💬｜general-chat

[past robin]

I hope 1.0 will be better at photorealism. I've tested quite some negative prompting, but struggle to get good photo's. Looks like the model learns to do it all, from anime tot cgi to screenshots of games and photos, but fails a bit when you just ask for 1 simple thing.

That's the great thing about open source though. They made it easy for geeks to finetune this one. Curious to see what we get in comings weeks.

[broken smelt]

what did you do for optimization besides the usual --opt- args in the webui-user.bat?

[distant swift]

cuDNN, specific pytorch and python versions

[broken smelt]

did you try out multiple versions to see which ones worked best or use a suggested "best" version?

[distant swift]

yes, this is the best setup if you have a 4000 series card: pytorch2.0.1+cu118, latest cuDNN, python 3.10.9/3.10.10, (if you like xformers) latest xformers

[strong sparrow]

What if they ask me if it's A or B and the answer is yes? Some of these generations are really amazing.

That happened to me during the beta, only then none of the devs knew anything about it. They said it might have been an automated suspension.

I still to this day have no idea why I was suspended. They thought maybe I had a generation that was flagged later that I didn't report. But not sure.

[silver iris]

its just too much people, if entire governments cant handle this "too much people" problem , discord servers and mods are even less effective, relax , dont invest emotions on such a minor incovenience

[astral goblet]

government regulation is the only mechanism that can stop capitalism from steam rolling the populus

governments need to act fast because multinational corps sure are

[strong sparrow]

For me, it was during the period where it didn't auto filter results. So I may have had some generations that were a bit too risque. But yeah, I was using SFW prompts too. I think my suspension was 48 hours though. Or maybe it just felt that way...

[astral goblet]

3 sides to every story

[karmic brook]

You were muted for prompting little girl in a bikini.
Please don't single out staff and be aggro towards them.

[silver iris]

the blame i've put was on "too much people", sorry if i made you think i was blaming you directly

[tough shell]

Will the Lora with created with the 0.9 model be compatible with the 1.0 version, given that the model comes out tomorrow, I guess the question is solved?

[astral goblet]

totally sfw

[silver iris]

it depends, i would not be at my work seein "little girls in a bikini" , so not really safe for work hahaha

[strong sparrow]

You'll probably want to use swimsuit or swim dress.

[astral goblet]

rules shouldn't explicitely define what consitutes child sexualization because then pervos come along and walk those rules right on the line. in this case it looks like mod discretion was a huge success and they succesfully disciplined a user for trying to "walk the line".

f around and find out.

[broken smelt]

little girl in a bikini is not NSFW?? LMAO

[karmic brook]

Yes, we filter through especially when it comes to specified "little girl" or "young girl" in bikini, because typically this phrase is often abused to yield specific results, and is deemed as not very sfw friendly. There are filters in place for this and we are alerted of these phrases. Our server has rules for this and is a safety net for pedophilic content, thus we crack down on it.
Picking at our devs who are NOT mods to reply to your dm is not the way to go, and I apologize for any confusion, but the above is likely why you were flagged.

[astral goblet]

it's about a bigger picture than just you

[upper coral]

Why are users having to use qr codes without any other means of authentification offered?

[strong sparrow]

Like they're trying to say, it's not just against you. It's hard to moderate so many people, and so they need a deterrent. People getting suspended is that deterrent so that people are more hesitant to do the same thing.

[waxen geyser]

can someone tell me what aesthetic score exactly is?

[broken smelt]

does the automod not inform a user of why administrative actions are taken against them?

[topaz breach]

where can I test my own lora online?

[broken smelt]

if not, might need to do so otherwise disciplined users dont improve and are simply confused

[astral goblet]

an AI that classifies a score of an image according to it's training. many of them are trained towards a "chad" kind of aesthetic. many of them are trained on laion's aesthetic subsets.

[distant swift]

wait, SDXL is nsfw filtered?

[broken smelt]

that was specifically about the bot in #1100170312106127410

[astral goblet]

i don't use them because so often the popular recommended ones look for chads and donnas

[karmic brook]

Filters for words used on our moderation end

[distant swift]

ah, I see. so most of this server won't be upset

[astral goblet]

the bots are. obviously. they're using discord as a platform afterall. that's even before any ethical considerations

i contend only a small vocal minority are upset about censoring. they seem like a bigger crowd than they actually are. less than 1%

[waxen geyser]

can someone help me with the install of sdxl, i have the base and the refiner model, is there needed anything more? the vae for sure, but there are some other high GB data.

[sudden ruin]

Thats all you need

[strong sparrow]

Some of the A and B options are so close now, I don't know how to answer.

[sudden ruin]

You dont have to vote if you dont want to, its just to improve the outputs down the road. If youre happy with both thats fine too

[strong sparrow]

Which is a good thing, I guess. Means it is generally looking very nice.

[upper coral]

Anyone else being requested to use an authentification application now? I can't use it, so I can't use Dreamstudio now.

[waxen geyser]

mhh im not really able to generate something with a1111 , my gpu isnt the best one but in comfy it works perfect

1 image takes like 40-50minutes in a1111

[sudden ruin]

What GPU?

[waxen geyser]

gtx 1070, in comfy it works fine

[sudden ruin]

Yeah comfy is a smoother experience for me as well with SDXL

[waxen geyser]

Downloading VAEApprox model to: D:\stable-diffusion-webui\models\VAE-approx\vaeapprox-sdxl.pt
100%|███████████████████████████████████████████████████████████████████████████████| 209k/209k [00:00<00:00, 9.49MB/s]
17%|█████████████▊ | 5/30 [03:49<18:00, 43.23s/it]
Total progress: 17%|███████████▏ | 5/30 [02:56<15:30, 37.24s/it]

[strong sparrow]

Yeah... when you can commission an artist to make you something faster, that's not optimal.

[sudden ruin]

1070 isnt a very capable GPU these days tbf

[waxen geyser]

i could do anything with it on 1.5 so its fine, i cant upgrade bec. i would need a new monitor aswell, mine dont have a displayport

[sudden ruin]

XL is more demanding

[waxen geyser]

sure i know, but on comfy it worked out really nice with the 1070 😄

might get a 3060 in the future, or the 4060 ti

but for gaming my 1070 can play anything i need so its kinda useless upgrade in that case 😄

[sudden ruin]

Yeah for 1080p its probably fine

[waxen geyser]

using a benq 144hz monitor, and dont wanna upgrade its fine, and i hate ips panels 😄

wanna stay with my tn

[sudden ruin]

I upgraded to the odyssey g9 oled recently and having a blast

[waxen geyser]

i just dont like ips bec of the reaction times

had 2 monitors here, and they just perform kinda same with my tn when they had terrible ghosting

like i had to activate all that overdrive stuff to get a similar result to my tn

[sudden ruin]

I switched from 2 old 1080p 60hz screens

Its a whole different world

[waxen geyser]

oh i bet 😄

[sudden ruin]

together with my new 4090 liquid x i am very much gaming

[waxen geyser]

i just play csgo, valorant and some retro emulator stuff, so it would be just overkill especially on 1080p 😄

i still hope they bring out some fixes, comfy works great for me, but auto just really takes hours

[sudden ruin]

There already is a hotfix for auto in the works 1.5.1 from what ive seen

It will only improve

[waxen geyser]

yeah i heard, but no idea if it improves the overall performance

[sudden ruin]

We shall see

[strong sparrow]

True or false. More art has been made in the last year than in the entirety of human history before then.

[sudden ruin]

Depends how you define art

Most of the stuff AI spits out I wouldnt consider that

[strong sparrow]

Sorry, you must answer true or false. 😉

[sudden ruin]

tralse

[strong sparrow]

Actually you didn't answer true or false, so clearly my last statement is false.

Good point though. I guess I should say anything resembling art. But you can count in bad art people make too.

[sudden ruin]

Updated my A1111 and slightly overclocked my GPU, 1.0 can come

With torch, xformers and all the fun stuff that spits errors at you

[strong sparrow]

You're allowing the update now? lol. Alright, guys. Boto is ready for SDXL 1.0

[warm junco]

prepare for cuda launch blocking 😛

[sudden ruin]

Noooo

Took me like over an hour cause I tried to also update python to the newest version, but it refused to work

[warm junco]

python 3.10.11 ?

normaly its just running the installer and reboot. sometimes deleting the venv helps too

[worn trail]

with my new desktop Ill finally be able to run SD at a good speed

[sudden ruin]

I tried 3.11.4 and then the webui refused to launch whatever I tried

[warm junco]

SD wont work with 3.11 xD

not supported

[sudden ruin]

Yeah I learned it the hard way

version: v1.5.0  •  python: 3.10.9  •  torch: 2.0.1+cu118  •  xformers: 0.0.20
Went back to 3.10.9, since that worked for me before

[desert onyx]

He i use a111 and can i use multiple lora models in 1 image or not

Just got it so still new

[sudden ruin]

Yes you can

[waxen geyser]

yes u can

[desert onyx]

Oh ok nice

[waxen geyser]

i really hope a1111 gets close to comfy performance, would be really nice, 25mins vs like 2-3 mins is huge xD

[sudden ruin]

Comfy is just way more lightweighted, hard for A1111 to go that path now

[desert onyx]

If you talk about generation apeed mine generate in a couple of seconds

[waxen geyser]

but u are probably not using the sdxl model

[waxen geyser]

i dont think that has something to do with the overall generation speed

[sudden ruin]

Not sure, im not super deep into that

[wild steppe]

Join us for the Stage tomorrow 👀

https://discord.gg/RJspPRUm?event=1130919343941758978

[waxen storm]

Is the 1.0 public model available tomorrow?

[fervent thunder]

definitely

[waxen storm]

are you definitely being sarcastic? 🙂

[fervent thunder]

no, i'm pretty sure they're releasing it tomorrow

[waxen storm]

oh ok lol, had to ask 🙂 thnx

[midnight verge]

What's the best resolution for realistic photos of people?

[limpid thistle]

does anyone know what min gpu I need to run Kohya_ss

[urban dome]

@wise stratus I already have 2 factor authentication setup on my Google account. Adding this QR code step to dreamstudio is unnecessary and will just be a pain. Now I need to use the authenticator app and save the recovery code somewhere. Extra complication for no benefit. There should be an option to skip it.

[twin dirge]

anybody got any experience with lora training? how long it usually start after clicking start training?

[open marsh]

ten minutes to a week

[wind current]

Do discord bots that generate pictures on this channel work on the sdxl model?

[chilly echo]

Why does my inpaint keep drawing small versions of the overall pic, instead of just redrawing the thing I colored? I've got only masked. Got masked content original. Etc.

I inpaint a hand and it puts a face.

Maybe cause I had a seed set by accident?

[cosmic light]

Do you have it set to "inpaint only masked" or "inpaint whole image"

[chilly echo]

Only masked
No seed
Masked content Original

Restarted Stable and it's still doing it

[cosmic light]

Set it to inpaint the whole image. That way it will use context from outside the mask,
Or if you don't want that context, increase the padding on the mask, so it can see enough of the original image to realize there's supposed to be a hand, etc

Lmk if it helps

[chilly echo]

Comp is slow. Trying. Inpaint whole picture seemed to work. Thank you. Never had this issue before with only masked.

[cosmic light]

Best of luck

[chilly echo]

👍

[gilded hearth]

Can someone tell me why I got this kind of format ? #1100170312106127410 message

[silent stone]

🥬

[dark plover]

can someone help me edit

because i don't have idea for where start

[dry tendon]

Hi everyone, I'd like to know if it's possible to download the models we use in the bot channels, and where can we get these models? Thank you

[wild steppe]

Bot models are SDXL 1.0 candidates (not yet released): #🗣｜artisan-support-feedback message

@faint nova Please don't self advertise in general chat, you can post in #🌶｜off-topic instead

[dry tendon]

thanks

[elfin pivot]

https://youtu.be/QP2E-wEM748

[spring talon]

Guys where I can find the notebook file setup in civitai / github pages?

[wanton vessel]

any tamil people here?

[delicate iron]

hey! does anyone know how to update Gradio to 3.23.0?
I got an error message when trying to use an extension that told me to update it. I have searched online which brought me to GitHub where it shows me a specific workflow/pull-request, but I have no idea how to work it ;-;

[sudden ruin]

Did you update your A1111 install?

[languid cedar]

Hi, any news about official release of SD XL 1.0?

[sudden ruin]

Its enough to ask in one channel
Check out the next event and find out more

[languid cedar]

okay thankjs!

[midnight verge]

In almost every image with a person on stable diffusion the background is blurred out as the focus is almost completely on the person but that makes a very unrealistic picture, is there any way I can get a detailed background with a detailed person also to make it look like a real picture

[stable pawn]

cherubs was monsters with a lions body, eagles wings and a human head, but uneducated people started to use that name on Putto so now when I write "cherubs" I get ugly babies. So what are the original cherubs called and how do I get them in AI SDXL?

[honest berry]

Hello I want to know can I use the bot of stable diffusion in other Discord servers??

[honest berry]

Use Adobe Firefly

[sudden ruin]

Nope

[bleak matrix]

Good morning, everyone! I hope we are all well today!

[robust coral]

Hey Sunny. The same to you!

[bleak matrix]

Thanks! What are you up to?

[robust coral]

I've made a reddit post with some of my SDXL experiments including the prompts: https://www.reddit.com/r/StableDiffusion/comments/159rio2/some_of_my_sdxl_experiments_with_prompts/

I've also shared my SDXL prompts here on the discord server over the last couple of days if you search for my name.

[bleak matrix]

I shall take a loook!

I really love the one with the spell casting!

And that you shared all your prompts!

that's really cool!

[karmic brook]

https://discord.gg/nHMkvhsD?event=1130919343941758978

soon soon soon~

[hushed plover]

Woohoo

[regal quartz]

Right? I joined this server after the delay to see what was up. Now I'm glued to the channels hoping there isn't another.

[bold solar]

It’d be hilarious if that presentation is an announcement of the future release date of SDXL 1.0

[regal quartz]

I heard that they are also doing some new tool kits for training. Is this true? And will they be released at the same time?

[sacred musk]

is there a way to /describe with SD like midjourney does?

[regal quartz]

I'm not familiar with /describe or MJ that much. But initial search seems that it works like img2img

So yes.

[sacred musk]

its basically img2text, just going in reverse trying to predict the prompt

[regal quartz]

Oh! Yes

Clip interrogator

[sacred musk]

ah ok cool

[regal quartz]

fuck i can't spell

There we go

[rugged patrol]

Image prompting would be nice which is kind of like a blend of controlnet reference and clip, but dang Mj does it well

[regal quartz]

I think there are some more extensions that could make it better? Like a search and test token thing...

Who knows. I don't really focus on image prompting as much as controlnet and then inpainting.

[shadow halo]

I'm still struggling with the character consistency, any advice? 😬

[regal quartz]

Lots. I think the easist way is to find textual inversions on civit.ai. You find a few that are close and start blending them in the prompt

[shadow halo]

Is it better than img2img textual inversion?

[regal quartz]

Textual inversion is a specific way of fine tune training to create a token that represents a thing. So I don't really understand the question. I'd say it's about the same.

[rugged patrol]

Same. But I have used MJ image prompting though to make fairly consistent material for rounding out Lora training sets.

[regal quartz]

I've got some examples one sec

[shadow halo]

I mean Img2Img have deapboru and the interrogator which should do the job

[sacred musk]

https://huggingface.co/spaces/fffiloni/CLIP-Interrogator-2 this worked pretty well just now

[shadow halo]

I'm suring Roop for the face, which is great and grant me the same aspect for the character, but keeping consistency on cloth is a real, real, mess 😕

[regal quartz]

Okay here's a short tutorial using automatic1111.

Say I want a consistent character. I find 2 to 3 TI(textual inversions) on civit.ai that i think are close to what I want the character to look like.

In this example I'll be using https://civitai.com/models/36147/veronica-cipher and https://civitai.com/models/23511/shay-mitchell

[forest vector]

not sure if they've announced but im wondering if theyre working on freewilly 13b

[regal quartz]

Then I use the prompt to blend between them. like [x:y:#] x is shay, y is cypher.
https://media.discordapp.net/attachments/1060269382728683671/1133783524407787741/image.png?width=1268&height=106

[regal quartz]

the # is when it switches from one training to another. You can see the progression from one TI to anther here. I find a spot that looks good and then I use that.

[edgy finch]

Hey guys! Any tips on how to upscale images to 3x and keep realism?

Mine get too oversharpened (noticeable in the eyes, curls in the skin look sharp, double eyelid is like a line).

[regal quartz]

Turn down cfg, up the downsampling, or go in stages(like 1.5x then 1.5x that)

[regal quartz]

Then I added in some https://civitai.com/models/48104/sarah-shahi to switch inbetween steps.

Then finally I get [[VeronicaCipher:sh4ym1tch:0.79]|s4r4hsh4hi] as a good merge for a consistent original character

https://media.discordapp.net/attachments/1060269382728683671/1133785430609895465/image.png

https://cdn.discordapp.com/attachments/1060269382728683671/1133786139841548500/image.png

This is not the most straight forward but it is def the easiest with no training required

[regal quartz]

Also might be an upscaler issue... Try other ones?

[dim nest]

So today is the day they announce the release date of 1.0?

[tropic sequoia]

Either they announce the release or they announce the delay. Well seeeee

[sinful owl]

[midnight verge]

Is there anything I can use to generate prompt from an inserted image

[robust coral]

Thanks, Sunny! Really appreciated 🙂

[regal snow]

hmm... announcement says 1.0 was released on github, but the Releases still only has 0.9 🤔

[mild temple]

wheres the download

still uploadin?

[latent jetty]

fr

[verbal osprey]

Was it all just a lie?! 😭

[olive dagger]

how do i download SDXL model? i can't see it on github!

[regal snow]

I suppose I can wait a few more hours XD

[verbal osprey]

They probably don't have gigabit internet. Takes a while to upload. Might be on 56k modem.

[regal snow]

until its, u know, "released released"

[verbal osprey]

Back in my day you had to wait 60 seconds for an image to load on a website.

[olive dagger]

given it succeeded to load*

you remember logging into net, opening as many websites as you can, and disconnecting from the internet

[wild steppe]

It will be made public in the links mentioned in #📣｜announcements soon! Please be patient ❤️

[olive dagger]

👍 👍

anyway i still have some time to watch the UAP congress hearing so thanks for delaying it!

[verbal osprey]

Oh yeah! I forgot about that! Insane.....

[olive dagger]

i would be torn in two halves if you released it now

[bleak matrix]

np!

[timid gate]

Disclosure happening now so i quit my job because soon we will join the galactic federation and live off of alien tech

[fickle vault]

wait so sdxl 1.0 will not have a public release? only api and amazon?..

[timid gate]

also, link please

[timid gate]

(apparently they are uploading it now)

[fickle vault]

where to

API?

[timid gate]

No clue

[fickle vault]

lol

[timid gate]

ive been looking at github nothing yet

[wild steppe]

Soon! Things are rolling out

[verbal osprey]

If Aliens are revealed to be real, as exciting as 1.0 is, the alien reveal would be 1000x more interesting.

No offense to the stability team but we're talking about a world altering event.

[obsidian adder]

When using the active bots, is there an automatic CFG value applied to the prompt?

[real zodiac]

I see a new vae on hf, will there be a new refiner model too?

[rose wave]

what is sdxl

[signal dome]

will sdxl compatible with rtx 2060 6gb

please

[olive dagger]

post scarcity civilisation, here we goo!

[timid gate]

SD but big

Vanilla SD is small in size, probably SDXL is midjourney V3 level

[chilly crow]

https://cdn.discordapp.com/emojis/886398602153439283.gif?size=96

Soon

[rose wave]

so its a model or a totally new environment?

[timid gate]

new enviroment i presume

[oblique pine]

Excited to hear Emad

Big day today

[crimson belfry]

ALIENS

This update is amazing

[vestal dew]

So it is SDXL today?

[broken smelt]

God I love Emad

[vestal dew]

Emad?

[verbal osprey]

I'm trying to download the juice, but I'm getting a 404 error.

[wild steppe]

That's what the prophecy says

[vestal dew]

Ah, now it's a prophecy

[wild steppe]

https://twitter.com/EMostaque/status/1684251295623020544

Emad (@EMostaque)

CreativeML Weights, API & more coming today 🤞🏾

Fine tuning API preview launches today, #SDXL tunes really well, great job by team.

▶ Play video

Twitter · Jul 26, 2023, 5:16 PM

[karmic brook]

https://discord.gg/nHMkvhsD?event=1130919343941758978

JOIN US

[verbal osprey]

The hugging face pages are still 404'ing. Anyone else?

[fervent thunder]

What exactly is SDXL and is there any reason for users of Automatic 1111 to switch over at this time?

[wild steppe]

Head over to #1005545148211527800 and find out 👀

[fervent thunder]

Is there some information that I can read about it instead? I'm not good with the whole talking thing.

[wild steppe]

We have an announcement & a blog post: #📣｜announcements message

[cobalt nexus]

i have this too

[tall pivot]

where model?

[distant rampart]

does SDXL change any of the requirements for generating images, like system resources (16GB VRAM etc)?

[wild steppe]

Very soon!

[light solar]

Are there anyways to load the info to find the prompt for a zeroscope generation, my computer crashed and didnt save the file i had written it in. Is there a way to bring up the data either in the auto1111 UI or on cmd where it appeared when it generated the initial video so i can use zeroscope XL?

[olive dagger]

404

[light solar]

?

[verbal osprey]

Still 404'ing on huggingface.

[olive dagger]

oh hi again castor

[verbal osprey]

yes good evening

[olive dagger]

nice pfp, i'm not sure if it's a pokemon or an astronaut helmet

[verbal osprey]

Ha, it's the ship from Heavy Metal, the movie.

[rustic wadi]

Waiting on the model drop while listening to the alien hearing is indeed a very interesting experience

[olive dagger]

right?

[coarse warren]

I ran this scripts: while true; do curl --fail https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0 && say Model Released; sleep 5; done

😃

[olive dagger]

the timing with AI is quite bad, now that we can have perfect UAP AI generated photos.....

it will be hard to believe any of the legit photographs are real!

[verbal osprey]

Haha definitely. Creating an image out of text isn't going to seem so cool when we learn that teleportation is possible or that there are vehicles that can defy physics as we know them. I feel bad for the stability team today.

The aliens probably have tech to type something and then have it 3d print that thing.

Real living lifeforms or inanimate objects.

[fervent thunder]

https://medium.com/@s1610.2003/sdxl-1-0-a-great-leap-towards-outperforming-competitors-in-the-mid-journey-of-image-generation-bce322dace9e

[verbal osprey]

Link at the end still 404's

[wild steppe]

<t:1690399800:R> 👀

[fervent thunder]

it will be activated soon

[olive dagger]

LOL

[fervent thunder]

does anyone think the title of the blog might piss off mid journey

[olive dagger]

i will still be using midjourney for specific stuff. it's the right tool for the right task situation

[verbal osprey]

Ehhh I don't think so. Mid-journey is still extremely good for certain things. I find the discord stuff annoying, but it can produce cool images.

[olive dagger]

discord stuff?

[verbal osprey]

You have to create within discord. No dedicated app.

Sharing stuff is great, but they are limited because they don't have a customizable UI.

[true nacelle]

Seeing a 404 error on the Hugging Face download page for 1.0

[verbal osprey]

25 min

[true nacelle]

Got it!

[verbal osprey]

Stay strong

[forest vale]

@wise stratus https://media.discordapp.net/attachments/1128636013573574686/1131336054637600940/00149-1558615466.png?width=895&height=671

[haughty cloud]

Waiting for weights. https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0

[broken gull]

#1029055412764422214 message

@elfin pivot InvokeAI can

was answer to if auto1111 now works with SDXL

[elfin pivot]

On my screen it says No Access

[broken gull]

ye cuz it was a link to stage chat which is now closed for us

[elfin pivot]

Ah ok

I'm just currently using Automatic 1111 because I'm kind of use to that

so I'm just gonna wait a bit if it's not working like in the next 2 weeks I'll try out Invoke or comfy

[broken smelt]

#1029055412764422214 message

ah crap

we cant access the stage chat anymore

[wild steppe]

gogogo

https://github.com/Stability-AI/StableSwarmUI

[fervent thunder]

Why is there no channel for FreeWilly

[olive dagger]

what's sd_xl_offset_example-lora_1.0.safetensors?

[solar swan]

StableSwarmUI have a local API like auto1111?

[olive dagger]

hmmm i'm trying to use SDXL in auto1111 the way i use other checkpoints, but it doesn't work, is there a different way to do it?

size mismatch for model.diffusion_model.output_blocks.5.1.transformer_blocks.0.norm2.weight: copying a param with shape torch.Size([640]) from checkpoint, the shape in current model is torch.Size([1280]).

[wind crag]

i was using the SDXL .9 branch of A1111, had to use --no-half-vae in the webui commandline args, might still need to use it ? not sure

and 1024x1024 res

[coarse warren]

its works for me

I've update it to our 1111 website. works fine.

[olive dagger]

it just won't load so 1024x1024 won't mean anything i guess

[wind crag]

oh dang

[olive dagger]

do you do anything differently ?

[wind crag]

lemme load up the sdxl branch of a1111 and check it out see if it works

not really, paste into models and voila

[coarse warren]

What version did you install?

[wind crag]

but i did use a different brnahc

[olive dagger]

i just got update, let me check

[coarse warren]

You can checkout to v1.5.0

[olive dagger]

i'm running it with --no-half-vae now

[olive crescent]

Anyone had luck with using a colab?

[olive dagger]

how do i check version of automatic? 😅

[wind crag]

down the bottom

[olive dagger]

oh

it says 1.4

[bitter maple]

Stable diffusion model failed to load, exiting.... IN AUTOMATIC1111, WHY?

[olive dagger]

[wind crag]

version:

my sdxl branch is 1.4.1

[tight mist]

SDXL 1.0 model is out ?

[spare plume]

Hi, for general use, would you rather recommend base or refiner? Or do both go together?

[tight mist]

both

[wind crag]

SDXL to load will use a lot of memory, like a TONNE mine used 29gb just then, so it will take a while if you have less, by packing it into local storage once memory is full, you may have to wait but i just launched it fine

[olive dagger]

ok, so when i try to update it says:
error: Your local changes to the following files would be overwritten by merge:
[here is a list of some files]
Please commit your changes or stash them before you merge.
Aborting

any idea?

[wind crag]

https://www.youtube.com/watch?v=0u8idds-Y_Q

thats the video i followed to get the SDXL specific branch of A1111, not the main one

[olive dagger]

ah thank you i will check

[wind crag]

you have to use command line to get the specific branch from github (still a1111's github), if he has updated it im not sure if it will still be available

[olive dagger]

i'm using automatic1111 easy installer https://github.com/EmpireMediaScience/A1111-Web-UI-Installer
it has updating option in it which i'm trying to use

[bitter maple]

set COMMANDLINE_ARGS= --opt-sdp-attention --no-half-vae --theme dark --autolaunch
git pull
set ERROR_REPORTING=FALSE

IS THIS RIGHT?

[main sphinx]

any website where people can use and host multiple stable diffusion models?

[wind crag]

hmmm not sure, i think i remember mattvidpro mentioning it in a vid, but i think it wasnt a showcase video so not searchable. Might have better luck with google sadly or maybe the stable diffusion subreddit

[wind crag]

sadly there was no availability of the refiner for that branch but the generations were/are still amazing

[bitter maple]

Stable diffusion model failed to load, exiting, please help

[wind crag]

not sure, i use the separate branch sorry mate, my command like args are --xformers --autolaunch --no-half-vae

iirc --xformers reduces vram usage and speeds up generation

[tight mist]

https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0

[wind crag]

i am also using a gpu with 24gb vram and have 64gb of normal ram, so not sure if its a technical issue youre running into related to hardware, so many variables

but during generations this model only uses around 9gb vram in combination with windows also using some

scratch that, increasing steps and resolution it jumped to 16gb

[verbal osprey]

Maybe I'm a bit confused but are there safetensor versions of the model to download? I'm not quite sure where to download the actual models on the huggingface page.

[wind crag]

yes

https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/tree/main @verbal osprey

[turbid plinth]

I have downloaded the 2 files XL and VAE XL in the usual folders of my Automate 1111 but when i try to generate : it write : TypeError: must be real number, not NoneType

[wind crag]

the safetensors 2nd from the bottom, click the download icon in the middle of the row

[olive dagger]

thank you friend, it worked

[wind crag]

i think it was devil that did mention some issues with A1111 as of up to a couple of hours ago, i think for the next 24-48 hours the updates for it will be touch and go

[olive dagger]

i'm not sure why it's not straightforward though

[wind crag]

youre VW, glad i could help someone else xD bigger thanks goes to coderx for the walkthrough video, shame it was so hard to find it

[olive dagger]

😩

i would spent today and whole weekend trying to figure this out if you didn't tell me

and would probably get sick of it and went back to the old version

😂

[wind crag]

xD yeah that was how it was for me when using 1.5, nearly bricked my pc multiple times because im not versed in using git lol

[olive dagger]

i will try to run animatediff now, it didn't work for me before either

maybe it's just that

but wait... it probably doesn't work with xl

[wind crag]

i would say within the next month there will be some crazy stuff coming, especially if 3rd party devs got early access to the model, by the end of the year our minds will be blown i reckon

[olive dagger]

yeah

it's interesting how people get used to the current state very fast

so, even though the progress is incredible

people just get used to it

[wind crag]

there will be logo specific models, vector models, realistic models, waifu models, camera-specific models. Exciting times 😄

[olive dagger]

"oh yeah of course we can create amazing art in almost real time"

and noone remembers disco diffusion that was taking like 1.5h to create an image

😄

which was already mindblowing

[wind crag]

i remember my mind being blown with disney using the LED projector dome for the mandalorian, then this came out....just huge strides bro lol

but what i really like (personally) is people and even some websites now, using parallax separation with generated AI images, gives such crazy depth

that method is used a lot on stills for intros and stuff but for AI it makes me feel like a kid again with a pop-up book lol

[olive dagger]

do you know if after preparing automatic1111 for SDXL it doesn't support 1.5 anymore?

because 1.5 don't show in my drop down menu

[dire shell]

hi everybody. does anyone know how to get perfect tiled pattern? I try but the images doesnt match

[wind crag]

i created a separate install folder and basically did a new install for the branch, if you didnt back up that folder i would recommend doing maybe a system restore to a previous date/time, that will restore those files that have been lost/changed. Thats the best method i can think of if backups were not made

[olive dagger]

yeah i have a separate branch

so it's all good

[wind crag]

but yeah i keep 1.5 and XL separate

ok sweet

[olive dagger]

i learned my lesson multiple times in the last 3 decades 😂

[wind crag]

but im pretty sure either works fine by itself but i imagine the updated A1111 as of today might have issues because it is trying to provide flexibility between those models

[olive dagger]

right

[wind crag]

anyhoo my dragon (wife) is now awake, gotta go tend to her 😄 very glad i could help out 🙂

[ornate flame]

Is there a way to save jpgs instead of pngs using ComfyUI?

[olive dagger]

thanks again !

[wind crag]

sorry returned because i think i remembered where might be able to do that, runpod iirc

[olive dagger]

do your generations also have a weird texture in all images?

[wind crag]

only if i dont use the 1024x1024, because thats the minimum it was trained on

people in comfyui got pretty good results at 4k with some tinkering, because it uses less vram with certain settings and nodes, but ive never used it

[lime glen]

hi boys

i did upload my lora for alex from minecraft and i would like to share it with u all

how can i share my lora ?

[karmic brook]

#1047197565365538826 !

[lime glen]

thx

[wind crag]

there may be a channel here, but people also have shared stuff on the models community page of the huggingface site

[lime glen]

thx homie i did share it

[iron fossil]

im glad you said something i was doing gens in 512x512 and was like what the heck is going on lmao

[wind crag]

haha yeah reminded me of when i used low res in sd 2.1 xD

[iron fossil]

what sampler you using?

i was addicted to unipc but its not supported yet for sdxl

[wind crag]

just euler a, because im base as hell lol, but Olivio Sarikas has a great video showing different steps and samplers for SDXL, quite interesting with the difference in generations, high steps can make some of them look cooked as hell

https://www.youtube.com/watch?v=SVBOL_ukFqE&t=601s

[iron fossil]

I'm seeing that as well, the usual 30-35 steps seems too much.

[wind crag]

yeah i noticed in longer prompts you had to find the "goldilocks" range of samples, too much, too little then after an hour or 2, just right 😄

i really need to upgrade my gpu though, get about 3 it/s on a 3090 with 80 steps when batching 100

[proven mulch]

how do you correctly use sd_xl_offset_example-lora_1.0 lora in comfy...? There's no trigger word or anything

[wind crag]

might try comfyui, some dude said he was getting much faster results but i dont like speghetti that nodes-based rendering is lol

[iron fossil]

my 4080 is flexin hard on these gens i can feel the heat lmao

[wind crag]

hey, thats a good thing, free space heating during winter 🙂

bet theres some lurker in this discord with a 48gb card just having a grand old time generating 5120x1440 wallpapers

[pseudo finch]

Can SDXL 1.0 be used for commercial?

[wind crag]

I know there was something akin to a eula/licencing for the closed beta of SDXL, you might need to check out huggingface if they dont have that provided here

but iirc it was more along the lines of "dont carbon copy known brands" etc

[dire shell]

can anyone tell me how to get perfect tiling with sdxl? the tiling doesnt match

[halcyon kelp]

https://youtu.be/LwsI-0dqZ6w

[iron fossil]

would the refiner work on images generated with sd 1.5 ? my peasant brain is trying to understand the research surrounding it lol

[bitter maple]

whats the difference between base and refiner for automatic1111?

[iron fossil]

the way i'm reading it , sdxl generations are a 2 step process equivalent to essentially a text-to-image from the base model followed by an image-to-image by the refiner

[verbal osprey]

Is there a guide on how to use this new stuff with automatic1111? I guess it will take some time to implement.

I downloaded four files. The base model. The offset-example-lora (no idea what this is), the refiner, and the vae

[iron fossil]

i just grabbed a fresh copy of a1111 as i dont want to mess with my current setup too much and the only thing i had to do was add the no half vae flag thats mentioned in the latest release notes

but so far ive only got the base model working not the refiner

[wind crag]

https://huggingface.co/stabilityai/stable-diffusion-xl-base-0.9/tree/main/vae cannot get that bottom "diffusion_pytorch_model.safetensors
" to save as default vae in A1111 (newest version), keeps defaulting to the 1.0 SDXL vae, as soon as i click apply settings it says 0 settings changed. Any idea what is causing that ?

[real zodiac]

that's a vae in diffusers format, won't work with auto

[wind crag]

ah gotcha xD

[iron fossil]

you mean Xeet

[frank spruce]

my stablediffusion automatically tried to install SDXL just now with the "git pull" in my bash file, is that normal

[formal minnow]

Hello guys, one question. Is sdxl 1.0 live on discord. tY

?

[frank spruce]

idk but i will switch to sdxl just as soon as they support 4gb vram or i get a new computer

[dark escarp]

Can you generate nsfw with the released models now?

[abstract ice]

sdxl 2.0 when?

😋

[wintry grotto]

Yes

[shell venture]

Hi, i got Sdxl 1.0 but i cant seem to make a good anime style but the bot could? How do i add theses Styles

[sleek dragon]

does anyone know the analog film style prompt?
like what to type in to get the same results as using the style

[long wolf]

anyone has a comfyui workflow that works well with sdxl

[brittle tundra]

sent

[long wolf]

gracias

[brittle tundra]

no worries, glad to help. paying it forward, as it were.

[long wolf]

how would i go about adding the new sdxl vae to the nodes? hmmm

ahh i think i see it now let me see if this works

nope didnt work lol

[brittle tundra]

That particular workflow has blocks. Put the Base model in the bottom one, and the refiner in the top one. You will need to link the refiner model and the VAE. Also the clip to both of the upper joints.. I think that one was modified and broke a bit.. I may try to fix it real quick.

[long wolf]

gotchaa

[old pagoda]

Hello Everyone!

[dire sleet]

asked this on a few reddit threads and couldn't get an answer - is there a reason I can't see the two safetensor files in the git repo?

I'm just looking to download the two new safetensor files for 1.0

[brittle tundra]

not sure.. I found the on huggingface

[dire sleet]

oh are they not just in the git repo? that's what everyone was referencing on reddit

ah ok yeah I see them on huggingface, thanks

[verbal osprey]

What does the sdxl lora file do?

The file that you can download on hugging face.

[real zodiac]

they said it's supposed to add offset noise (ie better contrast) since 1.0 has less of it than 0.9

[frigid orchid]

Can anyone please help me install SDXL? I am pretty good at all of this. I know how to install extentions. I know about this video here which helped to install the demo. Via this link https://github.com/lifeisboringsoprogramming/sd-webui-xldemo-txt2img
https://www.youtube.com/watch?v=SVBOL_ukFqE

But is there a new link since this is no longer the demo?

I think I just put in an extension link into Automatic 1111

And than go download the models with my huggingface account. Do the activation for the permission or whatever generate the activation key etc.

But if anyone has a step by step guide, or can walk me though it quickly I would love that. Thank you.

[verbal osprey]

So I have to use that lora with all of my creations?

[real zodiac]

ehh not really, I tried it but didn't really feel like it improved anything

[verbal osprey]

Cool thanks for the help!

[real zodiac]

you might still find it useful, I dunno really

[frigid orchid]

Wait. SDXL 1.0 does not need an extention anymore unlike the demo?

It installs like a regular model?

[raven pewter]

i think you can just generate with the model and then image2image with the refiner https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl

[frigid orchid]

I can't wait until we can generate an image with SDXL Automatic 1111 and have them automatically refined in one click.

All in one process.

[verbal osprey]

Yeah, it is too much of a pain to split it up into two steps. I'm not wasting time doing that.

I have no issue waiting days/weeks until it's a one click option.

[frigid orchid]

Yes I still have not figured out how to set it up. Found two guides that give completely different steps. lol

[raven pewter]

in comfy ui you can do it

[frigid orchid]

I am not learning comfy UI yet.

Sorry. lol

I can't wait for Aitrepreneur to put out a SDXL setup video. As I find his guides are quite good.

[raven pewter]

can i ask a question how much do you now about tiled diffusion/multidiffusion

[frigid orchid]

Me? I don't know if I know anything about that.

If I do I don't know those terms.

[raven pewter]

ok

its like an upscale methode i just dont understand the purpose of it

[frigid orchid]

I see. I know the alternative methed of hiresfix.

[raven pewter]

that would be

[frigid orchid]

Were you can use like latent upscale with image to imagine.

image to image*

[raven pewter]

tell me more

[frigid orchid]

It's like an alternative way of using highres fix in Automatic 1111.

[raven pewter]

the thing is i dont exactly know whats in latent space and what isnt

[frigid orchid]

It's best you watch a video as I am not completely fresh on it. I can't teach it. Also it has a few ways to do it.

Let me find you a very good video

[raven pewter]

im a mad man i want all the detail

just cant get it

im currently upscaling an image to 9k just to see where diminishing returns begin

[hollow basalt]

im so sorry but would anyone be able to DM me and help trouble shoot me through whatevers not allowing me to start generating images?
i tried my best to install SD from the instructions on github lol

[raven pewter]

whats your gpu ?

[hollow basalt]

how do you check again?

[verbal osprey]

I've spent too much time setting up automatic1111 to make the switch. All of my extensions are set up in automatic.

[raven pewter]

uh device manager or task manager will do

[frigid orchid]

I have a RTX 4080. Here is the video I was referring too. https://youtu.be/A6dQPMy_tHY ULTRA SHARP Upscale! - Don't miss this Method!!! / A1111 - NEW Model by Olivio Sarikas

[raven pewter]

comfy also doesnt support all extensions i think

[frigid orchid]

ULTRA SHARP Upscale! - Don't miss this Method!!! / A1111 - NEW Model by Olivio Sarikas

[hollow basalt]

it says NVIDIA GeForce GTX 1660 Ti

thats sort of middling, right?

[verbal osprey]

1660 is on the lower end.

[raven pewter]

bearly making it

6 gb is bearly enough

[raven pewter]

i know all of that

just need someone to explain the latent space to me

[frigid orchid]

Yes that's my peak of knewledge likely more tricks now since this is 4 months old but useful.

[raven pewter]

yes

[hollow basalt]

okay well can someone trouble shoot me through it?

[raven pewter]

the best way i found currently is image 2 image with controlnet tile resample and sd ultimate upscale

[hollow basalt]

im sorry lol

[raven pewter]

multidiffusion doesnt work for me

[frigid orchid]

Quick question can all that be done on 16GB of VRAM? I have not tried that yet in control net etc.

[raven pewter]

thats it it can be done one any vram

[grave stag]

Hey I have some refs either I found them online or I had them custom drawn since I either still can't or wasn't at least then or didn't know I could get an A.I. to create the refs exactly how I want for an erp. Now I created those refs have a good PC and have found stable diffusion but idk after setting it up I can't seem to find a way to get it to make something that looks like my character, nothing that could pass as a sister. I was told by someone in DMs it be easy to do that can I DM someone some images and get some help with setup and prompts?

[raven pewter]

im currently upscaling to 9k

it takes 15 min but im doing it

[frigid orchid]

Yes 9K is extreme.

[raven pewter]

and i hav 24 gb

[hollow basalt]

aay fr can someone help me lol

[raven pewter]

bro

check dm

[raven pewter]

maybe later currently helping sanzaru

[fickle igloo]

any one know what to do with Installing gfpgan

it's stuck

[grave stag]

Lol then I'll go 16k

[grave stag]

Ping me but just warning the refs are 18+ and she isn't fully human half dragon but not like half humanoid like I've been able to get it or weird skin and not right you know?

[humble panther]

using stuff like automatic1111 in the past, where did it install everything? including like github related things

[jolly crystal]

How can I know the best promot for the model I choose? For example I chose mistoon_anime how can I know the best prompt?

[wooden fjord]

How do I use a specific style or modal? I was going through the galleries and would love to test a style with a prompt.

[fervent thunder]

??

[misty veldt]

🧠'nt

[grave stag]

?

[rotund dagger]

Hi guys, does the checkpoint selected affect img2img?

[fickle igloo]

i have a problem

nothing is generated

[verbal osprey]

In automatic1111 it does. Because normally that's where I will choose an inpainting checkpoint.

[silk gulch]

i was using sdxl 1.0 and whenever i type in a prompt it generates a sign that says "no" (different picture every time but same idea) what do i do

[lime glen]

yo boys

i did a simple guide on how to use stable diffsuion on google colab for free

i would like to share it where to share ?

[thick trout]

how is this server’s banner generated? The 2 astronauts

[wooden fjord]

Can we generate images to the bot in our own private server?

[marble knoll]

This is a serious question and I would greatly appreciate an answer.
#✨｜sdxl message

[gentle heath]

Hi, I am looking for a trained model to do product photography. What civtai pre-trained models will be best for that?

[craggy geyser]

Hi goodnight

I have a question

How would you prompt something like this?

Ow

Where can I send images-

Nvm I'm dumb when I'm sleepy

[stable grail]

Hello, everyone! How do I use the refiner model for sdxl in automatic1111's webui?

[silent fractal]

Is SDXL 1.0 allowed to be used commercially?

[coarse idol]

Yeah

[heavy violet]

hello!

[fervent thunder]

Can I use the same ComfyUI workflow of SDXL 0.9 with the new model 1.0?

[near valley]

Can anyone tell me how the training data for sdxl was captioned? Captioning isn't mentioned in the sdxl paper on arxiv.

There was a lot of interesting work done on variations, mixing and clip inversion on the previous architecture and I'm just looking for more information.

[dusky stream]

who wants to play chess?

;-;

[near valley]

lichess is the app my man, it's great.

[dusky stream]

you want to play?

[near valley]

thank you but no, I mostly just do random pairing, 3 min blitz, sat on the loo.

[radiant nacelle]

Did anyone saw that new upcoming dalle

It's something else man so powerful

[tight mist]

Guys we all agree that the model understand way better our prompts than 1.5 based models ?

[blissful wasp]

Guys!!! Wtf am I reading in this article?

https://www.analyticsvidhya.com/blog/2023/07/openai-teases-open-source-gpt-model-release/

GPT 3.5 for open release?

Isn't gpt 3.5 there's whole business?

[quick girder]

.

[digital kindle]

I am new here, and I would need help to update my automatic1111. I am stuck on v1.2.1. I have setup the "git pull" in my "webui-user-bat" and disable all my extensions but still it is not updating.

[warm junco]

What does it say in the console? A Screenshot in #🤝｜tech-support pls

[gusty totem]

I posted my images on Adobe's Discord and was informed that sharing work generated by other engines is not permitted. Isn't it interesting?

Could it be a hint of apprehension about acknowledging the true potential of FREE STABLE DIFFUSION?

[trim ocean]

does anyone use comfyUI? i dont find DPM++ 2M Karras sampler? Or what is it called in ComfyUI?

also does anyone have a good workflow for comfyUI i kinda wanna get into it

[regal snow]

I asked this question yesterday but no one answered (perhaps no one actually knows)
when using the SDXL discord bots, you can specify a separate "Style" tag from the "Prompt" tag.
Is there a special syntax for that when writing prompts? Or does the bot literally just add for example "photography style" to the prompt? or ...?

@trim ocean pretty sure there is a dropdown for DPM++ 2M as sampler and a separate dropdown to select Karras as scheduler

[urban atlas]

why do i need to wait for showdown to end to generate images thats yucky. most of the time its just 2 bozos comparing catgirls

[trim ocean]

jep i found that out lol ty!

[empty grail]

hello

[gusty totem]

In Ksampler node:

sampler_name = dpmpp_2m
scheduler = karras

also you can finde soe tutorial on my youtube chanel for ComfyUI https://www.youtube.com/@ArchAi3D

[quiet dagger]

Is it possible to use refiner in SwarmUI?

[trim ocean]

awesome ill check it out i wanna get good with comfyUI!

PS: is there a way to easily open the folder of outputs with comfyUI?

[gusty totem]

I think No

[trim ocean]

ah too bad but thats ok i love comfyUI its so performant

[gusty totem]

but you can config the output folder name and file names easily

Display the pixel image in the node frame and output/save it as a PNG file on the server below. It cannot handle latent images.

The input port is named images , but only one edge can be connected. If the number of batches of input is >1, each will be saved as a separate file.

Saved files are named with a 5-digit sequential number, and by default ComfyUI_01234_.pnghave a format similar to . filename_prefixComfyUI allows you to change the beginning (part) of the filename .

Within filename_prefix you can use some preprocessor variable strings:

%width%: image width (number of pixels)
%height%: image height (pixels)
%date:<format>%: Date and time of enqueue. It becomes like Nara %date:yyyy-MM-dd%. The contents to be replaced with are as follows, and the rest remains as it is. 2023-01-23<format>
yyyy: year (4 digits)
yy: year (two digits)
MM: Month (2 digits)
dd: day (2 digits)
hh: hour (2 digits)
mm: minute (2 digits)
ss: second (2 digits)
:doesn't work on Windows or Mac (Linux is mostly fine). to make sure.
%<node>.<widget>%: Widget value of the specified node. For example: %KSampler.seed%. <node>If there is a node with a matching Node name for S&R then that node is used, otherwise the node with the matching title is used.
TitleAfter creating a CLIPTextEncode titled

%date:yyyy-MM-dd%%Title.text%/%date:yyyyMMddThhmmss%%Title.text%%width%x%height%%KSampler.seed%_%Load Checkpoint.ckpt_name%
It is also possible to specify an overkill, such as Directory separation /can be used without problems on Windows. If the directory does not exist when saving, it will be created.

https://comfyui.creamlab.net/nodes/SaveImage

[trim ocean]

oh ok not sure i understand that all but i start watching your videos 😄

is there a way to easily set another folder as output?

or was that the explanation?

[gusty totem]

yes, change the name in (save image node) to this: (XX/image) it will creat a folder in output folder with name of (XX) and it will save the images with name of (image.png) in that XX folder

[trim ocean]

ah ok but i cant tell it to use a folder outside of the comfyui folder like on my desktop for easy access?

[fervent thunder]

TensorArt is sketchy AF lol. they paying model creators / lora trainers on their ko-fi/patreon to go upload their model on their site.

one employee just dm'd me asking if i could upload my models, pretending he's new to AI and would like to learn from me despite having no idea who i am and what my images look like. 3 day old account too.

[gusty totem]

no, i don't think

[fervent thunder]

evidence https://i.imgur.com/KwxBAUO.png https://i.imgur.com/V44vuJc.png

[trim ocean]

All good i make a link to the folder to the desktop 😉

[fierce fox]

Playground for Hugging Face models, GPT4, Stable Diffusion, Whisper: https://lastmileai.dev/workbooks/clkjugr5g0036phc1ri5g517n

[severe crown]

the photoshop plugin download link is broken on your site. Can you guys please point me a mirror link? Also, where do I pay for the SD api usage?

[charred sail]

What do we need to do to use SDXL 1.0 with Auto1111? I tried just putting it in my model folder but it results in errors when i try to load it

nvm, if anyone gets the same issue just update auto1111 to latest release, that solved it ^

[weary smelt]

is there a hotkey for comfyui to cancel current prompt?

lol

[arctic sedge]

Hey guys. My 3080 10Gb pauses between image gens, while running SDXL 1.0. Is this normal?
(using Auto1111 btw)

[pulsar basin]

how can I delete my image?

[grave stag]

If I generate an image with stable diffusion and it looks good but the hands and faces don't look quite right how do I tell it to go back and just fix that?

[rotund portal]

is there a way to load custom nodes into comfyui in a colab setting

[frigid orchid]

I have a general question. Back when SD first released I somehow was told or suggested to use --no-half. I have an RTX 4080 16GB. So I have been using --no-half. SDXL 1.0 does not work well with --no-half. So I removed it. I noticed if I re-run some of my seeds without --no-half my output is different. But I'm not sure if it's worse or just different.

My questions are should I be using --no-half with an RTX 4080 for 1.5 based models? I also heard of --nohalf-vae I do not and have never used that. What are the pros and cons of --no-half? Should I do away with using it?

I think I used --no-half as a way to eliminte black generated images that were all black?

[dapper ferry]

Whats your thoughts on the room temp superconducting material? is it B.S. or is the world about to change as we know it.

[fervent thunder]

bro this is a chat about ai image generation

there's #🌶｜off-topic

[waxen storm]

where do you put the refiner model fro auto1111?

[fervent thunder]

you can only do a img2img which dosent give you the same quality

wait for an update or learn comfyui or the other one

[rotund portal]

is there a way to put the comfyui manager in colab

[near wagon]

I don't see why not

[tribal fjord]

Is there a recording of yesterday's launch somewhere?

@rotund portal comfy colab notebook:
https://github.com/comfyanonymous/ComfyUI/blob/master/notebooks/comfyui_colab.ipynb

[near wagon]

if there is I hope it's like that video of the windows 98 launch with steve balmer and the gang dancing enthusiastically.

[crystal patrol]

Fast Stable Diffusion got updated and can run SDXL now but as a Google colab user im pretty disappointed: the compute units per hour went up from <2 to 5.45, because it need more ram and vram, and this makes colab too expensive (at least for me). So the options are to switch to runpod/paperscape or to go with Comfyui(it should run on lower ram machines, correct me if im wrong).

I like colab because of the ease of storing everything on Drive.

I would like to hear from users that run SD on web services what do you use, what are your experiences, etc.

[dapper ferry]

I take it its pointless trying to run SDXL on a GTX 1080?

[lime glen]

sup bois

i made a tutorial on how to import models from civitai straight to SD and how to run Sd on google colab for free where to share this link ?

[dapper ferry]

i use clipdrop and pay monthly it takes only a second or two to generate the images.

i just timed it about 8 seconds for 4 images.

i hate google colab. its so clunky and slow.

had such a bad experience on VQGAN+clip a few years ago on google collab that i would never go back to that crappy thing

all that programing code and shit

[tight pewter]

hey guys, any recent advancement in generating proper hands?
or any techniques to bypass bad hands? i'm already using the badhandv4 embedding btw

[lime glen]

bruh

i just asked where to share my guide,and i feel bad for u but there is people who deos not mind google colab because they can't afford to buy a new gpu

but i agree with u kinda

[dapper ferry]

i understand... But whats the point in running on a google collab notebook when you can use a webservice? like clipdrop

[lime glen]

because i didn't know about clipdrop

[dapper ferry]

why would i want to hire a virtual machine from google and have them spying on everything i do

[lime glen]

XD

[dapper ferry]

oh right... ive been using clipdrop for ages, £7 a month for about 5000 images every day.

[lime glen]

:0

too bad i am from egypt so i can't pay outside the country

;-;

[dapper ferry]

must be a way

[lime glen]

i tried almost everything

[dapper ferry]

what about using cryptocurrency?

[potent spire]

Im thinking about using SD (again) for very specific usage

[lime glen]

banned in egypt :((

[dapper ferry]

governments hate cryptocurrency with a vengence.

[lime glen]

to generate segs XD ?

[lime glen]

i hate my life

wait no i hate my country

[dapper ferry]

i had crypto on binance and then the British government threw a tantrum at them and practically banned binance

but i funneled my crypto over to coinbase

[lime glen]

welp screw it i will just accept the fact that i can't pay outside

btw if u are planning to visit egypt the first thing to do is leave

[potent spire]

No, more like filling the gap of Adobe Firefly for fun stuff with celebrities

[dapper ferry]

i always wanted to see the ancient runins.

[lime glen]

sheeesh nice fr

[lime glen]

u will see it

[dapper ferry]

your country has massive potential for tourism if its ran correctly

[potent spire]

Literally just for fun

[lime glen]

but u will see hell

[potent spire]

For serious i stay with Firefly ofc if

[dapper ferry]

Well there are going to be new solutions popping up in the near future imagine, for running SDXL

like peer to peer, i imagine.. because they are doing peer to peer for LLMs

[potent spire]

You talk to me? 😄

Or generally

[dapper ferry]

i dunno, i guess to you 🙂

[potent spire]

I dont have plans to switch when it comes to FF, but SD is free and i can play with copyrighted stuff and therefore its only for fun for some stuff

I mean i use FF also more for fun atm tbh

[dapper ferry]

does FF ban copyrighted characters???

[potent spire]

Yes, unless they are part of their Stock and therefore basically also theirs

[dapper ferry]

knew there would be a catch with Adobe.. they think they own everything.

control freaks.

[potent spire]

That is for a reason

Its targeted towards artists and professionals and companies, they dont want that stuff there

[dapper ferry]

let me guess, everything anyone creates with adobe software, Adobe owns the rights.

[potent spire]

No

[dapper ferry]

but have you read the TCs

[potent spire]

They dont own your stuff

They can train on those tho depending on some circumstances

For generative AI they dont own any of your images copyright

[dapper ferry]

But companies like Autodesk own your stuff, that you create with their software like 3DSMAX

[potent spire]

Nah

[dapper ferry]

its true mate.. its all in their terms

[potent spire]

Assets that you make are dours

Yours

[dapper ferry]

yea you might own a copyright but Autodesk own it as well.

its a known thing.

[potent spire]

I use 3DS Max from them

[dapper ferry]

you may think you own your work like renders.. But if you dig through the leagal terms you will find they own the rights.

[potent spire]

Its actually my main tool besides of ZBrush and Photoshop

[dapper ferry]

i know its true because Ton the creator of Blender talked about it in an interview.

[potent spire]

Its limited on what they own on your stuff

[dapper ferry]

you still use Zbrush after the aqquisition then

[potent spire]

They dont own your assets in sense that they can resell it etc

They might make ads with them tho

Etc

Or likely train on thel

them

[dapper ferry]

did Zbrush survive the Maxon thing?

[potent spire]

Yes

Easily

Just like Adobe after jumping to subscription modeo

model

[dapper ferry]

doesn't generative AI make you disenchanted with creating in 3D?

[potent spire]

They could afford themselves that

[potent spire]

Disenchanted?

[dapper ferry]

The text to 3D model is looming

[potent spire]

Sry have to look up the word

[dapper ferry]

np

[potent spire]

Generative AI made me appreciate those more actually

I just integrate gen AI eventually as part of the workflow

In one way or another

Because i want and need far more than generative AI tools are capable of right now

[dapper ferry]

But you know its going to catchup to all the stuff it cant do

then what.

if i had a talent for sculpting right now, i think i might consider sculpting out of physical materials and selling the models.

[potent spire]

Thats speculative to say the least and id not want to put a bet on that and wait for something that might not even come

[dapper ferry]

sculpt something great out of marble, i bet you'd make a fortune, and then wonder why the hell you were doing it in zbrush for years

[potent spire]

because im more a digital artist 😄

For money i will look later but there are (good) ways to do something and eventually earn on the side at least

Because as of now i already have a full time job

Otherwise i couldnt afford myself the huge amount of money spent in hardware and software

[dapper ferry]

Got to have plans for when AGI starts taking most jobs.

yea, thats a lot of software subscription you are paying.

But i worry about the future... although i have one thing up my sleeve.

[potent spire]

If it comes

[dapper ferry]

And that is im a guitarist and ive been in a band for a while, and when AI takes all jobs,, nobody is going to want to go to a concert to watch robots play

[potent spire]

But as said i dont want to bet and wait for something that isnt necessarily coming

[potent spire]

But if you enjoy playing it you will continue playing?

[dapper ferry]

well its just that AI wont replace that job..

watching a live band is a human thing, and an artform people appreciate, and dont want to see machines do

so when everyone is running about like a headless chicken because AI has replaced their job role, i know i have the gear, and the years of practice and experience to keep playing in a band.

[potent spire]

But eventually getting paid back with secondary incone

income

But id do it without that still

[potent spire]

Kinda like established artists

[dapper ferry]

just my advice on what i think is coming with AI.. i would switch to physical artwork, because art collectors are always going to appreciate art made by human hands... But digital stuff might get swamped by AI generated content

[potent spire]

Im between 2.000€-3.000€/year right now for software costs

[dapper ferry]

Zbrush looks addictive.. it looks very rewarding.

[potent spire]

Or more like at least 1.700€

[dapper ferry]

and ive tried Blender sculpting, it cant touch Zbrush, but i see why people like it

[potent spire]

Digital stuff is still not replacable by AI, the only issue is spamming of those AI images

But for the industry itself it isnt relevant since they want skills from artists

When they look at your portfolio they will know the potential candidates

[dapper ferry]

yea but you have to be realistic about automation.. The game comapnies the movie compaines, will automate as soon as the technology is possible.. which will be soon.

[potent spire]

Blender is free which is already a reason for many to use and its still powerful

[potent spire]

Automation is already a thing, but its not there yet and some things are extremely hard to automate and replace

[dapper ferry]

self driving cars are hard, but look at Musk's company

hard is not a barrier.

[potent spire]

To compare, some no name company would more likely do that compared to those like Riot Games

[potent spire]

For art and what i mean it is

[dapper ferry]

art wont get replaced... that is true. like i said it dont matter how much AI content there is, people want to buy human made art.

[potent spire]

The generative AI cant even touch the established artists in the industry. Thats why you see beginner artists being much more scared than those

It can be used there tho

But not to replace the complex workflow

Where you might see those are low companies

[dapper ferry]

its only years off though.. a decade flys by

[potent spire]

That are amateur like

Have less money

And dont give that much damn abour details

about

I personally am not realls worried about myself there and i also like art without earning money as well

Otherwise i wouldnt pay this much

[dapper ferry]

Theres not many skills that people can be confident will survive the AI apocalypse.

[potent spire]

Btw which SD model do u use?

[dapper ferry]

Clipdrop so its still 0.9

[potent spire]

Oh okad

Okay

I will have to look up agaik

i dont need fancy tools for SD because i will edit in Firefly and Photoshop anyway

[dapper ferry]

its just a gimmick to me really, just something to do to relax..

watch what it comes up with

[potent spire]

I just need to generate specific stuff

In a easy matter

No more no less

For SD

Ideally id use some web UI for SD without installing anything

[dapper ferry]

yea why bother with the hassle.

[potent spire]

I have already enough to bother with my pipeline xD

Back and forth between softwares and add ons

[dapper ferry]

well have fun.. sounds like you are in a phase where you are really deep into creating...

im currently bored with life and have creative block...

[potent spire]

I always loved being a creator, but my motivation was crushed by other people and partially myself

I did paint in youth for example and as a kid, but got demotivated by others

Plus had a very dark phase of life

[dapper ferry]

yup.. that can easily happen.

[potent spire]

Oh, i hope you find the track again 🙂

[dapper ferry]

i hope so.

[potent spire]

There are people who find enjoyment in art even late in their life 😄

[dapper ferry]

of course, i think the mind just has to be ready for it. sometimes life gets in the way

[potent spire]

Im not the youngest either but still young id sas

say

[potent spire]

Agree

[gilded trail]

When i try to generate images in clipdrop in mobile browser the file size is coming out around 1.5 mb where in the pc browser its coming out in 200kb. Why is this happening with the change of device thats being used to generate? Any idea guys?

im using the same prompts that i used for 0.9 in the latest 1.0 , the images looks bad and low in quality. speaking about clipdrop

[dapper ferry]

in PC browser when your account has lots of credits you click on a generation and it will download full size image.. then after a few thousand images it seems to disable upscale and gives a smaller image size.

then when it renews again the next day you can get the larger images again. either that or the way you are downloading you are getting jpg rather than PNG

[gilded trail]

ok i just noticed this after your explanation, that when i download the image in my mobile browser its in PNG in 3-5 mb and when i download it on PC its in JPEG in 200kb

[dapper ferry]

On PC, if you right click save, even before you left click to enlarge, then it will do the PNG thing

[gilded trail]

in PC theres no way/option that i can download the image in PNG, i tried right clicking to save the image and also the direct download button that appears on top of image...both ways are downloading the image in JPEG but not PNG. This is wierd

[dapper ferry]

odd because in chrome, if i right click save from the grid it definetly saves as PNG

PNG just gives a bit more color gamut.

[gilded trail]

im trying it in every possible way to download the image in PNG from my PC browser and its not happening🥲 . still downloading in jpeg only. In my mobile its downloading in PNG but the issue is that earlier with 0.9 the default images were downloading in 2048x2048 ~ 5MB. But now after the 1.0 update even in my mobile the default image is downloading with 1024x1024 resolution ~2MB. Not sure what i must do to get the best quality...

[dark plinth]

I'm new here guys - can somebody help me understand what the main difference(s) are to midjourney? sorry if its a stoopid question?

[dapper ferry]

ignore what i said, i just refreshed my browser and it changed to 1.0 from 0.9 so its probably different now.

[waxen storm]

in Auto1111 is there a way to save a template with resolution,neg prompt etc so you dont have to retype after loading everytime?

[dapper ferry]

1024x1024 is native... 2K was just their to be honest crap upscaleer

or mediocore upscaler.. that any upscaler could do the same

[gilded trail]

lol... can you now check the downloaded image is in png or jpeg? and also if there any difference in resolution and file size?

[gilded trail]

nahh, with 0.9 the default download resolution was 2048x2048, im sure abt that. with the update it dropped down to 1024x1024.

[dapper ferry]

i clicked the blue download button and it gave 1024x1024 184kb PNG 96dpi 24bit

so thats small for a PNG must have been maximum compression

i guess

looks good i dont see any artifacts or color banding

[gilded trail]

is it! now im confused🥲

[dapper ferry]

actually the edges of things do look compressed. but it could just be the image

[gilded trail]

can you compare with your previous generated images that were in default aspect ratio 1:1 and see the resolution difference?

[dapper ferry]

SDXL is native 1024x1024 anything above that is just a standard upscaler. so put it through what upscaler you like

not got anymore time its midnight here

[potent spire]

The most obvious difference is that Midjourney is basically ready-to-go/out of the box AI image generator. You have a few options for generating. With Stable Diffusion you have much more customisation options but its also way more harder to setup

also you can use SD locally

with MJ that doesnt work that way, you only generate via Discord as of now there

[gilded trail]

is there any way that i can use the clipdrop 0.9 back again?

[fervent thunder]

[dapper ferry]

dont forget clipdrop has its own upscaler

[gilded trail]

i does but the 1024x1024 images that are generated after this update are not great. before the update im sure the default download resolution without upscaling was 2048x2048 and the images lookd more crisp and sharp.

[trim ocean]

Guys do i use the same sampler and steps for high res fix in comfyUI? Whats a good denoise strength, 0.3?

[dark plinth]

thanks mate

[potent spire]

You are welcome 🙂

[waxen storm]

can anyone tell me where to put this in Auto1111 'detailed_eye-10' its not a Lora?

[proud herald]

hey can anybody tell me how to set remote local access for stables warm with windows?

[waxen storm]

also what is this when using Loras RuntimeError: The size of tensor a (2048) must match the size of tensor b (768) at non-singleton dimension 1

[next sigil]

is anyone able to help me?

pytorch_lightning should be imported after torch, but it re-enables warnings on import so import once to disable them

[astral flicker]

whassup

does someone know the aspect ration of yt banners?

[waxen storm]

2560 x 1440 pixels

16:9 aspect ratio

[astral flicker]

lovely thanks can't open firefox rn; misconfigured my proxychain

[vast berry]

discord is dying right now how long until someone thinks it's just this server

[next sigil]

anyone have troubles after installing temporal kit? im getting this error:
ImportError: cannot import name 'auto' from 'tqdm' (C:\aiii\system\python\lib\site-packages\tqdm_init_.py)

and i cant find anywhere on how to fix it

[dense phoenix]

anyone have issues getting sdxl base to load in Automatic1111? The progress bar loads and then it goes back to the previous ckpt I had loaded

[broken smelt]

I'd ask in the github of the extension, the developer is your best resource for 3rd party extensions

[fervent thunder]

Anyone know what SD puts in the watermark on images it generates? I can't seem to find a definitive answer anywhere...

[next sigil]

what things need to be in the extensions file? do we need moviepy in there if we are going to be using temporalkit?

[foggy spear]

Is any1 using kohya on Collab to train SDXL content yet ? I have a question

[vast berry]

me sitting and waiting for the generation to come back on because i already voted on everything:

[waxen storm]

guess theres no point in asking questions here...nothings been answered in hours FYI

[red skiff]

Anybody still using colab? Just me?

Like it works, its free (0$), and uh it has alot of features, it runs on your phone

Maybe its laggy as hell

[eternal geode]

Colab is the way for the people who can't afford a 4090

[red skiff]

I sure can't

[eternal geode]

or a 3090, 80, 70, 60

[red skiff]

But like its free you dont have to pay a cent

[eternal geode]

People tell me "don't worry you will get one in the future"

But in the future the 4090 will be obsolete

8 years ago my dream build was a GTX 980 + i7 6700

And look at that now

Completely obsolete

[red skiff]

Well I don't even dream of that tbh

[eternal geode]

Plus nvidia prices are getting higher each year

With each new gen

[red skiff]

Ok cool

Why do people choose runpod or stuff over colab? Is it less laggy or sumthing?

[eternal geode]

Idk

Never tried runpod

[red skiff]

Me neither dont have us cash

[eternal geode]

You know

Idk what to say lol

It's been 11 months since sd came out

[vast berry]

it's because you're doing it in the wrong channel lol

[snow carbon]

I pay $10 for colab and another $10 to Google One for 2tb's GDrive . I know I could colab free but I think it's worth $

[next sigil]

where is controlnet on stable diffusion now?

[thick dragon]

are the bot -1 to 10 now running SDXL 1.0?

[warm junco]

Dont use 1.5 loras with SDXL

[junior peak]

comfy ui looking kinda basic rn guys how do i upgrade it to show the refinery options and base instead of the "load checkpoint"

[burnt haven]

does anyone have any negative prompts for architecture/buildings?

sxdl 1

[mighty pond]

For peoples info. The lowst gpu yet I tried to succesfully run sdxl fluently is a 3060 laptop/mobile gpu with 6gb vram!

and it was on auto

[empty star]

does InvokeAI have a feature similar to auto's hires fix?

[dapper ferry]

$10 per month for a 2 TB drive?

For $190 approx, you can get a 10TB segate external hard drive.

For $60 approx you can get at 2 TB external drive.

So after about 6 months google should send you the physical drive, because basically you've paid for it.

[potent spire]

So RTX 2070 would eventually work?

[pallid herald]

How, i get not enough vram on R5700 8gb

Also question

If buying new gpu for SD, probably better gp nvidia, instead of amd, even amd might get more vram?

[dapper ferry]

Dont buy a new GPU its a scam currently.

Nvidia are still messing around scamming consumers.

Its hard to know when it will change but they plan on sticking to the same B.S. of releasing a gigantic over priced 5090 and then probably wait an entire year before the affordable varients come out.

[mighty pond]

well technically speaking yes

[jolly gate]

does anyone here use vlad?

i need help

[mighty pond]

amd gpus are mostly not useable or at least harder to use

[dapper ferry]

Intel is adding new instructions to their CPUs... would this allow SDXL?

word is the RTX 5070 will cost $1500 and have 8GB VRAM

and will release xmas 2025

[maiden cliff]

is there any workflow on comfyui with roop and controlnet together?

[pallid herald]

If vram is true, its just insult to buyers

[pallid herald]

And suggestion is to do? Buy second hand?

I think maybe a 3060 12gb v4am would b3 good enough for now for sdxl

[dapper ferry]

My suggestion would be just to wait. until a true option becomes available due to consumer demand

[potent spire]

I will likely get RTX 4090 next year if 5090 doesnt come out

[dapper ferry]

Nvidia still needs to wake up and smell the coffee after the bonanza they had during the crazed pandemic era buying of gaming cards.

People are struggling to afford to pay the energy bills let alone go out and buy an overpriced gaming card so they can game all day

[pallid herald]

But then issue is, my r5700 seems not running sdxl, option is using something like runDiffussion, and paying monthly and its cloud, so dont like it as my generations are there :/ so maybe a 3060 12gb is a good placeholder, till.good options come.out or amd gets better support for ai generations

[dapper ferry]

is it really worth it? just generate in the cloud. and if you want NSFW stuff just view others work.

[pallid herald]

Its not about nsfw stuff, its general all workflows, like i generate image, in cloud i need to download or might get lost in rundiffusion or pay 35eur a month, not great price.

[dapper ferry]

clipdropio is £7 a month for me, and its like 5000 images every day and loads of other tools

https://clipdrop.co/stable-diffusion

i already have 10500 SDXL images on my computer now.

it would have been drastically more expensive to buy an nvidia card and generate it all manually

Im on a GTX 1070 and still waiting for Nvidia to release an appropriate viable upgrade. But still nothing worth a buy.

[tidal bolt]

invokeai and comfyui are great tools for inference. is there something with a ui for finetuning/training?

[iron fossil]

kohya_ss gui, onetrainer

[tidal bolt]

thank you. i will have a look at them.

[red skiff]

Hear me out instead of paying a cent
Use colab (except if you dislike lag)

[formal mauve]

guys how do I install stable diffusion on google colab

I've been following several tutorials the entire day

https://www.youtube.com/watch?v=0GXGnGp-2g0&ab_channel=EXNOZGAMING ---This one doesnt work for me

[finite cloak]

depends what UI for it you want to use? If you like SwarmUI, https://colab.research.google.com/github/Stability-AI/StableSwarmUI/blob/master/colab/colab-notebook.ipynb

[formal mauve]

https://www.youtube.com/watch?v=sHxv24ySgyk&ab_channel=AIMarvelverse -- This one works for me, but the problem of this is I cant put in my own models

[vast berry]

out of all the things that get in pantheon it's my stupid little gamer chihuahua

[formal mauve]

im using stable diffusion

[finite cloak]

stable diffusion is a not a UI

[formal mauve]

oh fr

[red skiff]

The hochiminh logo webui?

[finite cloak]

both the videos you posted are Automatic1111 WebUI

[formal mauve]

eyyy patrotic:))

[formal mauve]

ah alright

I didnt know that

[formal mauve]

so what do I do now? 😭

[red skiff]

=)) dm me

[fervent thunder]

Does anyone know if i can use Clipdrop: Stable Diffsuion XL for commercial uses?

[waxen geyser]

soneone here using a 3060?

[red skiff]

How to use free sd, if u no have pc:
Option1 : Colab
Option2: Playground ai (1000 imgs per day theres no fking way u would run out)

[narrow epoch]

Hi, do anyone know where to download SDXL beta (2.2.2 version)? I found good results from that model

[prisma pulsar]

You wouldn't download a car

[dusty tree]

what are the new best anime models

[red skiff]

I would

[snow carbon]

nah it's a service. my own HDD wouldn't be as convenient especially as a colab user. models that take minutes to download to Gdrive would take hours on my home Internet

colab is more then SD too. RVC, local LLM's. lots of models and training files to store in my Gdrive

[fervent thunder]

Can I save files of ControlNet Canny model into a folder?

if i use batch

[broken smelt]

are you on a paid google collab account? google stopped allowing SD for free accounts a while back

[formal mauve]

Nooooooo

[waxen geyser]

how is ur performance with sdxl?

[willow locust]

Outta curiosity. I was having a discussion to someone on GPUs and noticed that the AMD 7000 series performed quite good on this benchmark relative to 6000 series, is there a specific reason for this? https://i.gzn.jp/img/2023/01/26/stable-diffusion-gpu-benchmarks/01.png

[trim ocean]

whats your guys favourite sampler for base image and high res fix for SDXL? Using the new Dreamshaper model

[shy thicket]

where do i put the refiner ? In with the model?

[iron fossil]

good question lol, let me know if you find out

[pallid herald]

Would a 3060 12gb be enough to run sdxl at 1024x1024?

As my r5700 8gb is not enough it seems, failing with mem out error also its very slow

As 4070 cost 2x has same 12gb vram :/

[dry tendon]

Hi everyone, I need to find a model for flat illustrations. I tried using lora:Drawing combined with deliberate_v2e, and lora does a great job with faces and animals. However, it doesn't work well for objects and other things. In the example, I want to illustrate fruits in a classic style using hatching and outlines. Currently, the test model's bot does an excellent job with the hatching style, but it's not available yet. Does anyone know of a model or lora that works well for this?

[iron fossil]

style LoRAs should be possible, but the captioning techniques might be different than they would be for objects. not sure if you accounted for that / how much you've played around

failing that, maybe Dreambooth ?

[dry tendon]

I`ll try to add my prompt to the images, give me a few minutes

[warm junco]

Yes easily

[barren raptor]

Hi, I am trying to understand if SDXL can be used to create art that can be used commercial purposes like online ads. Please clarify if you are certain of the answer. Thank you

[dusty tree]

how does the vae fr mse 840000 ema pruned ckpt work

[karmic brook]

cut your prompt down!

[dapper ferry]

what is the token limit?

[pine fulcrum]

so

i put somthing in the negative prompt and it shows up more often

the fuck

[near wagon]

what specifically did you put in the negative prompt that shows up more now?

[proud mountain]

hi @karmic brook i wanted to make a link to Stable Foundation in the deforum discord

we were thinking of sharing the invite link and a cover image

[karmic brook]

[proud mountain]

would you be interested in helping me with this?

[karmic brook]

Sorry if I'm misunderstanding-- What exactly is there to help with?

The link and img together are both here: https://discord.gg/stablediffusion

or do you mean formatted something like we have in this channel: #1072229057707659404

[proud mountain]

yeah the #1072229057707659404 stuff looks really good

we were thinking of using the new gallery channel

so it would be like a cover image then info inside when you click

[karmic brook]

oh, thanks! I made that channel using https://discohook.org/