🏞｜general-with-images

[jovial tiger]

cascade

in the style of Tomer Hanuka and Tomasz Alen Kopera, A massive mutant with razor-sharp claws and glowing red eyes sits calmly sipping a venti-sized cup of coffee from a Starbucks cup as it gazes out the shattered window of a ruined skyscraper, its multiple arms cradling laptops and notebooks amidst scattered papers and broken office equipment.

[clever oar]

attack of the Titans

[robust path]

Cool, not sure if I can run it in Draw Things but I'll definately give it a try, thanks!

[nimble mason]

oh shit i forgot about that!

[clever oar]

sd3?

[brave summit]

.

My art is going too far

[royal monolith]

not only that, it's breaking this discords rules

[jovial tiger]

I was curious to recreate your picture. Sd3 2b should be repealed.

[nimble mason]

z

[languid pebble]

Good morning coffee ...

[brave summit]

Model??

[celest sigil]

https://tenor.com/view/its-bacon-gif-27063121

[wispy nest]

thats what I got

not sure about those chairs, think they have outlived their usefullness

[nimble mason]

That's stable cascade... Pretty much the best model imo

[wild sorrel]

I just got interesting thing, which makes me think if we can actually do duo's in suno? Have you tried it?
Like male + female voice

[languid pebble]

Haven't tried to make suno do this. Sometimes it adds voices to the chorus, sometimes the gender changes between the verses ...

[wild sorrel]

yea, I just got this, which never happened to me before, usually she was just starts whispering or losing her voice midway, but this is something else lol

I think it should be possible even if suno doesn't understand "duo \ duet" thingy, like get 2 voice seperately and hope it bugs out eventually and overlaps them lol

[languid pebble]

Sometimes it just don't listen if you say: male or female voice ... we'll have to find out ^^

[wild sorrel]

yea genres \ keys \ mood often weird too

minor key sounds too happy soooo often , I hate it 😦

[languid pebble]

It's like a mulish child sometimes 😄

But it also gave a cool twist ignoring my lyrics ....

[wild sorrel]

yea 😄

But...sometimes weirdness and not following instruction as strict is what makes song more interesting

[languid pebble]

I'm working an a few lyrics before I restart my abo ...

[covert zinc]

getting this isssue on a1111

[languid pebble]

Tried tunecore to sell my songs. But only find them in my youtube artist channel they created ...

[wild sorrel]

I don't think we can profit from it much, I think that could be fun addition to something else you are doing on youtube...as a big thing on it's own I think songs needs to be exceptionally good to interest alot of people...

or maybe if you make them copyright free and try to attract more people this way...

[languid pebble]

I know a few celebs ... but I more want to put them on sale for fun ...

[wild sorrel]

I see

[languid pebble]

It's like leaving a footprint ...saying: I've been here and done that years before everybody did 🙂

[wild sorrel]

hm, fair

[languid pebble]

Otherwise you need marketing and the right friends ... and I am not a fan of marketing ^^

[languid pebble]

Area 52 Restaurant 🙂

I've done some cool pictures with Area 52 warehouses and museums ...

[languid pebble]

😄

[candid surge]

D*oo||* loading

[pallid ruin]

Almost!

[pallid ruin]

So doom style is possible in sd3, i gotta use examples of other games

[candid surge]

hell yeah that one is good!

prompt plz?

my last trial was "classic doom 3d game, 90ies 3d game, doom the game, in game screenshot, 3d shooter, holding a shotgun, blasting monsters, dark dirty nuclear waste, bins"

[pallid ruin]

"voxel style
first person view,
FPS Shooter doom,
POV holding a gun and there are demons far away in mars, cinematic 1990 game inspired in doom counter strike and wolfstein 3d,"

[candid surge]

thanks LeoB!

[heady anchor]

Some fresh locations from https://Hexagen.World

[frank drift]

Is it common to get something like this?

[royal monolith]

depending on your settinngs, yes.

[frank drift]

Where might you predict I went wrong?

[royal monolith]

ther are a number of things you might have set to get that. what are your settings?

[frank drift]

Let me send a screen cap

[royal monolith]

okay

[royal monolith]

based on that, i think your content is violating some filters and is blocked.

[frank drift]

Huh, that shocks me.

[royal monolith]

there's a lot of tokens in your prompt that could pull up exactly the wrong data

you might try dropping the word vampire out of the prompt

[frank drift]

I'm not quite sure where to start. If I am running it locally, how can it violate filters? What's more, I got the model I was using from Civitai, they have a crazy amount of nsfw stuff.

Hmm. Trying it again seems to have worked

[royal monolith]

if you're running it local, then it's not going to be filters.

[frank drift]

I would have suspected

Running it a second time seems to have produced results

[royal monolith]

anything in the console log?

[frank drift]

Where can I check it?

Oh, the command line, dug

duh*

[royal monolith]

yeah. that's the console

(cause ancient history)

[frank drift]

Lol you're right, I should have known that. I know words >.<

It could honestly be my model, it's working with 2 different models

[unique condor]

still feature creeping on my crappy standalone diffuser for pixart sigma. if i get around to it, i'll figure out which dependencies it actually uses, make a requirements.txt and put it on github(I use this venv for a couple other projects). god i hate python...

[white pivot]

A teacher using AI images of her students to help motivate them

[vast parcel]

A photorealistic image of various colorful candies and sweets (including gummy bears, lollipops, jelly candies, hard candies, gumdrops) and glossy green and red apples floating in mid-air against a gradient background transitioning from light pink and blue to warm yellow and orange. The scene is highly detailed, vibrant, and energetic with a dynamic and chaotic arrangement. Emphasize glossy and shiny textures, playful and whimsical atmosphere, high-resolution, perfect lighting.

[royal monolith]

you can't generate in this channel, or most of the others on this discord. read the information at this link -> #🗣｜artisan-support-feedback

[nimble mason]

Here is the image you requested.

[past pelican]

Here is the image you requested.

[clever oar]

perpetual motion device

[steep sleet]

I used SD3-Medium (and some embeddings/textual inversions from A1111). The embeddings are actually SD1.5!!!

[languid pebble]

Hi Dicordos 🙂 Good luck you don't see what I typed first 😄

[clever oar]

hi what ?😎

[languid pebble]

I think there was an additional "k" 😄

[clever oar]

how could you do this?🙁

[languid pebble]

Switching between different topics and languages can be hard ... 😄

[clever oar]

I need time to come to my senses and recover from this

Mojo

[languid pebble]

YaY! Didn't have the time today ...

[clever oar]

in next day you need two cup

[languid pebble]

Need inspiration ...

[pallid ruin]

https://tenor.com/view/chicken-fbi-skeptic-chicken-funny-gif-14153035

[pallid ruin]

So cute 🥺 🤗

[clever oar]

thx)

[hearty violet]

stable diffusion to cpu https://github.com/rupeshs/fastsdcpu

anyone test it?

can I used? I have a notebook

[quick notch]

Create an illustration of a 3D wide-anime-style watercolor cartoon scene. ann, a lively 20-year-old beauty with waist-length black hair and reddish-brown eyes full of character. She wears a white T-shirt, and black jeans, adorned with a cross earring. tom, an intelligent 18-year-old teenager with a slightly thin face, trendy light golden short hair, and vibrant blue eyes. He wears a yellow T-shirt with interesting cartoon patterns and casual pants. A pair of black-rimmed glasses.

Ann and Tom are skiing down a snowy mountain slope, with Ann in a white ski jacket and black pants, her hair flowing behind her, and Tom in a bright yellow ski jacket and blue pants, his glasses slightly frosted. Snow-covered pine trees and a clear blue sky complete the background.

[royal monolith]

read the information here #🗣｜artisan-support-feedback

[nimble mason]

the real SD3

[nimble mason]

@onyx isle you def might want to check out stablecascade then

i'm pretty sure you could do this on 8gb

[onyx isle]

How about stright up comfy? 😄 <my fave>

[nimble mason]

that's the only thing i use

comfyui is the way to go

[onyx isle]

Do you have a workflow you are willing to share? 😄 Or is this all from a model you created?

[nimble mason]

yep of course i share 100% of everything i figure out

yeah go ahead and get cascade installed and lemme know what you're able to run... if you need lite or the full version

you should be able to do all this stuff regardless - right now i believe my lora only works with full C, but honestly the cascade base model is pretty damn good

[onyx isle]

It looks so innocent at first, from a distance 😉

[onyx isle]

How come SAI created cascade, when they already had SD?

How are you folks doing these? I went to the free online huggingface SC page and all my images sucked lol

[nimble mason]

i'm doing this on my computer with comfyui

with my custom nodes

https://github.com/ClownsharkBatwing/RES4LYF

[onyx isle]

You're the one who made that fried eggs SD3 stuff arne't you? 😄 (that one is awesome)

The download size of SC is kinda huge and my proper "modem" won't be here for a few days.. so I'm stuck with the huggingface for now

[nimble mason]

hahah yeah

the fried eggs and ham lora?

i didn't realize anyone ever messed with it

[onyx isle]

Windows 11 and I (I'm new to windows) had a disagreement about what "temporary" files are, so I can't show you, but the images turned out awesomely lol

Someone sent me a workflow with your lora in it...

and the images were amazing!

[nimble mason]

wow lol i'm surprised they were any good

i called it fried eggs and ham cuz i felt like i'd burned it lol

[onyx isle]

You should see some of the images that were made with your models, which people posted in the SD3 section of this forum!!!!!

[nimble mason]

awesome... had no idea, that's cool to hear 🙂

[royal monolith]

research what cascade actualy does 🙂

[royal monolith]

your comfy generations should be inside your comfy output folder

[onyx isle]

Nah, I deleted all my "temporary" files....

[royal monolith]

those should be in a folder labeled 'tmp'. not the same folder.

[onyx isle]

Windows seemed to feel that my downloads folder was all tempe, since it's set to go to google drive nonstop...

[royal monolith]

you can't genereate in this channel

[royal monolith]

that's not where the comfy output files go.

[onyx isle]

That sort of ai art is illegal

[royal monolith]

#🗣｜artisan-support-feedback

[onyx isle]

Hey don't get the good servers shut down due to that crap

[royal monolith]

no.
Lea la información aquí. #🗣｜artisan-support-feedback
Y no se puede crear imágenes con un mensaje como que en este discord. que le conseguirá eliminado de la discordia

[nimble mason]

Here is the image you requested.

[royal monolith]

@wispy nest

que no importa.
Ese tipo de aviso va contra las reglas de esta discordia.

gracias

haga clic en este enlace #🗣｜artisan-support-feedback
Y luego copie el texto de la página y póngalo en un programa traductor

[wispy nest]

https://media.discordapp.net/attachments/931925914470711346/1061076535525785680/321666B2-CC36-467F-B254-56055DEB1C8C.gif

[onyx isle]

So people with great GPUs can create models right within SC?

[royal monolith]

??

um, what?

[onyx isle]

So Getmini lied to me again! LOL

[royal monolith]

https://www.ikomia.ai/blog/stable-cascade-image-generation

read through that

[royal monolith]

when you prompt an LLM, specifically tell it not to guess and to tell you if it doesn't have 100% confidence in the answer to the request you are about to make - otherwise it WILL make stuff up as it MUST give an answer.

[steep sleet]

Did you use a Cascade LoRA on this one at all?

[nimble mason]

yeah that's my lora

workflow should be embedded

@onyx isle you seem to like this kinda stuff... 🙂

[fleet goblet]

Hm.

Well that isn't Fallout.

But it is power armor.

[finite lynx]

(Regarding this message: #💬｜general-chat message )

[steep sleet]

"When shall we three meet again? In thunder, lightning, and in rain?!"

[languid pebble]

Good morning coffee... I know ... SD3 and hands ... ... ...

[nimble mason]

zoom in

90 steps, directly generated, custom sampler... crazy, crazy detail

[steep sleet]

Cool!

Ouch! Detail so sharp it hurts!

[languid pebble]

The first one is pretty crazy ...

A bit like the black block in Odyssey 2001?

[steep sleet]

I knew the guy who did the sound on Kubrick's 2001

[languid pebble]

Wagner? 🙂

Oh no ... Strauss ...

[steep sleet]

"Bill Roe"

I mean the studio sound-desk mix

[languid pebble]

There's a fun story why Strauss became the Theme song. It's said that some sound ingeneers have been listening to it while Kubrick passed and so he got the idea to use it as theme song ....

[steep sleet]

"Daisy! Daisy! Give me your answer do...!"

Kubrick is ultimately a poetic director

So no doubt he would delight in 'poetic music'

[languid pebble]

I've used it as topic for some stuff 🙂

[steep sleet]

All space music seems to side with majesty and poetry - listen to Hans Zimmer's "Interstellar"

[languid pebble]

My A.I. 🙂

[steep sleet]

Visual poetry too!

[languid pebble]

SciFi Soundtracks started my interest in classic music ...

[wispy nest]

nice black octopus

[languid pebble]

Daisy Daisy gives me a shiver ... ^^

[steep sleet]

Poor Hal! I wonder what he's doing now?! 😄

[languid pebble]

Training Stable Diffusion Models ... 😄

[languid pebble]

SD3 is pretty HALish: "I'm sorry Dave! I'm afraid I can't do hands ..."

[languid pebble]

"I'm sorry dicordos. I'm afraid I can't do text ..." 😄

[clever oar]

hehehe

[languid pebble]

Latte art 🙂

[clever oar]

I think such a coffee shop would be in demand😎

[languid pebble]

Japanese are doing 3D Latte art ...

[languid pebble]

Hi HAL 🙂

[clever oar]

in the cascade they create mainly abstract images with a psychedelic style:?

[languid pebble]

I think it's more Clownsharks thing ...

[celest sigil]

No, that's just what Clownshark creates

@nimble mason https://huggingface.co/ostris/vae-kl-f8-d16

Looking forward to that SDXL version!

[nimble mason]

i want that with stage B so badly

[clever oar]

I have enough psychedelic in my life so I like a softer style 😃

[celest sigil]

It says that it's not compatible with SD3, so I'm not sure what it's for.

[nimble mason]

a new model project

you start with images, encode to latents, then noise them and train going backwards to the image

currently there's no models for them

[celest sigil]

https://tenor.com/view/travis-neil-primrose-still-waiting-im-still-waiting-waiting-gif-27380282

[nimble mason]

pass the time with some of these

[clever oar]

its looks soft

[nimble mason]

ah yes, the soft earthy psychedelia vs hard acid lol

[languid pebble]

I need to go! Enjoy the day/night/whatever!

[celest sigil]

Cascade

[steep sleet]

Another Lector-Snack™ Product!

[celest sigil]

A CSBW prompt 🙂

[clever oar]

I noticed an interesting feature of the model that makes people in the animation style listen better promt

like there are more details and the images are more interesting

[clever oar]

what they eat in desert?

[celest sigil]

Dried fish

[steep sleet]

SD3 via SwarmUI

[boreal crow]

Generate skibidi toilet eating cameraman

[nimble mason]

2880x1728
Prompt executed in 24.74 seconds

[celest sigil]

A "crispier" version 😄
3480x1992

It did take 130 secs though 🤷🏻‍♂️

[nimble mason]

ooooo

you found another resolution that works!

uckin awesome!

i'd tried something between that and what i've been running and it gave some horrendous mess, like two images half overlayed

[celest sigil]

That was using my SDXL i2i stepped upscaler

[languid pebble]

I wonder how long it would take on my computer 🙂

[celest sigil]

https://tenor.com/view/cpu-on-smoke-cpu-blasts-potato-pc-gif-24592403

[languid pebble]

hehe Most od the time I only run it in energy safe mode 🙂

https://estore.asus.com/de/rog/90nr0ip2-m00650-rog-strix-scar-18-2024.html

[languid pebble]

She doesn't need a light at home ....

[languid maple]

how to make an image

[clever oar]

first you need install sd

[languid pebble]

Or you can pay for: https://discord.com/channels/1002292111942635562/1237461679286128730

[clever oar]

Mojo

I eat oatmeal cookies

[languid pebble]

Hi dico 🙂

[wispy nest]

aww

[cloud wolf]

Made myself a gruvbox theme to match the rest of my stuff

Need to figure out what the variable for the sliders are though

Everything matches 😆

[languid pebble]

Phew ... youtube is slow ^^

[covert zinc]

Non upscaled vs Upscaled 2x
Model Animagine 3.1

[clever oar]

hey everyone give me image 20 k ultrarealistic ultradetailed and very very very cool

[wispy nest]

this one is really great, how did you get it so wide

[languid pebble]

It's a special Model in 2048x512 ....

[languid pebble]

ComfyUI-LuminaWrapper

[wispy nest]

ah yeah I know lumina

I forgot they trained it so wide

[languid pebble]

The composition example workflow ....

[cloud wolf]

Very cool, what upscaler are you using out of curiosity? The ones I've been using thus far seem a bit bad

[languid pebble]

It's a picture with 5(or 4) prompts ... I didn't upscale .... Magnific.ai should be able to do it ... or SUPIR

[wispy nest]

SUPIR is amazing yeah

[languid pebble]

Ohhh... I answered a question not asked to me 😄

[languid pebble]

Pretty cool! 😍

[clever oar]

thx😎

[languid pebble]

I have been cheated by an company ... they've made a competition but said: "Not enough entrees so canceled" afterwards ... so I've made a song 🙂 https://suno.com/song/2f746e8d-e515-4bcb-a0d9-7e601db54d99

They owe me 1000.- € ....

[clever oar]

go to court

[languid pebble]

Will be next step ... I hope they get the idea they go wrong with showing them what I can do ...

I am so stupid ... I don't play dirty ... I always hope to be treated OK ....

[wispy nest]

life is too short to play dirty

[languid pebble]

But some just do ...

[clever oar]

play harder

throw feces under their door

[languid pebble]

The price they have to pay will be much more ...

[candid surge]

night chat!

[royal monolith]

never pay to enter a competition

[pallid ruin]

Nice images! 👏

[pallid ruin]

cute! 🤗 🙂

[covert zinc]

Was using stable swarm's 2x upscaler.

[wispy nest]

its interesting that Clownshark Batwing probably makes the best images on the server
but he is using Cascade
maybe we slept on Cascade too much...

[valid urchin]

Guys, what is the possible PROMPTS to make this art

[celest sigil]

Lumina

[nimble mason]

what an AR

[wispy nest]

is lumina the best model for wide now

seems to be much wider than Chinook

which was the best SDXL model for wide

[celest sigil]

It is made for 2048, and you can prompt regions easily

[wispy nest]

ah I didn't realise- the lumina people also made SPHINX
its a good VLM

[valid urchin]

what's lumina here? how do i get this prompt ?

so is it just for wide image only?

[wispy nest]

not just wide

its got a range of sizes

https://huggingface.co/Alpha-VLLM/Lumina-Next-SFT-diffusers
its got a difffusers version
if you like that

[wispy nest]

I like the paint colours

[valid urchin]

so what's exactly lumina? an ai ? i dont get it

more likely a realistic dataset ?

[valid urchin]

how do u make this bruh? looks good

[nimble mason]

using a lora i trained on stable cascade, and some custom nodes and a heavily modified version of the RES sampler

[wispy nest]

lumina is a whole model

[languid pebble]

I didn't pay ... only with my time ... they will have to pay ... maybe not me directly ... https://suno.com/song/2f746e8d-e515-4bcb-a0d9-7e601db54d99

[wispy nest]

nice I love wide sci fi shots

[celest sigil]

Feed it into an LLM...

The image is a digital illustration of a futuristic cityscape with a large spaceship flying over it. The sky is filled with clouds and the city is made up of tall buildings and skyscrapers. The spaceship is in the center of the image, with two smaller spaceships flying above it. On the left side of the spaceship, there is a tall tower with a blue and purple light emanating from it, and on the right side, there are two smaller planes flying in the same direction. The city is surrounded by mountains and there are a few people standing on the ground, looking up at the spaceship. The image also has a quote at the bottom that reads "The last stronghold of all knowledge. The keep it where all the value reaches. A place to wonder, protect, and fight for."

Then use i2i to reproduce it

[wispy nest]

thats better lol

[valid urchin]

you meant, i upload the orginal image to LLM first right? which LLM? can u refer me ?

[valid urchin]

so i can learn faster and how the ai thinking here.

[wispy nest]

wow nice is that lumina

[celest sigil]

I use Florence2 locally, in ComfyUI

[celest sigil]

No, that's SDXL

[wispy nest]

ah ok

[wispy nest]

do you have like 6 passes

so many

[nimble mason]

cascade... this thing thinks it can spell lol

i got sdxl doing text pretty reliably but not cascade (yet?)

[celest sigil]

Look at the size of the output. It's adding a lot of detail while enlarging the image.

[valid urchin]

is this image, you recreate based on this prompt

The image is a digital illustration of a futuristic cityscape with a large spaceship flying over it. The sky is filled with clouds and the city is made up of tall buildings and skyscrapers. The spaceship is in the center of the image, with two smaller spaceships flying above it. On the left side of the spaceship, there is a tall tower with a blue and purple light emanating from it, and on the right side, there are two smaller planes flying in the same direction. The city is surrounded by mountains and there are a few people standing on the ground, looking up at the spaceship.

completely? that's insane if it looks that similar . im hardly to believe it
or u just enhance it?

[celest sigil]

It's a mixture of both. It is recreating the image.

You can drop any image in there and it'll recreate it

[valid urchin]

ok, i tot from zero the prompts. im tryn to see the result from assitant stable lol.
is prompt has limit characters?

[valid urchin]

even if it's just a sketch ?

[celest sigil]

The prompt is automatically added, and it'll try to recreate it from the original.

[celest sigil]

Should do, but not tried it

Some other examples here #✨｜sdxl message

The images below too

[valid urchin]

can u test with lumina, what's the result looks like.
it will be realistic I assumed ?

[celest sigil]

No, I don't think the nodes for Lumina can do it the same.

I've not seen good clarity in Lumina either

[valid urchin]

do u know dataset from hugface for illustration/painting above ??? imnot sure.
oh that's why i want to know how the AI detect it, is it detecting as painting or some kind of stuff (obviously not anime, or realistic)

[celest sigil]

You can just take the LLM text output and manually adjust it

[valid urchin]

what do u mean by this? the lumina seems has realistic feature but a lot of flaws which make it obvious as AI

[celest sigil]

It's just not sharp and clearly defined.

[celest sigil]

The first line from the LLM is The image is a digital illustration...

The workflow then uses other "magic" to transfer the style and shape of the image.

Then more to improve details and scale it up

[languid pebble]

Good morning coffee

[wispy nest]

nice

[crystal osprey]

kid

[eternal sand]

plane in city

[languid pebble]

\ /
+----o0o----+

[rugged dagger]

https://tenor.com/view/american-flag-red-white-blue-flowers-god-bless-america-independence-day-gif-15951756209039022096

Happy 4th of July to all of my fellow Americans.
Have a great Independence Day and be safe!

[deft bison]

Thanks a lot ! ...

[fair kiln]

Can anyone help how to make art like this

[royal monolith]

i'd probably use the term "3d" at the start of my prompt, and possibly also use "cell shading" and "cartoon" in the prompt

[candid surge]

something like this? Not exact style copy, just some trials manually prompting:

POS: painted semi-realistic cartoon illustration, two Turkish men riding in a wooden flying bathtub, sunny, kid's show, (home trees and blue sky:0.8), 2d hand painted style, lowres, jpeg artifacts, low quality jpg
NEG: legs, (naked, nudity:0.85), (3d, real, photo, render:1.3), (elf ears, pointy ears:0.9), line art, sketch, masterpiece, best quality

[wispy nest]

Cat

[fair kiln]

Hi this is awesome! I have a little story line for you. If you could give me an output for it.

There was once a poor man whom a mermaid gave an enchanted sea water. With help of this enchanted water he opens a magical water park in village. Once a man in village knows about this. He then steals mermaid but in the end mermaid is saved and the man i s punished.

So based on this story can you make a similar output?

[opaque torrent]

This is a superhero I've been working on I think this image came out pretty good. Shes named Madam Mercurial. She's a psionic superhero

[wispy nest]

А

[fair kiln]

Like i want a mermaid giving an enchanted sea water cup to poor man image. What would be most closest prompt would it be?

[candid surge]

im working on it, as I image the poor man being punished by the powerful mermaid lady

a bit like the last two episode of too old to die young

[fair kiln]

Poor man given an enchanted water cup by mermaid

[candid surge]

im going for the end scene

[fair kiln]

Alright

[candid surge]

Saved mermaid taking her time punishing the man from the village who first stole her. Like tit for tat, action equals reaction.

sd3

[fair kiln]

This is great! What are the prompts? My bad i cant run sd3

Other good 1.5 checkpoints for similar art?

[candid surge]

not sure, I guess sd15 doesnt have the prompt flexibility

maybe ill try later on

need to go out to the store now

[fair kiln]

Alright

[candid surge]

store is close by, im back

[languid pebble]

I think User @crisp stream knows a good checkpoint for you. I always forget the name ^^

[clever oar]

best model for art like you want uber realistic porn merge

[candid surge]

oh how i wish sd3 could do a bit more naughty stuff

[languid pebble]

It can ....

[candid surge]

oh yes i know 😄

but, it has a very high tendency to remove stuff like nipples, explicit views, etc

im sure it would be better with horror gore than nudity (not tried, not planning to)

sd3 is still good enough for some sfw sexy woman

[languid pebble]

I need to try my fallen angel again ...

[candid surge]

Go ahead, just you stay away from my metal armor guardian angel 😄
and green haired green eyed Rin
and dark skin black haired blue outfit woman
and Caitlin

Caitlin, a high class brown haired female with purple eyes. Studied behavior management, but also loves to use the spiritual forces. Tool and methods matter only little for Caitlin compared to results. Cute looks hide her cold calculated brain. Wears red earrings and you won't see her without expensive custom made glasses. It's not for her to see, but for her looks.

Caitlin, back in the facility, home sweet home as she just loves her work. Current project active water supplements on inmates to stimulate behavior adaptation. Some subjects needed to be terminated but the test group is still large enough to present positive result to upper management. Who needs free time anyway, when you enjoy the work!

[languid pebble]

Oh yes ... my fallen Angel is to hardcore to post it here ....

[candid surge]

just order her to put some clothes on

you might need to cut open the back of the outfit twice, for the wings

ai might have some issues with that

also, side view with front view wings ... common ai thing

im going to test if SD3 is any better in side view angel wings

[languid pebble]

Nah ... it's a fallen Angel inside a slaughterhouse ... it's not naked but to bloody and some organs lying around ...

[candid surge]

ohhw. I see.

make it rubber toys!

wait . that might be worse ...

[languid pebble]

I didn't prompt the organs 🙂

[candid surge]

I love to prompt heart

[languid pebble]

Eyes can be fun, too ....

[candid surge]

going to try "cute bloody real heart" ... im scared !

lol! turned out now so bad. could be english liquorice candy 😂

[languid pebble]

The heart thingy ...

[candid surge]

This is what i mean!! Damn you ai, get those wings on her back in the correct orientation! How hard is it to get SIDE VIEW WINGS?!

[languid pebble]

😄

[candid surge]

And all the cats jump up and say "HEEEY!!"

[languid pebble]

The try to create a cover for my song ...

[candid surge]

did you make an AI song, or do you make music yourself?

[languid pebble]

A.I. Song ... hope youtube releases it fast ...

[candid surge]

I tried some stuff on suno

[languid pebble]

It's pretty good ... for my stuff

Your angel 🙂

[candid surge]

damn you.
jealous

[languid pebble]

Easy prompt and SD3 "Angel of Spring by Tim Burton" maybe I just had luck?

[rugged vector]

i liked the way this image went and wanted to fix up the weird railing supports on the right. using comfyui, is there a way I could inpaint that section using the correctly spaced supports as a reference?

[karmic peak]

running AI on a budget build

[hearty violet]

why sometimes my controlnet dont is generated?

[hearty violet]

guys can I put a unique prompt to each controlnet?

I want use controlnet with masks to generate each architectural element with a unique prompt: floor, ceiling, wall, with a differrent prompt

it's possible on a11?

[tidal sandal]

I trained my own model, but for some reason I am not able to select it in the web UI

[spice wave]

Did you try turning it off and on again

[tidal sandal]

yeah I did
got the same result

[unique condor]

Change settings to bring your vram usage down. You don't want models to even use a megabyte of shared memory. What app are you using for diffusion?

[clever oar]

bullet fingers😃

[karmic peak]

uhhh

funny stuff, I have comfyUI running which has a ton of models loaded (CLip vision, Ultralytics face detect, some SD model, ipadapter models, ...)
and then I also have a python server that loads a ton of image background remove models (u2net, u2net-human-seg, bria, ...)
everything runs fine for some reason

[wispy nest]

these flower ones are some of the best

this workflow does greens really nicely

[languid pebble]

Good morning!

[quiet current]

Regional Prompt Control might be a good place to look at : https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111/wiki/Regional-Prompt-Control

[hearty violet]

well I do a thing

hahaha thanks a lot ^^

@quiet current can I make a region more detailed, like this?

with some kind of masks, or colours

[wispy nest]

generally there are two types of regional stuff

1. control net, which includes inpainting

2. modifying attention scores

[hearty violet]

can you talk more about it? I dont understand

[wispy nest]

do you know what control net is?

[hearty violet]

yes I'm using it

[wispy nest]

ok so control net is the first way to do regional stuff

the second way involves attention scores

the denoising Unet or the diffusion transformer has cross-attention layers and self-attention layers

with attention scores between tokens

and these scores can be manipulated

[hearty violet]

ok, first part

how I use controlnet to do regional stuff with inpaiting? I never do it I think

and how I relation a prompt to various controlnet?

[quiet current]

There was "latent couple" but it hasn t been updated in a while and I think it s broken by now. You can try this fork of it which seems to be the most recent one https://github.com/ashen-sensored/stable-diffusion-webui-two-shot no promises it will work tho.

[wispy nest]

there's a couple of different forms of inpainting
cos some models are trained with inpainting in mind
but you can also do inpainting on models that are not trained for it, by substituting in different latents during the generation process
neither of these methods require control net
however you can combine this with control net perfectly well, in the same way that you do with non-inpainting workflows

[quiet current]

and yes you can also achieve somewhat similar results with controlnet even tho if it s a little bit more complex imo

[hearty violet]

nice thanks, I will try it

[wispy nest]

another type of regional generation is modifying attention scores

there was a library called Omost that went viral recently that does this

https://github.com/lllyasviel/Omost

this sort of method waits at each cross attention and self attention layer until they have generated their attention scores
and then it changes them

to boost attention towards certain tokens in certain regions

[hearty violet]

yeah I heard about it, I don't try I will check it thanks ^^

[wispy nest]

there's so many good methods
another nice one is Differential Diffusion
which is another one based on latents
https://github.com/exx8/differential-diffusion

[hearty violet]

man I started to study about the theorical part of sd

you use so many terms, I recognize some, know others but now all

but I'm on the way

[languid pebble]

It's pretty hard work ...

[wispy nest]

yeah its really hard

[languid pebble]

Things are always changing ... and there are so many technics and models I sometimes mix it up ... that's normal

And comfyUI uses different scales that A1111 or SDForge ... that can mess up stuff, too

[wispy nest]

its probably easiest to learn inpainting first, including Differential Diffusion
then control nets after that
and then attention map injections after that

[hearty violet]

so many things to learn haha its cool ^^

[hearty violet]

working 🙂

[languid pebble]

It's fun, too!

[hearty violet]

I'm a bit addictad in it rss

some times trouble my work... I'm a home office architect

[languid pebble]

Yes sounds like A.I. ist becoming a tool for architects. You are not the first one I meet.

[hearty violet]

I use almost 95% for it

helps a lot

[languid pebble]

I can imagine that. Changing a sketch into a photo like picture, etc. Or just adding some creativity. It's an interesting time.

[hearty violet]

exatcly 🙂

well, I unlocked a new way to do things 🙂

with your help, thanks

[languid pebble]

I didn't help much. That's NeonNinja work. But it's also fun to help others when you see they want to learn and get their ideas. Many just come asking to do the work for them cause they don't really want to lear 😄

[hearty violet]

first time I understand batter the AND

the extension create it throught mine prompts

I like that one

wow I'm better, I'm happy haha ^^

[languid pebble]

Thank god NeonNinja has been here. I'm not really using this part of A.I. a lot.

[hearty violet]

it's for a psychoanalyst client, she wants a clinic with a batcave concept

relax thanks for your attention

[languid pebble]

Don't forget the Batmobil 😄

[hearty violet]

hueheuhwu

I will talk to her

[unique condor]

Ahh, it might be pushing some of the inactive models into that shared memory then, instead of system memory(they both use the system memory, but one has a lot of weird hoops to accompany it). Thought that was for a single task. If it was for a single thing, you'd know real quick because a normally 30s generation might take more like 300s

But still, I wouldn't let things spill into shared memory like that, it can randomly spell disaster. There's a reason why Nvidia finally added in the cuda sysmem fallback policy option in the control panel a while back

God I just saw your ram usage as well lol, probably thrashing the hell out of your page file on your drive as well

[languid pebble]

Good morning! 🙂

[clever oar]

😃

[languid pebble]

Need to get something to eat. See you later and have a nice day!

[clever oar]

kofee on all year XD

[hearty violet]

https://tenor.com/view/weezing-pokemon-pantallahumo-gif-7569194

you say coffing?

[clever oar]

i try to generate him but fail

[hearty violet]

haha ^^

[hearty violet]

red pikachu rss

[quiet current]

12gb should be plenty enough of vram, most likely you re not using it correctly (eg : asking for native 4k output)

[clever oar]

i have 4 gb

your gpu not budget

[unique condor]

I work with an old 2080 FE that has 8gb vram with zero problems. The only time I have dumb things down is I'm using ipa and a cnet like depth. Even then, I just drop depth down to 128 instead of the 256 for a tiny loss in quality

But I stay in the 1 megapixel range, any higher would spill out of vram

[grave quest]

can i use 生成

[nimble mason]

.

.

[candid surge]

For Chris, to show a point. Anime female in red bikini. SD3. 1st 2 hits.

[languid pebble]

Arms a bit long ^^

[candid surge]

hm, yes. But no bad hands!

[languid pebble]

😄

[wild sorrel]

just do hands behind back, no hands = no problem

[languid pebble]

Or: sitting on hands 😄

[candid surge]

hmm, best hands, good arms, white fillers almost popping out!

[snow hatch]

https://civitai.com/models/558233?modelVersionId=621417

[clever oar]

need to cook food

[languid pebble]

Style transfer?

[brave summit]

It's so good

[languid pebble]

ISS will be killed ...

[clever oar]

Mojo

[languid pebble]

Spacecat 🙂

[clever oar]

I cook potatoes with paprika in the oven

[languid pebble]

Yummy!

[clever oar]

Do they eat oatmeal in Germany?

[languid pebble]

And some Tzaziki?

Yes 🙂

[clever oar]

chili papers

[languid pebble]

Almost impossible to get a good result with SD3 ...

[clever oar]

with object possible

with animal and people lottery

[languid pebble]

But I need a dancing man and Text ....

[languid pebble]

0 of 20 pictures good ....

[languid pebble]

The only seat he can't lean back 🙂

[languid pebble]

Need to sleep! Good nite!

[brave summit]

Gg gn

[crisp stream]

Dreamlike Art Photoreal 2.0

@fair kiln

[hearty violet]

what is the difference between safetensor and yaml?

[ripe bluff]

woah

[royal monolith]

from huggingface "safetensors is a safe and fast file format for storing and loading tensors."

[hearty violet]

I consufe safetensor, with prickle, yaml, xformers, transformers, diffusers

I will google it

[royal monolith]

might start a document, write down what you're finding, so you can refer to it in the future

[hearty violet]

I'm recording a architecture course of sd

I started with the theorical part, wanna see?

[royal monolith]

sure

[hearty violet]

but is all in portuguese rs

but you will can see

must I upload here or dm you?

[royal monolith]

DM me

[hearty violet]

ok

[pallid ruin]

Amazing images bro @nimble mason 👏

[wispy nest]

yet fan boys are defending sd3 trash

[royal monolith]

why are you here?

[wispy nest]

that model which was released is under trained

[royal monolith]

pretty sure he's using cascade - since that's what he uses mostly

[wispy nest]

keep yapping

[royal monolith]

blocking someone so you can't see what they are saying is not necessarily in your best interest

[unique condor]

Sd3 is great... If you're not trying to make people with visible limbs. Personally, idgaf about making people, so of the stuff I like to make, it's awesome

[wispy nest]

if you dont give a f about anatomy then dont bring half baked crap

that sd3 model has some advantage over sdxl when it comes to object and scenic renders but anatomically its horrid garbage

and anyone refusing to admit that is simply foolishly ignorant

[pallid ruin]

sd3 is a good model, or at least a good arquitecture, of course it has some problems, but it can generate good images

[wispy nest]

from what i explained tha'ts a pathetic display of your reasoning

[pallid ruin]

undertrained + bad anatomy, that´s what you said

[wispy nest]

im going to ignore you

dont wanna get into shitty argument

[pallid ruin]

https://tenor.com/view/yoo-that-crazy-did-i-ask-i-dont-remember-askng-dr-fate-injustice2-gif-16579106

[wispy nest]

i also said you just portrayed your level of poor reasoning, the whole SAI team admits sd3 2b is not fully trained and here you are regurgitating what the fan boys been yapping about, i've tested sd3 when it came out, so have many others, its total garbage with human anatomy, i can't believe i have to repeat this to you

[pallid ruin]

Where did I say I was good at anatomy? I mentioned a lot of things he was bad at a few weeks ago, anatomy, it lacks knowledge about a lot of styles, games, and aesthetically it's not the best

[wispy nest]

then stop defending it

rather make correct crtique and help the dev know it needs to get better

[unique condor]

Reread what I said... If English isn't your primary language, I can gladly reexplain it Español, Deutsch or 日本語, if needed

[royal monolith]

SD 3. just plain sd3 medium, and no negative prompts. anatomy seems fine

[wispy nest]

your ego is in your way of clear thinking, lets quit talking about this nonsense on a loop

and just to make things clearer for you, i'm talking about anatomical problems with sd3 and if you dgaf about it then dont say anything

[royal monolith]

love how he continues on about bad anatomy right under an image of perfectly good anatomy

[wispy nest]

@royal monolith you are special kind of brain dead

people like you do more harm to SAI impression than good

[unique condor]

".......and anyone refusing to admit that is simply foolishly ignorant" when I had literally just said that sd3 is great if your not trying to make people with visible limbs.

This made me think that either you had some kind of reading comprehension issue and/or that your primary language isn't English. So no, it's an ego thing if I'm asking you if you want to communicate in a language you're more familiar with; out of the list of languages that I'm conversationally fluent enough in to converse with you in...

[wispy nest]

if you have basic education, re read what i said what that implied

[unique condor]

Practice your writing skills then and learn how to address things in a more complete manner. You know, the skills you learn in sixth grade...

But I'm not going to argue with an obvious troll that has stunted emotional maturity levels and that's just going to stonewall any attempt of discussion. You've made up your mind, we get it...

Have fun making your waifus with 1.5

[wispy nest]

you are being a nag

blocking you for sanity's sake

[languid pebble]

Looks like a good nite Brandy ...

[steep sleet]

SD2 Openjourney_V2

Openjourney_V2 is excellent at grunge, dystopia, the darker side!

Photoshop and Nik Filters

[languid pebble]

Prompted "cyberpink" instead of "cyberpunk" 😂

[hollow bane]

This is my latest project, artworks are done with ComfyUI and custom-trained Loras.

"The Kumma series, comprising 400 unique 1/1 editions, unveils a surreal tapestry of textures and forms. It blends organic elements with AI glitches, elevating them through evocative representations of bear-like mythical creatures."

the link to the NFT-collection: https://objkt.com/collections/KT1VW2BHWQ28toXqVyPfT1W7EFTFNUosDD8X , current floor price is 0.5 tez (approximate .60 euros).

[steep sleet]

NightVisionXL checkpoint model and sdxl_refiner_1.0_09._vae - SDXL prompt = dustbowl, okies, american gothic. homeless people, style of andrea kowch, norman rockwell, dystopia, chaos, despair, frail children of dust, anger, shout, shriek, anguish, vintage cars, shacks, wind, dust clouds,

[steep sleet]

NightVisionXL refiner, AUTOMATIC1111 (SilverFX filter for B&W efect) Prompt = dustbowl, okies, american gothic. homeless people, style of andrea kowch, norman rockwell, dystopia, chaos, despair, frail children of dust, anger , shout, shriek, anguish, old rusted vintage trucks, shacks, wind, dust clouds

[steep sleet]

Dustbowl SquareBob!!!

[languid pebble]

You make me wanna post kitten pics 😄

[clever oar]

Are you by any chance from Japan? They love this stuff

[nimble mason]

nope, from the US

but you are right on that 🙂

[languid pebble]

Big market 😄

[clever oar]

What is your inspiration for creating all this anime?😃

Mojo

i buy cherry

[languid pebble]

Yummy!

Found an interesting service ... it's creating websites on demand ... I asked for an A.I. Artist website and ir created one including a picture generator ^^

[clever oar]

soon people won't be needed😃