#💬｜general-chat

1 messages · Page 50 of 1

vestal dew
#

but the post was from December

frigid orchid
#

I'm likely already running all that than. Unless you have to follow special install steps. Which does not seem right.

#

I read and follow a lot of guides etc. Never heard of anything. Except those things I don't use called deformers or something?

#

I don't even know what they are or if I should use them. lol

buoyant jacinth
#

the 'boost' was to download libraries from nvidia and drop them in the pytorch dependency folder, at the time automatic1111 (I think) didn't have the ability to distribute them. But a newer post said they're included now.

frigid orchid
#

But likely will soon.

#

Month is like 25 years in AI time.

#

But so much was changing I was like... I don't want to update every few days. So waiting for a bit longer.

past edge
#

yup thats why all of these txt2video gizmos are dropping out right now and I'm like 🥱

#

I'm waiting a good 3 weeks for us to have the real thing

frigid orchid
#

txt2video is going to be awesome in the future. Like next year.

#

But for now I don't like it.

past edge
#

r0fl thats like 200 years in AI time

#

I give it 2 months max

frigid orchid
#

Exactly!

#

lol

#

If someone makes an app that alerts user of every AI advancement or news. It may just be better for the app to send an alert when 20 minutes pass without an update.

#

Less alerts.

past edge
#

I just wait for aitrepreneur's videos although even he can't keep up

frigid orchid
#

Yes I follow like 3 youtubers and he is one of them

past edge
#

I'm waiting for him to drop a vid on comfyai since I dont like olivio's vids

frigid orchid
#

I sadly get the mixed up.

#

Olivios I also follow.

buoyant jacinth
#

after testing the same prompt/seed/config on 30 different models I can say, without a doubt, that standard diffusion loves butts.

#

So those videos will likely be a lot of twerking.

frigid orchid
#

Vanilla 1.5 is not that good of a model.

#

It kind of is.

#

But when you look at user models it's not even in the same like region. Let alone ball park.

past edge
#

exactly

#

base models are trash

#

merges are where its at

frigid orchid
#

1.5 vanilla is better than 2.0 and 2.1 vanilla.

past edge
#

2.x is still trash all these months later

frigid orchid
#

Sure 2.0 and 2.1 have some exclusive like features or bonuses. But overall are worse.

gaunt radish
#

how do you combine concepts

buoyant jacinth
#

I don't even have the vanillas. Anymore. I used to have 1.4 and 1.5, not sure what happened to them.

frigid orchid
#

2.1 that one user model is pretty good!

past edge
#

sadly theres no indication that 3.0 is gonna be any better. thank the big corporations and snowflakes for censoring AI

buoyant jacinth
#

Probably deleted to make room

frigid orchid
#

But it's not as good as 1.5 models

gaunt radish
#

like im trying to make a pokemon turtle made of vines but I can't it just gives me regular turtles

frigid orchid
past edge
#

damn has it really?

#

ur right. it was only a few months ago that I discovered mj and went into this rabbithole

#

I've come a long way in such a short time lol

frigid orchid
#

Yes AI as we know it is not even 1 year old

past edge
#

fuck

frigid orchid
#

I think AUG is about 1 year

#

Now sure there is like private betas etc.

past edge
#

All bad, especially teasers like Google's music AI

frigid orchid
#

But I mean going public for MJ or SD. If I am not wrong it's like AUG or something like that?

#

I think MJ came out sooner.

past edge
#

Once we get true AI generated music I can say for sure it's a good time to be alive

#

but not music that sounds like stock music

frigid orchid
#

Yes good AI music does exist now but it's closed to the public. That's my understanding

past edge
#

I mean AI that can copy an artist's style. That would be the shit

#

yeh but it's unknown whether it just does generic or if it copies styles

frigid orchid
#

Also probably 2 text to image modesl that can surpass or match MJ/SD User models that are also closed to public.

past edge
#

I mean, V5 is essentially closed behind a paywall rn

#

not that I care, MJ can go fuck itself

frigid orchid
#

Yes I have used the V5 Alpha

#

It's good!

past edge
#

Yeah its good but what good is it without dreambooth?

#

No inpainting?

#

Its trash.

#

But so much wasted potential though

buoyant jacinth
#

when music AI gets released, the lawsuits will start flying. the music industry does not like any form of competition

frigid orchid
#

My view point is MJ V5 is the best text to AI on the market. Paid or free. However with limitations that you don't have with SD. So SD overall I think is better and of course has more max potential.

past edge
#

ofc

#

It's the fact that it runs on discord that grinds my balls

#

If they let us have an interface like SD's, I'd be like $300 poorer rn cuz I'd be subbed since it came out

#

but fine, let em do it their way

frigid orchid
#

MJ as far as we know will never be 100% private, allow gore/nudity etc, and it will never be free... and likely never have nearly the same level of customization. And likely won't be open source. All of these are major victory for SD.

past edge
#

ugh yes I forgot about that

#

no boobs, no ass, and I think they nerfed celebs as well

#

nty.

frigid orchid
#

Oh and lets not forget that MJ is struggling to release inpainting, outpainting etc... don't eve get me started on things like controlnet.

maiden bough
#

What is SDXL?

frigid orchid
past edge
buoyant jacinth
#

So, uh, I tried that prompt test on 1.5. OMG. It's terribad. Like art seen on the back of a redneck truck bad.

past edge
#

MJ is like the Mac of computers.

maiden bough
past edge
#

If people wanna waste money on something so inferior then let them

frigid orchid
maiden bough
#

I'm guessing MJv5 is an 8B param model

#

It's just that much better than anything else out today

maiden bough
buoyant jacinth
#

I've seen a few messes, but none this bad.

frigid orchid
# past edge ugh yes I forgot about that

MJ V5 I think has some of the best celebrity functionally of any model! Otherwise your normally right they try to hold back on some of that. But a lot of people pointing out that MJ V5 at least the alpha has been pretty amazing for that.

buoyant jacinth
#

ah, wait, don't have a vae for it. let's see if that helps

past edge
#

its insane in the amount of detail it can generate

#

I wish someone could suck that model right out of their tentacles and find a way to export it to SD without the censors

#

can you imagine the utopia that would emerge

frigid orchid
# maiden bough Oh hell yeah we need that

Yes I agree. I want it so badly. As long as it does not flop like 2.0 and 2.1. Here is a message I sent to my friend. This is NOT SUPPOSE to be serious. This is just me giving a rough opinion. lol

Here is how I see quality of AI models and what open source can do.
If vanilla SD 1.5 is == 100 in quality on a scale.
Vanilla SD 2.0 and 2.1 are both like 70 on the same scale.
User models for SD 1.5 are like 300 on the same scale.
Dalle 2 is like 150 on the same scale.
And MJ is like 325 on the same scale (edited)

So if Vanilla SDLX can be like 150 on the same scale. I'm sure it will easily be open sourced to like 350-400+ on the same scale.
Oh and to finish this scale off I would put user models for 2.0 and 2.1 at like 250ish but of course with some unique bonuses over SD 1.5 models.
And Craiyon is like 20 on the same scale. lol

past edge
#

in sheer quality yes

#

but theres more to it than that. controlnet alone makes MJ be like a -100

frigid orchid
#

Absolutely. And I agree 100%

past edge
#

seriously I dont even bother generating without controlnet these days

#

good luck having those fools integerate that into a clunky discord prompt

frigid orchid
#

Sometimes though certain generations are impossible with controlnet.

#

So it's all about what you want!

#

However thats were SD shines. You can fine tune things to what you want.

maiden bough
#

Do we know how many params they are targeting?

#

Are they going back to CLIP and porn?

#

We need clip and porn both in the base model imho

frigid orchid
#

If you want everything without really having to change to many settings at all. MJ is the best overall model and system likely. But like we all have all discussed. It has major issues. That can make it pointless to use at times.

maiden bough
#

Unless openclip is a lot better

frigid orchid
# maiden bough Unless openclip is a lot better

Sadly all of that goes over my head. But I do want NSFW in base model. Just like 1.5 etc. However I don't know how important that really is? Like can user models force that in if they need too? OR does base model need some of it anyways. I agree if it's has no NSFW ability than the SDXL will still have a purpose! But it won't surpass 1.5 overall in my mind. I may go a long time without generating NSFW content at all. But you need the ability too in order for a truly great model.

#

To some point it's not important. To others it's very important. But not having any ability to do so is a problem when it comes to free expression of art.

#

Like I clearly still love MJ. And it has none of that. SO it's not like my use of text to image is dependent on it. But at times it's very good.

maiden bough
past edge
#

so is it gonna be base 512 or not?

#

cuz im tired of 512 dawg I want that 1024

maiden bough
#

It's not about the porn, it's that making an SD base model with no porn is like making an Olympic athlete that can't eat protein. Just fundamentally really disadvantaged

past edge
#

can we just agree that censorship and cuckification are destructive not protective.

maiden bough
frigid orchid
#

yes the great thing is that having NSFW content in a model makes more people put effort into it. Even it's PG content are likely going to get a buff too. PG content enhances NSFW I don't know if you know what I mean but... people often want like you know amazing looking things in there NSFW content. You can still use many of those models in non adult ways.

past edge
#

for people with small vram yes

maiden bough
past edge
#

why

maiden bough
#

You want more parameters, not more resolution

past edge
#

why not both?

frigid orchid
maiden bough
#

High-res fix will get you to 2048x2048 with no issues

past edge
#

yeah while the imgs look plastic af...

#

mj I can guarantee is at least 768

#

I suspect v5 is 1024+

frigid orchid
maiden bough
# past edge why not both?

Because they both take ram. If you want the thing to run in 24 GB of VRAM, you can do either 1024x1024 and 2B params, or you can do 512x512 and 8B params

past edge
#

oic

maiden bough
#

And I'm highly confident that 8B params is going to do better than 2B

buoyant jacinth
past edge
#

Why not 768 4B? the sweet spot lol

frigid orchid
#

I agree with Sen that we need higher than 512 at some point. I can be fine with 512 with more paramaters for now. We have seen how good 1.5 models can be. But I would like 768 or more with more paramaters and 1024 is likely all we need. But I'm fine with waiting a year or 2.

past edge
#

768 in theory is already spectacular if it wasn't cucked into oblivion

maiden bough
past edge
#

interesting

#

are the exact specs of MJ models known?

frigid orchid
#

I agree taek. But both would be my prefer when possible.

past edge
#

or is it all speculation

#

cuz having some reference point would be better

frigid orchid
#

I think it's mostly speculation. But rumor has it that MJ needs like over 40GB VRAM to run.

maiden bough
#

Is this the official stability ai discord btw?

past edge
#

yeh they changed the name for some dumb reason

#

and the pfp for an even dumber reason

maiden bough
#

It was very hard to find lol

past edge
#

u know how those mods are. always out of touch

frigid orchid
frigid orchid
#

lol

past edge
#

r0fl

frigid orchid
# past edge r0fl

One time they went on for like an hour about how they are worried about people using the Bikini prompt. And are secretly trying to nerf basically good looking woman from being generated.

#

IT was more than that. But that was the gift of it.

past edge
#

because muh feelings

frigid orchid
#

Yeah

past edge
#

because muh unrealistic beauty standards

#

all the while true artists get royally fucked

frigid orchid
#

Lets create an ART AI. And then judge our customers and shame them.

#

And nerf our product.

past edge
#

well same thing happened to chatgpt sadly

#

and I dont think we'll ever get a decent chatgpt competitor.

#

as soon as an uncensored gpt-4 model comes out, it will get sued into oblivion as soon as 1 person learns how to make meth

frigid orchid
#

I do agree and I'm completely fine if they want to "hide" or reduce those kind of images from the public view. Like filter them down on the top image boards etc. There home pages etc. But don't nerf the model itself.

past edge
#

that's why chatgpt still gives you the disney-approved answers on anything remotely creative

frigid orchid
#

Yes we need an open source system to fix that.

#

But I think that is a while away.

past edge
#

lawsuits dont make money, but they sure hurt freedom

#

also

untold ocean
past edge
#

have you seen the average twitter moron

#

there were people sabotaging sd cuz they were scared people would use it to generate kids

frigid orchid
past edge
#

yea I have no faith in a corporate-sponsored AI product. If it happens, it's going to be crowdbased

#

the internet is too controlled these days bro. 1984 in the flesh

#

I wish I was joking

#

1.5 might be the last taste of true AI freedom we got to lick off the floor lmao.

frigid orchid
# past edge there were people sabotaging sd cuz they were scared people would use it to gene...

Honestly I a lot of twitter hate is just degenerate furry artists just want there gig to not be stolen by AI. Not that I blame them. lol But I don't think for a split second that they really care about those kind of images being generated. Sure some do and those are legit concerns. However they don't have there priorities straight. They may say that is there reason. But they don't like AI art in general.

#

Most of them also don't know how AI art really works. And get image to image mixed up with text to image.

#

So there IQ is far to low to take seriously. But sadly they reeeeee loud enough that it still causes problems

maiden bough
frigid orchid
#

Yes it is.

#

I use Llama to remove stuff

#

One of if not the best text removers etc.

#

Or removing an extra limb etc.

maiden bough
#

Text removal!?

frigid orchid
#

Like know how an AI may spit out garbage text that you did not even ask for

#

But the rest of the image is godly.

#

Llama can like remove stuff

maiden bough
#

Llama is an LLM though

frigid orchid
#

I don't know what that means.

maiden bough
#

It's a language model not an image model

frigid orchid
#

lama-cleaner

#

Oh we may be talking about something else

maiden bough
#

Oh that's completely different

frigid orchid
#

Oh okay sorry.

maiden bough
#

Llama 65 is Facebook's GPT competitor that got leaked so the weights are freely available

frigid orchid
#

Neat

warm jasper
#

Are you using llama.cpp? I have the 65B downloaded in another computer that I'll have access only in a few weeks, and I'm wondering how much memory I would need to convert it to the format llama.cpp uses

frigid orchid
vast ingot
#

🐮 COW FACT TIME 🐮
Did you know that some of the most experienced moderators on Discord are actually cows in disguise? That's right - these clever bovines have learned to navigate the internet and communicate with humans in order to keep their servers running smoothly.

Despite their large hooves, cows are surprisingly adept at typing and using computers. They've even developed their own secret language of moos and cowbells that they use to communicate with each other while monitoring Discord channels.

Of course, it's not always easy for cows to blend in with human moderators. They sometimes accidentally reveal their true identities by randomly mooing during voice chats, or by expressing a sudden interest in grazing on Discord's grassy green color scheme.

But for the most part, these bovine moderators are experts at keeping their cow-ness under wraps. So if you've ever been moderated by someone with a mysterious affinity for milk and grass, there's a good chance you were being watched over by a clever cow in disguise!
🐮 END OF COW FACT TIME 🐮

untold ocean
#

🤔

frigid orchid
#

Alright? Is that a chat GTP generation or something? lol

vast ingot
#

I would never

frigid orchid
#

lol

vast ingot
#

ok I would

#

ok I did

foggy wadi
#

Was this server bought?

#

Why are there so many kids here

maiden bough
#

10 t/s

vast ingot
#

the server hasn't been bought no, and population changes quite a lot, depending on the weeks. we do have a very active anime community that tends to be younger sometimes, but there are quite a lot of representation of all ages, from out last "sensus"

warm jasper
#

GTX 1650 here, so no way it's going to the GPU
But the Ryzen 7 5800H can do something, I hope 🙏

frigid orchid
maiden bough
#

So there are certainly more improvements to be made

frigid orchid
#

Nice Taek!

maiden bough
#

I'm hoping to release the code soon, still making a UI

warm jasper
#

I am getting kinda good results with 7B and 13B models on CPU, 7B can get around 4 t/s, with 13B it's around 3 t/s

frigid orchid
frigid orchid
warm jasper
#

LLaMA, 7B and 13B

maiden bough
#

Have you tried alpaca?

frigid orchid
#

Ah okay. I was like.... SD barely uses CPU at all. I misunderstood.

warm jasper
#

Not yet

charred mesa
#

I tried 7B and 13B alpaca

#

both online stanford and offline 4-bit version using llama.cpp/alpaca.cpp

frigid orchid
#

I can't wait till we see something like Kobald AI but using something as powerful as chat GTP. IF that makes sense? I don't know if anyone understands what I'm even talking about. But for role play, adventure and even erotic stuff.

charred mesa
#

have you tried pygmalion for that?

frigid orchid
#

No

charred mesa
#

it's made for the latter

frigid orchid
#

Kobalt AI is the best I've seen/used.

#

I may look into it.

maiden bough
#

Llama 65b is really good at erotica

charred mesa
#

ugh 👀

frigid orchid
#

I may look into that too.

maiden bough
#

I'll make you a story if you ask nicely :p

frigid orchid
#

No I'm good. lol

maiden bough
#

😂

frigid orchid
#

What the best way to get started with Llama 65b on my PC. Can I use it directly on my machine? Like SD? If so am I good with 12700K and 4080?

#

And 64GB RAM

#

Sorry

#

64GB regular ram

warm jasper
#

I want to know how much memory is needed to convert LLaMa 65B to 4-bit too

#

Because my laptop has 40 GB and thanks to Lenovo's buggy firmware, it doesn't work with 64

charred mesa
maiden bough
maiden bough
frigid orchid
# maiden bough Fair enough

Wait so you can't run it locally? Stack of 6 A100 sounds insane? Is that a one time compute? I'm lost. lol

#

Nvm I looked into it. Yes Llama 65b is not a regular user product.

maiden bough
#

Has the stability ai team made any comments on porn in SDXL?

frigid orchid
#

Yes again I'm curious as well. 2.0 and 2.1 did not have that right?

#

If they did not I don't think SDXL likely would? However hopefully it does and they learned of their mistake.

maiden bough
#

I'm hoping they realized:

  1. We figured out how to make really good porn anyway
  2. It really stunted the quality of all their non-porn images
#

As long as users have the weights, they will fine tune porn into the model

vast ingot
#

well, the position on last models (2.1) was to kick NSFW out of the base models, yeah, and I don't think this is going to change in next models, but I didn't see anything specific on that XL version specificaly for that.
My guess is that, yes, they know we can train it in. it's what open source and responsible use is there. and yeah, some will do as they like watever the rules.
But as a leading AI thing currently (not even sure how to characterise all those blooming currently), it needs to set the bar quite high for the official model, in order to prevent very damaging litigation, and to be able to come into each and every country following local laws.

elder cradle
#

The only thing that SAI only tries to remove is hard porn, not general nsfw iirc

vast ingot
#

yeah but it has some impacts on quality of NSFW for sure too, even if that isn't the main target

#

but yeah, base model are called based for a reason, they get trained. they aren't MJ good in any subject, but they are good in everything and we can train in the rest

#

and it's hard to spit on free cookies tbh too, when there are big technological steps crossed each model

maiden bough
#

2.1 is clearly inferior to 1.5

vestal dew
#

In outpaint in Auto 1111, I could overlap the dream only go one box beyond the original image, and it would generate just that one box new. Now it's regenerating the image where it's overlapped, but it's not generating anything beyond that. Anyone know about outpaint, what setting this could be?

#

nevermind. I had img2img selected rather than dream

iron hazel
#

how to get updates from a git repository to files in my pc in the terminal?
'git pull' or 'git clone'?

vast ingot
#

git pull to update

#

git clone is to download a new repository

iron hazel
#

git pull doesn't do it
fatal: not a git repository (or any of the parent directories): .git

vast ingot
#

you need to be inside the right directory

buoyant jacinth
#

git offamylawn is to chase away hoodlums

random shale
#

hu

vast ingot
#

using CD, move into the directory

iron hazel
#

im in there by using cd

vast ingot
#

ok, then maybe you didn't install using git clone ?

#

if you installed by downloading the code as a zip folder

wispy totem
#

my automatic111 web-ui is crashing everytime I switch the model :/

vast ingot
#

then you need to update by doing the same

iron hazel
#

by using git clone?

buoyant jacinth
#

update by downloadin a new zip

vast ingot
#

if you install using "git clone", that makes a git directory.

#

download the zip, overwrite your old files

iron hazel
#

the thing is I moved the whole web-ui folder from C drive to D drive by copy pasting

#

it used to work before by typing git pull i think

vast ingot
#

check if there isn't a ".git" folder in your old install ?

#

it's an hidden folder

#

it should be inside automatic if you did a git pull

#

but the error "not a git directory" seems to say that no, you don't have it

iron hazel
#

.git is there

vast ingot
#

ok then make sure you are inside that folder in your console prompt maybe ? and run "git status" to know more on your git state
(inside that folder = inside webui, not inside .git, that wasn't clear)

iron hazel
#

is there a faq somewhere that states how to update, automatic only lists how to install

buoyant jacinth
#

can try: make new directory. git clone into it. copy your changed files over from your install dir. do a git pull to fix anything terribly wrong. then run and see if it works.

iron hazel
#

wait a minute

#

doing cd in terminal doesn't make the directory the main one

vast ingot
#

click in the adress bar inside Explorer, in a windows that is in the directory you want, and type "cmd" then Enter

buoyant jacinth
#

if cmd, think you have to change drive before cding?

vast ingot
#

in windows, you may need to do "cd /d F:"

#

to change directory first

#

F or another letter

iron hazel
#

that worked

#

the cmd thing

vast ingot
#

but yeah

#

the cmd thing is quite useful to know

buoyant jacinth
#

or just F:

vast ingot
#

then maybe even git pull works now ?

iron hazel
#

because before i only have to copy the address of directory inexplorer

buoyant jacinth
#

or use powershell (:

iron hazel
#

git pull worked after the cmd

vast ingot
#

nice

#

yeah, changing drive through CD in the basic command prompt in windows may require to use the "/d" option

iron hazel
#

apparently the terminal wasn't directing me to the directory using 'cd' and i didnt notice it

#

so = cd /d (directory name)?

vast ingot
#

yes

#

from what I see here```C:\Users\berti_o7k3ry8>cd /d F:

F:>```

#

sorry no, not directory name, but drive name

#

then cd path

#

my bad both at the same time work

# 

F:\AI\Software\Python>```
wispy totem
#

will manually updating SD break AUTOMATIC1111's ui?

vast ingot
past edge
#

is there a controlnet that lets me isolate the background so I can just add a background to a subject? seems like something so simple I forgot how to do it

vast ingot
past edge
#

I managed to do it with the classic midas script. How would one tell the canny to just paint outside of the subject edges, instead of inside?

vast ingot
#

the problem I see is to keep the subject intact honestly

#

canny doesn't seem the best to just create things out of white background

#

scribble would be

#

but scribble will destroy your main subject

#

maybe using a mask on it ?

#

canny will invent things for the background too, but it will try to do so while keeping the same canny preprocessed image

#

so in your case, it would mean the background would just stay fuzzy, to prevent edges

#

I'm not sure there is a lot you can do to change canny on that side. the best would be to push thresholds to their maximum while still having the outlines of your object

#

but that would allow it to modify the subject quite a lot too since so few details would be set in stone

#

to me, it mostly depends on :

  • do you want to do a mask to preserve the subject ?
  • how important are the fine details of the subject ?
past edge
#

Nah forget about all that G

#

just use midas invert

#

on general-with-images. invert depthmask does wonders

warm jasper
#

Realistic_Vision understands 4bpp on the prompt, that's cool

devout vine
#

hello?

#

i came to see what was the general opinion regarding ai

wispy totem
astral goblet
round kraken
#

queue: 14/14 | 71.6/699.4s
Dang, the wait times for ModelScope Text to Video are getting pretty long.
I wish my PC met the minimum requirements to run it locally, but I only have 8GB vram

vestal dew
#

So wait, Wally West and Aqualad are now black?

past edge
#

or at least nothing that conventional sd to video cant already do

round kraken
past edge
#

does anyone know if it does nsfw?

vestal dew
#

the dc movies have ruined dc characters....aquaman is that awful hawaiianm guy, can't get SD to give the original Aquaman or Aqualad

shy prism
#

Is there a way to check what objects and styles a specific stable diffusion model knows?

vestal dew
#

anyone know how to get green eyes? I tried (green eyes) to different strengths, but it makes the clothes green

lavish current
#

green irises or green pupils maybe

vestal dew
#

ill try it

#

also trying inpainting\

round kraken
#

Huh, it seems that Hugging Face wait times seem to be specific to my IP address. I had a 1000 second wait to use ModelScope Text to Image, but after turning on my VPN it's only a 92 second wait

vestal dew
#

eh, green iris and pupils also does green clothes

#

I guess I have ti inpaint

#

it will not do green eyes

#

it at best makes aqua blue

void citrus
#

Hi, I have a general question about hosting SF on my main computer but accessing it on my laptop via a1111. I know that gradio isn't right but I can't seem to get Spaces to work either

#

Apologies if this is the wrong place to ask

#

Gradio works for like a few hours and I want to set up something more long term

#

I am using the --listen tag in my .bat

inner iron
#

my first few steps with AI, im exited about the results... https://prompthero.com/BalrogDx and the wolf is so cute as i described 🙂 if anyone interested in these prompts, here u go!

brave matrix
#

Hello

#

Can I ask for some feedback on my project?

#

I can’t tell if people like it or not, is it too long maybe?

void citrus
#

Definitely too long but not bad nice

#

make it 5 mins of the best

brave matrix
#

Maybe upload it as clips?

#

So four segments

void citrus
#

edit the clips and do 1 upload nobody is going to click on that 5x

#

but you'll be good

#

make it 1 project

brave matrix
#

What do you mean one project

void citrus
#

well you need to edit for comedic timing first of all

brave matrix
#

My goal is to replicate a real talk show though so around 15min makes sense

void citrus
#

for sure, but I'd start with 5 good minutes and start stretching little by little. That's how they do it in the paying world

brave matrix
#

Are you saying it’s not high enough quality yet for 15min

void citrus
#

I'm not saying anything, just offering my criticism. I might be wrong

brave matrix
#

I’m feeling pretty discouraged at the lack of attention it’s getting

#

I thought it’d be more of a hit

#

It’s entirely AI

#

No one else is doing something like this to my knowledge

void citrus
#

How long has it been since you uploaded?

brave matrix
#

Only an hour but it’s usually a good indication

#

And the Reddit post has no likes

void citrus
#

Well there you go you're freaking out over an hour of not getting likes on the internet. You're gonna have to get over that pretty quick. Sorry to be the one to tell you.

brave matrix
#

I know how that sounds lol

#

But still

void citrus
#

But still it's time to grow up and stop worrying about that shit. You'll be fine. Go do things you like

brave matrix
#

This is what I like

#

Spent all day on this one

#

I feel like from a technical perspective it’s a notable achievement

void citrus
#

Awesome, so trim it down to a concise, journalistic 5 mins and see what people think. Nobody will spend 15 mins listening to someone they don't know. You have to ease your way in

brave matrix
#

Entirely GPT-4 written, stable diffusion animations, all visuals are SD and Midjourney, ElevenLabs voice cloning, even the music is all AI

broken gull
void citrus
#

Fair

broken gull
#

If you create the vid in a interesting manner to keep watching that helps as well

#

just use the stability.ai tag on twitter Xd you'll reach plenty of people ¯_(ツ)_/¯

brave matrix
#

What tag ?

#

#stabliltyAI?

broken gull
#

Also anyone that knows how you can easily generate the exact same character but in different poses?

broken gull
brave matrix
broken gull
#

It doesn't generate my character exactly the same and I need to put denoising strenght on 1 to have changes in poses and it doesn't even match.

young palm
#

I need to create an animation use my photos.

inner iron
#

https://prompthero.com/BalrogDx someone has tipps for the snow fox of this gallery? to get a good looking flufffy fur is hard. sometimes its like spikes sometimes its to animated lookalike. The 2 Pics in these Gallery are the 2 of 30 which was looking nice

#

if someone has expierience or ides to get a nice fluffy fur, im glad to know that knowledge 🙂

void citrus
#

I'm working on "Tyrannosaurus Rex runs a kitchen in 1972"

inner iron
#

and maybe a more natural forest

#

lol, sounds nice aswell

void citrus
inner iron
#

do u have a gallery to look?

#

ahh ok

#

nice pic, love the black white retro style-maybe a tipp , try to describe the shirt collar or the neck with a additional scarf.

#

only a idea cause the point from neck to shirt is to big, but i think its hard to describe it

broken gull
#

It will use your photos as reference points

#

And animate the characters inside

void citrus
#

Totally agree. I'm running a new one that's like 20 mins. 30 steps, batch count 10

#

Well nevermind Grdaio failed

#

*Gradio

inner iron
#

maybe with weight?

#

so that the collar has more weight?

void citrus
#

starting a new one soon

#

had to reboot

#

A dinosaur created a successful restaurant. 35 steps

inner iron
#

ok, so its must be true^^

#

will try that, its a funny idea

pliant parrot
warm junco
fervent thunder
broken gull
fiery pecan
#

hello frens, i ran into a problem with inpaint section

fervent thunder
#

hello everyone, i have a question. Can i generate an image from a clip vector using stable diffusion, if yes how ? I tried to use ddim sampler but it generates some images full of artefacts

fervent thunder
#

bing image creator seems like its dalle-3 or something. but more new ai stuffs with mj5. good time to drop a new stable mode lol

#

Idk if we'll get actual new model until lawsuits will be figured out and what data sources we're able to use and whatnot

#

But I'd take huge 768x768 or 1024x1024 model for sure...and wait for community to do more things with it

#

adobe got some model coming too

#

going to be a lot of options in time. chatgpt is showing them how profitable this stuff is. the models (and cash) will flow

sleek osprey
#

Emad sure taking a while to drop new models, maybe hes just gonna drop distilled XL and blow the competition out of the water

fervent thunder
#

yeah they been quiet

spring marlin
#

Hello, I am a high school student that has read a decent amount of deep learning computer vision papers and would like to ask you if someone could give me advice on how I can start learning pytorch to implement these papers and code my own when I write my own research paper in the future. I am already profficient with python

sleek osprey
#

on youtube

#

try that

muted skiff
# spring marlin Hello, I am a high school student that has read a decent amount of deep learnin...

Well let me tell you somethin', brother! It's fantastic that you're already proficient in Python, dude. That's like having a headlock on the programming world, man! Now, to get started with PyTorch, you're gonna wanna check out their official website and documentation, brother. They've got tutorials and resources that'll help you body slam those deep learning concepts, just like I did to André the Giant back in the day, man!

Now, when you're ready to implement those computer vision papers, think of each one like a wrestling match, brother. You're gonna need to study your opponent – in this case, the paper – and understand their moves and techniques. Break it down step by step, and don't be afraid to ask for help from the community, man. Those folks are like your tag team partners, and they'll help you bring home the championship belt!

silk cipher
#

is there an implementation of multi-controlnet in python/jupyter or a pipeline in diffusers? (i ran it in auto1111, but can't for the life of me understand where to start at, if i want to implement it from a python script)

frank delta
buoyant jacinth
#

right next to the model, with the same name except .vae.pt at the end

#

if you want it used automatically

mild olive
#

How do I report users spamming DM's?

vast ingot
lilac reef
#

Does anyone know of a way to change the lighting of an image either from the prompt or in post? prompt would be best as I get a lot of warm and mostly cold lighting in mine, but I don't mind post either.

astral goblet
#

it's sort of stable diffusion since Stability owns CD now

lilac reef
#

I see that now. I remember this from a while back

#

my thing is I train a lot of images and I don't want to go through all 100+ of them to tag them with cold_lighting, warm_lighting, and what kind of lighting is this, for each one.

#

lighting colour affects the faces mostly

astral goblet
#

automation would help yeah. i seen some 2.1 negative embeds that may help

lilac reef
#

Naw, it gets baked into the training lora

#

can't be removed it has to be prevented at the source

#

can tag it away but that is hideous as I tried it yesterday and too much for me

#

lost all my hard work too when the extension decided to zap it on me. I ticketed that issue.

astral goblet
potent sand
#

Does anyone have information regarding the legal regulations pertaining to the use of stable diffusion? I am currently working on a project that involves the use of the Dreamlike Diffusion 1.0 model and I am unsure whether I can share it on YouTube or use it for commercial purposes. Could someone please guide me on this matter?

astral goblet
#

can't you preprocess the images before training?

lilac reef
#

Sure if I had something to do but 100+ images in PS is a long time

astral goblet
lilac reef
#

if all images were too cold, or too warm then easy peasy but you get glows, and too cold, or too warm and sometimes both in the same image

astral goblet
#

outside of that, the legalities fall under regular copyright usages. Fair use exists. be transformative and you're golden.

stable pawn
#

In Automatic1111 WebUi, is it possible to disable the (i) button? I have some LoRAs and saw that button, clicked on one that should make cute creatures (CuteChibi), and there was prompts and tags in there that I do not want my friends to see, like... I can not write any of that here.
I am not sure the LoRA creator made or used those tags, sometime that list can be rather long while othe LoRAs have a very short Info.

astral goblet
#

if you've got photoshop, there may be a fancy batch script that allows for dynamic color grading you're looking for

astral goblet
potent sand
stable pawn
#

Good morning Flo, trolling again this wonderful evening, feeling still the need to answer on everything, have an opinion that is not relevant.

lilac reef
#

tbh, I am not sure what I am looking for. I mean trying to get everything to have 5500k studio lighting is impossible.

pliant sequoia
#

Hey everyone, I’m just quick question. I am using local stable diffusion and I want to use the interpolation feature, but I cannot use it. Can you help me with this?

astral goblet
astral goblet
#

There's a lot of improvement to be had in this area

tiny mica
#

im new to ai

#

can someone explain me the difference between using () or {}

#

?

wheat hazel
#

who thinks auto1111 is an absolute piece of crap when it comes to code cleanliness ?

astral goblet
warm junco
astral goblet
#

its a very feature rich and cutting edge tool set that started as a MVP and has been wrangled together rapidly as new technologies come out. the field is moving far too fast to keep the code clean. It's the bleeding edge on that repo.

If you want a professional product that paid engineers to take their time to design UX features, you're probably wanting to subscribe to creative cloud

astral goblet
tiny mica
#

the speed at which the whole AI field has expanded these last few months is incredible

wheat hazel
#

I'm just trying to create an extension to process videos, and... there seem to be some really bad interactions between gradio video input and the extension system made in 1111. So bad that you simply can't put a css to the video with gr.Blocks(css=...) and consequently, if you put a video whose height is pretty huge, then the video control goes all the way instead of fitting in the video in itself...

#

maybe 2-3 hours I'm trying to fix that stupid crap, I'm so fed up

scenic thicket
#

where do i make art?

astral goblet
#

it probably has nothing to do with the code cleanliness or skill level of the people who work on automatic1111's repo

scenic thicket
astral goblet
#

deflecting blame for the problem onto nearby bystanders never fixes the problem, but it's something frustrated people will do consistently, for so many problems

lilac reef
tiny mica
wheat hazel
#

Maybe. But I've worked on 1111 code before, and ... how did they decide that 1000+ line code files and functions with 30+ parameters were a good idea ?

astral goblet
scenic thicket
warm junco
astral goblet
#

now now. y'all can make art all day. calm down. this server doesn't have a generation bot anymore. #1072229020520947753

#

already is publicly available through a few services 😉

buoyant jacinth
fervent thunder
#

I just do (word:1.1), (two words:1.2), (word, another word:0.5) instead, easier to read then lots of different () {}

shell mortar
#

Is there a bot where we can do prompts?

shell mortar
fervent thunder
#

waow make a room for this specific question I can throw people at

shell mortar
#

I guess it's just the midjourney thing

#

They have rooms for art prompts, so i thought we had it somewhere as well

fervent thunder
#

we had it too, but I don't find discord bots comfortable to use anyway

#

it's good to use someone else's gpu, I guess tho 😄

shell mortar
#

The best!

fervent thunder
#

Hello , how can i starts creating those AI made painting

fervent thunder
#

Yes thank you kind user

thorny sorrel
#

guys quick question, when I use control net does the image i put on the IMG2IMG part matter?

#

or is it just the prompt that affects the end result?

manic flint
#

/hello good afternoon someone please tell me how to create the images here, please

thorny sorrel
exotic latch
#

Free webinar on how to integrate your own information into generative AI systems. Maybe you're an author who wants to bring their book to life, maybe you're in a company and want to write using your internal information. If you want AI to write on your content exclusively, this webinar is for you.

Where: 🫴  https://lu.ma/rg32qsbx

cursive gate
#

Does anyone have any update on 3.0?

hallow plaza
#

Anything negative happen if you use different resolutions on dreambooth?

past edge
#

What is the best tool currently for outpainting? I havent caught up

#

Is it still InvokeAI?

dry trellis
#

the best one I've used is a Photoshop plugin but its not free

fervent thunder
vast ingot
#

invoke stays top tier though

fervent thunder
#

Painthua is more comfortable to work with imo, but invoke is fast

#

probably cause it uses DDIM nomatter what you chose...

past edge
#

Ty. I remember using painthua but I liked InvokeAI. Time to do a git pull on that bih

fervent thunder
#

I wish invoke had same button just to remove image I don't want from generated image choice, so I can choice between good images only back to back
just this one thing makes painthua so much better

#

Did invoke add DPM++ samplers btw?
It's been some time since I used it

past edge
#

I'll lyk once I get it to work

rose olive
#

Hey 👋 guys, Is anybody successfully feeding / training their own AI data set on the cloud for their own version of chat gpt, personalised with their data set? Like their own AI assistant

old imp
#

.

rose olive
iron flare
#

excuse me, how do i use my seed to recreate the same photo

#

i dont seem be able to enter the seed anywhere

fervent thunder
iron flare
fervent thunder
iron flare
#

Thank you. Do I need to install anything else to use this?

fervent thunder
hallow plaza
#

Anything negative happen if you use different resolutions on dreambooth?

past edge
#

u gotta do one rez at a time

#

preferably stick to the rez of the model

hallow plaza
#

@past edge bummer. Makes the process a lot of work. Thanks for the feedback

whole garden
#

just curious, what is the general consensus as"too hot" for laptop gpu's?

#

70-80 c?

past edge
#

anything above 82c I would say

#

for prolonged periods of time hell naw

young ivy
tough isle
#

Hey guys, please does anyone know if there is a Jax implementation of SD v2.1? I can only see one for v1.5

iron flare
night cobalt
#

!infractions

vestal dew
#

Watching this tutorial for Controlnet and Blender. https://www.youtube.com/watch?v=ptEZQrKgHAg

How in Blender do I get the hands and feet to show as in this video? I see bones, so it's hard to tell the orientation of the hand.

#

It's yet another simply do video. It's very simple....as long as you skip all the steps

vestal dew
#

cant get any of it to work. When I export, I only get the hands and feet for the depth map. In the video, he gets the whole body

leaden mural
#

guys

vestal dew
#

Ok, now Im trying openpose...no good luck there either

rare geyser
#

I just released a prototype of my game that uses stable diffusion as a key element. I could use some testers..... free beer! https://aigameboy.com

vestal dew
#

the game does nothing

#

the game, she does nothing

devout vine
#

what is this communities opinion about ai in general

vestal dew
#

it's terrific

crisp hatch
vestal dew
#

in other words, it changes nothing

forest trout
#

Hey if you had to prune your model libraries down to 5 models, which ones would you keep? What's your top 5 models?

vestal dew
#

wow, this controlnet openpose is garbage

dawn sparrow
#

Hey guys is there a place where I can advertise a freelance project for prompt wizards / advanced users ?

vast ingot
#

and I see you've been here a small bit over the time but not a lot, so let me extend the welcomes from here 🙂

sand hazel
#

Hey @vast ingot I'm looking for something similar

vast ingot
sand hazel
#

👌

vast ingot
#

we get quite some of that recently, so having tickets for it also helps us see the need for dedicated channel for that maybe.
Tbh it's a complicated subject, but I think having a common place for it would be best for everyone

sand hazel
#

Sounds good, thanks

shut lantern
#

How can I install a model that isn’t on hugging face in google collab stable diffusion 1111?

vast ingot
# shut lantern How can I install a model that isn’t on hugging face in google collab stable dif...

you can usually connect your google drive to your google collab, and automatic installs in there. Run it at least once like that, and you'll have all the folders in your google drive
then you can just download the model at home, put it in the stable-diffusion-UI/models/Stable-Diffusion folder in your google drive, wait for it to synchronise
and run again your notebook, it should start a lot faster since all will aready be downloaded, and you should also see your new model, if it's in the right folder

#

outside of that method, you could also download the model at home, and just upload it to google colab once you have automatic running in there

#

but that would take ages off your free time

shut lantern
#

Why?

vast ingot
#

why what ^^

shut lantern
#

Why is take my free time

vast ingot
#

lots of things ^^

#

well, google colab gives you some free time to use it each day

#

and uploading a 2GB+ model from your home connection, that can take a while

shut lantern
#

Can’t you pay for more?

vast ingot
#

yep sure

shut lantern
#

Is it worth it

vast ingot
#

depends how much you want to do SD honestly... 2h free per day more or less is a lot imo, but I wouldn't be satisfied with it anymore personally

shut lantern
#

Also

#

If a model on citiv says no selling content with it

#

Can I still do it anyway

#

Without consequences

#

I am not selling the model I’m making pictures with it and selling nsfw content

vast ingot
#

if it has "no selling" on it, it's not about selling the model but the pictures, so you would be infringing on the copyright of the model you are using I guess, and they could technically sue you if they discovered it

#

I would need to read the conditions of civitAI to be sure, I haven't

#

but the no selling part is about pictures, from what I see when I publish a model myself

lavish current
#

I can't stop bunn-ing! 😭

#

This is too much bun

fervent thunder
#

well, unless you do it in collab directly

vast ingot
#

But i guess if you intend to sell, and pay for colab, the question is a little different

lament narwhal
oblique fable
#

If it isnt a very unique model with a super distinctive style

tranquil sleet
#

Hmmmm...
Where can I use the bot friends?
I can't find it.

fierce charm
#

What are NCNN and ONNX and how are they used in upscaling? I see extra folders in upscaling models for these files, but i'm not sure where you would use them.

elder cradle
#

ONNX is a way to accelerate machine learning

elder cradle
#

Both ONNX and NCNN are high-performance machine learning frameworks

fierce charm
#

for mobile phones? do people render AI images on their mobile phones?

elder cradle
#

I know ONNX, but I'm not sure about NCNN

tranquil sleet
#

Where can I use dreambot?

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!

elder cradle
#

Bot is down now

tranquil sleet
fierce charm
elder cradle
#

Maybe the inclusion of ncnn is an oversight

#

OH, NCNN actually has Windows support

fierce charm
#

Since i reinstalled my A1111 on a new computer (using a 4080) a lot of my images come out desaturated. Any ideas why?

vast ingot
fierce charm
gloomy night
#

Can you run SD using ths PC ram and not the graphics card VRAM?

proud nova
#

Colab has issues again

#

That "fastapi" thing

warm junco
gloomy night
warm junco
upbeat tangle
#

whats the best config to train loras on faces ?

#

how many repeats/total steps/epochs ?

gloomy night
warm junco
#

Fun fact: you can run Windows XP fully installed on GPU vram xD

vast ingot
#

🐮Fun fact: You can fit an entire herd of cows in GPU vram! Just imagine, no more crowded pastures or messy barns - simply store your cows in the digital realm. Plus, with virtual reality technology, you can even milk them without the mess! Just be sure to give your graphics card a break - all that mooing can be quite the strain.

buoyant jacinth
gloomy night
lavish current
#

I like bunny girls

#

🐇

sullen oasis
lofty gate
#

What was the trick/prompt symbol for "switch between these 2 prompts"

#

It was like [dog,cat] but I think that was blending

vast ingot
lofty gate
#

Thanks :)

vast ingot
#

you're welcome, it does a little more than you wanted though, I can't find the exact one you'd want

#

looking for it

vast ingot
#

the first I linked would try all combinations of the words

#

this one will just use each word as an alternative

#

damn not even

#

lol

#

this changes each step

lofty gate
#

Lol that's interesting. It's cool though thanks for the help!

tulip beacon
#

Hey guys im new to this and using automatic1111 stable diffusion and when i want to start the webui-user.bat then i get an error "Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check". Can anyone help me? 😦

warm junco
tulip beacon
#

its a amd vega 56

warm junco
#

And hi

tulip beacon
#

tried to delete the venv folder and reinstalling it all but doesnt help

warm junco
#

Do you use the directml version if webui ?

#

Thats for AMD

tulip beacon
#

directml version? whats that 😅

warm junco
lavish current
#

I think I'm gonna go back to using Chatbots mostly

warm junco
#

@tulip beacon
Here is a short help for AMD copied from @still glacier
AMD GPU users have the following options :

tulip beacon
#

oh yea u sent it i will try that thx ❤️

astral goblet
#

DirectML is what enables AMD cards to work on Windows for ML purposes. It's the less good cousin of RocM on Linux. If you've got an AMD card, even without ML considerations, use Linux. It's just infinitely better. Features like FSR 2 on every game was first on Linux

lavish current
#

Why are images considered inappropriate if there's no nudity in them

#

Using a filtered model none the less

astral goblet
#

sexualization is one of those things that "you know it when you see it" and if you start to define it, you end up with people doing ridiculous japanese censor bars from people trying to skirt the rules

#

mod discretion is best practice here

lavish current
#

That's highly subjective imo

astral goblet
#

yup

#

it sure is

lavish current
#

I get what you mean, but it kind of ruins it for the people not trying to do that

astral goblet
#

there are ai communities that are all about nsfw imagery. this isn't the place for that

lavish current
#

The model I'm using doesn't allow NSFW

vast ingot
#

the 2 ones I had a problem with, one of those had nudity and boobs on a young girl.
The other was not direct nudity like that but was more borderline like you were told by the other user, and I think they were right on this.
the rule 4 goes a little further than just nudity. sexual intent can also lead to that, or gore for example.
Mixing with younger subject, we are more proactive when getting close to the line, following a big increase in rightful reports, like the last few pins indicate there.

I am sorry I needed to stop some of those, I didn't want to just stop the party, but I need to act on those when they get reported, and they do, to be able to still have a party

astral goblet
#

I agree that some lines shouldn't even be tested or even approached

#

i have a little room for understanding here, since many users are 15-18 themselves, but it's important to set expectations here.

tulip beacon
#

What is the problem here? :

Installing requirements for CodeFormer
Installing requirements for Web UI
Launching Web UI with arguments:
Traceback (most recent call last):
File "C:\z\AI\stable-diffusion-webui-directml\launch.py", line 377, in <module>
start()
File "C:\z\AI\stable-diffusion-webui-directml\launch.py", line 368, in start
import webui
File "C:\z\AI\stable-diffusion-webui-directml\webui.py", line 15, in <module>
from modules import paths, timer, import_hook, errors
File "C:\z\AI\stable-diffusion-webui-directml\modules\paths.py", line 26, in <module>
assert sd_path is not None, "Couldn't find Stable Diffusion in any of: " + str(possible_sd_paths)
AssertionError: Couldn't find Stable Diffusion in any of: ['C:\z\AI\stable-diffusion-webui-directml\repositories/stable-diffusion-stability-ai', '.', 'C:\z\AI']

vast ingot
#

your venv seems broken. 3 steps :
1/ delete venv folder inside AUTOMATIC
2/ restart automatic
3/ ask in #🤝｜tech-support , they know quite a lot their stuff 😉

#

sorry I don't know that directml version, I may be wrong, I missread

tulip beacon
#

it seems like amd users need to use directml version no?

astral goblet
#

yup. or install it with ROCm on linux. Guiz likely is right. Venv folder shenanigans. delete c:\z\AI\stable-diffusion-webui-directlml\venv the entire folder. Run start bat again and it'll rebuild the venv fresh

tulip beacon
#

Okay i will try that

#

Can you recommend like models or checkpoint how they are called that are working good, with no restrictions to experiment? im new to this and rly dont know, i got the 1.4 original version from hugging face but idk if its good

astral goblet
# lavish current The model I'm using doesn't allow NSFW

That's often not the case. Anything based on 1.5 will allow NSFW even if the author says it's not capable of it. There's also the case of the nsfw filter which blacks out images that meet a threshold detection. That won't catch everything that can subjectively look nsfw though.

tulip beacon
#

Still got the same error... deleted venv

#

Couldn't find Stable Diffusion in any of:

astral goblet
tulip beacon
#

dont seem to be an active channel.. so i hoped to find help here lol

#

but i asked there

astral goblet
#

gl

tulip beacon
astral goblet
#

delete the venv

vast ingot
tulip beacon
tulip beacon
astral goblet
#

there are so many models. 1.5 is an upgrade over 1.4 imo.

vast ingot
astral goblet
#

Runway ML did well there. I wonder what the drama with the release was all about. Nothing really was revealed about that. Looked like damage control went on behind closed doors though

thorn hornet
#

Football edits with messi and ronaldo logo

tulip beacon
astral goblet
#

yup. All of the checkpoints on civit are for Stable diffusion

vast ingot
astral goblet
#

sometimes for their checkpoints, it'll tell you an additional configuration file is needed. a .yaml file that goes in the same folder

astral goblet
tulip beacon
astral goblet
tulip beacon
astral goblet
#

not only can people refine models with their own image sets, you can breed models too. like they're some kind of pokemon

astral goblet
tulip beacon
#

yea i didnt know that i can host an text2img ai on my own

astral goblet
#

hurray technology!

vast ingot
#

quite fun to do

#

and yeah, a model can be made a lots of ways

tulip beacon
#

wondering if there will be a way to host chat ai like gpt so there is no need for openai giving access to the service

astral goblet
#

yup!

vast ingot
#

base models, official ones, are trained "from scratch" using a giant dataset
most other models are refinments of those generic models in a given direction, like anime or photorealistic

astral goblet
#

stanford released alpaca! it can run on people's home pc's with 32gb of system ram

#

just the other day

vast ingot
#

there is already a way to run a GPT like at home, and fine tune it to your needs yeah

tulip beacon
#

oh for real? so it does the same job as chatgpt?

astral goblet
#

it gets close! it performs very well for being a much smaller model

vast ingot
#

it's like DallE and MJ are "the same", it does great text too, but will be differently tuned

tulip beacon
#

but alpaca is for image generation and not chat no?

tulip beacon
astral goblet
#

yeah thats right

tulip beacon
#

but alpaca is waitlist i guess..

astral goblet
#

afaik you can get the model today. people have trained it using the data in the github. or it leaked. i haven't looked too into it. i only have 32gb of memory

tulip beacon
#

in chatgpt i still got the problem that my history is missing.. hope they are fixing it fast

vast ingot
#

it's there for me right now, but it's missing 90% of the time for me too

astral goblet
#

i didn't think it was keeping my history at all then one day i used it and stuff from the very beginning was there. so it does keep it , even if its not accessible most times

tulip beacon
#

just wanna get stable diff work but i dont find any solution..

vast ingot
#

if there a hardware reason you are using this directml version ?

#

I'd love to help you but that is the part I don't know about, never used that one

tulip beacon
#

idk someone told me i need to use directml cause i have amd graphics

vast ingot
#

you are on AMD, ok

tulip beacon
#

so idk if i rly need to

#

cause the normal version didnt work too

vast ingot
#

the normal version won't work on AMD without adaptation yeah

#

SD came out on NVIDIA exclusively at first, and has been adapt, thus AMD lacking a little right now

#

so, you went with the directml version

#

downloaded it

#

you checked that you had both Git and Python 3.10.6 in the path ?

#

you can check by opening a cmd prompt window and typing : bash git python --version

tulip beacon
vast ingot
#

that seems to be wrong yeah, you should try to downgrade to 3.10.6 to be sure

#

I'm following this part

tulip beacon
#

so i remove python 3.10.10 and install 3.10.6?

vast ingot
#

yep

#

being the above version, in machine learning, isn't always best

#

it needs all dependancies to be working on the other version too, and can break things

#

like 3.11 for sure breaks things here

tulip beacon
#

okay im installing it rn

vast ingot
#

you will need to delete the "venv" folder again too

tulip beacon
#

ye doing it too

vast ingot
#

since python version is duplicated inside it

#

also, one thing that breaks : don't run the webui-user.bat as admin

#

brb

tulip beacon
#

oh okay good to know

#

okay will try it and come back 🙂

still glacier
#

did you ping me ?

vast ingot
#

😉

#

MVP

still glacier
#

Well I guess you handled things :p

#

Sorry things are kinda hectic at the moment on my side. I don't do that much SD customer service at the moment :p

tulip beacon
tulip beacon
still glacier
#

what's the error message ?

tulip beacon
# still glacier what's the error message ?

Installing requirements for CodeFormer
Installing requirements for Web UI
Launching Web UI with arguments:
Traceback (most recent call last):
File "C:\z\AI\stable-diffusion-webui-directml\launch.py", line 377, in <module>
start()
File "C:\z\AI\stable-diffusion-webui-directml\launch.py", line 368, in start
import webui
File "C:\z\AI\stable-diffusion-webui-directml\webui.py", line 15, in <module>
from modules import paths, timer, import_hook, errors
File "C:\z\AI\stable-diffusion-webui-directml\modules\paths.py", line 26, in <module>
assert sd_path is not None, "Couldn't find Stable Diffusion in any of: " + str(possible_sd_paths)
AssertionError: Couldn't find Stable Diffusion in any of: ['C:\z\AI\stable-diffusion-webui-directml\repositories/stable-diffusion-stability-ai', '.', 'C:\z\AI']

#

using directml cause i got amd hardware

still glacier
#

you might want to git clone the repository again in a fresh folder.

tulip beacon
astral goblet
# vast ingot like 3.11 for sure breaks things here

3.11 will break things yeah, but 3.10.6 also really broke things for me. I havne't investigated further on this matter, but i think it had to do with the e-cores that my alderlake intel has. 3.10.9 got rid of the NAN errors that would randomly appear. nothing else other than updating 3.10.6 to a newer 3.10 version fixed this

#

i've not found any instances where 3.10.9 or 10.10 break anything

tulip beacon
#

so .11 is bad and .10 is good?

still glacier
#

3.10.9 is generally safe, we've seen instances of install getting borked with 3.10.10 however

tulip beacon
#

installed .6 but doesnt change anything in my error then maybe i can install a newer version

astral goblet
#

yeah. they update both python versions too. neat development crew over there. it's been going for quite some time. i've never bothered to learn it

astral goblet
still glacier
tulip beacon
#

$ cd C:\z\Stable Diffusion
bash: cd: too many arguments

what is the problem now?

#

in the git bash

still glacier
astral goblet
still glacier
#

yup

#

but people are lazy and "but mah youtuber told me to do so !"

#

they don't know the risk.

astral goblet
#

yeah its a very lazy troubleshoot. if it's eveyr done, it should only be done to see if it does make it work, then turn it off right away before it fudges up folder perms or allows some remote code to execute

#

I think it comes out of cracked games culture a lot. People download a crack. nfo file says "run it as an admin! The virus scan is a false positive. TRUST ME!" and then it works so they learn that it works

#

and the cycle continues

still glacier
#

That's certainly part of it.

#

@tulip beacon so how is it going ?

tulip beacon
#

4gb takes short time

rotund blade
#

Hey all. Sorry I'm really out of the loop, what's the state of the art with respect to making reasonable hands? Has anyone trained a model that inpaints fixed hands? That would be awesome. My workflow for generating art works well except that 90% of the time is fixing hands!

tulip beacon
#

whats fixing hands?

rotund blade
astral goblet
#

you'll learn soon enough that SD is really bad at generating accurate hands

tulip beacon
astral goblet
#

there are a lot of good quality memes about this problem

rotund blade
still glacier
astral goblet
rotund blade
#

My current approach is to manually paint in the general shape I want and then iteratively allow stable diffusion inpainting to create detail and lighting match for the scene. Except it takes many iterations and is really slow and painful.

astral goblet
tulip beacon
#

A new release of pip available: 22.3.1 -> 23.0.1

is that important? @still glacier

still glacier
rotund blade
astral goblet
#

One limitation i've found with that, is that a lot of the depth detail is lost when you scale the image. they could do with better interpolation / scaling code here

#

it does help a lot though

tulip beacon
#

still the same error... @still glacier

rotund blade
#

One of my problems is that I don't want to/can't afford to spend dozens/hundreds of hours constantly staying up to date on new features and their infinite myriad of ways of being used. I have a bespoke script-based workflow and it's really hard to fit new stuff in since these things tend to get tightly integrated with specific web based UIs.

still glacier
#

there were no errors during the git clone ?

still glacier
#

screenshots are ok

tulip beacon
#

Creating venv in directory C:\z\StableDiffusion\stable-diffusion-webui-directml\venv using python "C:\Users\thoma\AppData\Local\Programs\Python\Python310\python.exe"
venv "C:\z\StableDiffusion\stable-diffusion-webui-directml\venv\Scripts\Python.exe"
Python 3.10.9 (tags/v3.10.9:1dd9be6, Dec 6 2022, 20:01:21) [MSC v.1934 64 bit (AMD64)]
Commit hash: f64ad25386ad77d1e07986a204dac923b9a46894
Installing torch and torchvision
Collecting torch==1.13.1
Using cached torch-1.13.1-cp310-cp310-win_amd64.whl (162.6 MB)
Collecting torchvision==0.14.1
Using cached torchvision-0.14.1-cp310-cp310-win_amd64.whl (1.1 MB)
Collecting torch-directml
Using cached torch_directml-0.1.13.1.dev230301-cp310-cp310-win_amd64.whl (7.4 MB)
Collecting typing-extensions
Using cached typing_extensions-4.5.0-py3-none-any.whl (27 kB)
Collecting pillow!=8.3.*,>=5.3.0
Using cached Pillow-9.4.0-cp310-cp310-win_amd64.whl (2.5 MB)
Collecting requests
Using cached requests-2.28.2-py3-none-any.whl (62 kB)
Collecting numpy
Using cached numpy-1.24.2-cp310-cp310-win_amd64.whl (14.8 MB)
Collecting idna<4,>=2.5
Using cached idna-3.4-py3-none-any.whl (61 kB)
Collecting urllib3<1.27,>=1.21.1
Using cached urllib3-1.26.15-py2.py3-none-any.whl (140 kB)
Collecting certifi>=2017.4.17
Using cached certifi-2022.12.7-py3-none-any.whl (155 kB)
Collecting charset-normalizer<4,>=2
Using cached charset_normalizer-3.1.0-cp310-cp310-win_amd64.whl (97 kB)
Installing collected packages: urllib3, typing-extensions, pillow, numpy, idna, charset-normalizer, certifi, torch, requests, torchvision, torch-directml
Successfully installed certifi-2022.12.7 charset-normalizer-3.1.0 idna-3.4 numpy-1.24.2 pillow-9.4.0 requests-2.28.2 torch-1.13.1 torch-directml-0.1.13.1.dev230301 torchvision-0.14.1 typing-extensions-4.5.0 urllib3-1.26.15

#

[notice] A new release of pip available: 22.3.1 -> 23.0.1
[notice] To update, run: C:\z\StableDiffusion\stable-diffusion-webui-directml\venv\Scripts\python.exe -m pip install --upgrade pip
Installing gfpgan
Installing clip
Installing open_clip
Cloning Taming Transformers into C:\z\StableDiffusion\stable-diffusion-webui-directml\repositories\taming-transformers...
Cloning CodeFormer into C:\z\StableDiffusion\stable-diffusion-webui-directml\repositories\CodeFormer...
Cloning BLIP into C:\z\StableDiffusion\stable-diffusion-webui-directml\repositories\BLIP...
Installing requirements for CodeFormer
Installing requirements for Web UI
Launching Web UI with arguments:
Traceback (most recent call last):
File "C:\z\StableDiffusion\stable-diffusion-webui-directml\launch.py", line 377, in <module>
start()
File "C:\z\StableDiffusion\stable-diffusion-webui-directml\launch.py", line 368, in start
import webui
File "C:\z\StableDiffusion\stable-diffusion-webui-directml\webui.py", line 15, in <module>
from modules import paths, timer, import_hook, errors
File "C:\z\StableDiffusion\stable-diffusion-webui-directml\modules\paths.py", line 26, in <module>
assert sd_path is not None, "Couldn't find Stable Diffusion in any of: " + str(possible_sd_paths)
AssertionError: Couldn't find Stable Diffusion in any of: ['C:\z\StableDiffusion\stable-diffusion-webui-directml\repositories/stable-diffusion-stability-ai', '.', 'C:\z\StableDiffusion']
Drücken Sie eine beliebige Taste . . .

astral goblet
# rotund blade One of my problems is that I don't want to/can't afford to spend dozens/hundreds...

yup. i fully understand the need for a consistent workflow. i'm still redeveloping mine to a point where i can start "working" uninterupted. I keep getting hit with new tech and i'm like "ooooo"

but we hit limitations on our workflows, like you've run into with hands, and we redevelop them and integrate new systems. i would fully recommend you investigate controlnet in your workflow. it's such a powerful tool that it will likely be as default as something like negative prompts. Novel additions that are basically default SD now.

tulip beacon
#

@still glacier just dont know what it means that it cant find stable diffusion

astral goblet
#

professional animation studios will use long outdated software because thats where their workflow is established and thats what works

#

no rockets as big as the Saturn V have launched since the apollo program. it's because all those old workflows pre computer aided design, were lost and the new tools were not re established until lately. Now we got Starship up on the same pad that Saturn V sat on. Nice to see

#

there is so much value in having a consistent workflow. i could draw 10k more examples

rotund blade
#

My specific use case is to generate characters for a game, where each character has 9 variations representing increased "level" of the character. The specific challenge is on getting 9 versions of the same character that look like they've age-progressed.

#

As a result I don't focus as much on having the most "beautiful" images possible since it's so much effort just to get age-progression that infinitely tweaking style just isn't something I have time for.

#

I have a process that works, it generates ~1,500 images overnight for me to sift through, and they're usually good enough that it only takes 30 minutes or so to sequence a few characters. But then -- hand fixing ends up taking another 2 - 3 hours per character. It's by far the most time consuming part of the workflow.

#

For example:
#🏞｜general-with-images message
Almost all the hands in those images were gnarled lumps that had to be fixed "manually". I'm not an artist, I'm a software developer, and so I don't ever draw final pixels, I use a combination of rough manual sketching followed by inpainting. Although I'm actually starting to get reasonably good at just drawing hands outright I've done it so much now.

hallow plaza
#

in a1111 Is there an easy way to find all your embeddings? I've installed a few but didn't write down all the key words.

warm junco
hallow plaza
#

But what about the trigger words?

warm junco
rustic seal
#

https://glaze.cs.uchicago.edu/
So anyone seen this? It does something to the images that doesn't just make it difficult to replicate but poisons the models weighting system causing it to corrupt the rest of the model.

astral goblet
#

with lora's, theres something i'm annoyed by. I often check if an embedding is working by looking at the generation details and seeing that the embedding was used. Is there a way to enable this for loras? Also, when using loras, A1's ui puts the lora file name into the prompt, but do you also have to use the trigger word?

#

like lora:name:1 will often work for many of them i download, but if i train one its like i need to use a key token to make it work. it'd be nice to see in the generation details if it was being loaded at all

shy parcel
#

I've got access to Google's Bard. It's amazing ..... how bad it is

verbal vale
#

I have been focusing the past month on fantasy character portraits, but now that I have switched to landscapes of high-fantasy worlds, I am having difficulty with finding a good model/checkpoint or Lora Style. Not sure if anyone has any suggestions, here.

#

Ideally something where I wouldn't have to make very bloated prompts in order to achieve a somewhat consistent result.

#

bard was a rush/panic job.

cursive axle
#

Bard would have been cool a year ago, but OpenAI put the bar pretty high

strange aurora
#

So I was wondering how viable combining stable diffusion and 3d animation software to make 2d animations would be:

Taking something like Cascadeur (I think that’s the name, the 3d animation software that essentially uses AI to make a lot of the finer details of animation) and essentially make a 3d animation with simple mannequin models that are roughly the body shape of the 2d character you want, setting up the camera angles and color coding each model. Then using diffusion to essentially go through each frame and creat the 2d animation, using the color coding to assign different “models” that would be trained on those specific characters. Making it so that you made the animation in 3d with only guidelines to convey different character and set pieces, then diffusion put down the actual world, almost tracing over the models.

astral goblet
# shy parcel I've got access to Google's Bard. It's amazing ..... how bad it is

i dont even think it's that bad. it has a few obvious fail cases and theres lots of stuff for the antiwoke people to hate and meme about, as they usually do, but "how bad it is" seems a bit hyperbolic. I'll likely use it for a lot of different tasks when it comes out. As well as the other branded systems too. Brand loyalty is just stupid. Use the tools that work for the task instead.

Same way i use windows sometimes, linux other times, chrome, firefox, edge, i alternate. blind loyalty to one brand is just a bad habit

shy parcel
astral goblet
astral goblet
#

i mean, the antiwoke mafia are loud and proud about how angry they are with these new systems. I only brought it up to recognize that major criticism that goes on, not to accuse you of it

shy parcel
#

you're either a troll... or you're piping your responses through bard. either way you fail the turing test.

astral goblet
#

it'd be like not addressing the elephant in the room

lavish current
#

I wonder if I can img2img those pics that were wrongfully removed

astral goblet
#

nope. it's just weird that you're suddenly this enraged about the antiwoke aspect being mentioned at all

astral goblet
lavish current
#

That's why I want to img2img them

#

I'll make clean versions and an actual NSFW version, lol

#

Everyone wins 🥳

astral goblet
#

it's probably just smarter to not test those rules instead. they described good reasoning for why it was removed and i think that you should go read those again and respect that decision, instead of casual accusations of wrongful action

fervent thunder
#

thinking That's too much effort just to throw your image into discord channel where it'll be forgoten in a 5 mins lol

lavish current
#

ikr

astral goblet
#

it's not about posting at this point. it's about getting past the filter somehow. it's a puzzle. i get it. it's just... don't

fervent thunder
#

filter is pretty strict btw, it seems to think it found boobies when dress is too white color or something, but eh, whatever, I don't really care if some images are ignored

lavish current
#

That happens to me a lot too

astral goblet
#

discord's own filter is very tenacious

#

they could do to improve it

fervent thunder
#

With AI doing better and better I'm sure they will eventually

lavish current
#

And then they'll be filtered by... Humans! 🤪

astral goblet
#

humans are all out of jobs. get real

#

we've exited the information age and are entering the cat lifestyle age

boreal dove
#

heyyy

past edge
#

cat lifestyle age lmfao

#

that would be true if the people running this world were anything like cat owners but nope.

#

they will rather let us starve than give us handouts even if robots took all of our jerbs

sweet turret
#

how to use bot stable diffusion?

vestal dew
past edge
#

profits from substituting humans with machines?

haughty turtle
#

ok

wispy jasper
#

Man MJ5 is spectacular, they just seem to take everything released for SD and make it better lol.... No controlnet though, that's a big downside right now.

surreal perchBOT
#

@sweet turret

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!

past edge
near silo
#

MJ V5 is fine, but it's not as good as people hype it up to be

#

What do no for sure is that adobe fire fly is trash currently lol

#

Though I can't say if I dislike Adobe or mid journey more

wispy jasper
#

On the contrary the results Im seeing in the MJ5 beta feed are out of this world.

near silo
#

I've been look as well and meh, I don't see it

#

Like I see it, but it's not that impressive IMO. And most of the results can be easily mimicked or surpassed by stable diffusion

past edge
#

someone who has a good eye like me

#

yeah me and this dude got into a flame war yesterday over which model is best

#

As much as I loathe MJ, that model is incomparable

#

the amount of fine detail is insane

near silo
#

It really wasn't a flame war, and It wasn't an argument either. You sent a very lackluster generation from MJv5, I along with the rest of the server all said that it wasn't impressive, then you sent another one to which I replied immediately after with a higher quality generation of the same prompt in stable diffusion

vestal dew
#

AI image generation is dying. Civitai is receiving DMCA notices over images generated with LORA's.

near silo
#

"dying" is a bit dramatic lol

vestal dew
#

Im trying to figure out controlnet to pose hands, get better hands and feet. Any tips on that? I tried some blender thing, but despite following the tutorial couldn't get it to work

wispy jasper
#

I'm an SD Master, and it would take me a TON of time to get the type of output MJ5 produces.. alas it's always been this way, it's a game of catch-up. I still think SD is far more powerful over-all, but for quick beautiful renders nothing like MJ.

vestal dew
#

If all the LORAs and models have to be taken down, can't be distributed, that's very bad

near silo
vestal dew
near silo
#

It will be fine lol

vestal dew
#

oh master

wispy jasper
#

There's a lot of info on this if you search Youtube, there are already control-net hand models. Most of the time you just need a template. Lots of Auto1111 tools for it.

vestal dew
#

openpose, for example, can do only basic poses. walking, runniong? not even that

near silo
vestal dew
#

I have searched youtube. I tried the blender..,.cant make it work to even generate the depth map image

wispy jasper
#

You're not doing it right then.

vestal dew
#

I followed it step by step. The simplest par

#

simplest part doesn't come out right...

#

the open pose in a1111 111...that image is so simple. It can't do much re poses

near silo
#

It's really such a shame that midjourney doesn't benefit anybody

#

It is really good tech, but for the benefit of nobody

#

I used to support them until I realized they are as malicious as they sre

#

*are

fervent thunder
#

What is a LORA?

near silo
#

You slap it on top of model and it teaches it how to do a new style/character

fervent thunder
#

Oh wow, thats incredible

#

thank youy

near silo
#

Yeah, I train LoRAs

fervent thunder
#

how do I apply them? I downloaded one and put it in models/LORA, I am guessing the "trigger word" does soemthing?

near silo
#

Trigger word is what you have to put in the prompt for the LoRA to activate

fervent thunder
#

i see

near silo
ocean sphinx
#

Hey guys , is there something like rundiffusion which offers automatic1111 and invokeai but allows me to upload my custom model ?

fervent thunder
#

is there a place where I can see prompts others have used and their corresponding images?

astral goblet
#

all over civit

fervent thunder
#

right thanks

astral goblet
#

i dont like a lot about civit but its really a good resource if you don't mind all the smut there

fervent thunder
#

huh alright

#

@delicate oxide

delicate oxide
#

whats up

fervent thunder
#

???

#

@delicate oxide

past edge
#

yoo no way

karmic brook
past edge
#

any plans for a 1024x1024 model pretty please

untold chasm
#

Please share if you enjoyed 🙂

#

also let me know what you think xD

compact nova
#

anyone here who could maybe help me out with deforum?

sand hazel
#

hey guys I'm trying to install URPM, do i put vae-ft-mse file in the "VAE" folder?

astral goblet
#

i don't like to see the inherent value of a word watered down

short tapir
#

So I need some help with a prompt. I'm trying to make a prompt where there's one mail and one female in the image. But no matter what I do it only puts 1 or the other. What do I do?

astral goblet
#

unfortunately, it's a natural language algorithm, not a technical implementation. it's easy to confuse the focus of an image by giving it too many subjects, like, more than one. Something i've done is instead of "man and woman" i describe them as a set like "couple portrait" "marriage photo", so that they're one subject. And since SD models have an inherent bias in them due to cultural forces, you don't have to stipulate man & woman

#

more to the point, i find one subject prompts are easier to work with

#

i saw some some research being done on programming languages to create models, while it's work being done elsewhere, not on diffusion models, i think it could at some point apply and i can see a vision of function calls being available to generations

#

it's stuff like that which i think could benefit from a programming language being inside the base model i think. extensions like that could just be something native to the model and so many other possibilities

wispy jasper
short tapir
astral goblet
#

i don't get the mj5 hype and you saying you can't get that, tells me you've peaked then. if thats mastery, you've capped your level. maybe i treat the word too literally and no one else does. i get that often

astral goblet
# short tapir So if I'm using the couples portrait thing as a prompt for example would I not u...

the "standard" 1boy 1girl thing is a booru tag thing and only works on the knowledge that was trained using that language. typically anime. Not to mention, the people who did that training really did not put nearly as much work into the 1boy side of the knowledge base. It's "mostly" a natural language model, and the reason keywords like that work is because they are heavily refined into the anime training

#

anime fans have been tagging images for decades and they're called "booru" tags.. its a .. its a dark space of research

short tapir
warm junco
#

Try a man and a woman ....

#

Or 1boy and 1girl holding Hands... Should work with anime models

astral goblet
#

holding_hands is the booru tag there too

#

oh, thinking about it too, when i was showing off stuff for my brother and his daughter, she wanted to see a tea party and that always had two subjects.

glacial spire
#

This seems fun, what's this server about?

fervent thunder
#

Stable Diffusion...mostly?

near silo
# short tapir So I need some help with a prompt. I'm trying to make a prompt where there's one...

You can utilize different extensions in order to achieve stuff like this. Something that you can use is zone focusing, or composition split.

What it does is it allows you to add multiple splits and different sections onto one image, while also allowing for multiple prompts in each one

So basically you could have one long composition and have it be the described background, then on the left side you can have it be a person, maybe a woman. And then on the right side You could have it be another person, like a man

#

So then it will make the background, the man, and the woman all together

short tapir
#

That's .. all gibberish to me

old bough
#

#imagine

fervent thunder
#

well Id never heard of that before and it sounds pretty well explained

whole garden
#

has anyone tried a Tesla M40 gpu? I cannot find many people talking about it for stable diffusion

near silo
grand path
#

/subscribe

buoyant jacinth
#

the m40 has 24gb ram wichis nice, but is fairly slow overall compared to a more modern gpu.
from reddit:

I have a Tesla M40, and a 3080Ti. Using SD to produce a 512x512 image @ 100 steps, the M40 finishes in just under a minute. The 3080ti takes roughly 12 seconds.

wispy totem
#

Why am I getting low quality renders with a custom uploaded mask? :/

My mask is properly black and white.

past edge
#

bru just get a 3090 lel. best bang for the buck

#

who gives a fuark about tensor management. its all about that sweet sweet vram

young ivy
#

hmmm, where do I put pylint commands and lines to have it working and imported or is pip install wrapt typing-extensions tomlkit tomli platformdirs mccabe lazy-object-proxy isort dill astroid pylinto enough to get --upcastattn ?

fervent thunder
#

I saw some posts about a "text2video"

#

anyone know how I could use that?