🆕｜sd3

[dull star]

please don't make it super stylistically aligned or whatever 🙏
I love the current look of 8B no joke

[frail shoal]

There are very knowledgeable people in the open source community. They are really great. If they can't fix it, but are doing fine with other models. It simply means that Sd3 has a problem.

[craggy crest]

you just asked a question you don't want an answer to, you'll be here all day reading the response.

[bitter hearth]

lick Steven

[odd basalt]

How quickly it turned into hbo 😂

[desert garnet]

yea unfortunately the one you were talking to wasnt one of them 😔

[dull star]

https://github.com/redhottensors/ComfyUI-ODE people are already working on samplers for SD3, these work perfectly

[frail shoal]

You are a fanboy. You are not thinking clearly

[mortal mesa]

the boffin's dont even know how this works, it "just does"

[craggy crest]

i am a programer. try again

[desert garnet]

the new BS programing language

[frail shoal]

Doesn't prevent you for being a fanboy

[craggy crest]

how many hours do you have in on working with stable diffusion since 1.5 was released? how much of the code have you read? how many loras and other things have you trained?

[bitter hearth]

SD3 does balls perfectly, it's the best model out there at the moment

[dull star]

what colour is your bugatti?

[desert garnet]

how many confirmed kills do you have?

[craggy crest]

bugatti colored

[dull star]

wait... no joke? really?!?!?! (not being sarcastic, I want to know)

[mortal mesa]

European or African swallow

[dull star]

oh you mean normal balls 😅

[bitter hearth]

You can do pretty much any ball

[dull star]

totally wasn't thinking of something else

[odd basalt]

Try madballs prompting

[bitter hearth]

[craggy crest]

yeah, it does spheres very well

[dull star]

did previous models have issues with spheres?

[odd basalt]

It does landscaping perfect

[gusty trail]

I want the sun hits the earth

[bitter hearth]

Icey has a lora for them, the model does know them a bit

[frail shoal]

I only use models, create merges. Don't have experience with fine-tuning though. But I have been using open source gen AI for more than a year and a half now. And I have tried and worked with almost all models. I have tried to do the same with Sd3 but I felt it was not a skill issue. Something was broken , but I can't pinpoint what is causing that

[dull star]

oh I forgor

[craggy crest]

i see. talk to me after you learn python and have programmed for a while then

[dull star]

SD3 and SC users seeing the license changes: https://tenor.com/view/frank-reynolds-danny-devito-oh-my-god-i-get-it-its-always-sunny-in-philadelphia-gif-1719969854406009304

[odd basalt]

There’s public data sheets for them on huggingface

[desert garnet]

https://tenor.com/view/bad-ass-uh-oh-ndt-neil-neil-degrass-tyson-gif-14802463180959376587

[bitter hearth]

But then we'd die

[frail shoal]

I will not, talk to me when you grow up a bit.

[gusty trail]

Just in SD3, we won't

[dull star]

its unfortunate that SD3 2B is still more intelligent in my testing than pixart, even though pixart was such a massive improvement over SDXL (and its just a 0.6B DiT model, not ~2.6B Unet!)

[bitter hearth]

Don't think this should be the size

[dull star]

correct scaling compared to the earth 👌

[craggy crest]

put the sun back where you got it from, please.

[bitter hearth]

[desert garnet]

nah leave it there he hasnt seen it in a while

[bitter hearth]

What is the sun

I got one even smaller, that's funny

Size difference

[frail shoal]

It's more intelligent in some prompts but pixart is better in some other prompts. Also pixart gives better dynamic pictures like midjouney. When Sd3 is stiff. Align your steps does wonders with pixart sigma. It literally changes everything.

[desert garnet]

fake news created by the media,earth is a cube there is no sun 🧠

[dull star]

mostly compositionally? that makes sense, as SD3 was probably finetuned on typical portraits

[frail shoal]

I wanted to implement align your steps with Sd3 too but it didn't work

[bitter hearth]

Earth is gone no more Steven Seagal

[craggy crest]

vampires rarely see the sun...

[dull star]

according to alex (mcmonkey) they could easily retrain it to get the correct values for SD3 (don't exactly know what he said)

[mortal mesa]

something money money something

[dull star]

AYS won't work out of the box, but can be retrained for SD3

for AYS they published a unique list of sigma values to use per model arch, so they'd need to rerun it and get a list specific to SD3

this is what he said

[odd basalt]

Ppl can say anything but it’s about Action and seeing it done…. Talking about it gets nothing done …

[frail shoal]

Yeah but pixart sigma authors did not train it to be compatible with all these tools. I literally showed taught the authors that it was compatible with so many things . They did not know

[mortal mesa]

ah yes he and others are part of a direct action group

[frail shoal]

So I'm wondering why is Sd3 so rigid

[dull star]

because it has flow matching

[bitter hearth]

Rigid for what

[dull star]

for other samplers and schedulers

[craggy crest]

a glass jar full of a dark horse with a fiery mane in the desert. The jar is placed on the sand and we see the ocean behind it

[dull star]

it needs ODE samplers

[gusty trail]

You just need to learn more about the architecture

[fleet meteor]

@bitter hearth you use the 8b version right? If so can you test this prompt "photo on the street, its raining potatoes", sd3 2b couldn´t understand that concept

[bitter hearth]

@desert garnet

[dull star]

schlatt beard

[bitter hearth]

He's alive

[fleet meteor]

I wonder how it would be possible to train more complex concepts like this, its dumb but other AIs can generate it

[frail shoal]

Anyway, I really hope people figure out what is wrong with Sd3 and fix it. I'm really eager to use it.

[odd basalt]

Just use it…. It’s fun

[bitter hearth]

weird

[fleet meteor]

F its still don´t know that concept

[frail shoal]

I have, but there were so many problems. And I could be brothered anymore when there were other alternatives. Soon we are going to have 16ch vae for sdxl

[gusty trail]

There is no wrong about SD3 couldn't use something others could. It is just because the architecture is different

[frail shoal]

And most people are going to use sdxl with 16ch vae

[frail shoal]

I hope so

[bitter hearth]

Probably better to not use the word rain, that is hard trained to mean water and stuff

[odd basalt]

I don’t have a issue with it

[desert garnet]

wait theres a 16ch vae for sdxl? give 🫱

[fleet meteor]

https://tenor.com/view/cat-gimme-that-gif-3302967658968126593

[frail shoal]

Sure, but I mostly do characters, and weak composition is a letdown for me. Also bad anatomy. If it had good inpainting than yes. But it somehow feels wrong to inpaint it over with sdxl when Sd3 has such a good vae. You would crush all the details

[bitter hearth]

[frail shoal]

Ostris just finished training a 16ch vae for sd1.5 and is finishing doing the same for sdxl. You have to be patient a bit but is going to come soon

[torn wharf]

star balls

[loud monolith]

A currency note with Nita Ambani’s face on it

[frail shoal]

16ch vae with a lora adaptor which will work on your favorite finetunes.

[torn wharf]

is that finished? i heard the vae was complete but then he plans to train sd15 as a new base model that uses it. which will then have to be refined again

[odd basalt]

Art is how u make it… it’s not weak to inpaint anything or fix by drawing by hand…. Ai in general is not 💯 percent…. So we have other tools like we do with photography to fix those errors

[frail shoal]

No he is creating an adapter as a lora. The lora will make every model compatible

[torn wharf]

interesting. wonder how that's going to work. that ssounds like architectual black magic

[bitter hearth]

Everything is black magic for my brain

[odd basalt]

Sure does sound voodoo server magic 🪄

[frail shoal]

I use photoshop to fix all my pics. But then again , when you have better alternatives for what you are looking for you just go to those. It's just easier

[torn wharf]

so the lora will extend the 4channels x4? i think the base model having the knowledge, not the lora, would be better

[loud monolith]

A currency note with this face on it

[odd basalt]

That’s why I keep saying SD3 and Photoshop are the best other alt is procreate

[craggy crest]

why are you guys tossing potatoes at people

[desert garnet]

average rainy day in Ireland

[torn wharf]

this server doesn't have a bot. you'll have to find a place to generate that yourself

[odd basalt]

Shh we are the bots 😂

[mortal mesa]

man wrong chat

[frail shoal]

They found out that you need only to train 2 layers. In and out. Than making a lora adaptor to make it work with everything. I'm not sure how the lora works. But they guys says it's working well. They have had trouble doing the same for pixartsigma but still experimenting

[frail shoal]

I don't know. You have to talk with the dude.

[torn wharf]

sounds hacky if the base model doesn't really know more than 4 channels

[odd basalt]

Same I have no idea 🤷‍♀️ 😂

[gusty trail]

same and I don't know

[frail shoal]

It's going to extend sd1.5 to gen pics at 1k res

[torn wharf]

i feel like you could get a vga port to work on an hdmi monitor with an adapter, but it would still only put vga over the wire

[frail shoal]

I didn't really pay attention to it. But he is sharing his progress in the Pixart discord. Also his Twitter. So you could ask him

[gusty trail]

I think the point is, you need to adapt the vae first then train more data with the new vae to get better, might be?

[torn wharf]

the apadter will apparantly work with any refine, and you can't use it while training

training scripts will have to be rewritten to use a lora adapter

seems like a lot of smoke and mirrors right now. i've been watching how it develops so far, but the lora adapter is new and is concerning

[supple gulch]

A surreal scene where potatoes are falling from the sky like rain. The backdrop is a cloudy sky, and the ground below shows a cityscape with people looking up in amazement. The potatoes vary in size, and some are mid-air while others are about to hit the ground. The lighting gives the impression of a real, bizarre weather event.

[lavish osprey]

Prompt for the last 2 cats?
2b or Ultra?

[torn wharf]

the death ball is fully operational

[fleet meteor]

Nice!

[bitter hearth]

""i am bot" cat with formal clothes says, the cat is a robot"

Normal 8b, yes my prompt doesn't make much sense im in bed very lazily

[torn wharf]

lazy prompts give great results

[bitter hearth]

What is your prompt

[torn wharf]

your prompt with my lora loaded. it makes themlook foam an scuplted

[torn wharf]

added "monster skull" to the prompt for better lora evocation. ❤️

[bitter hearth]

Pineapple pizza textured ball

[frail shoal]

Apparently adapting sd1.5 to 16ch vae it's easy.

[bitter hearth]

Is 16ch vae gonna make sdxl use a lot of vram or something

[fleet meteor]

Probably, sdxl uses more vram than sd3 in my PC

[desert garnet]

1.5 getting better and better

[bitter hearth]

No ponies for you on 1.5

[desert garnet]

there are ponies there but they look thicker

[bitter hearth]

Sale

[craggy crest]

What i find funny is that someone will come into this channel, see all the really good images being created with SD3 medium - and then post how sd3 medium can't create images that are any good

[bitter hearth]

I really wanted to build on that rust

[fleet meteor]

Niiice

[odd basalt]

I

[bitter hearth]

Angy

[odd basalt]

Ghostly

Cute

[lavish osprey]

[torn wharf]

thats a spiky ball

[bitter hearth]

Admit you love them

[uncut river]

ModelSamplingSD3 set to 4.0 or something

[craggy crest]

wrong sampler/scheduler

[uncut river]

no

sd3 + sd3 vae + normal euler. always works. not this time with strange prompt and high setting for ModelSamplingSD3

[craggy crest]

to quote comfy "shift numbers that are lower focus on details. shift numbers that are higher focus on structure" - euler does not work well with SD3

[uncut river]

"dark lines art, line art, many lines, flat colors, colorful, abstract fractal art, small houses left over after the apocalypse, apocalyptic"

[craggy crest]

try heunpp2

[uncut river]

euler works fine for me with sd3

try my prompt plz?

neg part is: "monochrome, (3d, real, photorealistic, realistic:1.1), frame, paper, painting"

[craggy crest]

i've got it running, give me a second

[bitter hearth]

(:1.2) stuff doesn't work from what I know

[uncut river]

with heunpp2

sure it does

[craggy crest]

if etch-a-sketches were in color, you could almost have made that with one

[uncut river]

no, that is not enough. we need at least TWO, 2, balls

[bitter hearth]

Didn't prompt for a ball and got a ball, that's a win

[craggy crest]

this so needs to be a style lora

[uncut river]

if it can be done with prompting, for sd3, no lora required !

[frail shoal]

Sd3 it's a great model. However how great is a model It is always in comparison with other models. Look at imagesys. It's not only me saying this.

[raven fern]

yea but maybe people want it for non sd3 stuff too 🙂

[bitter hearth]

Loras are fun, you can do weirder stuff with them on

[craggy crest]

sure, but for people that don't want to have to prompt, just say something like "house" or "cityscape" - a lora would be good

[craggy crest]

are you seriously trying to continue this argument?

[bitter hearth]

Sd3 is the best model

[uncut river]

sd3 sucks in many ways, but its also better in other ways. we'll see in about 2 weeks,

Still at the moment I love to generate SD3 images

[raven fern]

[bitter hearth]

[craggy crest]

nothing's going to change in 2 weeks

[frail shoal]

You were asking for it

[craggy crest]

i'm not asking you for anything. once you actually put several thousand hours into learning how to talk to stable diffusion so it understands you, can get what you want out of it every single time, don't have to use negative prompts, and have learned how to program, then i'll talk to you - and i mean all SD models, including SD 3

[uncut river]

negative prompt is a feature, not a requirement

[craggy crest]

sure, but someone that actualy knows what they are doing doesn't need to use a negative prompt

[frail shoal]

So everyone in imagesys is wrong and you are right. Good job

[craggy crest]

maybe

[uncut river]

weight as in (night:0.01) or (night:0.95) still works, maybe less than before

[fleet meteor]

AAAA what was the prompt?

[bitter hearth]

To whoever says bad stuff about THE BEST model: this this

[uncut river]

thought prompting always wins hands down!

[odd basalt]

Now u all are just showing out 😂

[bitter hearth]

Just lykon said, prompt engineering is not a job, you really won't expect my prompt kek

[uncut river]

set to (weird dark evil cyberpunk, warped sci-fi style, darkness shadows and evil:0.1)

[bitter hearth]

@fleet meteor

The prompt is: Ultra, giga, graphics on ultra, gaming, absurd

[frail shoal]

In my opinion healthy criticism is good for any project. It pushes the project to be better instead of stagnating. Also these tools are for the community to use, so if they don't know how to do it well then what's the point. If you have a secret sauce for Sd3 than please share a guide. I'm looking forward to using it as my main model.

[raven fern]

nice Kolors

pun intended

[craggy crest]

sure. providing you were actually providing healthy criticisim instead of obsessive negative comments and ignoring everything in front of your face. you could be using it as your main model right now, if you really wanted to

[lavish osprey]

please no fighting, just share cool images

[craggy crest]

sorry

[raven fern]

not even pillow fight?

[lavish osprey]

only if sexy enough

[raven fern]

🙂

[bitter hearth]

not sure if the image elo arenas are as useful as the original LMSYS Chatbot Arena

"muscular men fighting with pillows"

[raven fern]

take those elo arenas with a grain of salt tho, it's not always necessarily accurate or representative

[bitter hearth]

LMSYS Chatbot Arena actually does seem to be pretty accurate
i.e. it correlates well with other benchmarks
but it is done by a large and good team
not sure the image ones have had the same amount of effort

[raven fern]

im sure there are some good ones, but i also heard about some bad ones, i think at the end of the day, it's up to the end user to decide what is best for them

(in other words, trying multiple models anyway, not just "the best one")

[raven fern]

now where is the bikini version?

[bitter hearth]

making a good elo system is actually really hard

[uncut river]

"Oh wait, dont land here! that thing looks sharp. Plz float this thing somewhere else!""I cannot. The wind""Ohw. No."

[raven fern]

meow

[raven fern]

i remember floating islands being a struggle for some models

[uncut river]

its classic fantasy art!

should be mandatory for all models

[raven fern]

yea

[bitter hearth]

youre food now

[raven fern]

i know im delicious 🙂

[bitter hearth]

food for the dogs , tastes too bad

[raven fern]

[torn wharf]

they said pillow fight , not pillow flight

[craggy crest]

no cooking kitties

[uncut river]

sorry, forgot about the pillows

[keen rose]

generate photo of a white dog

[craggy crest]

you can only do that in the artisan channels. click here #🗣｜artisan-support-feedback

[uncut river]

just throwing random chat stuff into the prompt, see what sd3 can handle

[craggy crest]

here https://docs.google.com/spreadsheets/d/1bdidA4w5pB2BQMyxhkFu710Nu5bzKM1E-Wh_38ZZlT0/edit?usp=sharing grab stuff off that

[raven fern]

https://tenor.com/view/ups-cute-kitty-lets-play-gif-22863023

[odd basalt]

Thanx I’ll check them out I mainly do horror art

[craggy crest]

make a copy for yourself, i change it just about every day

[sage burrow]

@torn wharf can I ask you which program you used to create your lora, and what you used for the tagging? I finally want to create one now 😄
Though I'm going to use a rented gpu since mine is only 8gb. I'm guessing you have 12 gb vram or more?

[compact niche]

'

[odd basalt]

One of ur prompts

[craggy crest]

came out really nice

[torn wharf]

one trainer is what i used. i prefer working with bmaltais ui for kohya but onetrainer got support out first

[odd basalt]

Thank u

[torn wharf]

the tagging i just do myself. i tried a couple different vit models and a blip thing, but they were crap captions. 150ish images in the dataset

[sage burrow]

Thank you 🙂
Does that mean kohya does as well? I remember trying kohya for sd3 before but the github branch was broken 😦

I have to say that how much you got sd3 to meld to your style is inspiring

[torn wharf]

i haven't checked on the progress of kohya for a few days. i like the bmaltais ui and will wait for him to update

[sage burrow]

I figure I can create and tag inages while I wait for my new modem to come in the mail ☺️

[craggy crest]

this time, don't spill coffee on it

[torn wharf]

i'm not even really trying. i just threw a high lr at it. i think it trains better for classes that are not so broken. like i can't train a person but i've trained balls

captioning i sort of went with a mix of comma seperated tagging and descriptive language

[sage burrow]

I used my cell phone auto to tag a bunch for an sdxl lora before.... I thought I'd take it more seriously this time lolololol

[torn wharf]

when 'm more motivated i might take a second pass on captioning. expand the long descriptiosn. see how that works.

there's an extension for auto1111 i've used to fiddle with captioning. its useful once you figure out it's quirks https://github.com/toshiaki1729/stable-diffusion-webui-dataset-tag-editor

[strong wagon]

Лев

[sage burrow]

Thank you 🙂

[dull star]

didn't astralite make a finetune or custom image captioner

oh they are making/finetuning one

[indigo tangle]

why

i wanna see

[dull star]

so that it can do natural language with characters that most models couldn't recognize

[sage burrow]

I put that name into Google and just hot reddit posts 😭

[dull star]

such as.... ponies....

[uncut river]

sd3 medium

[lavish osprey]

strong and powerful dungeons and dragons epic movie poster barbarian woman with cape charging into battle violent roar riding a vicious ice [wolf|tiger] beast leather and fur boots warriors and red banners (windy dust debris storm:1.1) volumetric lighting fog depth mist pass z pass great stone castle very bright morning sunlight from side

His wings are large, white, and feathery, spreading out behind him in a display of divine grace. He wields a glowing, celestial sword,```

anime screencap standing on the top of a building, short hair, bangs, black hair, long sleeves, 1boy, closed mouth, standing, collarbone, jacket, white shirt, outdoors, open clothes, collared shirt, belt, pants, open jacket, coat, black jacket, grey eyes, dress shirt, night, black pants, ground vehicle, motor vehicle, black coat, hands in pockets, shirt tucked in, car, lamppost, neon lights, cinematic, retro

[uncut river]

did someone say pony?

(its sd3)

[sage burrow]

SD3 made me one of those today as well (unprompted)

[torn wharf]

problem with any captioning model and ball photos, is these aren't your normal characters and image captioners don't really know how to describe them

[odd basalt]

I have a clashing one I made in sd3 with venom love that universe

[sage burrow]

That's a good thing, imo.

[craggy crest]

search for that name on here then

[torn wharf]

here's copilot's take on a ball

[torn wharf]

it does recognize it as a mad ball at least

[craggy crest]

no more coffee for you today

[torn wharf]

fixed

[bitter hearth]

I think ultra has noise injection

[bitter hearth]

look at the fine, yet chaotic details on the flowers lykon posted

seems like noise injection to me

its doing a lot more than that but that's one part I think

[dull star]

uhhh

[craggy crest]

why are all the images with a moon in them that stable does always a full moon?

[bitter hearth]

even GANs like to do a full moon

[bitter hearth]

or transformers like SWIN-IR, they like to do a full moon too

[sage burrow]

Others beside SD have that problem as well, I remember it being difficult to get other moon phases.

Btw, blood moon results are fun! 😄

[craggy crest]

i can just imagine what you might get with blood moon in the prompt

[bitter hearth]

gibbous phase moon will be hard

almost never see art with gibbous phase (3 quarters)

[bitter hearth]

I noticed that clownshark's comfy workflow has a lot of noise injection stuff too
I think this is the way to go

[craggy crest]

lots of public domain photos from nasa

[bitter hearth]

but the question is will the model be able to generalise that to other art style

sometimes it can do this with a subject and sometimes not

[lavish osprey]

link?

[bitter hearth]

I think the images he posts on this discord have it

but its two nodes

[sage burrow]

Usually amazing dark clouds and dark red ponds through the forest 🙂

As long as you don't add zombies to it that is

[bitter hearth]

clownsampler and sharksampler

[odd basalt]

He gave me his links

[torn wharf]

egg shaped moon

[lavish osprey]

heh happens

[bitter hearth]

ah i found the repo
https://github.com/ClownsharkBatwing/RES4LYF

[craggy crest]

different prompts to each encoder

[lavish osprey]

ty

[torn wharf]

🪄

[bitter hearth]

Clownshark has by far the best colours I have seen from any model/workflow
so its something in that workflow but not sure what LOL

[odd basalt]

https://github.com/ClownsharkBatwing/RES4LYF

[bitter hearth]

ye that's it

[bitter hearth]

#nsfw

[odd basalt]

@lavish osprey ^

[torn wharf]

only because mlp has implications

[torn wharf]

#eggmoon

https://tenor.com/view/moon-dr-who-gif-14327794

[uncut river]

is that some kind of img2img or pure ai?

[torn wharf]

thats dr who capaldi season

[bitter hearth]

this noise injection example is good
https://old.reddit.com/r/StableDiffusion/comments/1clg4i3/improving_sd_precision_with_more_noise/

but there are better examples out there
I saw one in japanese somewhere, the noise made it way more detailed
its hard to get it right

[upper snow]

Mode collapse? On a GAN? Say it ain't so!

[bitter hearth]

I'm one of those people who still thinks GANs will eventually overtake diffusion models

[hazy kestrel]

futurama?

[lavish osprey]

a cute baby chicken snuggles up in a blanket of short fur. idyllic background, out of focus

[torn wharf]

you can sorta see the shape of leela's boobs its sewwww hot

[craggy crest]

That's no moon. That's a space station

[bitter hearth]

StyleGAN2 is a GAN from 2019
and its so much more realistic than any diffusion image I have ever seen

[sage burrow]

via glif

[craggy crest]

there was a paper on GANs recently...

[torn wharf]

ultraball

[bitter hearth]

its a hotter topic than that, there's over 100 GAN papers per month

[hazy kestrel]

best practices for prompting with three text encoders? different prompts in each section l/g/t5?, which words/tag should go where?

[upper snow]

All I know about GANs is that they are annoying and fragile.

What discriminator should I use for finetuning the SD3 VAE? Call it a frivolous waste of GPU time if you want, but if I can get a small reduction in reconstruction loss, I want it.

[torn wharf]

experiment. it's the frontier of prompt engineering. i dont think there is best practices.

I've done style prompting on L and composition on G then a novel on t5

[upper snow]

Would be even better if somehow the weird high frequency noise that's prevalent with SD3 across most images I've seen and made can be eliminated.

[torn wharf]

lately i'm just doing balls so i just prompt one to all three

[craggy crest]

turn time_step shift up?

[upper snow]

I don't know if that's the issue. I think it honestly might be more related to CFG, since unconditional images show it a lot more.

[fleet meteor]

So cuteeee

[torn wharf]

apetite warning

[craggy crest]

this is what i use for settings, they seem to work well

[torn wharf]

ballz 400!

[bitter hearth]

they found some new tricks for GANs over the years

for example you want to avoid mode collapse, where the generator has low diversity, so what you can do is have the discriminator compare in batches, which lets it know if the generator is using low diversity

[torn wharf]

actually really love the logos that sd3 can do

[bitter hearth]

[torn wharf]

not for baby weak stomachs . APETITE WARNING. Only for monster eaters WARNING

[craggy crest]

needs cheese

[uncut river]

eater ill eat that or it 'll eat me

[bitter hearth]

there's lots of weird advantages to GANs like they are better for style transfer
https://junyanz.github.io/CycleGAN/

this GAN is from 2017 and is way better than any IP adaptor or control net

[bitter hearth]

Wait a minute

[torn wharf]

ballz with cheese order up

[sage burrow]

I think the AI has been partaking today

Well I guess there's this approach as well...

[sage burrow]

Wings go on the back 😦

Thought I'd join in the umbrella theme 😄

[bitter hearth]

[bitter hearth]

Ball

[odd basalt]

Whats up with civitai… site seems broken

[bitter hearth]

It is always broken, what do you mean exactly

@odd basalt your ball

[craggy crest]

they've been having site issues for several days

[craggy crest]

3d popup book: mountain landscape: 3d rococo string art Whimsical sunstreaks, flower petals, symmetrical blue sky, fluffy puffy sunset clouds; by artist "Karol Bak"; by artist "Jacek Yerka"

[odd basalt]

😂 they are leaving red notes saying they are aware their site is broken basically… I was gonna download some SDXL and I am greeted with all their broken sticky notes ,… awwwe I love it thank u 😊

[odd basalt]

I think it’s been on going … they over charge and ever sense they did some stupid site change… their site has gone down hill …

[oak hamlet]

is it even possible to download sd3 right now?

[uncut river]

https://huggingface.co/stabilityai/stable-diffusion-3-medium

[oak hamlet]

damn

i thought i couldnt with the civit ai announcment and stuff

[uncut river]

depending if civitai has more than 1m revenue, they might need to discuss terms for sd3 license

https://huggingface.co/spaces/stabilityai/stable-diffusion-3-medium

[craggy crest]

not too surprising, really.

[craggy crest]

you're assuming that civit will be around in 6 months

[uncut river]

yes

[bitter hearth]

how would a popular site eever die out

[craggy crest]

businesses kill themselves all the time

[bitter hearth]

More balls

[torn wharf]

balls

[sage burrow]

Myspace.

[sterile pendant]

Wish dot com was probably exponentially more used until temu knocked them out of the cheap flea market quality item market

[craggy crest]

if they didn't sell junk, they wouldn't have lost their customers

[sterile pendant]

basically, but it was a hit it and quit it type business, they knew it was just a bubble and had no intentions of staying in the game

oh damn, wish peaked to around 22 billion and myspace peaked to 12 billion(somewhere around 18 billion if you adjust for inflation)

it really was bigger than myspace

[craggy crest]

should probably move this to #💬｜general-chat ?

[sterile pendant]

who cares

they can delete or ban if it bothers them

[craggy crest]

i was thinking more of the rabid users that might jump on you. shrug

[sterile pendant]

oh i dont care about online hecklers. i just assume 99% of them are either people like asmongold or people like the WoW southpark guy

[supple gulch]

SD is still amazing,Civitai It is reasonable to pay after more than $1 million for SD3, if SD3 does not survive how to provide the community with a basic update. Either make a mutually beneficial deal, or civitai moves on to something else. Unfriendly slogans are unnecessary such as too big BAN word

[hallow lion]

even facebook could disappear, history is full of too big to fail companies that are gone

[bitter hearth]

literally looks like a pixar movie 🙂

[torn wharf]

yahoo was a monolith

[craggy crest]

yahoo still exists

[mental field]

#🏞｜general-with-images tiger gold ring design

[tight bay]

Window in to life

[mental field]

#🏞｜general-with-images tiger gold ring design for men fancy unique style with diamonds

[mental field]

#🏞｜general-with-images tiger gold ring design for men fancy unique style with diamonds

[limpid drum]

In case you didn't know, you can't create images in this channel. I did it locally for you because I was bored 🙂

[mental field]

oh sorry didn’t know… thank you by the way

[dreamy ether]

Cat

[limpid thunder]

Thank you for using comcom analytics.
"comcom analytics" supports all community managers (moderators and server owners) by stats, visualization, and analytics.

If you have any questions, feel free to ask us!
Your dashboard
Help
Support server

Other languages
en: help
ja: help Japanese

[dreamy ether]

Help

[limpid drum]

All good, try the Artisan channels (I believe it's a trial first, then you pay)

[odd basalt]

It’s actually not that popular, u can look at the net traffic and greed with buzz… other sites are popping up like Temu effect that are free… site has been broken for a while so ppl move on…

[verbal epoch]

Can I use it A1111 with SD3

[hallow lion]

i tried - it did not work 🤷‍♀️

[desert garnet]

what a skill issue 😔

[hallow lion]

my skills have failed me

[desert garnet]

its because you didnt caress and respect a1111

[sage burrow]

Ouch! lol

[craggy crest]

It was announced yesterday that auto1111 now supports sd3

[cursive mica]

mulher

[lucid swift]

animated this sd3 image

[verbal epoch]

Which model do I need to use it

I have the one where you need to download the other files manually

[craggy crest]

https://x.com/Lykon4072/status/1809583921526387078 read through Lykon's annoucement on twitter

Lykon (@Lykon4072) on X

SD3 support has been added to Auto1111 webui.
Only Euler sample (for now) and t5 disabled by default (can be enabled in settings).
Good job!
https://t.co/uI9hrPPYwP

Twitter · Jul 6, 2024, 1:43 PM

[verbal epoch]

It does not say which model needed

[verbal epoch]

I dont know which I need

[craggy crest]

i don't know how much Vram you have, but the fp16 is for 16g vram, the fp8 is for 8g vram. the ones that say "inc_clips" are models that have clip included. i would suggest you start by loading sd3_medium_incl_clips.safetensors, run it, and see if you get errors

[verbal epoch]

I have 24gb vram

[craggy crest]

still use that one, it's the most generic, and should run without issues. try it, if it gets errors, post in #🤝｜tech-support and someone'll help you figure out what isn't set up right

[verbal epoch]

Okay thx you

do I need to activate this then

[bitter hearth]

Booba

[lavish osprey]

how did you get this specific style with 2b?

[proper lichen]

Booba

[sage burrow]

https://glif.app/@LadyLalita/glifs/clybqmfud000481x3ljvzhmwo

I think I need to add SDXL at the end to fix fingers 😦

[bitter hearth]

Balls don't have that problem, balls are superior

[sage burrow]

Cuase ballz just leave the hands out! 😄
I Haven't done a lora yet, though I fear for the hands when I do!

[bitter hearth]

[bitter hearth]

Lying on grass meme

[uncut river]

(modern anime style:1.03), cute 24yo woman wearing intricate anime dress, white dress with black waist band, intricate fabric, clothing ornaments, nose, long wavy blonde hair, black hair ornaments, orange eyes, pink roses, blossom, classic japanese buildings, (foggy sky, sunny:0.8), high contrast, (sharp lines:0.76), (2d anime illustration:0.55), (realistic lighting:0.86), (shy, blushing, head tilted:0.7), (half body shot:1.05), (big boobs, cleavage:0.76), soft glow, (face close up:0.6), 8k, intricate masterpiece, highly detailed, (fear, surprised, big eyes, disgust:1.2)

[sage burrow]

gloves make all the difference in the world!

[bitter hearth]

now less furry

[sage burrow]

I should make a gliff, where I sneak things into the code that guarantee furryness 😉

[bitter hearth]

[bitter hearth]

MOAR LIGHTNING

Mont Saint-Michel

[sage burrow]

My SD3 img2img glif is ready 😉
https://glif.app/@LadyLalita/glifs/clybum9x8000314e2uym1es80

[sick cedar]

Any reason why the T5 encoder is disabled by default?

Does it support offloading to the CPU? I wonder if Forge will update to this..

[sacred jewel]

Try them all until one breaks? 🤷‍♂️

[sacred jewel]

Balls have Spherical Problems.

[odd basalt]

I would def fix it to perfection in ps or procreate

[odd basalt]

My scene girls

[terse belfry]

Batman Gameplay
SDXL vs SD3

[lavish osprey]

sd3 2b or 8b?

[muted dove]

Looks like 2b

[sage burrow]

Generative fill to the rescue soonish

[fleet meteor]

Yooo what version of sdxl? it looks good!

[terse belfry]

Left image is SDXL

The right one is SD3 flash

[fleet meteor]

Yes but i mean are you using a finetune model?

[terse belfry]

No, its basic SDXL, no finetune

[fleet meteor]

wow

[muted dove]

What is SD3 "flash"?

[lavish osprey]

distillation

similar to turbo

[muted dove]

Oh, I didn't know there was one 😄

Based on 2b though?

[lavish osprey]

yes

only 2b has been released, so

[sage burrow]

LOL

[muted dove]

I didn't know if there was an API flash version 🤷🏻‍♂️

[craggy crest]

not a clue. @lavish osprey might know

[sage burrow]

https://huggingface.co/spaces/jasperai/flash-sd3 a handy version

[signal shuttle]

Is there a use case for SD3 flash?, in my case almost all of the images are worse then normal 2B SD3

[dull star]

is it at least still more intelligent than previous fast models? (sdxl turbo/lightning/hyper for example)

that would be the only reason

[terse belfry]

Batman gameplay screenshot

SDXL x SD3 x Dall-E3 Comparison

[dull star]

I like SD3

is that 8B or 2B

[terse belfry]

2b

[dull star]

nice, it looks like some console game screenshot

[terse belfry]

I like the mood of sd3, but the anatomy seems off. The batman is kinda ...short.lol

[craggy crest]

that's the warping and shortening effect that any subject not right in front of the camera suffers from. you might be able to sort of fix it by putting "tall and skinny" in the prompt

[terse belfry]

Thx. I'll try that!

[bitter hearth]

No, he is just short and fat. its not a camera angle thing

[odd basalt]

Def looks like a prompt issue

[bitter hearth]

he needs to lose some weight!

[odd basalt]

Older comics

[bitter hearth]

fatman

[sage burrow]

Now I want to make a glif that turns everyone into zombies 😉

[fleet meteor]

sd3 2b lighting is so good

[odd basalt]

2b SD3

SD3 does gore so well for us horror fans

[uncut river]

yeah, it does body horror by default 😄

[vestal cypress]

they accidentally made a model for scp generation

[terse belfry]

I wasn't even trying. I just use same prompt for all 3 models for a test.

No need to be toxic

[upper snow]

add the king charles portrait to it

[uncut river]

Is that also an English saying ... translated "with the knife on the table" ?

"Next game gentlemen!"

[odd basalt]

I’m chill did not mean to to sound harsh… ❤️

[vestal cypress]

idk what this is

playground-v2.5-asethic

[vestal cypress]

she like me fr

[sick cedar]

@lavish osprey Do you know why that decision was made? Other than high Vram usage? #🆕｜sd3 message

[terse belfry]

Looks nice, love it.

[terse belfry]

Models test comparison with the prompt : Batman character concept all angles

SDXL x SD3 flash x Dall-E 3

SDXL wins for me.

[lavish osprey]

probably, but you should ask the authors of the tool

[lavish osprey]

I mean, sd3 flash shouldn't be a contender for anything

[signal shuttle]

I feel using SD3 flash over normal SD3 is not the best way to compare SD3 to other models

[terse belfry]

SD3 with gpt4 prompting

Prompt: Batman character concept all angles

[terse belfry]

SD3 wins that test if using GPT4 prompt preprocess

[torn wharf]

i like the variety i got from this one

[odd basalt]

SD3 for the win

[terse belfry]

Nice, love it

Left to right SDXL x SD3(2b) x Dall-E 3
Prompt: Batman returns ps4 gameplay screenshot, unreal engine, dark gotham city

[bitter hearth]

he is not fat...

... he is just big boned

I do not want him rescuing me, he will have a heart attack

[terse belfry]

SD3 with 16:9 aspect ratio, changin AR works well.

[sick cedar]

When it comes to Batman game renders, SDXL does it really well. Which by all logic means that later finetunes of SD3 could do even better.

[bitter hearth]

very cool

[terse belfry]

Yeah, rly love what SDXL does here! Changing aspect ratio on SD3 generate different results, much better than 1:1

[fleet meteor]

SDXL is my favourite , the atmosphere and the "camera position" makes it look like an actual game

[bitter hearth]

ok now I am just trolling, sorry!

[terse belfry]

SDXL version reminds me of Arkham Origins game

[fleet meteor]

I´ll try the same prompt in a sdxl finetune

[terse belfry]

Share the results when u r done!

[marsh violet]

So, no finetunes?

[terse belfry]

Super Mario PS5 gameplay screenshot, walkthrough, third person view

[fleet meteor]

I know it can be better but this is the one I most liked, this is leosamsHelloworldXL finetune

[raven fern]

Mario on playstation 🤣

[sacred jewel]

I like the DALL-E 3 best then the SD3

[terse belfry]

Like a real gameplay, love it

[terse belfry]

Lol 😆 I hate nintendo

[sacred jewel]

Looks sterile 🤮

[raven fern]

Nintendo releasing some hd remaster of very old games... $60... smh my head

[odd basalt]

SD3 2B medium

[fleet meteor]

I know ,it can be way better, but its all i get in 4 tries

[odd basalt]

Uncensored Dalle

[sacred jewel]

Yeah I like this better.

[odd basalt]

“High-resolution cinematic screenshot of Batman Origins gameplay on PS4, created in Unreal Engine, featuring a dark Gotham City at night. The scene captures a zoom-in camera shot with Batman in the foreground, standing on a rooftop, overlooking the city. The lighting is dim with streetlights, neon signs, and spotlights creating dramatic shadows. The color palette includes deep blacks, dark greys, navy blues, with highlights of neon green and purple. The atmosphere is gritty and intense, with light rain, mist, and fog. The background shows gothic architecture, tall buildings, and dark alleys with subtle reflections on wet streets.” Remixed the prompt

[terse belfry]

Opendalle?

This looks better indeed

[odd basalt]

Thank u

[terse belfry]

Spongebob squarepants 2d platformer gameplay screenshot, playthrough

[odd basalt]

Yes and sorry about earlier hun …. I remixed ur prompt try that version I wanna see ur results

[odd basalt]

And yes also to open Dalle

[uncut river]

sd3, medium

[terse belfry]

I like how u changed the prompt! Very detailed.

[odd basalt]

You are very welcome and thanx for urs … I love the games so much as well

[sacred jewel]

What is this? I only know DALL-E via ChatGPT-4o

[odd basalt]

https://huggingface.co/spaces/mrfakename/OpenDalleV1.1-GPU-Demo

[sacred jewel]

Oh SDXL. Never mind. I thought it was a new version of DALL-E 🤭

[odd basalt]

😂 but it’s better ❤️

[sacred jewel]

🤭🎉👊👍👌

[sage burrow]

Claude helping. 1st image is sd3 no sdxl. the 2nd image is with sdxl at the end. I used the prompt above

[terse belfry]

Oh, these are good! SDXL one is like a screenshot from Arkham Asylum game

[odd basalt]

They are very good and very detailed

SD3 and open Dalle

I am using Apple iPad 10 so results will be much different from my gaming pc

Cascade

[bitter hearth]

that vest does not match

finally, batman loses some weight

[terse belfry]

👌

Batman returns 1992 platformer game, 2d platformer, SEGA GENESIS Gameplay screenshot, snes

[bitter hearth]

omg i love batman

[fleet meteor]

Aaa I love that style

[bitter hearth]

how many enemies does batman have

[terse belfry]

All my attempts of generating actual penguin villain failed 😢

[fleet meteor]

I changed your prompt a little (well, just the title XD), Matrix PS5 gameplay screenshot, walkthrough, third person view

[bitter hearth]

Maybe try one of the penguin actors?

which begs the question, who was the best penguin?!?

[terse belfry]

Tried. It gives some weird results 😆

[bitter hearth]

catman

I got some weird results for Clooney once, it was the last time I tried an actor

[bitter hearth]

too cute

[sage burrow]

Needs work, but I like the concept 🙂

[terse belfry]

I love Matrix! Wanna try to generate smth from this universe.

[bitter hearth]

oh yea Keanu

from before he had to shave

tiptoes away

[cunning lintel]

the penguin according to kolors, fun challenge :p

A sinister, umbrella-wielding Oswald Cobblepot, aka The Penguin, stands amidst the dark, misty alleys of Gotham City, his tuxedo-clad figure illuminated by the faint, flickering glow of a nearby streetlamp. His pale, bulbous nose twitches with malevolent glee as he gazes upwards, his beady eyes gleaming with madness. A top hat, slightly askew, adds to his air of eccentricity, while his gloved hands grasp the ornate handle of his iconic umbrella, its tip menacingly pointed towards the shadows. The crumbling, gothic spires of Gotham's architecture loom in the background, their dark silhouettes a fitting backdrop for the Penguin's twisted schemes. The image is rendered in a dark, film noir-inspired style, with bold brushstrokes and heavy shading, evoking the gritty, atmospheric works of artist Tim Sale.

[terse belfry]

Keanu reeves as Neo from Matrix, platformer game, snes, SEGA GENESIS Gameplay screenshot, dark

[sage burrow]

Anyone want to try my SD3 img2img glif? 😉
https://glif.app/@LadyLalita/glifs/clyc08kox0005o8lokxrpdmcx

[terse belfry]

Lemme try! I still have some credits left lol

[sage burrow]

It shouldn't take any credits at all/ It's 100% free.

[terse belfry]

I mean, daily rate limit

[sage burrow]

I had no idea! Mine says 23

[cunning lintel]

creative idea!

[sage burrow]

Innocent look 😉

[cunning lintel]

i think i broke it https://glif.app/@Aliquip/runs/f4cncnekpomx6qyb6qr5dco3

[sage burrow]

It doesn't work for all images I've found.

@sacred jewel I made a glif for ya 😄

[bitter hearth]

oh george

[bitter hearth]

I can't even get close to a young neo with 2b

[sage burrow]

is that the straw horse upgrade? 😉

[terse belfry]

Prompt: Kratos God Of War 2D snes game platformer screenshot, gameplay, side scroller,

[bitter hearth]

I have a painting in my room thats kinda like this

[odd basalt]

That has to be frustrating

[bitter hearth]

nah, its just for laughs

[odd basalt]

Ur coded 😉

It does give daily’s

[sage burrow]

Really innocent look 😉

[odd basalt]

I used one of my Harley’s in that bubble gum pop color attitude ….

[sage burrow]

That blonde haired one is cute too!

Has anyone else made any SD3 glifs? I made some more serious ones as well, but they are pretty much remixes of others, so kinda boring.

glif has this btw, but I think it's only for sdxl. Handy for enhancing SD3 images at the end though.

[odd basalt]

No I’ll hop on pc tomarrow after work and make some there is a huge difference in me using my iPad vs me using gaming pc

It makes a huge difference in photo’s

[odd basalt]

Thank u

[torn wharf]

🦊

[sage burrow]

I made a special SD3 img2img glif 😉 Maybe you can make a ballz one 😄

[bitter hearth]

I come back

To see furries

[craggy crest]

aren't cats furry?

[bitter hearth]

No.. They are cats

enough explanation

[sacred jewel]

oooohhh. me go play

[sacred jewel]

Am I doing this right?

[sacred jewel]

@sage burrow

[hallow lion]

this model would be soo good if it was good

look at those flowers...

i really like the color, composition and prompt following

everythign else..... ughhhhhh.............

[hallow lion]

wtf lol

not the cute cat youre looking for

[craggy crest]

well - if cats aren't furry, that's what they look like

[hallow lion]

what about the demonic eyes tho

[sacred jewel]

@sage burrow

[hallow lion]

Foxfest.

[craggy crest]

it's a cat

[sage burrow]

The first lady if my fave 🙂

[sage burrow]

Oh that one is cute ❤️

[sacred jewel]

a few more before bed

[sage burrow]

I'll admit it, what looks like a regular img2img glif, I snuck in something to make all (most) characters turn into furries ROFL

[bitter hearth]

LMAO

[sacred jewel]

You didn;t fool me... hahahahahaa it has your "signature" all over them

[bitter hearth]

I like the white robot ones

[sage burrow]

The wizard is cute 😄

[bitter hearth]

its good at super blue eyes

SDXL couldn't do colours this good

[sage burrow]

I did a version with sdxl at the end, to improve the images, and for whatever reason it made them worse. Though perhaps i added it in wrong

In case anyone wants to remix it, or start your own from scratch, all I did was add a couple of words for GPT to include 😉

[sacred jewel]

Leve it as is... it works fine!!

[fleet meteor]

Do you have the prompt? Those flowers are not focused but they look beautiful 🤗

[low stone]

you guys should really get on the lumina/kolors bandwagon. fish king demands it.

[sacred jewel]

• my SD3 LoRA latest version...

[sacred jewel]

Zero prompting... It was img2img using @sage burrow 's Glif recipe.

[fleet meteor]

Aaaa okey

[bitter hearth]

wow really nice lora

[sacred jewel]

Thanks... It's a mess... too sentsitive, overfitted, etc.

Literally did a blind traning using SDXL template... SD3 needs a lot more info from SAI to train it.

[sage burrow]

If you mean my glif, it can be used for so many things with a slight change of wording... for exapmle if someone wants to do a zombie theme, or say a robot theme, etc. etc. 😄

[bitter hearth]

ah yeah overfitting is a shame

but there is use for heavily overfitted Lora

if you just want something that looks like the sample images

[sacred jewel]

There is a prompt field? I didn't see one. I just upload image, pick A:R and hit GLIF

[sacred jewel]

If only I knew what I was doing.

BUT, I am learning. I challenged myself to learn training with an unknown like SD3. So far, it is like drinking from a firehose.

If anyone wants to play

https://huggingface.co/gonzalu/YFG-SpyWorld50

[bitter hearth]

thanks ah it uses the ballz method

I love how ballz was the first SD3 lora on here

[sage burrow]

It's basically just chatgpt looking at whichever reference image YouFunnyGuys used, and describing it to SD3 (with my addition to make the main character into a furry lol). https://glif.app/@LadyLalita/glifs/clyc08kox0005o8lokxrpdmcx

[sage burrow]

Only if you remix the app. GPT does the prompt for you based on the image. I however snuck in a bit about making the main character a furry, into the GPT code part 😉

[bitter hearth]

LOL he did a spyworld prompt and it made it into an awesome fox

no wait maybe that's a wolf not a fox

[sage burrow]

There's also a glif option, where you can have a particular image, or sevearl particular images, that are always built into any workflow. I accidentally did that when making my first one.

This one looks interesting. I added SD3 at the end though, so I don't think it's taken into account much. https://glif.app/@LadyLalita/glifs/clyc8bxpw0004519bvnbrb166
Basically with glifs, entire glifs, or chunks of glifs are just copied from others, then customized.

[sacred jewel]

\

[craggy crest]

when dinner fights back

[low stone]

wow you've really got the text down

[odd basalt]

“Create a detailed image of an anthropomorphic cat girl with a punk aesthetic. She has long, wavy silver hair tied in a high ponytail with a black headband, and large, expressive purple eyes. She wears a purple crop top and a black leather jacket decorated with pink heart patches and studs. Accessories include a black choker with silver details, multiple silver necklaces, silver hoop earrings, and a black studded belt with a chain. She has a small bat tattoo on her forearm, a butterfly tattoo on her waist, and a script tattoo that reads ‘Always’ on her hip. The background features a bokeh effect with soft, circular light spots in shades of purple and pink, and the lighting is soft and diffused, creating a dreamy and whimsical atmosphere. The overall mood is confident, rebellious, and stylish.” Note chat gpt is outta date they need major upgrade use more up to date bots that are not over filtered u get better detailed results example gif method and any bot of ur choosing will give you a prompt

Rev, Dalle and sd3

[craggy crest]

actually, just ask @sage burrow for a prompt for this

[dawn ravine]

типичный завод в Пскове

#типичный завод в Пскове

[edgy kelp]

Where's the balls outputs? I'm missing my balls

[sage burrow]

I just created a SD3 text to image glif! 😄 It works really well 😉
Glif is free btw, but each person gets a max # of daily credits/images, around 100.
https://glif.app/@LadyLalita/glifs/clycy3ugs0006dfk4a72wklco

[empty robin]

images

[wary coral]

Add an attractive background based on this image that allows for changing perspectives

[teal fossil]

Computer says no.

[odd basalt]

I used her tool and converted into a prompt with prompt engineering skills I dislike chat gpt its out of date and behind

[sage burrow]

I used Claude for my others.

Which is your fave, if any?

@%*&($( glif censorship arg!

[odd basalt]

I use meta a lot but I use opens source and old school bots that understand the language models better … also the un censored bots

When I get on pc hun I’ll shoot u links

[sage burrow]

I looked up Meta llm 😄

I do only have 8gb gpu though...

[odd basalt]

It’s def up to date

[sage burrow]

darnit no free use online with their GPUs trial! https://huggingface.co/meta-llama/Meta-Llama-3-8B

[bitter hearth]

[sage burrow]

oh but wait!

looks free. Wonder what else it'll do 😉

[edgy kelp]

Yeah, nice but I don't see my balls

[odd basalt]

https://github.com/oobabooga/text-generation-webui

[edgy kelp]

With the 8gb VRAM of Becky, she can barely run a 7B LLM I reckon

[bitter hearth]

[odd basalt]

Girly u might need bigger ram ….

[edgy kelp]

At least this was true when I used Oobabooga some months ago, dunno if there were giant steps in the meantime

[sage burrow]

I just paid $1700 (CAD) for my system 2 months ago, not rich! Fortunately there are a heap of online options

[odd basalt]

Very true lots of options and I run nice on pc but can run a lot on Apple iPad just pictures are a different better on pc than on any mobible or iPad devices thou

[bitter hearth]

@edgy kelp

[edgy kelp]

Nice balls!

[bitter hearth]

Colorful and cute balls

[sage burrow]

Fortunately SDXL or SD 1.5 via comfy works great on my system. SD3 is a bit challenged though with complex workflows

[edgy kelp]

As the youngsters say: "F*ck it, we ball"

[bitter hearth]

So this is a ballroom

[odd basalt]

I’m lucky mine works considering it’s a few years older gaming system… that’s fantastic it works for u sd3 😭 but it’s okay because some ppl have to result into using art bot or buy the gpu because nothing works…

[sage burrow]

I want to use the fancy workflows evetually though, so perhaps GPU rental eventually. For now, I just keep creating custom glifs 😉

and there's some nice huggingface ones also

[bitter hearth]

Kinda borky but cool

@odd basalt

[sage burrow]

Just remixmy txt2img glif and replace all the mentions of furry with Guy Fawkes mask 😄

[compact niche]

very nice

[edgy kelp]

Finally I have my balls back

[odd basalt]

Hey u do really good at those… it’s tottally fine because art is how u make it and u don’t need to go out and buy a million dollars worth of hardware … if that makes sense

[odd basalt]

It’s cool I was at that party with u back in the 60s 😂 jk

[odd basalt]

If he don’t make it I’ll def make one \m/

[sage burrow]

Here's the glif I made, you can do any theme by remixing, and just find/replace any mention of "Animal" or "furry" with your own concept.

Or you could always use the original I remixed it from. Though with mine I did try about 10 different versions until I got it to work 80% or more of the time. The trick is getting Claude to focus on what you want in addition to the original focus.
https://glif.app/@LadyLalita/glifs/clycy3ugs0006dfk4a72wklco

[odd basalt]

I’ll try it

[sage burrow]

Doing these on somewhere that is so extremely censored is annoying though, I'm going to hunt for other options. Perhaps google collab. (though I might use up my google collab with lora making)

[sage burrow]

I'm looking forward to it 😄 The text is pretty obvious, about where you should insert your own style/etc.

[sage burrow]

Link? 😄

[odd basalt]

a gray cat with black stripes is devastated when he saw his town being destroyed by a group of snakes, unreal engine, greg rutkowski, loish, rhads, beeple, makoto shinkai and lois van baarle, ilya kuvshinov, rossdraws, tom bagshaw, alphonse mucha, global illumination, detailed and intricate environment …. Ur own link 😂 those were generated in urs here’s the prompt