🏞｜general-with-images

[pale garnet]

idk, because men aren't supposed to beat women?

unless the feminism women, I'm gonna ignore

[nimble mason]

This is one of the main problems I see with SD and tornadoes espec cascade

Overly defined lines where it looks like a screw when really there's a highly diffuse debris field that makes it look smooth

The others are things that don't look like a funnel at all, or obvious vertical air movement instead of horizontal

Or an overly defined shroud at the top of the tornado

The second one looks like a rising smoke plume rather than a tornado due to the structure... It's clearly not rapidly rotating

Idk why it has so much trouble with something there's so many photos of on the Internet lol

[clever oar]

why is this the only one censored?

[jade summit]

A set of cartoons showing people looking in front ,in the style of monochrome portraits, catcore, minimalist,traditional vietnamese,sarah andersen --v 5

[jade summit]

#🏞｜general-with-images A set of cartoons showing people looking in front ,in the style of monochrome portraits, catcore, minimalist,traditional vietnamese,sarah andersen --v 5

#💬｜general-chat A set of cartoons showing people looking in front ,in the style of monochrome portraits, catcore, minimalist,traditional vietnamese,sarah andersen --v 5

[steel forge]

#🏞｜general-with-images 坚持

[gusty cloak]

@lapis quarry Dall-E 3 + SDXL Canny, exact same prompt

[lapis quarry]

That is epic

Ima use that controlnet

[gusty cloak]

Dall-E 3 -> SDXL Checkpoint with Canny -> Canny + Inpainting

[gusty cloak]

And my attempt to colorize an image with Canny! The "workflow" is more involved and requires manual editing though.

[quiet current]

we gonna need a bigger cheese for this one...

[clever oar]

he only eats cheddar

[quiet current]

ok so no cheese then

I refuse to call cheddar a cheese

[clever oar]

why

[quiet current]

over my dead boddy...

get jumped at the throat by a giant mouse

[wispy nest]

[quiet current]

because it s pasteurised

[lapis quarry]

Awesome. What inpainting model did you use?

[gusty cloak]

I actually don't know if the model I used is optimized for inpainting but it does work well enough 🤔
Hephaistos NextGen DPO! Combined with Canny though!

[pallid ruin]

So cool bro

[pallid ruin]

🤗

[crisp stream]

Isn´t almost every milk product pasteurised?

[quiet current]

in USA, yes

[clever oar]

i eat week ago chedder is not bad

[covert pagoda]

fr

[clever oar]

i like all cheese)

[pallid ruin]

x2

Oof im hungry now

[clever oar]

energy drink)

[clever oar]

mmm
blue mold

[crisp stream]

in Europe as well, though there are a few exceptions regarding cheese like Camembert/Roquefort in France and Gorgonzola/Mozzarella/Ricotta in Italy. And you can buy raw milk in basically every country, though it has to be labeled as such and usually isn´t available in standard supermarkets.

[eternal comet]

When I'm home

I'll generate stuff people

[quiet current]

most cheese aren't pasteurised in France.

[pallid ruin]

I ran out of ideas/prompts😭

[clever oar]

its easy go to chat gpt and ask

or use promt randomizer

[pallid ruin]

Wait that´s true

[crisp stream]

At least Claude (2?) says, 30-40% aren´t pasteurised in France, though I´m not sure how reliable that info is coming from Claude 🙂

[quiet current]

info from "just looking at the cheese alley in my supermarket" tells me otherwise :p

[crisp stream]

😄

[quiet current]

But sure there are a lots of industrialized stuff. And if it s exported, then it s pasteurised.

[crisp stream]

yes, read about the export rules as well

[cloud sorrel]

I have never heard of half of those. I know standard models and what LCM, turbo, lightning, and cascade are, but what are those other things?!

[cyan shoal]

Pony is an extensive finetuned of SDXL

SD 2.1 unclip is for replicating images

SVD and SVD XT are for very short video and short video generation

Playground V2 is similar to SDXL but its trained to be very aesthetic like midjourney

[rich pollen]

look at this

[gusty cloak]

Looking

[cyan shoal]

Pixart A uses T5, in my experience its not that much more intelligent than generic models

[rich pollen]

this is my prompt

an Egyptian pharaoh looking out into the horizon from a balcony, ancient Egypt, crowd of Egyptians below, (pyramid and Sphinx statue) in background, daytime

negative prompt: female, woman, sexy, deformed, blurry, extra limbs

[gusty cloak]

What model is this?

[rich pollen]

flat2D animerge

but even with other models it's similar

[gusty cloak]

I think (SPECULATION) it's because your model is being dumb so when it sees the long hair it instantly defaults to denoising a woman

[rich pollen]

this is text to image

[gusty cloak]

I know 😅 But in any case I'm uncertain as to what could be wrong

Try a casual picture of a man

[rich pollen]

that prompt generated male characters

anyways im going to try switching to another model

any suggestion for a good one that produces highly stylized/cartoon images?

im downloading toonYou at the moment

[gusty cloak]

🤔 Off the top of my head... Hmm

If you want males I think there was one that was abbreviated as "MMM"

[rich pollen]

it's no better with this model

it's actually worse because the composition is not as good

[gusty cloak]

"mmm_cartonnstyle" But it's an SD1.5 model

[rich pollen]

ok i'll give it a try

thanks for the info

[gusty cloak]

I'm not sure if it can do Pharaos but 🤔

Let me dig up other ones

So a cartoon style is preferred correct?

Not anime?

[rich pollen]

i need something that can do i wide variety of subject matter

no i dont want anime

[rich pollen]

this style is pretty close to what i want

highly stylized cartoon but not anime

im trying to produce a series of scenes from various moments in history; stone age, egyptians, mayans, chinese, japanese, rome, greece, england, france, wars/battles, city scenes. all that kind of stuff

[gusty cloak]

What about this one? https://civitai.com/models/184414/3d-cartoon-vision

[rich pollen]

looks like it's worth a try. downloading it now

[grizzled sage]

pixart-a was trained on a 10th of the dataset that sdxl was so it takes a 10th of the time to train

[cyan shoal]

exactly

[grizzled sage]

pretty good prompt comprehension in the latent spaces it does know, but it doesn't know a lot

[cyan shoal]

idk I tried a relatively simple prompt and it couldn't do it

[grizzled sage]

honstly, in my experience, most specialized models can be a lora. regarding 3d cartoon vision 1.5 model

[cyan shoal]

then again the huggingface demo had bad results

no matter the settings

[grizzled sage]

i find descriptive natural language works the best on pixart-a. cooking this prompt right now on the hugging face demo's default settings a wide shot of a colossal alien temple carved into th side of a black granite cliff face. multiple alien moons in the sky. a squad of marines approaching the gigantic entrance of the temple. in the style of H.R. Giger

it missed the moons but basically caught the vibe i was going for

[cyan shoal]

ooh nice

this could benefit from superprompt-v1 then

[grizzled sage]

i'm more excited about the latent spaces that t5 can dive into. pixart-a is just a lens into the possibilities. it's not my daily driver, but its a big part of why i wanted to see if that lavi-adapter would actually work.

i'm surprised nobody has turned their code into a node or extension yet.

the code works. some numbnut like me could make it work. i don't get it

[grizzled sage]

cute

[arctic laurel]

they look like bad architectural renderings

[grizzled sage]

yeah like one of those billboards you see outside a future development

[grizzled sage]

https://github.com/vladmandic/automatic/blob/master/README.md new sdnext is out. it has a diffusers back end and even he is sleeping on the lavi adapter code. that should be the easiest one to port it to

[cyan shoal]

@river quiver here's another reply though

[river quiver]

thanks

[rain gazelle]

Hi there -- this one was lightningbloom 🙂

[grizzled sage]

i think they're cherry picking the list for people they want testing. They certainly aren't doing a general rollout. Lots of people signed up the very hour it dropped, but youtubers are getting priority.

I signed up 8 hours after the gates opened, and i'm in here saying their crypto partnership is a dead canary, so i'll probably be last invited, if at all

[cyan shoal]

sounds about right

I wonder if I ask him on twitter once I get an invite

some people outright asked and got it instantly?

its odd

[grizzled sage]

Yeah i seen Dr Furkan ask and was given one

[cyan shoal]

I guess those were either model fine tuners and researchers as well

[grizzled sage]

creator of proteus keeps asking and isn't getting one

stable staff outright called his model a virus in the scene. afaik the entire rundiffusion team hasn't got one.

[cyan shoal]

thats cause he uses pony or something

and he made a reddit post about "should I remove my proteus model"

[clever oar]

dune

[grizzled sage]

was going for more of a Prometheus vibe, though ridley did take a lot of inspiration from Lynch's vision when designing his alienverse

[grizzled sage]

"should i remove" came after the dogpile that resulted from stability employees painting it as a virus.

[cyan shoal]

yes

its interesting how stability itself made all these claims

they must know something

[grizzled sage]

https://old.reddit.com/r/StableDiffusion/comments/1bf0943/proteus_v04_from_rundiffussion_just_dropped/kvbvhcn/ here's the staff member saying he used explicit kid images in his training data even

[cyan shoal]

that we dont 👀

Lykon: so sometimes I forget that the world is full of delicate flowers who can't live with different opinions.

[grizzled sage]

lykon likes gingers so...

[clever oar]

house underwater

oh it not for your post)

ai know about prometheus?

[grizzled sage]

i just told it style of giger

thats why the entrance looks so familiar! 👀

[clever oar]

ai doesn't know many extinct animals

[grizzled sage]

just had a conversation with the dev of sd.next and he shot down lavi-bridge saying it couldn't work and is just a concept.

am i crazy and just imagined running their test script with the downloaded weights?

[clever oar]

i dont understand about bridge)

and dont know about sd.next

[grizzled sage]

its a webui with diffusers back end. the bridge is a small adapter and lora that allows large language models to embed into sd15 models

[clever oar]

control net distore faces

[grizzled sage]

nah

[clever oar]

yes

bad faces with control net

open pose

[grizzled sage]

no porn here. also, controlnet isn't causing it. prompts and settings

[wispy nest]

btw I was lurking and saw the mention of pixart. Can't thank you enough. That shi is GOOD 🙏

Never heard of it until now

[grizzled sage]

coo cool

[rain gazelle]

Just some randos.

[clever oar]

why without control net all ideal

[grizzled sage]

controlnet gives an edge to the latent spaces being explored. you have to adjust settings for it. lower cfg. different step counts. better promtps.

if you're convinced you're right though, no amount of explaining it is going to help you. I assure you that faces can be created just fine with controlnet. You don't seem to want to hear that though

[arctic laurel]

control nets are like little neural hubs, they got they own heterogeneous functionality. Konami’s VRC6 audio mapper for the Nintendo was like a controlnet

[clever oar]

what cfg i need? i use 9

35 step

[grizzled sage]

play around. i think the point i'm conveying is that you're incorrect about the source of the issue. Controlnet can do faces.

feels like i'm arguing it though.

[clever oar]

i play with all param

[grizzled sage]

i guess it's obviously controlnet. you win

[clever oar]

what standart setting i need

[clever oar]

show me something with openpose

common man i need expert training

[grizzled sage]

You started arguing with what i was saying. I only give 1 chance not to do that

if you need help, its not the time for an argument. thats 2 doors down

[clever oar]

this is not a dispute but my observations

deformed)

[grizzled sage]

Yup. Sure is.

[clever oar]

what need for normal fave and body

or i need special model for cn?

[cyan shoal]

Using Brian's superprompt-v1 offline on CPU (super fast) using #📝｜prompting-help message

currently it might not benefit us as much, but with SD3 it's going to help a lot with quality or emptyness of scenes

[wispy nest]

both sets look amazing

im wondering about your killer prompts 🙂

each of those images on top are distinctly artful

[silver pasture]

this is what i got

and idk where the image is

or how i can see it

[wispy nest]

😂

[vernal olive]

it's wonderful

[nimble mason]

Here is the image you requested.

[vernal olive]

Ancient Chinese wedding night，the woman and the man look at each other, the woman is sitting on a stool with a red hydrangea in her hand, and the man is holding the woman's hand。

[lapis quarry]

Hey @gusty cloak . Since you seem to know a bit about control net and stuff like that. What models do what things?

[gusty cloak]

Hi! Could you clarify a little bit? Do you mean ControlNet models?

🙂

[grizzled sage]

NOT THHE BEEES

[vernal olive]

请问应该从哪里使用sd

找了半天

[gusty cloak]

I don't understand a word 😂 We speak only English here

[vernal olive]

hey man,what can I say

ok,I got it

[grizzled sage]

Man who look long time live long life.

[wispy nest]

https://tenor.com/view/not-the-bees-gif-26245123

[vernal olive]

where should I use SD from

[lapis quarry]

Yeah. The controlnet models

[vernal olive]

I can't find it

[nimble mason]

Here is the image you requested.

[gusty cloak]

You can use Stable Diffusion in the WebUI of your choice, such as Automatic1111, ComfyUI or Invoke 🙂

[gusty cloak]

So, I'm not too familiar and comfortable with ControlNet, as I had my most successful experiments ever yesterday 😅 And I'm proud of the results!

Personally, I highly recommend using a SDXL Canny ControlNet Model if you want to achieve something like what I got!

Use the SDXL model of your choice and then SDXL Canny!

Are you using A1111?

[lapis quarry]

No. I am using Draw things ai. A ios app that runs Stable Diffusion locally

But I have access to those controlnets there

[gusty cloak]

Hmm 🤔 Well, I'm assuming the settings should be the same!

For starters, take a 1024x1024 photo and enable your Canny ControlNet. If possible, set the resolution from 512 to 1024 (there might be a "Pixel Perfect" setting instead)

[nimble mason]

Here is the image you requested.

[edgy shard]

wtf my message got eaten

ok. help with inpainting?

inpaint sketch skirt = good. inpaint sketch hair ribbon? no go.

inpainted the red skirt to green just using the sketch color. worked fine

now if i try to do the ribbon, it's a no-go LOL

[gusty cloak]

What if you use ControlNet + inpainting?

Canny specifically

[edgy shard]

What's that

[gusty cloak]

Yes you seem to have it! But do you have any models?

[edgy shard]

No

What do I need

[wild sorrel]

increase padding

and change prompt to be about only area you are inpainting - ribbon, only related to it stuff, not about whole image

[edgy shard]

well i just want to recolor what exists if possible

oh, well that actually worked

thanks xD

[tidal barn]

@urban lily nihao

[nimble mason]

Here is the image you requested.

[jovial tiger]

jeez, playground 2.5 doesn't need any help from dark arts, it's got it all going on by itself. this was just "tarantula barbie"

[nimble mason]

i need PG2.5 accessible on my phone somehow

[nimble mason]

whoa

it's dark without a lora

[humble oracle]

hello（你好），are ya conducting researches on stable diffusion models（老哥是不是最近也在研究扩散模型）

[humble oracle]

diffusion 2M dataset?

[humble oracle]

that reminds me of Jia Ling before she lost abt 130 pounds

[nimble mason]

Love it

[alpine sigil]

#🏞｜general-with-images girl 20 years old

[dull totem]

#🏞｜general-with-images The old version of the journey to the west pig leaves the picture

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[mystic apex]

#🏞｜general-with-images Generate a logo that is refreshing in color, using simple patterns and letters to reflect the concept of collaborative win-win and innovative development.

[pale garnet]

@carmine quail

is it the same?

[carmine quail]

Could be, I'm not sure

[pale garnet]

okok

[clever oar]

I imagined it on myself and wanted to itch

[young swan]

ffe

[ebon kestrel]

An 18-year-old girl, LUCY, sits by her bedroom window, surrounded by boxes containing her late mother's belongings. She sifts through the items, curiosity etched on her face. Suddenly, she stumbles upon two keys - one red and one blue. Intrigued, Lucy examines them closely, wondering about their significance.

[ebon kestrel]

She sifts through the items, curiosity etched on her face. Suddenly, she stumbles upon two keys - one red and one blue. Intrigued, Lucy examines them closely, wondering about their significance.

[solid geyser]

||一个美丽的女孩在一个反乌托邦的后世界末日色彩缤纷的城市长满了植被，反光水坑，阳光明媚的日子，现实，高度详细，8k，电影照明，辛烷渲染||

[quiet current]

Bots are still offline..... please read #1047610792226340935

And even if it was online, this is not the place to prompt for it.

[clever oar]

Aryetis

Aryetis I hope you ate cheese for breakfast today?

[quiet current]

its eyes reflections are betraying them

Now I can clearly see the true nature of this cat

It's a robot !

[clever oar]

oh I didn't want you to know the truth

[quiet current]

I'll let you know that I played through the whole ace attorney serie.

I'm basically a pro investigator at this point

[clever oar]

😲

is nature 100%

[eternal dome]

wooooow

[eternal comet]

💀

[clever oar]

I hope you ate cheese for breakfast today?

[tepid vault]

An 18-year-old girl, LUCY, sits by her bedroom window, surrounded by boxes containing her late mother's belongings. She sifts through the items, curiosity etched on her face. Suddenly, she stumbles upon two keys - one red and one blue. Intrigued, Lucy examines them closely, wondering about their significance.

[quiet current]

A tired guy responds on discord to someone who hasn't read the #1047610792226340935 . Reminding them in a sassy yet respectful manner that the bot is offline.

[robust path]

unless you're able to summon the minion @nimble mason bot

[tepid vault]

How do you draw it

[quiet current]

(not a proper bot)

[quiet current]

For now, either you install it locally or you use online services such as dreamstudio

[eternal comet]

I'll do it for you

[nimble mason]

Here is the image you requested.

[nimble mason]

Bot Image Quality Classifier Guidance Evaluation System (BIQCGES) score: 3.6/10.0. This image fails to adhere to the the guidelines described in our many high impact publications.

[glacial axle]

Anybody knows how I can fix that weird thumb? Maybe photoshop? Please I need a more experienced opinions. It has to look perfect

(the thumb is a little too thick)

@flat gulch @quiet current @dry crow @glossy herald @stark hazel ?

[quiet current]

please don't @ staff like that.

Also #📝｜prompting-help might be better suited for that

tldr : look into inpainting, adetailer, controlnet

[glacial axle]

I used controlnet a lot to get to these fingers but that thumb thickness is just there

[clever oar]

who give me right setting for controlnet openpose

i have distored faces in results

[cyan shoal]

ah you like AI Corn?

[grim bone]

Where can I AI painting?

[eternal comet]

Real I have stable diffusion 3!!!!

Omg

[grim bone]

Where can I love painting?

[clever oar]

resident evil 7)

[clever oar]

how you het it

[pallid ruin]

😍 that looks so good!

[clever oar]

thx

[pallid ruin]

What model are you using?

or maybe the prompts is too good xd

[clever oar]

idk its online sd generator(

[pallid ruin]

Aaa

[dull flame]

the era of anime girls is over !

[pallid ruin]

Do you remember the prompt you used? @clever oar

[clever oar]

room by Rebecca Louise Law AI Art,realistic,best details

[pallid ruin]

Thank you!

[clever oar]

I take the names of artists and combine them with various promts

[wispy nest]

how do yall have stable diffusion 3 😭😭

[pallid ruin]

We do not 😭

[wispy nest]

oof

nvm

[pallid ruin]

So cute

[eternal dome]

indeed !

[terse jasper]

https://github.com/Acly/krita-ai-diffusion

[wispy nest]

dalle be good

[idle jolt]

a round icon with a crown

[eternal comet]

Hello chat

Moofi always generates best images lol

[clever oar]

its sdxl?

skorchekd wassup

[eternal comet]

Nothing much

I might start using sdxl.again

Cascades ram requirements are ludicrous

[crisp stream]

yep

[crisp stream]

is it available offline?

[crisp stream]

Thank you for the f....ries 🙂

[exotic path]

Guys what makes the lora i train on the right tend to have similar results within a batch? (fyi, i applied glaze on the left one, the result actually made it much more diverse)

[round helm]

grounding example

[pallid ruin]

so chaotic but also cool 😁

[digital raptor]

||change to kayo||

[cobalt jungle]

Hey, i'm pretty stupid with this stuff, i purchased a subscription to midjourney specifically to use a stable diffusion generator to give me images of this building if it was rehabilitated; but since I am too dumb to figure it out i am blowing my fast GPU away
I put it in a server called Dr Toboggans Historical Rehabilitation
i'm totally lost in this server stuff; can someone help me? or tell me if it's even possible?

Here it is above; is there a way I can ask midjourney to run variations on this specific image? it's making me do a "describe this" where it prompts itself and tries to run it from the prompt it described to itself. Am I missing something?

[unique condor]

Use the unet+clip+vae setup with the lower bit versions of the models. You can even mix and match the stages.

I think the comfyui checkpoint versions are bf16(+clip and vae so they are bigger than the unet equivalents), but there are light and bf16lite versions of both B and C unets

If you use the unet setup, drop the C to lite and leave B at bf16 since it's smaller and is more of a "refiner"

If you're still having vram issues with the lite version of C, use the bf16lite version

[finite cobalt]

Hi everyone, have someone ever try a prompt to generate this Illustrator artwork like this? So difficult to explore the prompt

[desert kelp]

what kind of style is this how can, is there any tool out there for free where i can generate these kinds of images?

[sour ice]

looks beautiful! what's the prompt for the third image?

[crisp stream]

At least something happening for once 😄

[wispy nest]

gross.

Where do I download the installer for Stable diffusion 3?

[cyan shoal]

https://twitter.com/Lykon4072/status/1770856239204815357

Lykon (@Lykon4072) on X

These have been generated with a #SDXL Turbo finetune in 8 steps. Perfect prompt following and perfect hands 90% of the time. It shows how a good architecture is more important than early results. Remember that for #SD3: take the base model and project how good finetunes will be

Twitter · Mar 21, 2024, 4:53 PM

It shows how a good architecture is more important than early results. Remember that for #SD3: take the base model and project how good finetunes will be

[eternal comet]

💀

i tried recreating something off the reddit

💀

wtf

💀

[clever oar]

what you try to generate

[clever oar]

this?

[eternal comet]

💀 thats crazy

[clever oar]

what in reddit? anime?

[eternal comet]

women wearing hello kitty clothing

💀

[clever oar]

style reallife?

[eternal comet]

nahh

sdxl shits on cascade

💀

💀 this prompt is wild tho

this aint my prompt just incase someone pulls meta data

nah i aint gonna post it

[eternal comet]

badass

resepct that

how did you manage that

[clever oar]

what promt?

[eternal comet]

it cant do logos

[clever oar]

becouse you use xl

that bad for details

[eternal comet]

what do you use

1.5?

[clever oar]

yes

1.5 base models

epicphotogasm

i read about this prombem and dont start use them

details for me more important then qualioty

kitty mafia

[clever oar]

nice

how do now?

[eternal comet]

loras

i scraped a bunch of images

off google

put in lora

[clever oar]

you trained it?

[eternal comet]

ye

[clever oar]

nice i want too

[eternal comet]

💀

@clever oar

[clever oar]

can you show me something with control net and openpose without distored face

[pallid ruin]

so cute

[eternal comet]

cascade osmt tjat fgppd thnf

[deft bison]

@pallid ruin apparently... yes! ...

[dry stone]

Do you want to see the spanish crown princess

(Say yes)

[eternal comet]

ai is wild for this

ai wild

[clever oar]

horror

[eternal comet]

does that look good

i managed to get it by training it off discord egirls

yk in those discords with selfies channel

💀

[clever oar]

ai very hard generate -man kick woman

can you train it

[eternal comet]

ok

give me images of men kicking woemn

[clever oar]

not simple to find in google)

[calm canopy]

sometimes i question the people in this server and their intentions lmao

very strange

[eternal comet]

I will code a scraper

[clever oar]

😆 😆 😆

[clever oar]

how results?

[eternal comet]

Nothing

https://huggingface.co/dataautogpt3/ProteusV0.2/discussions

I'm gonna try use this

💀

My 3060 gonna dxplodt

Explode

[clever oar]

big % peoples use comfy?

[daring kernel]

Uh, I think so

[eternal comet]

@

@clever oar

[rugged ermine]

#🏞｜general-with-images cool

[willow ridge]

t

[eternal comet]

bro

this model is murdering my pc

it even crashed my fucking firefox

💀

[eternal comet]

💀

💀

[eternal comet]

"you look lonely i can fxi that"

https://youtu.be/tpV8HUMwfMU?si=6-UM5ny2lRmZRYEH

i fed it this lol

[eternal comet]

younger verson of postmalone

crazy

[clever oar]

she kick too

I fell asleep at the PC what happened

[eternal comet]

post malone as a women

[clever oar]

o shit need sleep)

I'll work like a zombie

good night

[eternal comet]

goodnight

[pale garnet]

The hand xDDD

[eternal comet]

"florida women solid abs sitting on crocodile wearing headphones wearing hat"

prompt following crazy

[eternal comet]

Insane

"I'm gonna make you my girlfriend"

[jovial tiger]

Looks like the stability AI api is up and running. gives back 6 2048x1152 sdxl images in 15 seconds.

[prisma hinge]

||city||

[nimble mason]

Here is the image you requested.

[surreal night]

Create a modern horror style comic book cover, titled first kill a African -American girl that resembles Taraji henson with long curly hair dressed as a vampire hunter holding a silver spear and also another teen that resembles fantasia African American girl with a short relaxed curly mohawk style, pixie, cut with silver vampire teeth in her mouth

[shy pawn]

Give me a green dinosaur similar to Godzilla, Japanese cartoon style

Give me a green dinosaur similar to Godzilla, Japanese cartoon style

[rich moss]

1 woman, special, (35yo), confused, expressive detailed eyes, camouflage uniform, standing in the middle of road, surrounded by running crowd, side view, (line sense:1.2), clean painting style, new sea city style, anime style, comic style, fire, smoke, rubble, best quality, masterpiece

give me a woman, special, (35yo), confused, expressive detailed eyes, camouflage uniform, standing in the middle of road, surrounded by running crowd, side view, (line sense:1.2), clean painting style, new sea city style, anime style, comic style, fire, smoke, rubble, best quality, masterpiece

[wispy nest]

😂

[clever oar]

picaaaaaaaaaaaaaaaaaaaaaaa

[wispy nest]

https://tenor.com/view/pika-dance-gif-22486082

[clever oar]

your model very good understand pokemon

[clever oar]

😲

my model dont understand pokemon

how you do this

[terse heron]

Design an innovative fashion atelier inspired by Sardinian heritage, featuring a woman at a loom weaving traditional Sardinian patterns, minimalist rugs adorned with Sardinian motifs, and sleek mannequins showcasing modern interpretations of Sardinian attire. This atelier embodies the essence of SRDN (Sardinian) design philosophy, showcasing the millennia-old legacy of Sardinian culture. Through our creations, we foster a dialogue between ancient textile artistry and contemporary, cosmopolitan sensibilities. SRDN Studio emerges as a bridge between the rich cultural heritage of Sardinia and the global fashion community. With an avant-garde vision, SRDN Studio collaborates with local artisans to reinterpret traditional Sardinian symbols and techniques into cutting-edge fashion garments and contemporary furnishings, presenting the world with a modern interpretation of ancient cultural aesthetics."

[junior sky]

Is that a prompt or a business plan? The bots are offline #1047610792226340935

[rich moss]

create 1girl，beautiful,[22yo]，charming.detailed face features, long hair, (hair length:1.3], blonde hair, summer, beach.sandy beach, ocean, blue sky, sunshine,walking on the beach, casual, swimsuit.bikini, sunglasses, flip flops, seashells,seagulls, footprints, waves, realistic.ambient light. [cinematic composition:1.2].HDR, Accent Lighting, wide-angle lens, best quality, masterpiece.

[clever oar]

josh teach how to generate pokemon

[proud orchid]

this upscaler is insane..

[shut sinew]

which one

[proud orchid]

https://civitai.com/models/330313/ttplanetsdxlcontrolnettilerealisticv1

[shut sinew]

Oh the one not meant for upscaling

[proud orchid]

actually it is

[shut sinew]

Nope the creator of it stated it wasnt meant for upscaling iirc

Does it work? Yes. But it wasnt the intended purpose

[shut sinew]

SUPIR also works

[proud orchid]

It does what I want, that's enough for me 😄

[heady anchor]

supir looks great the other one looks blurry

[shut sinew]

Supir adds random stuff tho, not perfect settings

[river quiver]

does this work with 1.5?

looks great, did you try?

[eternal comet]

Nah this model is insane with prompt following

[river quiver]

which obe

one

[eternal comet]

Proteus

Insane vram usage tho

[heady anchor]

1.5 already has its own tile controlnet which works better than the one from ttplanet

[river quiver]

ah..ive thought this was kind of a model/lora.

[river quiver]

example? please

[nimble mason]

Vram use should be the same per model with the same architecture and fp precision afaik

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[jovial tiger]

#SD3 for other chat

[cyan shoal]

these are so epi c

[jovial tiger]

Yeah really good. They're also way better than the first sd3 pics from a month ago, so it's definitely getting better over time

[cyan shoal]

exactly

they improved a lot in quality

[thin sable]

imagine how good will be the fine-tunes

[jovial tiger]

My only concern is that the fine tunes might have to get better about tagging to take advantage of the new engine, but I'm not familiar with specifics. Maybe put everything through gpt-vision or something.

[cyan shoal]

Well CogVLM isn't that difficult to run and might actually make things easier since that's what they used to caption half of the dataset

the only concern is for characters that CogVLM doesn't know of course

AAANd they might have to finetune something like llava on Corn or something if they wan't to try it

technically it's not necessary, but imagining the possibilities with prompt adherence for corn, could make it a very good option for those people

IF of course it works for them

[silent widget]

portrait of a pretty blonde woman, a flower crown, earthy makeup, flowing maxi dress with colorful patterns and fringe, a sunset or nature scene, green and gold color scheme

[covert pagoda]

what?

[south oyster]

It is original trained for my own realistic model used for Ultimate upscale process to boost the picture details.

says the author

[clever oar]

skorchekd how you train lora?

[south oyster]

can't stand small boobs

[south oyster]

so many fingers! still

[south oyster]

it says it needs 60gb ram and 60gb vram - is this what you are running?

[eternal comet]

cant stand bad prompt folowing

[south oyster]

agreed! can't wait to try SD3 for this reason alone!

[eternal comet]

i dont need sd3

model i have already does this lol

[shut sinew]

Yeah but in comfy it runs with less

Takes like 30s for this image on my 4090

Bigger images obviously take longer

[eternal comet]

look at prompt folowing on this model

perfection lol

this is the only issue 💀

black wido deepfake

[cyan shoal]

what model is this btw

[thin sable]

Dall-E is better for this one

[cyan shoal]

lmao

[eternal comet]

eh

i think its a draw

this model can go toe to toe with the big ones

with an extra push of course

]

proteus

[thin sable]

The problem with sd is the prompting

[eternal comet]

tbf i use a llm

so idc about prompting that much

[thin sable]

That's why Dall-E is the best atm. They used ChatGPT as text encoder

[eternal comet]

sd3 level pretty mcuh

just gotta train a texxt lora

and i got sd3 lol

[thin sable]

Hope SD3 is more close to Dall-E 3

[eternal comet]

with loras

i could make this dall e 3 level ig

[thin sable]

:)) alright

[eternal comet]

vram usage ludicrous

[thin sable]

Try to make a photorealistic image to see that you get so much blur

I already tried proteus

and deleted it

[clever oar]

skorchekd how you train lora to generate hello kitty?

what i need use to train

[eternal comet]

discord egirls

post malone

[clever oar]

ignore(

[eternal comet]

Not for me tbf

Maybe it's the version you use

[narrow panther]

A crisp and professional 4K HD LinkedIn profile photo of a 27-year-old arab fair skinned boy. He has short, styled hair and a lean build, exuding confidence and ambition. He is dressed in a stylish, well-fitting three-piece formal blue suit, with a white shirt and a blue tie. The background is a clean and clear professional studio setting, emphasizing the subject's strong presence and focus.4k, Ultra HD, realistic

[clever oar]

josh

use can generate all pocemon?

[gusty cloak]

I can't help but feel slightly irrationally irritated whenever I see someone try to prompt in a random channel 😶 But that's my problem

It's funny

[nocturne oak]

It's probably the most annoying thing that happens here. But I decided a while back to make it fun for myself by creating all of these response images.

Every single one is created via ComfyUI and the text is always prompted in; no tricks.

[gusty cloak]

And I sometimes question what people want to generate.

Like I have generated questionable things in private? If I'm using a public Discord I would never

🤣

Like things I could get eyebrow raises for

Meanwhile I used Canny to turn generated photos into digital paintings!

I may like the blue color

[solid oasis]

#🏞｜general-with-images

[eternal comet]

Who does this look like

[gusty cloak]

Reminds me of two people:
Post Malone (except way less decorated)
Talisman (the guy on the left on the second picture) but with a differently styled beard.

And honestly way more of Talisman

[eternal comet]

It's post Malone

With no tatoos

[gusty cloak]

As I said, less decorated 💀

[eternal comet]

Yeah

[nocturne oak]

So...pre-Malone.

[gusty cloak]

https://tenor.com/view/get-out-leave-gif-5785321051025239919

[gusty cloak]

Meanwhile I'm doing some fun generations tonight! 😂
I don't know what capreo means but slay 😎
In the last one I prompted "Nothing" and the result is a little creepy but fitting for the context.

Now the experience might be more bearable

Because I'll know I'll have an excuse to send one of these 😂

[nocturne oak]

Pretty much my thoughts when I made mine.

[gusty cloak]

The tree is on his right, and the monocle is on his left eye, though what's going on with his right eye? sweat

[gusty cloak]

This is also good!

[eternal comet]

yuh

its a vram eater tho

tamer now

[eternal comet]

rate prompt following

better than cascade and sdxl?

[gusty cloak]

I don't know about Cascade as I have never used it, but it's better than SDXL for sure 🤣

[gusty cloak]

#🌠｜show-and-tell Some of these are supposed to have blue stripes in the hair but they don't

[eternal comet]

cascade its better then that too

lol

[eternal comet]

is this doing stripes right

[gusty cloak]

Seems to be!

Sadly I don't think I can run Stable Cascade on my 4070 Ti...

12 GB VRAM

[eternal comet]

i can run it on my 3060

u want my setup?

[eternal comet]

cloned some guys setupo

u load first model

do it

then deload

then load the model again

[gusty cloak]

Isn't Cascade mostly usable on ComfyUI?

[eternal comet]

lol its usable for me

i can gen image in 90 second

[gusty cloak]

No like, A1111 has an extension but it's limited 🤔

[eternal comet]

oh

dont use that

youll overload cuda memory

[gusty cloak]

I guess I'm getting ComfyUI 😅

[eternal comet]

u dont need that

https://github.com/EtienneDosSantos/stable-cascade-one-click-installer

use this guys

[eternal comet]

"women with red blue and green stripes in her hair neon colours beautiful picturesque japanese women wearing cyberpunk clothes stunning. shes looking forward blushing extremely with a love heart on a digital screen on her chest showing shes in love. "

prompt following is insane

[clever oar]

how i train lora i have 4 gb vram

[eternal comet]

scrape pictures of discord egirls

this is where

i get 99%

of my training data

💀

[clever oar]

what is this discord egirl i try google)

[gusty cloak]

Egirls are so repelling to me

But then again who am I to speak

(I'm kind of not straight)

[eternal comet]

💀

bro

i see them as a win

but for their fathers...

nah kinda a loss isnt it

[gusty cloak]

🤣

No like for training, sure

[eternal comet]

yeah

for training

[gusty cloak]

👀

[eternal comet]

most of these egirls have severe self worth issues

theyd probably be honoured to be used for a lora

why else would they post their pics

they want attention

[clever oar]

kohuya ss

[eternal comet]

ye

[gusty cloak]

Could you rephrase that in English?

[eternal comet]

i download a bunch of their pics

lora "egirl hello kitty no daddy"

💀

or no father figure

my lora names weird af

seperate from modest people

to down bad people

to degenerates

to others

etc

[clever oar]

dont know how is name program for train lora

[eternal comet]

yo

also

i think i did it

needs more refining

but yes

i managed to make man kicking woemn @clever oar

[gusty cloak]

You can look up tutorials online, however I advise against training a lora on a 4-GB GPU

[gusty cloak]

Civitai allows you to do that, however it does cost 500 boosts if I got the name right 🤔

[clever oar]

close)

[clever oar]

mortal kombat)

[clever oar]

what you use for train?

i dont understand what is best standart for train it

[gusty cloak]

For my very first successful LoRA, I used Civitai (the website), and I made a LoRA that portrays a specific singer

[gusty cloak]

Because I couldn't figure out how to do it locally

[clever oar]

on civitai i dont see this

[gusty cloak]

Because this was my first LoRA, it's not the best, but it does look close to the actual singer! (Final photo is a real one)

[clever oar]

civitai is good way for it?

its safe?

[gusty cloak]

My LoRA seemed to be incredibly biased towards giving him a buttoned shirt 🤔 In the futuristic one he was supposed to be wearing futuristic clothes, but as you can see he's wearing something that resembles a black buttoned shirt

[gusty cloak]

Yes

However as I said you need boosts there unfortunately

[clever oar]

Today I’ll go to bed early for sure, now I’ll just look into discord for 5 minutes - you go to bed at 7 am

😃

[clever oar]

hard to train?

i need resize image for use it or civitai do this?

[gusty cloak]

No, generally, when training a LoRA, you don't need to resize your images nor crop them.

[clever oar]

its full image? ^^

[eternal comet]

💀

[gusty cloak]

https://tenor.com/view/bonk-gif-26414884

[crisp echo]

Pineapple

[jagged pelican]

A futuristic cityscape at dusk, where the architecture seamlessly blends organic and cybernetic elements, illuminated by the soft glow of bioluminescent plants and neon lights. In the foreground, a robotic owl perches on a metallic tree, observing the bustling life below.

[nimble mason]

Here is the image you requested.

[jovial tiger]

hardware comes this week

[nimble mason]

Awesome

[heady anchor]

redditors really cant understand sarcasm unless someone post a /s before the sentence

[round helm]

Clem is the CEO of huggingface

Unless your name is Elon Musk, CEOs typically don't just joke about that sort of thing on social media, not saying it will happen but I don't think it should be taken as a joke

[heady anchor]

even i can understand he was joking but as i said,redditors cant really understand that unless ppl post a "/s" before the sentence

[round helm]

sorry I can't help myself, too much irony in this: satya in realtime rn

[vernal otter]

People fear the truth

[night vine]

#💬｜general-chat

[covert pagoda]

what

[hazy warren]

he belongs on pizza!

[wispy nest]

👀

[hazy warren]

yummy

[hazy warren]

needs to go on Pizza

[hazy warren]

also could go on pizza 🙂

[hazy warren]

😮

[acoustic maple]

[gusty cloak]

StabilityAI internally:
https://tenor.com/view/fighting-star-wars-visions-sword-fight-sword-duel-disney-plus-gif-23428872

[blissful cairn]

I feel the urge to show Garfield to everyone

[gusty cloak]

Go ahead and show Garfield

[blissful cairn]

Oh, I already did. You can find it here #1207078178510872636 message

Good thing I searched before I went hunting for the picture

[gusty cloak]

🥲

It almost looks like Sailor Garfield except burned

[blissful cairn]

Iirc it was originally supposed to be a catgirl holding a tabby cat

Then someone suggested I ask for Garfield since it didn't understand stripes

[gusty cloak]

What model did you use? 😅

[blissful cairn]

Idr anymore lol

I'd say rev animated but it doesn't look like it

You can blame @pine jacinth for telling me to try Garfield. I found the discussion but not the prompt 🤷

[nimble mason]

Love this! And the Godzilla pineapple.

[gusty cloak]

(This will go on a brand new sentences thread will it not?)
Well I want to eat the sea now

🤣

[jovial tiger]

Can't, defended by pineapple warriors

[clever oar]

very nice

i want eat now

can you do frozen sea?

[clever oar]

mom pikachu

[eternal comet]

@clever oar

check this out

prompt following insame

"sitting in a ash tray"

[clever oar]

crippy

[eternal comet]

ye

wait