🏞｜general-with-images

[cyan shoal]

damn

[minor fractal]

the current main dev of wdxl

[cyan shoal]

ahhh makes sense

I wish I could request a prompt

[minor fractal]

if its weeb then they'll probably say yes

[cyan shoal]

also these look waaaay better than the previous early access images we've seen

[cyan shoal]

dayum

[minor fractal]

theres also a bunch i cant post (sd3 folks havent said "yea sure")

[cyan shoal]

I wish it was tested with stuff like minecraft

to see how much it has gotten its likeness improved

[pallid ruin]

x2

[grizzled sage]

just throw it out the window!

[minor fractal]

lol

[cyan shoal]

LMAO

[minor fractal]

i'd post the gens comfy made but i think they posted it here already

[cyan shoal]

ah hell yeah comfy made some nice ones, please post them

I mostly saw like 2-3

like the foxgirl one on the rooftop holding a gun

[minor fractal]

comfy only posts fennecs (for obvious reasons)

its not blood, its paint (trust them)

[cyan shoal]

damn that tail is super fluffy

like ridiculously

well these look nice so far

[minor fractal]

early sd3 example

[cyan shoal]

LMAO THOSE SNEAKERS

well its quite epic

I wonder if the model itself (not T5) will bear quantization better

[minor fractal]

it probably will

[cyan shoal]

idk if the rule applies for diffusers too

[minor fractal]

DiT is weird

[cyan shoal]

but 8B sounds a lot

[minor fractal]

it can barely run with 0 offload on 12gb iirc

[cyan shoal]

llama-7B and mistral-7B quantize well

lol

[cyan shoal]

at fp16?

[minor fractal]

if you can run llama7b with some offload you can run this perfectly

[minor fractal]

the DiT and CLIPs

t5 at 4bit

[cyan shoal]

llama-7B at 4-bit can be run without any offload with 12GB

[minor fractal]

lol

[cyan shoal]

LFMAOFAOFO

HELL YEAH

I want to make stupid crap like this with SD3

This is amazing, thank you

[minor fractal]

thank mcmonkey and comfy for the images

[grizzled sage]

i'm certain that fp8 will work without problem on sd3. maybe not at launch but it will

it'll be one of the immediate optimizations

I was using lavi adapter, the t5-large model, on sd15 the other night. Neat code. Wish i knew more about diffusers and i'd wing porting it to a UI

[cyan shoal]

🙏

[cyan shoal]

fp8 slows stuff down unless we use tensorRT, then it actually becomes faster

[grizzled sage]

only on older cards. Ada cards and newer are fine

[minor fractal]

fp8 only faster on ada

[grizzled sage]

fp8 is basically a free operation

[minor fractal]

tensorrt is the same

(for nvidia)

[cyan shoal]

yeah that's why it would kinda suck

it leaves 30 and 20 series in the dust

and uhh turing and older 💀

[grizzled sage]

well i mean, they will be. as is tradition

[minor fractal]

16xx is dead.

20xx is next

[grizzled sage]

executioner sharpening his axe while the head of 20xx lays on the choping block

[pallid ruin]

Lmao, almost but not quite there

[cyan shoal]

lol the hand

[minor fractal]

16xx can barely even do xl with comfyui

[cyan shoal]

yeah

[grizzled sage]

people with ampere cards should just git gud

[cyan shoal]

fr

[minor fractal]

16xx cant even do fp16

[cyan shoal]

im surprised turing could run these

[minor fractal]

box test #2
a red box on top of a blue box on top of a green box sitting on a sand dune

tag-based prompting is dead

[cyan shoal]

nice

[minor fractal]

NLP is meta now

[cyan shoal]

once failed guys!!!

this model is poopy doopy!

it failed once, therefore we should all go back to Stable Diffusion 1.2 and generate big boob ladies

[minor fractal]

"omg it put another box!!!!! delete model!!!"

[pallid ruin]

I hope that happens finally

[grizzled sage]

ada coming for turing

[grizzled sage]

you say that, but its not

people will still throw tags at prompts

the training scene is pretty tag orientated too. people love their shuffle tags and dropout options

[minor fractal]

bad habits live on

or something like that

[grizzled sage]

regional prompting . i messed it up and wrote blue on top though

AHA. i dont need no stinkin sd3 ||i actually do i really want an invite||

[cyan shoal]

fr

how did you do regional prompting

[grizzled sage]

regional prompt extension. prompt looks like this :
stacked cubes in the desert, sunshine, photography, kodak ADDCOMM
Sky ADDROW
red cube ADDROW
blue cube ADDROW
green cube ADDROW
sand dune

[cyan shoal]

what gui

[grizzled sage]

automatic1111 1.8

https://github.com/hako-mikan/sd-webui-regional-prompter

[cyan shoal]

ahh automatic

well thank you anyway

[grizzled sage]

theres regional setups in comfyui too

out of every 10, there's maybe 1 or 2 that are sort of good lol. regional prompting is still kind of finicky

[grizzled sage]

made these with regional too, thats why it gets two characters that look like alan turing and ada lovelace reasonably well. without making them cousins

SDXL we dont need no SD3!! ||nobutyeah, we really do||

[cyan shoal]

SD3 would replace the need for a region prompter

but not stuff like openpose and other controlnets

I will abuse img2img as much as it can go

that way I can get the closes to what I am imagining

[grizzled sage]

regional prompting will still be flexed with sd3. i'm sure it will have a lot of leverage over the model still

[eternal comet]

ok finger control increased.

ye its geting better

you can check meta data

for my prompting and setting etc

theyre getting better at hands

[eternal comet]

cool

[clever oar]

20 steps is your default setting for all generation?

[eternal comet]

tbf i put 30

if i want big details

it takes me 30 seconds to gen

or 20

[clever oar]

i check png info

[eternal comet]

yeah

my config there

in the meta data

my entire setup

[clever oar]

i use 30 steps

[clever oar]

because i have 1.5 base model

[eternal comet]

ah

which gpu

[clever oar]

1050 ti

[eternal comet]

how long

[clever oar]

1 image 30 sec

[eternal comet]

this model is good

it listens to my prompt

[clever oar]

why krea ai generate in 1 second

why sd cant this

[eternal comet]

image quality is shit tho

[eternal comet]

fingers are correct now

ye fingers are being consistent now.

🥶

most badass image generated

gotta love the control

it actually listens to my prompt now

wow

[wispy nest]

Runway haiper testing

[clever oar]

what is it

[wispy nest]

Runwayml ai video animation

[clever oar]

what need for animation

[clever oar]

1 image?

[wispy nest]

Done pretty much every scene possible

With pics bushing limits

Pushing

[wispy nest]

Test video is off one immage

[clever oar]

can you do this with real photo

[wispy nest]

Yes

Any photo

Or promt

[clever oar]

can you show something with photo

[clever oar]

oh is not sd extension ,sad

[wispy nest]

Cooking

[clever oar]

how long its make

[wispy nest]

A few minute

1 to 2

[eternal comet]

biden with blue hair

[clever oar]

emo)

[eternal comet]

real

[eternal comet]

what gpu

[wispy nest]

Am unsure there is a lot of settings

You can increase the intensity of motion and specific areas of motion or auto detect

[wispy nest]

I mean I use 3 devices running diffent versions of snapdragon and a chrome book so nothing too fancy

No m1 chip

[clever oar]

not bad but after second i not feel that it biden)

not him mouth

[wispy nest]

The thing is these video models are cool but they actually mostly don't work

I've tested almost all

And wouldn't pay to make it

There is small frames that are new creation a. Glitch in the code

Even from a photo

[clever oar]

skorchekd make for me something

[eternal comet]

fingies

[eternal comet]

tell me

give prompt

i make

[clever oar]

fat pig-faced woman with a bag of money

[clever oar]

no pig face(

[clear gorge]

a women,

[clever oar]

need pig face

[clear gorge]

a women with a little boy

[wispy nest]

lol

try a anime bank robbery

[clear gorge]

how can I use it

[nimble mason]

Here is the image you requested.

[clear gorge]

WTF

[clever oar]

what you mean

[wispy nest]

https://tenor.com/view/sunmanosdetijeras-gif-23377901

[clever oar]

close)

i need more!

[clever oar]

but all details

[wispy nest]

bruh

not bad

[clever oar]

yes its close

[wispy nest]

https://tenor.com/view/spirited-away-excuse-me-pig-gif-11747538

[clever oar]

best)

but need all details

[wispy nest]

someone showed me that movie like when I was 4 and it was mind blowingly creepy and unseeable

maybe was the time I guess I mean high quality animation

[eternal comet]

what

the

fck

[wispy nest]

horror film movie

[signal elk]

fill the shadow of the image with coffee beans

[nimble mason]

Here is the image you requested.

[clear gorge]

OMG haha, please provide some pretty ones

[clever oar]

nice

[nocturne oak]

Yo. Read the server rules.

[nimble mason]

Here is the image you requested.

[wispy nest]

is a snout

face swap mortf

[wispy nest]

why is lil peep sensord

I just tried to use him as a modification and it didn't go

[eternal comet]

💀

you can generate logan pauls wife

[wispy nest]

biden but its a robot chicken claymation

[wispy nest]

looks sus mang

[eternal comet]

💀 bro can generate logan paul wife

[grizzled sage]

who?

[wispy nest]

backround real

[eternal comet]

AHAHHAHA I MADE IT GENERATE A BITCOIN

ON HER

IRONIC BECAUSE LOGAN PAUL DID CRYPTO SCAM

[wispy nest]

crypto zoo

I've heard'

the bigger scam is open ai claiming they are open when it sits behind a pay wall

even Microsoft is figuring that out

[eternal comet]

ngl

im happy with my current setup

fuck open ai

my current setup is good enough

[wispy nest]

bros not building a ai chip for 7 trillion he's making it sentient

[signal elk]

please fill the shadow of the image with coffee beans

[heady anchor]

sorry i dont like coffee

[eternal comet]

prompt?

[clever oar]

fat woman covered in chocolate with the face of a pig holding a plate of chocolate sausages

🤭

[eternal comet]

let me try recreate this

im gonna use prompt expansion

if you use the word "hot" its basically prompt expansion tbf

[tawny widget]

what's the most photorealistic base model for portraits?

[clever oar]

epicphotogasm

[nimble mason]

depends

def rec taking a look at imperfectportrait

[grizzled sage]

city in a mug. beauty

[clever oar]

thx

[dull flame]

bottom right is pretty fire

"glass of water" right

[grizzled sage]

solid gold there

[heady anchor]

that can be fixed with any img editor

or keep trying with diff prompt combos

[dull flame]

yeah im trying in krita ai right now

lol yeah that didnt work

it rly likes the juice or whatever it is

maybe im not using the program correctly

[grizzled sage]

idea. negative prompt for corn. since corn has the juice

[heady anchor]

(glass of water,clear water,crystal glass:1.8)

[dull flame]

u can go 1.8? i though 1.5 was the highest

[heady anchor]

u can go as high as u want,keep going until the img becomes overcooked

[dull flame]

it tends to get cooked for me at 1.5, so im always around 1.2 or so

opening the oven

oh shiet

it seems like brackets matters a lot

soon as i removed crystal, it was back to juice lol

[heady anchor]

yea it depends on the model

[dull flame]

im trying different weights now

[heady anchor]

bad models can only create certain things

[dull flame]

i think its a good model, its just me not doing it right

im getting closer

[heady anchor]

yea u have to play with the prompt

[dull flame]

now glass of water has turned into actual glasses, prompting is pretty hard

[young lagoon]

/game_stats ：A little boy is studying hard at the desk. The background is the study room. His hair is messy and his expression is panicked.

[nimble mason]

What's your prompt

[heady anchor]

yea depends on the model and prompt

[dull flame]

(girl), (white kimono), black hair, (pink eyes), pale skin (detailed eyes:1.2), (full shot:1.2), Wednesday Addams, holding a glass of water,

(white background:1.2),(simple background:1.2),(glass of water:1.2),(clear water:1.2)

[heady anchor]

too many words,too many weighted words thats probably why

[dull flame]

perhaps

[dull flame]

feels like i need to weight a lot for it to work tho

[heady anchor]

(white background,simple background:1.5),1girl,white kimono,black hair,pink eyes,(full body:1.2),Wednesday Addams,(holding glass of water:1.3) try with this one

[dull flame]

will do

[clever oar]

i try it in my model)

[dull flame]

lol its going orange drink there as well = P

[clever oar]

what need without orange drink?)

[wispy nest]

🥱

[heady anchor]

try with holding cup of water instead of "glass of water"

[dull flame]

as u can see, as soon as u dont weight and use a lot of brackets, it gets pretty loose

i think some prompts just works better depending on model

[heady anchor]

yea it depends on the model

[dull flame]

ill try again with my original prompt, and use more negative prompt

[clever oar]

number in lora its power?

[dull flame]

what?

[clever oar]

is it strength or weight

<anime girgl:1> 1 is strength?

[dull flame]

its weight

but yeah, same thing

[clever oar]

oh ok

i thought is version)

[dull flame]

nope : )

[clever oar]

hm with 1 face distored

[dull flame]

@heady anchor turns out its mostly the pink eyes thats messing with the color of the drink

[heady anchor]

oh now it looks good ,u can try with "cutoff" extension,it helps to remove prompt bleed

[wispy nest]

chewy rice

exterior chewy rices

[dull flame]

i had no idea that existed, thats awesome

[nimble mason]

just fyi they're not kidding when they say it doesn't work well with sdxl

[dull flame]

thanks for the warning, im using 1.5, thankfully in this case

[nimble mason]

yeah i wasted a bit of time trying

[dull flame]

starting to get better results with that extension

[strong jungle]

Generate an image for the brand "Heritage Tea." The design should feature a realistic depiction of a Rajasthani fort gate adorned with intricate floral decorations. The scene should evoke the essence of Rajasthan's cultural heritage, with warm tones and vibrant colors. The brand name, "Heritage Tea," should be prominently displayed on the gate, colored in a striking shade of blue to contrast with the surrounding hues. The image should capture the feeling of embarking on a journey through time and tradition, inviting viewers to experience the richness of Rajasthan's architectural wonders and the premium quality of Heritage Tea.

[nimble mason]

Here is the image you requested

[strong jungle]

😂 lol

[crisp stream]

DPM++ SDE Karras is the best if you ask me.

[rain gazelle]

these are really neat.. reminds me of old marvel trading cards

[wispy nest]

ty someone finally noticed my art 🙃

[humble oracle]

guys, do ya use the 2M-image dataset or the much larger one in the competition

[nimble mason]

⚠️ as pointed out by flowwolf... These are racist memes @tropic shell @dry crow @wispy spindle @shut sinew

[cyan shoal]

thats a netflix/disney meme

but it's definitely off topic

[heady anchor]

just post a poster of the little mermaid or cleopatra documentary u wont need to edit anything

[nimble mason]

Yeah should just delete it

That is playing with fire

[cyan shoal]

yeahhh

[nimble mason]

https://cartoonresearch.com/index.php/the-censored-11-coal-black-and-the-sebben-dwarfs-1943/ "All of the characters have the standard ethnic caricature of big lips, and the prince has dice for teeth–an homage to the hoary gambling stereotype attached to African Americans." Yeah just delete that shit

Whatever the intent was, too much potential to be interpreted in a bad way

[nimble mason]

I don't think there's a best really

My go to is dpmpp_a but it's all so situational

I do have a worst though... Dpmpp 2m sde with normal scheduler is on that list lol

[eternal comet]

Good afternoon guys

Hope youre all doing epic

[cyan shoal]

I would be, if SD3 early access was given to more people

Yes thank you, I hope you are doing epic as well

[eternal comet]

Fair enough. Sd3 will be open source right

[cyan shoal]

the code will be, the model will be an "open model release"

[eternal comet]

Why do I feel like they're gonna pull a mistral moment

[crisp stream]

I feel like DPM++ SDE Karras is the "perfect" blend between the cleaner euler_a and for example DPM++ 2m Karras (that is a bit too edgy and flaky for my taste). Whensoever I check on other than DPM++ SDE Karras samplers I basically return pretty quickly.

[cyan shoal]

oh absolutely not

idk why people are worried by this, stability will 100% release SD3, even if it might be the last model

SD3 is guaranteed

[nimble mason]

I def agree that's a very good combo

[eternal comet]

Did stability work on dall w

Dall e

[cyan shoal]

work on dalle?

[eternal comet]

Are they working on open ai projects

Or just open source

What are the vram requirements of sd3

How many parameters

[cyan shoal]

for 8B, hopefully 8-12GB eventually

for 2B, possibly 6GB or less

and 800M should run on anything

none of these are facts

[eternal comet]

So I can run the largest model?

[cyan shoal]

its just estimates

[eternal comet]

Nice

Ah ok

[cyan shoal]

no no wait

[eternal comet]

I see

Yeah estimates

[cyan shoal]

its just my deductions

[eternal comet]

If it's 8b 7gb ram require probably

[cyan shoal]

since T5 will be offloaded, it might only depend on how much SD3 weights themselves will consume

[eternal comet]

If it's the same as llms

[cyan shoal]

(might even run on CPU entirely if its that fast)

but offloading is fast between RAM and VRAM

[eternal comet]

Offloading isn't tol fast

[nimble mason]

The thing is it's so situational, sometimes if I see something I like mid generation with dpmpp_a that disappears during preview I switch to the one you mentioned, or withthe exponential sechuduler etc and I get what I'm looking for then

[eternal comet]

It can slow down to a crawl

[cyan shoal]

wdym

[eternal comet]

Because of the GPU and sys ram going back and fourth

[nimble mason]

If I want a dramatically complex image res momentized is fantastic

[eternal comet]

This issue exists for llms

You'd only get really slow tokens per second

[cyan shoal]

Load T5 for conditioning to VRAM -> done -> offload T5 to RAM -> Load SD3 weights to VRAM and start generating

[eternal comet]

Ah ok

[cyan shoal]

not CPU offloading as in, it runs on CPU partially, I mean basically switching between active models

[heady anchor]

but then you would need more ram

[eternal comet]

I thought you mean full model ye

Ah ok

[cyan shoal]

yes, but RAM is way cheaper than VRAM

[nimble mason]

Ram is pretty cheap now though

So that's no biggie

[heady anchor]

oh ok buy me a few sticks then

[cyan shoal]

I think 12GBvram+32GBram might work with T5 at int8 and SD3 8B

that's just my guess though

we don't know how much SD3 ITSELF will take

T5 is easier to theorize about as its an LLM we have a lot more knowledge

4.7B isn't that much

[eternal comet]

Hopefully we have more prompt control

It's difficult for it to construct scenes not in it's training data sdxl

Unlike dall e it has to be trained on it super well

Dall e becomes cartooney and still constructs the scenario

[heady anchor]

because dalle has an llm and a scorer

[crisp stream]

Not saying there can´t be pleasant images achieved along other samplers, just noticing, for the most part, DPM++ SDE Karras is what I prefer in "99%" (exaggerated) of the situations. If I use anything else it´s mostly euler_a, then euler_a I find too clean in comparison. I´m usually doing painty stuff, reason I might perceive it that way

[cyan shoal]

absolutely, with T5 its guaranteed to understand way more

with SDXL, in a complex scene, it can mostly understand like 3-4 things and the rest of the concepts are merged into the photo

same with SD1.X

with SD3 we can probably take that up to like 8-10

[cyan shoal]

yes, DALLE3 has a painterly look to the images, even when you generate photos

[nimble mason]

Yuuuup very situational

If you want to swap in or clean up a deformed face without affecting proportions or lighting, best thing you can do is unsample to 50% noise with dpmpp 2m exponential, then resample with 50% denoise with dpmpp 3m GPU exponential

I don't use those very often in any other scenario but they whoop the others in that one

[crisp stream]

Sounds interesting, not familiar with unsampling though (using Easy Diffusion and only a bit of Comfy so far)

[eternal comet]

Bow and arrow

Hm

[nimble mason]

it's in comfy, i can drop you a json later

[nimble mason]

Arrow is aimed at her throat 😄

Cool bow though!

[wispy nest]

chewy chili rice

[nimble mason]

Trick shot

[crisp stream]

sounds good, thank you

[nimble mason]

If you like painting style stuff, def try the new res_momentized sampler on comfyui

My current go tos for generation are the one you mentioned, dpmpp_a Karras. And res_momentized Karras or sgm_uniform

[crisp stream]

Ah, good to know, thank you. Would you have a quick example image at hand?

[eternal comet]

💀

[golden pebble]

Hey guys I try to genrate with 2 loras

what do you think of this prompt?

a woman and one man, same hight, holding hand, super high resolution, bright backlight, best quality,Photos,4k,(Realistic:1.2) ADDCOMM

konoha-gakura park ADDBASE

1boy, Sasuke,lora:zuozhu-05:0.7,zuozhu,solo,1boy,male focus,uchiha sasuke,black hair,hand on girls shoulder,white background,simple background,closed mouth,full body,standing,zipper,zipper pull tab,rope,hair between eyes,braid,pants,black eyes,short sleeves,shirt,bangs,grey jacket,black footwear,ninja,toeless footwear,looking at viewer,black pants,spiked hair,
ADDCOL

1girl, karinuzumaki,lora:xiang-7:0.8,(face lighting:0.8), xiang,shorts,solo,thigh,red hair,navel,belt,outdoor,glasses,look at the audience,

i use regional prompter. I somehow can't get these two

you can see the sasuke lora i use and the creator also made a karin lora so i have two loras from one creator but still i get weird outcomes

[heady anchor]

what the hell is that negative, 258 tokens

https://tenor.com/view/pepe-the-frog-pepe-frog-shocked-what-the-heck-gif-17806223

[cyan shoal]

it can handle up to 300???

[heady anchor]

thats cool to know,ill make a 600 one then

[cyan shoal]

[heady anchor]

👍

everyone knows that the longer the neg,the better pics u get

[delicate hinge]

opinions on this render of Nami from one piece (also idk if this is the right channel for this so if it is let me know)

[golden pebble]

I copied the negative prompt from the creator of the lora

[heady anchor]

its too long plus your denoising strenght is too high at 0.7

it should be at around 0.3 to 0.4

[golden pebble]

I'm using this https://civitai.com/models/354816/naruto and this Lora https://civitai.com/models/338939 . It's by the same creator.

Okay I try to reduce the denoising strength.
But what base model should I use? The creator used Anylora, I used it too but it doesn't make it like them I guess she/he used a diffrent one.

[golden pebble]

I used the negative prompt from the creator, I thought it's good if I use their neg. but what should I use then? I used Lora neg. v2 and put them into negative prompt but the outcome is diffrent compared when i put long neg prompts

[heady anchor]

something like this is more than enough (worst quality, low quality, lowres:1.4),verybadimagenegative_v1.3,bad-hands-5,

too many words in prompt = bad results

[golden pebble]

Okay I try it with that too. But I wonder, the Loras itself look so amazing, putting them together it doesnt look a bit like them

[heady anchor]

it could be the model you are using,try with different models

or just test the lora if its working by makin a few pics with just one lora

[golden pebble]

Ahh you are right, something is off with the Sasuke Lora, it doesn not generate it good and it seems the creator retouched her final images

[heady anchor]

try with another sasuke lora theres a lot in civitai

[golden pebble]

I must admit, I tried every Lora on Civitai from Sasuke but they are not so good as I tried them.

[heady anchor]

it also depends on the model some not very good models get deformed when u apply some loras

[wispy nest]

Model?

[golden pebble]

I have over 100 models on my SD and I used those who are really good and harmony with every lora I used so far which are like 10 models but the Sasuke Loras are not good at all. I thought maybe this one is good but it seems like its sadly not

[heady anchor]

try to reduce the weight of your lora maybe like 0.5

[dull flame]

u can find it in #🍥｜anime , just scroll up tiny bit, someone else asked yesterday, dont rly wanna post it here as well, its nsfw

[wispy nest]

Alright thanks

[golden pebble]

great idea, I'll try it out

[golden pebble]

That was a great idea with lower Lora weight and lower Denoising strength it really now looks more like him

[dull flame]

thats beautiful

[golden pebble]

this is how he looks like when done alone, now I try two characters at once

[wispy nest]

Thank you a ghost type promt from over the summer

[dull flame]

i see im not the only one struggling with weapons

[heady anchor]

thats just normal,almost of all 1.5 models struggle with weapons

[golden pebble]

I just copy pasted a prompt and used it to see if it works but yes the weapons really look weird but there are loras for swords i once used and it looked good

[dull flame]

i figured i could find some weapon lora, but i failed

[robust path]

Sad dad and dog, because Emad said we need more of them

[golden pebble]

https://civitai.com/models/282753/sword-sakuraphotographic-style i use this two https://civitai.com/models/69323?modelVersionId=223864

[dull flame]

thanks, ive never used lycoris, whats the difference from normal lora?

[golden pebble]

I have no clue 😅 I'm very sorry. I just found it and it looked cool and i used it and it worked out.

[dull flame]

ims tarting to think its a somewhat good idea to seperate images and past them together later, like one character, then a background, then weapons

[golden pebble]

i dont understand why sasuke always gets generated more like a girl than the girl. here he looks " a bit more" like a guy but the others are always a girl

[covert pagoda]

cool

[umbral beacon]

美女

i dont understand why sasuke always gets generated more like a girl than the girl. here he looks " a bit more" like a guy but the others are always a girl

[wispy nest]

just gotta upscale it but

[harsh veldt]

Tang Dynasty concubine wears Tang Dynasty costume and drives a classic car to go shopping.

[distant vortex]

Captain Planet 😄

[eternal comet]

Generating soldiers is badass

[eternal comet]

a cyberpunk cityscape in the style of van gogh

[eternal comet]

"shell shocked soldier grinding teeth broken dead inside"

In the dystopian urban sprawl, where towering skyscrapers cast ominous shadows over labyrinthine alleyways, the city breathes with a palpable sense of decay and desperation. Neon signs flicker ominously, their garish glow reflecting off cracked pavement littered with the remnants of shattered dreams. The air hangs heavy with the acrid stench of industrial pollution, mingling with the distant hum of machinery and the whispered promises of false hope. Amidst the urban decay, hearts ache with the weight of betrayal, as the AI woman, cloaked in digital allure, prowls the streets, leaving a trail of shattered trust and fractured souls in her wake. In this bladerunner-esque dystopia, love is but a mirage, a fleeting illusion in a world consumed by lies and deceit, where pain is the only currency that truly holds value.

[pallid ruin]

That´s beautiful 🥺

[left rune]

man

[cyan shoal]

epic

[sour ibex]

Introduce Timmy, a curious young boy, wandering through a lush forest.

[eternal comet]

In the twilight realm of ethereal convergence, where whispers of quantum entanglement dance with the fractured echoes of singularity, a kaleidoscope of crystalline manifestations coalesce in a symphony of ephemeral resonance. Within this labyrinthine expanse of intricately woven obscurity, tendrils of shadow intertwine with shards of fractured light, each facet a cipher in the enigmatic tapestry of cosmic paradox. Time ebbs and flows like a spectral river, its currents twisting and turning through the boundless expanse of the ineffable, carrying with it the echoes of forgotten dreams and the whispers of unborn galaxies. In this nebulous realm of infinite possibility, the boundaries between reality and illusion blur into oblivion, leaving naught but the echo of a question unanswered in the void of eternal ambiguity.

[clever oar]

server room underwater ?)

[clever oar]

half life?

[flat kernel]

That's how you keep it cool

[fathom steeple]

just got a 4070 super and have to make this meme

using webui; generating at 1032x928, upscaling 2x with r-esrgan 4x+ anime 6b to 2064x1856, 20 hires steps, 0.55 denoising strength.

[crude sorrel]

I made a cross-eyed stereoscopic image with SDXL and ControlNet. I used a non-stereo image as the control image (the snow image, from here: https://unsplash.com/photos/a-road-in-the-middle-of-a-snowy-forest-Eu3iDH2HVeQ). I gave it a weight of 0.3 and it actually produced a stereoscopic image (the non-snow image).

It's a cross-eyed anaglyph

I bet finetuning sdxl with stereo images would make much better stereo images.

[eternal comet]

wow so good

[grizzled sage]

the ultimate redditor fantasy , big bazongas

[eternal comet]

BRO

I POSTED THAT

AND IT SHOWED NUDE

AHAHAAHAHAH

THIS IS FUNNY

[grizzled sage]

lol i didn't actually type big bazongas. i was going for two sandbags, but one giant one got the joke across too

[grizzled sage]

now with that kick your teeth into your throat level of caffiene

[compact chasm]

[SD3] Why does the FID score increase after 25 steps of sampling?

[nimble mason]

Here is the image you requested

[eternal comet]

Better now song lyrics

[nimble mason]

Those don't look like tears

[eternal comet]

Ik

These models aren't aligned lol

Oh for fuck sakes

I get it now

When I'm alone post Malone

NGL if you paste the lyrics in

It makes good covers

[nimble mason]

Tbh that was an epic image lol

[eternal comet]

Lmao

Making another

Xxxtentaction revenge lyrics as prompt

Let's see

[grizzled sage]

threw the forgot about dre chorus in. think it works imo. this is a sick cover for it

same lyrics in juggernaut xl lol. it really works

eminem - the way i am.. not so much but its a cute try

pretty fun experiments thanks for the neat idea @eternal comet

mr roboto lyrics but i used album cover by Shusei Nagaoka this time too

[crisp stream]

Discord removes Metadata

[crisp stream]

Don´t know, at least didn´t prompt for anything alike 😉

[rocky gulch]

A newborn Chinese baby eyes closed white skin,rosy face, with some light coming in from the side against a dreamy white background. The compositionis beautiful, rich, and bright.

@eternal comet A newborn Chinese baby eyes closed white skin,rosy face, with some light coming in from the side against a dreamy white background. The compositionis beautiful, rich, and bright.

A newborn Chinese baby eyes closed white skin,rosy face, with some light coming in from the side against a dreamy white background. The compositionis beautiful, rich, and bright.

[karmic fractal]

Hi Everyone, I'm thrilled to show you my latest experiment with Stable Video Diffusion and ElevenLabs: https://youtube.com/shorts/6EynYqYVTqE Enjoy 😉

[eternal comet]

Ping

Bro

What is that

[waxen coral]

a plane with a huge banner floating after its tail with the colorful word "dahmane !!!" written on it, seen from the beach, many jumping women looking at it with raised arms, blue sky, sunny weather

a plane with a huge banner floating after its tail with the colorful word "dahmane !!!" written on it, seen from the beach, blue sky, sunny weather

a plane with a huge banner floating after its tail with the colorful word "dahmane !!!" written on it, seen from the beach, blue sky, sunny weather

[eternal comet]

What

[wispy nest]

oh ok

drive/my-drive/PTFIles/sl1ng.pt

looks like a solid path ?

for this ?

the sl1ng file is inside PT FIles

is this better /

?

hello ?

[frail quarry]

@eternal comet draw a ballcat

[west crest]

White line figure with black back, vector graphics

[eternal comet]

I'm in college

And I can't draw but my GPU can!!

[late wigeon]

how to use stable diffusion 😭

like where

start

[nocturne oak]

#1080946152318443610
Google "Automatic1111 how to begin"

[flat kernel]

Also check out Fooocus

[wispy nest]

I took a photo of myself and put it thru ai video

Give it 2 years all music videos will likely to be done in ai unless the goal is vintage astetic like how we see fillm photographs now

[eternal comet]

💀 😂

Something you'd see in a dream

Looks liminal

[surreal zenith]

Beaches, beautiful women, bikinis, coconut trees, yachts

How can I use this bot?😫

[wispy nest]

can anyone give me an explanation to this ?

is the tool broken ?

I can't seem to be able to convert PT files to SAFETENSORS

[bleak iris]

Mariah carey

[distant vortex]

She-Ra

[eternal comet]

"generate me an image of someone with deep emotional pain hiding it and putting on a fascade"

[eternal dome]

ohhhh !!!

[eternal comet]

Wholesome

[eternal dome]

thx

[cold meadow]

Cascade?

[eternal comet]

Yuh

Prompt : don't save her she don't wanna be saved

[cyan shoal]

dayum nice

[eternal dome]

super nice !

[sly tangle]

What model is this?

[cerulean trail]

/dream:

[nocturne oak]

/no

[blissful jasper]

Imagine that super neuroscientists have created a machine that can simulate pleasurable experiences for the rest of your life. The simulation is ultra-realistic and indistinguishable from reality. There are no adverse side effects, and specific pleasurable experiences can even be programmed into the simulation. Regarding pleasure experienced, the machine offers more than is possible in several lifetimes.

Question:

Do we have any reason to not go in?

[neat blade]

come up with a logo with the name: empress. subject: esotericism, tarot cards. mid tones of color. png

[neat blade]

come up with a logo with the name: empress. subjects: esotericism, tarot cards

[neat blade]

come up with a logo with the theme of tarot cards. the queen must be present in the logo

[wispy nest]

a sentient ai robot launches him self into outer space to a space station to do repairs making spaceships obsolete

[neat blade]

make a logo with tarot cards and a queen in them

[neat blade]

Rofl?

make a tanos on tron

[wispy nest]

holgramicquantaseila

[random sundial]

How to generate pictures?

[nimble mason]

Beep boop. Prompt me!

[wild sorrel]

lol

[clever oar]

🖐️

[sinful island]

highly detailed and realistic 3D image depicting a 28 years old boy joyful celebrating the Holi with lord Krishna in the vibrant setting of mathura Vrindavan. The boy is wearing a t-shirt
on which “Akshay” name is written, their faces are visible, the “Happy Holi” are elegantly scripted on the top of the sky. The background features the picturesque landscapes of mathura Vrindavan, complete with colourful buildings and lush greenery. Children are depicted playing Holi.

[ocean turret]

create a chinese old farmer,thank you.

[clever oar]

[ocean turret]

create a picture about dragon and happy.

[cyan shoal]

SD3 Turbo failure cases

Although its fast and still coherent, it sometimes fails compared to regular SD3. It's still extremely impressive for 4 steps though.

[lost mango]

/dream/Solar Light Frog Bird Feeder

[ocean turret]

Please draw a picture. The theme is that employees wearing anti-static work clothes are working in an electronics factory. Thanks.

[ocean turret]

Please draw a picture with a blue background. A happy dragon circles the electronics factory. Thanks.

[robust path]

https://discord.com/channels/1002292111942635562/1040299965647437905

[pale garnet]

this thing can create 1080x1080 or 1280x720 without upscaler

and sometimes it's out of memory

the safest is 960x960 recommended

[earnest plank]

Please draw a cyberpunk style, 37-shaped, bionic handsome man with glasses

[warm vector]

Bedroom interior design sample

interior design

[viscid glacier]

@pale garnet

[pale garnet]

ugh, wow xDD

@viscid glacier

[cold nest]

https://tenor.com/bmmFW.gif

[eternal comet]

What

Why so many negatives

[pale garnet]

overprotective I guess xDDD

[eternal comet]

Hm

I put no negative and I get decent hands back

[pale garnet]

ugh, I like to cook some models (merge), and somehow the hands just broken out xDD

[eternal comet]

Oh

No sdxl models just can't do hands lmao

[pale garnet]

fr, idk what ingredients do I cook, i just cook it in

[eternal comet]

Lmao

Ok delete that

💀

Please delete that

Caseoh

[clever oar]

why

😃

[eternal comet]

@01011001_nx_

For "Hyundai N Vision 74 maybe? :v"

[eternal comet]

Nice

[pale garnet]

lmao, no it's not, but thanks for the effort xDDD

[eternal comet]

I don't think cascade can generate

Furry

Someone asked me to generate something for them

Didn't work

[pale garnet]

well, it simply don't have the furry model? xDD

[eternal comet]

Yeah that's probably it

That looks bad

Please don't generate that

It looks like she's eating her own feces

Or holding them

[clever oar]

yed

yes its what i generate

[eternal comet]

What model

[clever oar]

epicphotogasm

[eternal comet]

Ah

Alright

[pale garnet]

epic

[clever oar]

lol why delete

there no nude

[dry crow]

I thought its spam and breaks rule 4

[clever oar]

why spam

[dry crow]

Because we had bots that flooded the channels with disgusting images before

[pale garnet]

lmfao

[dry crow]

@clever oar Sry if you didn't intended to make it look like that.

But keep the images appropriate

[eternal comet]

Chocolate

[pale garnet]

is the character looks a bit small?

[eternal comet]

Nah

Art servers hate ai art

Ft

Fr

I posted it and they instantly were like nah fuck off

[pale garnet]

2 types of artist... the one who hate AI arts, and the one who uses AI to help them create their art

[eternal comet]

They're afraid of a GPU

[pale garnet]

ye know, this AI can helps artists too

[eternal comet]

They're afraid of cuda cores 😭

[dry crow]

Or the car is just big

[pale garnet]

hmm, since this is like, a mid to low angle shot...

idk tbh xDD

[pale garnet]

this one is jjust perfect

[eternal comet]

"chicken protesting" 💀

Free paste

[pale garnet]

pastetine

[eternal comet]

💀

[glacial pecan]

1girl,young beautiful girl,White skin,long legs, delicate skin, big eyes, ultra detailed, official art, unity 8k wallpaper,

[clever oar]

i cant post disgusting?

[eternal comet]

Fuck it

I'll gen it

[clever oar]

my models lose text

[glacial pecan]

go

"chicken"

[heady anchor]

https://tenor.com/view/when-will-you-learn-when-will-you-learn-your-lesson-when-will-you-get-it-when-will-you-understand-pewdiepie-gif-14185832

[pale garnet]

lmfaoooo

[gritty terrace]

Sea of red cherry blossoms, realistic, cinematic, macro, with bees on the stamens

[clever oar]

i cant post results)

[eternal comet]

Why

💀

Oh photogasm is trained on the funny

The funny content

[pale garnet]

HAHAHAHAHAHAHA

[clever oar]

young girl in very sex
clothes

[eternal comet]

My search results at 13 💀

[pale garnet]

defak xDD

[clever oar]

very young )

[eternal comet]

Nah nah nah

💀

[pale garnet]

hello fbi

[eternal comet]

💀

We making it into jail with this one

Crazy

[clever oar]

ups

[nimble mason]

Here is the image you requested.

[eternal comet]

What the fuck is that bro

💀

Why are you obsessed with spiders

Monke

[nimble mason]

Here is the image you requested

[eternal comet]

https://tenor.com/view/giantsbetta-small-monkey-monkey-backpack-monkey-gif-13385757124521273001

[wild sorrel]

add dragonscales to this, wanna see how it'll look lol

[nimble mason]

Here is the image you requested.

[eternal comet]

Hahahaha

Wtf

[wild sorrel]

I could see that creature being real existing thing

[eternal comet]

True monkey business

[eternal comet]

Merh

Meth

💀☠️

[eternal comet]

That's cold

[clever oar]

enimal planet

[pale garnet]

monkee

[eternal comet]

Sdxl isn't that good at following instructions

Cascade is pretty good ngl

[clever oar]

I don't like monkeys, I was bitten by one

[eternal comet]

💀

Nahgy

💀

[nimble mason]

Depends on the image of course but yeah pretty good

One thing it still sucks at is tornadoes

[eternal comet]

Eh it does them okay

[nimble mason]

That's not a tornado lol

[clever oar]

promt man beats woman with pig face

why woman beats every time?

ai dont listen promt