install dreambooth without sd? If so how? I'm having too many issues getting the extension to work on sd.

vast ingot
#

nope

#

NSFW rules guys

vast ingot
#

I would suggest Shivam or EveryDream for example

fervent thunder
glad jolt
#

IF when?

#

when if ever will IF release

polar sundial
#

Hello, what's the difference between the CompVis/stable-diffusion and the Stability-AI/stablediffusion repos please?

obsidian geode
#

compvis is sd-v1-4.ckpt

polar sundial
#

Thanks 👌

median light
#

Hello

#

Wow i can access the chat directly without any signup

#

cool

broken gull
median light
#

No but they have several steps where i have to answer a bunch of questions and where most chats are disabled by default

broken gull
#

oof

median light
#

Anyway, I just got started with stable diffusion. I downloaded 1.4 but i didn't really like the results. DALL-E was better according to my first impression

#

Now i am trying to get 2.1

broken gull
#

yikes

vestal dew
# wraith fern Workflow ?

A noble man come to my father, ask him to make a sword. My father work on that sword day and night, making a masterpiece. When the noble man come for the sword, my father give it to him, but he no pay for it. Instead, he kill my father.

median light
#

I have a question about the installation of stable diffusion. Is there a particular chat for this?

vast ingot
median light
#

Thanks

brisk cobalt
#

Is there a guide anywhere on how to use SafeTensor models in Automatic 1111?

#

i feel like im being dumb or putting something in the wrong place....

#

every guide is using the google colab thing but I dont want to jump through those hoops when I have a 3090 I can do stuff locally with

warm junco
brisk cobalt
warm junco
#

Its used together with models to get a style or a character

brisk cobalt
#

Sorry to bother you, how do I use it with models?

#

that would make sense with the one I downloaded! 😂

warm junco
#

Np, if you put the lora in the folder. Restart the webui. Then click on the third button under the Generate button and then select Lora and then your lora

brisk cobalt
#

Thanks! 🙂

crimson veldt
#

hey people

#

anyone here knows sd 3 is coming or not?

trail wing
#

Edge changed some put to sleep functions, was getting big lag from changing t2i to i2i back and forth, had to put the new options to off

#

they really eff it up at ms

warm junco
#

Firefox ftw

sick marsh
#

heyo, i have 2 questions about stable diffusion. If i had "(white skin:1.2)" in my prompt, what does the value 1.2 mean and what does the parenthesis do?

vast ingot
#

hey

#

the 1.2 is the "weight"

#

it will increase how important "white skin" is

#

in this case, () will help the 1.2 to know what to increase

#

but you can also have this :

a (cat)
or
a ((cat))

#

each () is basically like :1.1, multiplying a little more the weight

sick marsh
#

ty so very much

vast ingot
sick marsh
#

i was just about to ask about double parenthesis

vast ingot
#

here is the full doc on this

#

there is a little more

#

Cheat sheet:

a (word) - increase attention to word by a factor of 1.1
a ((word)) - increase attention to word by a factor of 1.21 (= 1.1 * 1.1)
a [word] - decrease attention to word by a factor of 1.1
a (word:1.5) - increase attention to word by a factor of 1.5
a (word:0.25) - decrease attention to word by a factor of 4 (= 1 / 0.25)
a (word) - use literal () characters in prompt

broken gull
#

my image ain't going to plan

broken gull
vast ingot
#

it should yeah, treat it as an escaped (. like if you used \(

timber solar
#

Excuse me, how much UNet training steps i should use for 1000 images via Dreambooth anyone knows?

uneven heath
hallow plaza
#

is there a reason I sometimes get two people in one image?

vast ingot
#

maybe because you use too high of a resolution ?

hallow plaza
#

Interesting. is 512x768 too high?

vast ingot
#

depends on the model. if using 1.X model, it's trained on 512x512 and can start to give up duplications on higher than 640 in my experience

#

you can fight this though

#

multiple solutions

#

the easiest is to use "highres fix" if you are using Automatic

#

the second one is to add more tokens to your prompt, and describe the background more

#

it will help the AI to fill the void with your background, and not create another subject

#

the duplication problem comes from the fact that the AI tries "too hard" to follow your prompt

#

and makes up a second subject to fill the empty space

elfin pivot
#

Are their any models that make really good weapons and things like that

#

I've tried generating images with the person holding a gun but the guns come out weird all the time

hallow plaza
hallow plaza
vast ingot
vast ingot
#

should work even better now with controlnet

hallow plaza
vast ingot
hallow plaza
vast ingot
#

the model you need to download to train on ?

hallow plaza
#

Is the point of dreambooth to get an output or to train a custom model?

vast ingot
#

train a custom model

#

and yes, you give it an input model to train from

#

usually 1.5 or 2.1

hallow plaza
#

Ok, so I'm looking for the model that I trained.

vast ingot
#

and the format will depend on what tool you use

#

automatic uses CKPT for example, or safetensors

#

and will give out those

#

how do you train ?

hallow plaza
#

in dreambooth I created a model,then I gave it inputs, settings and concepts. I then hit the train button.

vast ingot
#

yes, but like I said, "In dreambooth" can mean more than a dozen things

#

so what dreambooth do you use ?

hallow plaza
vast ingot
#

ok, the one in AUTOMATIC

#

I think it goes directly into your models folder, I haven't used it in a while now

#

but I think I remember finding my new model direcly with the others, in the same menu

#

and it gives .ckpt

#

or .safetensors maybe now

hallow plaza
#

Your right. it created several safetensors. now sure why or what that means

vast ingot
#

each has a number in its name maybe

hallow plaza
#

It does

vast ingot
#

usually, it does "checkpoint", quicksaves in the middle of the training if you want

#

those different files are this

#

different steps in the training

#

the highest number is the most trained

#

(because it's completly possible to overtrain, and need to go back to a less trained model)

hallow plaza
#

Thank you, that makes more sense now.

vestal dew
#

No SD, hiding hands and feet is not a solution to mangled hands and feet.

buoyant jacinth
vast ingot
#

sorry

#

that one

#

I'm really liking how this is turning

#

soon there will be full libraries of templates

buoyant jacinth
#

just spotted those nodes today. There's some PRs on a few to include in comfy

balmy jungle
#

how about the state not take so much money from us

graceful pewter
vast ingot
#

possibly

#

feels like deforum + custom edition

#

hard to know for sure

graceful pewter
#

any links to read more about it pls ?

vast ingot
#

for sure

#

so deforum is a tool made out of SD, specialized in making animations

#

it focus on using img2img on loopback

#

meaning it will feed the previous image output (previous frame) as input for the next

#

it also adds a depth map detection, and lets you "move the camera" around, zoom in and out, and let the img2img generate the new details

#

the custum edition, it's mostly just cutting and putting each sequence after the next in any video editor

#

they also have a nice discord

hard crow
#

Can you not generate images on this discord anymore?

#

Looks like the channels are gone that were there a couple weeks ago?

vast ingot
surreal perchBOT
#
FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

wraith fern
fervent thunder
#

Do we have a model to do creatures yet?

vestal dew
nova lily
#

Was wondering, how does "&" act in SD prompts, does it under stand it? does not?

vestal dew
#

havent seen it

astral goblet
viscid kite
#

Anyone see this ?

#

Open weights text2vid. No samples posted

fervent thunder
#

interesting

zenith olive
#

can you guys help me in installing st.d.2.0

#

pleasehabby

vast ingot
graceful pewter
#

i m afraid it doesn t work on a mac tho 😦

fervent thunder
#

Dm for teen videos

shadow blaze
#

-Hi! I´m a digital artist that has been working on VFX and 3D animation for 20 years,-I´m not a programmer, never dedicated time to learn code. -I'm totally aware of this revolution that is happening exponentially. -I have been studying and learning AI for the last couple of months, (mostly Stable Diffusion Auto1111 WebUI - Deforum - Dreambooth) and at this point I have some questions about the tools available.1- I have seen that Midjurney ( at this point ) is absolutely superior compared to Stable Diffusion but I'm not sure if that is because I´m using an older model: v1-52- Are there plans to upgrade SD models soon? I know that there are some models around the internet that you can download but is there a formal project to upgrade the standard model soon? 3- As a noob I'm not sure if one of the pros of SD Auto1111 WebUI is that you can create your own models and produce animations using extensions like Deforum. Is that the main reason many people are using it? Sacrificing quality, I mean....4- Noob question: can you load the Midjurney model into Auto1111 WebUI? 5- Why Midjurney results are SO SUPERIOR when you compare them to SD Auto1111 WebUIThanks!!!

woeful patio
#

Hey guys, I’m facing a problem which’s if my prompt token is > 75, the image will be totally black. Anyone may have a solution? It happens with all the model

vast ingot
#

but this isn't available on the base Stable diffusion repository

#

technicaly it's 77 tokens, but 2 are reserved for "beginning" and "end" of prompt

woeful patio
#

Thanks for your reply. I’m using A1111, and this happened recently.

vast ingot
#

this feels like a bug then

#

you should check with #🤝｜tech-support . I think doing an update may help ? maybe an extension would have broken it ?
hard to tell, they'll know more than me on that one

woeful patio
#

Thanks again. I’ll try to disable all the extensions then see

keen citrus
vast ingot
# shadow blaze -Hi! I´m a digital artist that has been working on VFX and 3D animation for 20 y...

Hey !
hard to read all on one line :p
1/ MJ does more than just push your prompt into its model. it adds flavor to it, it adds a fitting style based on what you prompt, and I believe it also applies some post processing hypernetworks or something, it feels a little more than basic generation for sure. They also use a series of custom models in my opinion. they aren't open on their pipeline so this is mostly my speculation.
2/ More models will come out. There is 2.1 already, but 2XL is being teased on #1084896022368624640 recently, and 3.0 has been talked about too in the past.
But already today, adding some randomizers to your prompt to have different, well working styles, each picture, and a good base model you could find on https://civitai.com/, you can do quite good in SD too.
here are some I did last night : #🏞｜general-with-images message
No idea on the timeline on official models though
3/auto has lots of extensions letting you do lots of things, but there are new good players around too. I would recommand looking at InvokeAI or ComfyUI for example
https://github.com/comfyanonymous/ComfyUI
https://invoke-ai.github.io/InvokeAI/
as well as the #1080946152318443610 channel here
4/if you find it somewhere, then yes, but I don't think it's public
5/ I explained it mostly in the 1. Automatic can also give really good results, but with more manual setup for sure

#

the thing is, official SD models don't try to be overly artistic like MJ does

#

they try to stay generic

#

they are intended to be trained on more, specialised

#

no model (on our current tech/parameter counts) can be good at everything, and SD base models, for now, have tried being OK at almost any subject

#

even MJ will have some problems tackling specific styles, especially if you don't want it to give you very "artsy" results

proper birch
#

I feel dumb please someone help. I join this server to make some prompt images but couldnt find where we do it. Is there any difference place to make it or im a blind 😦

vast ingot
#

hey and welcome first of all

#

we don't have a bot for that currently, sorry

#

I'll link the faq on that

surreal perchBOT
#
FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

broken gull
#

How can easily search for outpainting models on huggingface?

proper birch
broken gull
#

there is no section for it and when I search for outpainting I get two empty uploads

vast ingot
#

yep, I think it's inpainting models that are the best at that, no specific outpainting

#

outpainting is done by tricking the inpainting

broken gull
#

aha so I should look for inpainting? thx

vast ingot
#

np 🙂

broken gull
#

any favourites?

#

(yk what I am working on)

#

downloading stability v2 .safetensor inpaint one

unkempt hatch
#

Block coding artifacts, jpeg artifacts, aliasing artifacts, quantization artifacts, they're all different, right? I'm afraid I'll mislabel them in my captions

vast ingot
#

if you are not intending to prompt for them in the trained model, I would just go with "artifacts" for all those. It will divert the attention so they get less learned on the other tokens, while not hogging all the attention with too many different tokens either

unkempt hatch
#

or tag like "noise, jpeg, artifacts" so it'll cover them all?

#

also, is there some app that makes editing captions fast?