🏞｜general-with-images

[split nimbus]

playing a bit around

[pallid ruin]

so cuteeeee 😍 🤗

nice!

[eternal dome]

guys , any advice for chromaNeoFT checkpoint ? like prompting , CFG Scale , dionising

[wispy nest]

wtf

😂

💀🤷‍♀️

[gusty kite]

My SD1.5 minecraft items lora. Wanna upgrade to SD3 possibly

[cyan shoal]

interesting

SD3 is pretty good at pixelart by default

[gusty kite]

Yea it definitely looks better

I heard sd3 has like different sizes, maybe il use the smallest one

Still better then sd1.5 but cheaper to train

[cyan shoal]

SD3 image that I requested (I DONT HAVE ACCESS)

[jovial tiger]

I had asked for a dragon inspecting a toy while stuffed animals look on at a coffee shop. There's at least 2 in this group that are right on.

#sd3

[cyan shoal]

Ohhh so that was you

You were one of the people who prompted correctly imo

CogVLM (natural language) style

My prompt isn't perfect but it's still okay:
Old Disposable Camera photo of a toy box that has: "T5-Large" written on it. On the bottom of the toy box is written: "Quite memory hungry!". The toy itself is a piece of blue floppy disk.

I think I should have put the floppy disk part in the beginning rather than mentioning it again at the end, idk if that confuses it

Oh wait. I asked for a "Toy Box", this actually makes sense

I am dumb, SD3 is correct lmfaooooooooooo

should have asked for Toy Package or something

[jovial tiger]

Well, it's a reasonable ask for it to be able to understand what you wrote as is. The other services would have understood that.

[cyan shoal]

That's because those services usually modify your prompts

idk if that will be affordable offline

having an LLM offloaded, consuming even more RAM in the background

[main dew]

you know, those look really good, did you do post processing to get the pixels exact?

actually a simple downscale to 16x16 would work right

[jovial tiger]

Yeah but sd3 has a 10 gigabyte language model included. It should easily be able to understand various phrasing.

[cyan shoal]

And half the dataset was left unchanged

And the model they are testing are inferior apparently

[main dew]

the transformer is pretrained

the t5 is the most important part

[jovial tiger]

Based on what I'm seeing today on twitter, we're definitely still going to need controlnet and maybe even regional promoter to get the complicated stuff.

[cyan shoal]

If you say so

[jovial tiger]

Still no complex actions, still no ability to have one subject look into the face of another reliably.

[rocky hawk]

That's what I feel so far too from what I've been seeing

How high r the specs to run sd3 locally or is this just all through their service?

[jovial tiger]

They'll have various sizes so you'll be able to run some version of it

[rocky hawk]

Nice eager to see the fine tunes for anime

[jovial tiger]

And turbo models seemingly at launch which is neat

[cyan shoal]

idk the early preview model did subjects interacting with each other with an angry expression

then again this is a simpler prompt

[main dew]

sd3 will release with controlnet and all training scripts and compatibility with comfy a1111 etc

[main dew]

4 fingers..

[rocky hawk]

So sd3 I've yet to see human contact forms of interaction with perspective, most of that is heavy controlnet right now

[cyan shoal]

its the early preview model

the images I have seen of this are worse than what Lykon was posting

[jovial tiger]

Yeah, mouse had 2 tails

[main dew]

even the latest ones arent very good

[cyan shoal]

the Beach example was much superior

only left out the palm tree

[main dew]

yeah but those are simple prompts

[cyan shoal]

I swear the early preview looks like 8B undertrained or maybe the 2B model lmao

[main dew]

ill let them cook

[cyan shoal]

cant wait for mattvidpro to make a vid about SD3 early test

[wispy nest]

i am pretty sure im right that the preview is the 800m and 2b model

lykon has the 8b model which is leagues better than the 800m and 2b

if the 8b is that much better then 800m and 2b might be doa

[cyan shoal]

I hope so

then that means the 2B and 800M models are intelligent too

well, at least more intelligent than previous models

[wispy nest]

@cyan shoal imagine this workflow, use 2b model for quick "sketches" then 8b to do the final photo with hiresfix. The 2b model generates images that look close to 8b but js lacks the detail

that is from the paper with each column a different seed i think

[cyan shoal]

uhhh

[jovial tiger]

My question is, is there a difference in prompt following between those models? I'm less concerned about visuals and more about doing what I asked. SD is king for quality when it wants to be. Prompt adherence is the secret sauce right now though

[cyan shoal]

yeah..

if the current preview models are 2B or 800M then that means prompt adherence is 'good enough', as in its better than previous models

maybe we'll have to do highresfix before modifying it with other models if you want to do such a workflow

[wispy nest]

@cyan shoal on here it shows the 2b model runs at 568ms/it with a batch size of 32, so does that mean all 32 images took about 25 seconds to finish cause if so that's 0.78 seconds per image

assuming the step count is 50

[cyan shoal]

hmm

[clever oar]

Thanks!

[grizzled sage]

sometimes i prompt "boop" just to test.

[cyan shoal]

lmao

[dull flame]

its so hard to tell where the sweet spot is = /

[grizzled sage]

its more of a zone than a sweet spot.

[nocturne oak]

Definitely 2 steps and CFG 3. Go with that.

[dull flame]

great advice soul

am i allowed to post very sassy gifs here?

[rain gazelle]

dont you find it doesn't stick to your prompt at such low cfg?

[dull flame]

its lcm, supposed to be low, or did i misunderstand ur question?

[main dew]

.86-1.21 looks best, I hate the high saturation look of higher cfgs

[rain gazelle]

Yeah with the fast models, it's low cfg, but I feel like it is a trade off with prompts, still.

[dull flame]

0.8 certainly has a charm to them

does prompts rly get hindered a lot with lcm ?

[gusty kite]

sorry to be late, no thats pure output no scaling

[main dew]

very nice

[pallid ruin]

👀 is this sd cascade? it looks awesome!

[jovial tiger]

That's sd3. It's the 16 channel vae making everything look great

[pallid ruin]

The quality is amazing 🤗 I cannot wait to run it 🥺

[timid stream]

https://tenor.com/58Na.gif

https://tenor.com/bGciG.gif

https://tenor.com/bCcE0.gif

[versed cobalt]

@wispy nest those photos of the girls are nice. what model/loras did u use? Is that sd3 or cascade orrr something else?

[quiet current]

#1047610792226340935

[wispy nest]

@cyan shoal https://twitter.com/Lykon4072/status/1768405549333303587/photo/3 more proof lykon has a different model

Lykon (@Lykon4072) on X

#SD3

Twitter · Mar 14, 2024, 10:35 PM

[cyan shoal]

• he uses highresfix but yeah

these are extremely aesthetic

[main dew]

eyes still lookin weird

[cyan shoal]

reminds me of 2.X and base SDXL eyes

[pallid ruin]

x2

Someone give him a very good prooompt 😭

[cyan shoal]

I tried to ask him like twice now and he either avoided it or replied like a politician

https://twitter.com/Lykon4072/status/1768414855948611845/photo/1

Lykon (@Lykon4072) on X

#SD3

Twitter · Mar 14, 2024, 11:12 PM

now this looks very good

I wonder if SD3 will have some quirk like SD2.X (like you had to put "ugly" in the negative prompt to get good images)

🤔

Lykon replied with this image to someone's prompt: Dimly lit staircase within an old building. The stairs are made of worn stone. The light fades into darkness as the stairs descend, with a figure barely visible appearing midway down the stairs, partially concealed by the darkness.

there's no highresfix applied to this so image quality is inferior

it'd be funny if these images contained comfyui metadata so we could see the workflow

even if like quarter the nodes would be missing

[pallid ruin]

Yeah that one looks good

[wispy nest]

https://twitter.com/Lykon4072/status/1768419732070203710/photo/1

Lykon (@Lykon4072) on X

@gostyshev62516

Twitter · Mar 14, 2024, 11:31 PM

[cyan shoal]

very epic

https://twitter.com/Lykon4072/status/1768418557371523164/photo/1

Lykon (@Lykon4072) on X

#SD3

Twitter · Mar 14, 2024, 11:26 PM

geometry and reflections look clean so far

https://twitter.com/Lykon4072/status/1768420943347474478/photo/1

Lykon (@Lykon4072) on X

#SD3

Twitter · Mar 14, 2024, 11:36 PM

you can tell when it has highresfix or not

[hazy warren]

im using SDXL, is there a better one?

[pallid ruin]

I think sd cascade is better in some things, but I haven´t tested it yet, I only saw images

[hazy warren]

😮

[wispy nest]

test promt engernering complex randomneess

this features no mans sky and a kolidascope for weird modifications

[wispy nest]

sweet water color vibes

[pallid ruin]

Nice images!

[deft bison]

@pallid ruin THANKS !

[wispy nest]

my whole innternet just went out from jenortating images

bro

lol

https://twitter.com/EMostaque/status/1768416271429722186

Emad acc/acc (@EMostaque) on X

@1Gambz

Twitter · Mar 14, 2024, 11:17 PM

[calm canopy]

https://pbs.twimg.com/media/GIq0tFnXwAEeQVO?format=jpg&name=large

emad posted this an hour ago with someone's prompt of:
3 panels page about a fight between a gigantic irish mage redhead woman wearing a medieval outfit AND a black Dragon with amber eyes, near village with buildings burning. fireball magic, scared villagers are running (she is also colosal). boom sfx, comic style

this is actually crazy good

Pretty hyped for SD3 if thats just the base

[flat kernel]

That polkadotted shirt image Emad posted had crazy prompt adherence

A man wearing an orange shirt with polkadots standing in front of a large pink house with a big airplane in the sky that has text “ZOOMAIR”. Speech bubble out of man’s mouth saying: “WOAH, THAT’S ONE REALLY PINK HOUSE!”

[grizzled sage]

t5 is sauce

lykon's newest. just missing some drool

[wintry yacht]

a cat

[proud cosmos]

You can't generate pictures here anymore?

[pallid ruin]

Not yet, the bot is still down.

[wispy nest]

https://tenor.com/view/doc-brown-back-to-the-future-back-to-the-future-the-ride-universal-studios-universal-studios-florida-gif-25177148

[nimble mason]

Here is the image you requested.

[grizzled sage]

https://youtu.be/Ey_Bo5QLWt4

[main dew]

MORTY WERE STUCK IN THE THIRD DIMENSION MORTY WE GOTTA GET OUT HERE
oh jeez rick

[grizzled sage]

https://tenor.com/view/simpsons-homer-3d-dimension-falling-gif-18243159

[jovial tiger]

#sd3 I asked for some fork holding pics, and Lykon and another staffer responded. Very impressed.

[woven karma]

So, um... I was trying out some prompts with AAM XL and getting some really nice results

And then I added a couple more tags and suddenly started getting this

Does anyone know what could be happening? 😅

[grizzled sage]

way too high cfg

[woven karma]

Huh, the model recommended 5-7 so I just went with 7 to try and get a bit more prompt accuracy

I'll try to lower it, thanks!

[heady anchor]

no,thats an a1111 issue try to select another model then switch to AAM XL again and it will fix

if that doesnt fix it just restart a1111 and it will fix itself

[grizzled sage]

guess i don't really know aam. i've never run into that prob., just looked like blown out cfg to me

7 isn't high

[wispy nest]

this is untouched ya

isnt Rick and morty a acid trip

[split blade]

I think she's wearing makeup at least

[wispy nest]

well hair too

[wispy nest]

your more like a obese slopathen

[nimble mason]

Here is the image you requested.

[floral wind]

Please paint a big painted crane with zebra lines

[nimble mason]

Here is the image you requested.

[tired basin]

What model you using chief??

[nimble mason]

Here is the image you requested.

[nimble mason]

Beep boop. DarkArtsImagesXL

[tired basin]

Good bot 🙂

[lean talon]

anyone have any super fast (30FPS+) pipelines for SDXL img2img, or hints at what we can expect for img2img inference speed with the small 800m parameter SD3 model? the below was done with a sd 1.5 finetune running realtime through streamdiffusion

[tired basin]

Fps depends ultimately on hardware - are you trying to get 30fps on a a100 or on a gtx960??

[lean talon]

4090 + 14900K

i was looking at onediff + sdxl-lightning + taesdxl for a fast VAE but onediff seemed kinda broken to me

[tired basin]

I havent heard anyone that got onediff to make an improvement

Which lightning you using??

[lean talon]

well i was going to start playing with the bytedance one

[tired basin]

How many steps - im guessing the 1-step experimental one?

[lean talon]

was the plan

i can get 80+fps with sd-turbo and 30-40 with SD1.5 LCM + 3 steps

[golden pebble]

do you guys know the filter on tiktok chic cutie? how can someone create something like this? can this be done by sd?

[safe oasis]

The two little girls are playing on the beach.

[tired basin]

With that speed you could probably use the 2-step one which is much higher quality

[fiery shell]

what

[hazy warren]

type of art you see at hotels

[tired basin]

Where??

[gilded birch]

https://tenor.com/ba7fp.gif

https://tenor.com/LxVg.gif

https://tenor.com/bNssg.gif

[hazy warren]

Eye see what you did there

[reef mortar]

Ninja monkey with samurai

[hazy warren]

make it

[reef mortar]

how?

[umbral thunder]

future city planning, 3d grid

[hazy warren]

Person painting themselves into existence

[quiet current]

I am once again asking you to read #1047610792226340935

[cyan shoal]

I meant SDXL Base, the base model, finetunes are fine

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[cyan shoal]

hmm

[nimble mason]

Here is the image you requested.

[cyan shoal]

lmfao

[nimble mason]

Here is the image you requested.

[agile jungle]

Isometric portrait of elf, in pixel style

[dull flame]

those are beautiful Moofi

[crisp stream]

Thank you 🙂

[deft bison]

@hazy warren 👍

[gritty moth]

#🏞｜general-with-images

#🏞｜general-with-images flowers woven together

[shut sinew]

#1047610792226340935

[high bluff]

美女图片

[high bluff]

一个中国美女，牛仔裤黑头发，黑头发

中国美女

[topaz frost]

#🏞｜general-with-images women

#🏞｜general-with-images flowers

[shut sinew]

#1047610792226340935

[sharp lark]

a girl running with a lovely dog

[fringe kestrel]

man, yellow short hair, white short-sleeved shirt, Miyazaki style, pondering on a big bed, surreal atmosphere, shopping mall setting, girl, pink long hair, blue eyes, earrings, necklace, black collar,

[versed cobalt]

@wispy nest your images has that dalle style. Which model is being used in Sd to make that? Proteus? OpenDalle? or something else

I have a dalle style model I made and it is able to be recreated in fooocus fairly well but not quite the same. I have been struggling for sometime to get the model recreated in SD. Care to tell me which model/platform was used to get it that close to the dalle look?

[pearl trail]

where can I make the images?

[pearl trail]

where can I make the images?

[quiet current]

install stable diffusion locally or use online services (such as dreamstudio)

[wispy nest]

https://arvensis.capital

omg I just got accepted for cto

omfg

[ember sundial]

Congratulations 🫶🏻

[spark finch]

I really like the almost watercoloresque appeal of that.

[crisp stream]

here as well, especially in combination with the colours 🙂

[nocturne oak]

This one is really nice. Well done.

[crisp stream]

Thank you, appreciated 🙂

[clever oar]

there all use xl models?

[crisp stream]

if you are referring to the images I posted, yes, SDXL pure base model

[clever oar]

i have only 4 gb and use 1.5

[crisp stream]

Better than nothing 🙂

[clever oar]

sometimes I'm scared that the neural network constantly guesses the color I want

I think oh I forgot to write the color, but it gives the color that I had in mind

[crisp stream]

yes, making similar experiences here

[grizzled sage]

the one test prompt i did with t5 lavi bridge on base sd 1.4 that turned out half decent. code has no sliders to tweak for higher stpes or cfgs or negative prompts so i just wung it. this was the coolest only quality worthy one.

damn though sd14 really sucked by todays standards. lol

prompt is the file name

[clever oar]

i see you like cartoons

[grizzled sage]

reminds me of this one art style i used to see in magazine ads a lot

[cyan shoal]

YWS

[crisp stream]

not necessarily, just liking how it looks 🙂

[crisp stream]

interesting, wasn´t familiar with that style

[prisma estuary]

Hello, iam completely new to this and trying to generate verry similar image to the original one, because the og one is licensed, but iam getting completely different output. Is there an easy way to do this or i need to watch hours of tutorials ?

[grizzled sage]

its killing me to find it now but i have no idea what to google for lol. the style was popular for a bit. like intricate isometric scenes

[crisp stream]

no hurry 😄

would be interesting anyway

[crisp stream]

so the left one is what you would like to achieve?

[prisma estuary]

Yeah, with some minor adjustments

[grizzled sage]

https://www.mlmiotke.com/ art like this but more towards your image

[crisp stream]

ah, I see, the complexity is quite similar

[grizzled sage]

the one piece i'm thinking of had brands all over it. the guy who drew it, their name is eluding me though. covid brain. memory is swiss cheesed like sam beckett

[crisp stream]

🤭

[crisp stream]

😄

I know, it´s too painty, I like them anyway, gonna continue

[prisma estuary]

Yeah, cant really use these but it looks awesome

[crisp stream]

Thank you, gonna generate more realistic ones

[prisma estuary]

I wanna use it for a website as a background so it blends with the sillhouete

[crisp stream]

I see 🙂

[sly tangle]

So much meme potential here: https://cdn.80.lv/api/upload/content/fc/65f318da4cf4c.jpg

[crisp stream]

Pale glowing full moon shimmering through beautiful streaky clouds, dark blue night sky,Elsa Bleda,Marta Bevacqua
Negative Prompt: motion blur,realistic,,ground,woman

Dimensions: 1280x768, Sampler: euler_a, Inference Steps: 18, Guidance Scale: 7.5, Model: sd_xl_base_1.0,

you could play with colouring here as well, like writing blue clouds, etc.

[prisma estuary]

Thanks man, ill try it but i have 1050ti it take minutes to generate one image

[crisp stream]

you could check out a service, like Dream from Wombo (freemium)

ah, the free version merely has 9:16, 1:1 ratios though

the app that is, the web app only 9:16, so it might be not that suited

[prisma estuary]

Ill still give a try, or just buy new Gpu lol

[crisp stream]

😄

[prisma estuary]

Thank you for your time!

[crisp stream]

pleasure

[versed cobalt]

Does nanyone know of a model for Stable diffusion that coudl generate Dalle quality images?

[hazy warren]

Stargate

[hazy warren]

Needs to howl

[versed cobalt]

im trying to further build out my dalle images that i originally created in dalle but having some issues finding the right model. I see josh created some images that are almost exactly what i am looking for but havnt heard back form him yet

[versed cobalt]

im looking for this model so if someone coudl tell me that be great! I am tihknig proteus or opendalle but not sure

[deft bison]

@hazy warren 👍

[hazy warren]

😮

[deft bison]

@hazy warren 🤝

[clever oar]

[languid geode]

Dragon Ball Son Goku Super Saiyan 100 Form

[nimble mason]

Here is the image you requested.

[heady anchor]

Super Mystico Gohan Blanco 100 Por Ciento el Calvo Dorado SSSSJ100

[languid geode]

龙珠孙悟空超级赛亚人11种形态

[nimble mason]

Here is the image you requested.

[crisp stream]

naaah 😄

[hazy warren]

[hazy warren]

this would make a fun video game

[hazy warren]

me after a 12 course meal

[wispy nest]

I only just now noticed the heels

[grizzled sage]

it's because he wears lifters and thats his usual posture anyways

[jovial tiger]

i didn't even prompt this.

[jovial tiger]

I think that's how you get to deep space nine

[marble hawk]

Warrior cat with a blue head band , black and silver armour, anime style

[limber oxide]

hello

[plush forum]

1girl,longhair

[hazy warren]

@wispy nest @simple frigate Looks pretty realistic

[ebon zealot]

https://cdn.aibooru.download/original/88/59/__cirno_generated_by_conquestace_using_starry_night_model__8859e745c0f02c4504c25863a4aadba6.png

[wet rampart]

/Pale glowing full moon shimmering through beautiful streaky clouds, dark blue night sky,Elsa Bleda,Marta Bevacqua
Negative Prompt: motion blur,realistic,,ground,woman

[crisp stream]

@hazy warren

[hazy warren]

😮 😮 😮

[wispy nest]

Does second one looks more real to me but bolth are great nice work vortex

[wispy nest]

Oh wow love this

[wild sorrel]

yea style is good

a bit too shiny tho

[wispy nest]

Cloudcore mirrors

[kindred cape]

#🌠｜show-and-tell Plant Mond

[drifting trellis]

sunset

/create

[eternal comet]

Model one:

Model two:

[cyan shoal]

img2img will be fantastic with SD3

[clever oar]

android os in anime?

[arctic shuttle]

#best friende

[clever oar]

sd3 lisen promt 100%?

[eternal comet]

how do i fix blurry images

[cyan shoal]

it will probably listen from like 80% to 100%

depending if you use T5 and how long your prompt is

[clever oar]

sd now maybe 60-70%

[cyan shoal]

yeah it depends, if its just simple portraits, even current SD models do 90%-100%

for super super complex prompts probably 70-90%

🤷‍♂️

like there was a beach image and it left out palm trees but everything else was perfect

[clever oar]

sometimes sd does what I wrote in the negative

[cyan shoal]

weird..?

[clever oar]

doesn't listen to me

[eternal comet]

sd images are blurrty

for me

how do you fix

[clever oar]

sdxl model?

server room underwater

[cyan shoal]

ooh nice

[cyan shoal]

even when using fooocus?

try highresfix

and whatever that sharpening option is in Foooooooocus

[eternal comet]

no

just stable diffusion

[clever oar]

how krea ai generate so fast

with close result

[eternal comet]

DAMN

IT STEPPED UP

🥶

NAHH THIS IS AN INSANE SETUP

it followed prompt exactly.

[cyan shoal]

wow that's really nice

[eternal comet]

"devil on the battlefield

this is too op

[clever oar]

this nice what model and promt

[cyan shoal]

if these are sdxl then please post them in #✨｜sdxl, people might confuse it with 1.X

[eternal comet]

"solider with skullface mask pulling back his ak smoke coming out devil eyes dark powerful full block swat outfit full body shot red eyes smoke coming out gun. dark room filled with smoke soldier poking out the smoke red lights terrifying atmosphere "

[clever oar]

no no)

[cyan shoal]

the server room one

[clever oar]

yes

[eternal comet]

i just put what he put

[eternal comet]

alll i did was put this

[cyan shoal]

oh

[clever oar]

my model not give me same realistic result

[eternal comet]

what model u use

[crisp stream]

so this channel isn't for every version, merely 1.x?

[clever oar]

now dreamshaper before jugernaut

[cyan shoal]

well everything that is not SDXL if I recall correctly 🤷‍♂️

there's nothing stopping you from posting here, its just that people might think these are not SDXL, since there is a separate channel for that

you can post 2.X, Cascade, etc

[eternal comet]

you can just say its sdxl

also why does it matter cant people just use sdxl

[cyan shoal]

what

oh I mean post

[eternal comet]

also

[cyan shoal]

😅

[eternal comet]

sdxl juggernaught is very sus

💀

[cyan shoal]

damn

[eternal comet]

if you only put "women" half the time it generates nsfw

[crisp stream]

if nothing is stopping me, I continue here as well 😀

[cyan shoal]

yeah it probably has nsfw in the dataset

[eternal comet]

💀

[cyan shoal]

this image looks normal

[eternal comet]

too perfect

issue most ai models have

[clever oar]

For some reason the neural network doesn’t understand the names of ancient animals

[eternal comet]

💀

[clever oar]

I tried maybe 30 generations and she never drew an ichthyosaur)

mj cant do this for me too)

[clever oar]

trump demon

[eternal comet]

this is very cool

could be used in ads etc

[eternal comet]

ahaahahah

[eternal comet]

this makes me feel wholesome

for some reason i9 dont find cats animals or anythign else cute but i find this cute 💀

i find robots cute like wall e etdc

💀

[crisp stream]

lf you ask nicely, he may hand you some pancakes 🙂

[eternal comet]

yes

or tear off your head

💀

[eternal comet]

what prompt

[clever oar]

man with rooster tatoo)

[crisp stream]

naaah, he's chill 🙂

[eternal comet]

let me try this

my model has a shit ton of loras my pc is dying

[glossy remnant]

how do i correct morphed/ extra fingers after image generation, new here sorry if the question isn't worth asking

[eternal comet]

prompt disciplining

[crisp stream]

you could inpaint for one

[clever oar]

nice

[eternal comet]

"DO NOT DO MULTIPLE FINGERS OR INCORRECT ANATOMY. IF YOU DO THE EARTH WILL EXPLODE"

[clever oar]

but you need put full body man

[eternal comet]

aight

[glossy remnant]

i can't find that option in the stable swarm ui, i am aware of it in automatic1111 but my pc couldn't handle automatic

[eternal comet]

THE FEET AHAHA

[clever oar]

need face too)

[glossy remnant]

is there a documentation?

[clever oar]

in prison you have big problem with this tatoo lmao

[crisp stream]

have you checked on Easy Diffusion UI? Got a pleasant inpainting and also drawing interface. It's as well a 1-click installation.

[eternal comet]

lets try gen a anime girl

[glossy remnant]

ohh i never heard about it

[eternal comet]

💀

ok this model was trained off some weird stuff

[cyan shoal]

this looks like some negative prompt

[crisp stream]

it's what i mostly use so far.

[eternal comet]

nah i said add imperfections on purpose

[cyan shoal]

ah

[clever oar]

jill re3

[crisp stream]

the inpainting overlay

[eternal comet]

prompt?

the blood

what

[clever oar]

i use lora jill

[eternal comet]

why does she have cuts

[glossy remnant]

also i have only 4gb vram nvidia 1050, and intel gpu is never in use when running the UI why is that?

[clever oar]

after fights

with nemesis

chill

[eternal comet]

ah

i see my badd

fighting lora

like a deepfake

you could deepfake someone if you make a lora from them tbf

and put their name in the blip prompt thing

then mention it in the gui

with the lora

[crisp stream]

it uses the main gpu only in that combination from what I know. I dont think Intel GPU arent even supported, could be wrong though

[clever oar]

something like this

https://civitai.com/models/140120/jill-valentine-resident-evil-3-remake

i dont understand why she cut

[eternal comet]

ive wanted to paste song lyris

into prompts

and then see what they generate

[glossy remnant]

ok on the reddit people were saying it uses multiple gpu's, and tbh it sucks cuz i can't upgrade for a while i have a laptop.

[crisp stream]

4GB with SDXL isnt really enough, yet you can work with 1.x

[clever oar]

drug addict's dream)

[glossy remnant]

yea it works but when i upscale it takes like 20-30 minutes

[eternal comet]

ginseng strip

yeah let me feed full lyrics

[crisp stream]

yes muliple gpu possibly, just not with the second one being an Intel GPU I at least assume

[eternal comet]

💀

wtf it just started generating the artist

lets try a sad song

[glossy remnant]

i used sdxl 1.0 as base + photon(photo realistic model) as refiner again got a cursed result lol.

[crisp stream]

so you are running SDXL with your 4GB?

[glossy remnant]

yes

[crisp stream]

interesting

[glossy remnant]

and a normal image comes out within 5-10mins

[crisp stream]

didn´t think it was possible

[eternal comet]

💀

dido thank you

generated this

the lyrics to dido thank you

[crisp stream]

at what resolution?

[glossy remnant]

in the Automatic ui it didnt work but now i downloaded stable swarm ui, it actually works

[glossy remnant]

it says 512* 512 not sure

[crisp stream]

ah, sounds good

[eternal comet]

how long does it take you to gen

it takes me 30 seconds

[glossy remnant]

5mins but if i upscale it takes forever

[crisp stream]

yeah, ok that´s too low when using the pure basemodel here because it produces colourburn and cartoonisation at that resolution.

[eternal comet]

"priceless pancake"

[crisp stream]

yet your´s seem to work at least when looking at the lady

[eternal comet]

what resolution is optimal

[glossy remnant]

i will have to check how it changed the other day it was 1080

[eternal comet]

im hungry now

[crisp stream]

this is at 1280x768

and this at 512x512

[glossy remnant]

ok ok

[eternal comet]

what

why am i getting good images at 512x512

💀

[crisp stream]

I don´t know, possibly because of Refiner?

[crisp stream]

the training happened along 1024x1024. 640x640 looks better, yet not there yet:

704x704

[glossy remnant]

although i am generating at 512512 why does the output image says 10241024

[crisp stream]

Automatic upscaling of some sort?

[glossy remnant]

idk i checked in the output folder and it says

[eternal comet]

for fun

i generated an image at 1280x1280

[crisp stream]

768x704

[eternal comet]

wait

what if you just load a lora

could you use it as a model

🤔

[crisp stream]

not that familiar with loRAs then I would assume it needs a model to sit on top, don´t know though

768x768

[eternal comet]

if you have enough rtesources

maybe

[glossy remnant]

ohh after downloading it becomes 1536*1408 , double

[crisp stream]

nono, it´s merely upscaled 🙂

[glossy remnant]

noway XD

[crisp stream]

the original is at 768x704

[crisp stream]

🤭

[glossy remnant]

i am seem to be lacking knowledge about how pixels work

[crisp stream]

I simply generated at 768x704 and then upscaled it 2x afterwards, hence 1536x1408

[eternal comet]

1280x1280

[crisp stream]

1024x1024

[hazy warren]

Thanks man,
Yeah the 2nd one, added text like Pores and realistic skin to the prompt

I think females are harder to get more realistic skin probably coz the skin so so smooth

[glossy remnant]

but how do you blend her with background

it shouldn't look cut pasted

[hazy warren]

I didn’t add anything for the prompt for that, I just said “outside”

[glossy remnant]

are you using a specific model or sdxl

[hazy warren]

I used a version of a SD model, EpicPhotoGasm I think

[glossy remnant]

hell na the orignal on left and the upscaled on right

[hazy warren]

Yeah, you need to lower the noise to 0.3

[glossy remnant]

ok will try, thanks

[glossy remnant]

i actually changed the model when upscaling it's a model for photorealism that might be the reason

[hazy warren]

Would be the noise mostly

[cyan shoal]

upscale from lykon

[crisp stream]

@glossy remnant at 1280x1280 high-res artefacts are already appearing

[glossy remnant]

i can't even find the option for 1280

[crisp stream]

yeah, well, don´t know about your UI

[glossy remnant]

768x320 this is max i think

[glossy remnant]

what's the prompt btw

[crisp stream]

Usually I don´t hand out prompt, yet don´t care much this time, it´s also pretty basic:
Pascal Blanche,insanely weird pancakecooker golem giant warpinstruction,wes anderson

[glossy remnant]

are they from some movie or something or totally original?

[crisp stream]

what do you mean? If there is a movie that shows those? I don´t know, guess not 😄

[glossy remnant]

yea very unique characters/ robos

[glossy remnant]

reminds me of those solarpunk photos

[crisp stream]

832x832

[eternal comet]

with refiner?

do i need to use that resolution

with a refiner

[crisp stream]

nope, pure base model

[eternal comet]

if i use a refiner at 512

would it be the same quality

[crisp stream]

don´t know, not really using refiner here, how about testing?

[eternal comet]

its the same quality as your 1028 gens

[crisp stream]

example?

yeah, so like thought of earlier refiner might make it possible to generate good looking images at 512x512. With the pure base model, well 😄

[crisp stream]

You could check how it turns out when using Refiner at 1280x1280. Here, like said, with the pure base model it is already producing high-res artefacts

Not really a friend of Refiner here because it tends to make things look more generic, at least the stuff I do

could be pleasant for photorealistic stuff, don´t know

[glossy remnant]

ye it does produce higher quality than it initally says

just with base model

[crisp stream]

what do you mean exactly?

[glossy remnant]

like u said it upscales it or something, when it gives the output

[crisp stream]

no, I upscale it manually after it generated

[crisp stream]

I thought you were referring to your UI

[glossy remnant]

nvm i am just confused xD

[crisp stream]

😄

[glossy remnant]

with your prompt

it didn't got the robo legs and arms

[crisp stream]

yes, I noticed, there is the negative prompt missing though

so what was your original resolution you generated at?

[glossy remnant]

Resolution: 21:9 (1536x640)

[glossy remnant]

he fits in this solarpunk kinda environment

[crisp stream]

This is as well without the negative prompt at 1080x1080

natively, after upscaling 2160x2160

[crisp stream]

true

[glossy remnant]

unlike the ones i generated at 512 * 512 and came out at 1024 * 1024. This one came out at original resolution

[eternal comet]

ok

[crisp stream]

I see 🙂 Like said, not familiar with your UI, so I coudln´t really tell what´s going on 😄

[eternal comet]

i asked it to generate an image of someone eating a lolipop

[eternal comet]

i made a mistake using british word

💀

[glossy remnant]

okayyy finally my UI was tripping or something the resolution at aspect ratio 1:1 changed on its own, so indeed all the images are at relatively at high res

[glossy remnant]

what r u talking bout lol

[eternal comet]

Cant gen food eating

[eternal comet]

very nice

[glossy remnant]

pretty cool

[crisp stream]

ty

[glossy remnant]

is the man noise or you prompted him too

sometimes it just puts in elements on its own

curious

[crisp stream]

guess it´s coming from Wes Anderson

[clever oar]

who can make an ichthyosaur,?

sd not listening my promt

[glossy remnant]

what does it give as output

[clever oar]

like this

but ichthyosaur,

or underwater

[glossy remnant]

it probably doesn't know what it is

like not trained on enough pics

[clever oar]

also mj dont know too

strange

[glossy remnant]

mj?

try Lora

[clever oar]

midjourney dont know too like sd

[glossy remnant]

there is a lora for ichthosaur on civitai not sure if it works well

[clever oar]

downloading

nappa why you use xl with 4gb is so long

[glossy remnant]

it works just works lol, takes 3-5mins

[crisp stream]

it knows what it is, yet there are many varying depictions or styles how an ichthyosaurus is being imagined when you look at a search engine´s result (I at least assume that´s the reason

[glossy remnant]

or maybe it got confused with other fishes, cuz the mouth looks like that of a fish. totally not an ai expert here

[glossy remnant]

it just can't get the mouth right

[eternal comet]

did my pic get the mouth right

tbf i think it did

[glossy remnant]

nope

[clever oar]

not like this

[clever oar]

yes

like this

how you do this

[glossy remnant]

it just outputs a fish everytime

[crisp stream]

this is what discordos is going for I guess:

[glossy remnant]

idk its from google

[clever oar]

yes nice

[crisp stream]

that´s from ecosia search

this is also from ecosia search

This as well

[clever oar]

i do with lora

[glossy remnant]

how about simply describing it's anatomy in more detail

[clever oar]

with lora from civit

[glossy remnant]

cartoonish

[eternal comet]

ill send mine in a sec

im loading a new model

its taking the piss

[glossy remnant]

got that creepy smile

[eternal comet]

i use 2 models

one is base other is refiner

[clever oar]

https://civitai.com/models/257163/ichthyosaurus

this lora

i use

[glossy remnant]

i don't see anyone else posting image of ichthyosaur and this is the only lora

[eternal comet]

e

yk someone deepfaked me

using their iphone

lol

luckily it was too bad for anyone to beleive it

[glossy remnant]

what why xd

[eternal comet]

crazy person

very evil person

if i wanted to i could get revenge now but revenge is a fools game

[glossy remnant]

i made deep fake of my frnd as a meme to post in our group chat lol

[eternal comet]

lol

this person made a bad deepfake of me

using their iphone and they thought they were the shit

[glossy remnant]

how in iphone, google docs?

[eternal comet]

art

so they drew over my face

[glossy remnant]

ohk

[eternal comet]

they spent a long time drawing over my face

just to yk

it was obvious it wasnt real

shes lucky i didnt retaliate

https://tenor.com/view/memes-risitas-laughing-lmao-lol-gif-11500030

rate reflections

eh

[glossy remnant]

woah, u made this?

looks so real

[eternal comet]

yuh

fine tuned some rain loras

💀

[glossy remnant]

what's the base model

[eternal comet]

sdxl

[glossy remnant]

and which lora

[eternal comet]

"women shooting water from her hands"

[glossy remnant]

these are really good

[eternal comet]

lora i made trained off rain

[clever oar]

model?

[eternal comet]

sdxl

[clever oar]

oh(

[eternal comet]

ye

its beating my gpus ass though

[clever oar]

what your gpu

[eternal comet]

rtx 3060 12gb

[clever oar]

mt gtx 1050 ti

[eternal comet]

"spraying water from her hands!

[glossy remnant]

@eternal comet i have a question kinda off topic, i downloaded a model for photo realism and copied seed and various other parameters of the example image produced using that model but the difference in quality is huge

and all the images look quite bad with this model, its called photon but the images other users generated are wayy better

[eternal comet]

sec

let it cook

[glossy remnant]

DPM++ 2M Karras, they all are using this sampler i don't know if there's option for that in my UI

[clever oar]

with lora it cant make normal background

[glossy remnant]

do you use samplers as well? @eternal comet

[eternal comet]

ye

i sample it through two models

[glossy remnant]

like the base and refiner?

[eternal comet]

💀 wtf is this

ye

[glossy remnant]

lol u tried the same prompts?

or img2img

[glossy remnant]

with sdxl 1

lmao

[eternal comet]

cant tell

whos is better

[glossy remnant]

looks half decent now

[glossy remnant]

did u generate or image search on google, looks so good

[eternal comet]

generate it

[glossy remnant]

power of refiner??

[eternal comet]

idk

[clever oar]

make her fly into the fire from her ass

[wild sorrel]

You made her like 20 years older

[glossy remnant]

but it was upscaled 🥲

[glossy remnant]

anyways looks less animeish

[grizzled sage]

in my eyes that looks more like 14-24 aged lol

i'm 41 an she don't look anywhere close to women my age

[glossy remnant]

yes

[eternal comet]

sec

let me try again

[glossy remnant]

the imperfections are necessary i would say

[clever oar]

her ass is very hot

[grizzled sage]

you cant even see it! ooooooohhhh wait i get it

🔥

[pallid ruin]

🤣

[glossy remnant]

💀

literally

what she eating doe

[eternal comet]

rate the detail

[glossy remnant]

teach me the ways

[wild sorrel]

I went overboard with 20 years older, I'd say
20-25 first image (30 max), while skin on 2nd look anywhere from 30 to 40

But yea, that's just my thoughts ¯_(ツ)_/¯

[glossy remnant]

it just has more shadows the 2nd one

i just remembered this old meme of knuckles saying "do you know dae wae"

[grizzled sage]

vrchat ugly knuckles

[glossy remnant]

ye looks like that

[grizzled sage]

a big spoof of uganda's first publicly funded movie

[clever oar]

beans

[grizzled sage]

such a simple prompt. nice. juggernaut is a good model. i used your exact params, but i have fp8 turned on.

[eternal comet]

💀

they look the same

wtf

[grizzled sage]

same seed an params with juggernaut 9

[eternal comet]

we watching anime with the boys 🔥

whys it in german

[grizzled sage]

juggernaut 9 and 8 without fp8 turned on

don't sleep on fp8. the quality differences here are highly subjective.

[pallid ruin]

What was the prompt? 👀

[grizzled sage]

women showing off her best features
n: blurry, jagged, unrealistic, saturated, high contrast, big nose, painting, drawing, sketch, cartoon, anime, manga, render, CG, 3d, watermark,

[eternal comet]

i see no difference

[eternal comet]

what

HOW DDI YOU PULL THE PROMPT

crazy

is therre meta data

[grizzled sage]

metadata yup

https://tenor.com/view/anonymous-hackers-hack-the-planet-dade-murphy-jonny-lee-miller-gif-5405051

[eternal comet]

let me pull out the flex model

💀

my friend generated this

and this

[serene dew]

shadow is not shadowing

[eternal comet]

juggernaught doesnt do fingers good

[grizzled sage]

racist memes aren't appreciated anywhere i'm posting. wonder if the mods care about this shit.

[serene dew]

where is it racist?

[eternal comet]

how about we all calm down

and have chicken nuggets

!

[serene dew]

actress is not <snow white> and thats just a fact

[grizzled sage]

you know exactly what you're doing. playing dumb must come naturally

[serene dew]

PIZZA

[eternal comet]

tbh fuck disney

i dont really care what they do

[grizzled sage]

it's disney's story and they can tell it anyway they want... so.... it turns out she is

[eternal comet]

this

end of the day its their brand their story their movie

their choice

[serene dew]

okay, maybe shes not coal black, she choclete brown, i exagerated

but she is brown, not white

[grizzled sage]

and still classier than you

[serene dew]

its sad its no longer Walt Disney

[grizzled sage]

you're not really sad. you're just manufacturing outrage and you've never really cared about snow white before

[serene dew]

and how does those 2 corelate?

[grizzled sage]

https://tenor.com/view/stay-classy-classy-serious-gif-12273384

[serene dew]

i care for <snow white>, bc its a good movie from childhood

but only thing thats gonnna be snow white is title sadly

thats pretty sus

[pallid ruin]

wrong response

Also why the ai cant generate chicken nuggets? U.U

[grizzled sage]

fragile egos are fragile egos

[serene dew]

speaking from expirience?

[grizzled sage]

likes tendies better

[pallid ruin]

i´ll try that

[serene dew]

i think sauce is the problem, try nuggets with no sauce

[arctic laurel]

does your username spell “FARMACIST” or am I just being reductionist?

[pallid ruin]

I should use simpler prompts, this is strange af

[grizzled sage]

its a tendie cake!

[serene dew]

no, it does not spell like that

first letter is A

[arctic laurel]

almost looks like a koeign amann

[serene dew]

theres acutally cake that looks like thius

[grizzled sage]

fnrmcmsr

obsolete script from a dead society

[serene dew]

@grizzled sage now i wait for Michael Jordan Biography with main role being played by white man

i wonder what would be the reaction

[arctic laurel]

kouign amann* had to refresh my memory on the spelling of that bad boy

best pastry ever

[grizzled sage]

you've come up with some weird fantasies. MJ biopic was already out

[serene dew]

"better" remake

||" " = irony||

[grizzled sage]

the vikings were not a supreme race. they in fact died

[serene dew]

viking isnt fucking race

[grizzled sage]

a very primative culture. basically sucked at what they did. the polynesians beat them at every category

Remember when the Rock played hercules? awesome movie

better than arnold strong's hercule

[serene dew]

nordic race still exists, and it often shown as example of well being

[grizzled sage]

are they though? they're kind of not supreme

[serene dew]

wait for disney live action with black hercules

[serene dew]

neither race is supreme

[grizzled sage]

lol dwayne johnson is a black man

it already was live action

[serene dew]

very black

[grizzled sage]

nobody cared because youtuber outrage culture didn't exist

[serene dew]

ik, but it not first hercules movie

[grizzled sage]

https://tenor.com/view/success-rocky-johnson-young-rock-i-did-it-i-won-gif-20617505

[minor fractal]

sd3 beats the box test A photo of a blue box on a white background. A smaller red box is sitting on top of the blue box.

[grizzled sage]

that first one though

[cyan shoal]

the first one is not right, SD3 is so poopy doopy you guys

[minor fractal]

"oh sd3 sucks just delete it already"

[cyan shoal]

all hope lost its the worst model we have seen in the entire universe guys /s

[cyan shoal]

where did you find this btw

[minor fractal]

touhou ai

yet another weeb ai server

[cyan shoal]

ahh and someone has access over there?

[minor fractal]

yep