✨｜sdxl

[tawdry current]

"A young man in his early 20s standing in a dimly lit room, surrounded by shadows of tempting figures. His eyes are wide with longing, and his hands reach out towards the shadows, symbolizing his entrapment by desires and lusts. The background features a mix of dark, alluring colors, adding to the atmosphere of temptation and struggle. The scene conveys a sense of internal conflict and emotional intensity."

All the strikethroughs are just token noise that don't actually mean anything to the model. Though words like tempting, longing and alluring might be associated with NSFW concepts like nudity. Longing might also be associated with sadness in facial expressions.

[warm forge]

This Prompt feels like one of the "improved" DALLE-3 prompts that you get on the OpenAI discord server, lol

Minus the "make sure it's family friendly" part that the "improved" prompts have 🤣

[heavy sapphire]

hey people

[worthy orbit]

any tips to avoid these kind of seemlines in SD ultimate upscaler?

[vagrant tundra]

Draw a graph with Sinzh on it.

[tawdry current]

Use a tile controlnet with it

https://huggingface.co/lllyasviel/control_v11f1e_sd15_tile

There aren't really a good 1024x1024 sdxl ones, so you usually need to just use the sd1.5 one

So if you're upscaling a 1024 sdxl image to let's say 2048, with the sd1.5 tile controlnet, you have to do the upscaling with a 1.5 model

And the tile size will be more like 512x512

[worthy orbit]

Mmm yes, I didn't like the upscalling with SD 1.5 when I tried it, but I might aswell test it again

[tawdry current]

Just use a low denoise and next to nothing in the prompts. Only have like "high quality, best quality" and "text, watermark, worst quality" for prompts

Denoise should be like .25 or so usually

Should leave the image relatively unchanged and will get rid of any ai upscaling shartifacts

[worthy orbit]

what SD 1.5 model?

[tawdry current]

I usually just use realistic vision or dream shaper depending on the style

Then again, those are the only other 1.5 models I have other than the base lol

[worthy orbit]

ok, ty, I'll try it

[worthy orbit]

still I see the same seems

I use the CN model

[tawdry current]

what was your denoise set to?

[worthy orbit]

0.25

[tawdry current]

might need to bump it up a little, try like .3, then .35, etc checking to see if it changes your overall image too much. also, what are your steps?

and which sampler are you using

[worthy orbit]

30 steps, DPM++ SDE

[tawdry current]

try euler or dpmpp 2m, sde samplers aren't "stable"

[worthy orbit]

dpmpp 2m I don't think is in Automatic1111

[tawdry current]

well just use euler then, it's in the same family

i don't use a1111, so i dont know what all they have to pick from

and bump the denoise up to something more like .3 or .35

yeah i just tried on a BS test prompt and am not seeing any bands

[worthy orbit]

ok let me see

[tawdry current]

for instance

tried it with a gray background where it would really show up well

yeah its in comfyui, but the settings should still map the same to a1111

meh, i guess i see a tiny bit of a band here and there, but it's faint

[worthy orbit]

yes here aswell, but less than before

and some aberrations

[tawdry current]

Yeah it's always a pain in the ass to work with though, there's no real perfect solution other than using a model that can actually handle the higher resolutions. You might luck out and find a decent AI upscaler that can just scale it up without adding in a ton of artifacts or noise

4x ultrasharp is pretty well rounded tbh, but there are some other ones that work pretty well too. you can also use other programs like topaz(does ai upscaling, but on steroids with controls and settings)

[worthy orbit]

it gave a SD 1.5 style look, not bad, but I don't know

[worthy orbit]

yes I had use it but really I prefer using this. It enlarges the image with cheap artifacts

[tawdry current]

Well like I said, you'd need to find a model that can handle higher resolutions. Maybe check into SUPIR? It's demanding on hardware and I'm not sure if there's an a1111 implementation of it.

https://github.com/Fanghua-Yu/SUPIR

and https://github.com/kijai/ComfyUI-SUPIR for the comfy version

[worthy orbit]

oh I'm too lazy right now and I don't have much disk space

I have tried Euler and denoising of 0.35 but with SD XL model, no seems (it make the vertical seem as someething in the image)

Also changed the image a lot as it didn't know what it was

still some seem lines, but more inpaintable

I lowered the denoising and added the prompt, I like this one

still seems but It is worth the effor to inpaint them, I guess it is the best I can do

[twilit loom]

Hello everyone. Can anyone suggest how can I shift the floral motif near the chest to the centre of the t shirt. I am using inpainting and through controlnet, but not getting perfect results. 2nd pic is what i am expecting.

[stray warren]

Corporate greed

[civic gyro]

https://twitter.com/HikariUchu/status/1791143296703881433

Hikari Uchu (@HikariUchu) on X

A soul ascends, guided by ethereal beings into a heavenly realm. 🌌✨ #SpiritualJourney

#animation ＃デジタルアート #digitalart #美人 #beauty #hikariuchu #visualart #vj #aigenerativeart #aiartcommunity ＃コミュニティ#未来的な #futuristic #インスパイヤ #inspire #アニメ#anime

Twitter · May 16, 2024, 4:26 PM

Hopefully I go up and not down :p

[stray warren]

We have AYS now in Draw Things, very happy with it 🙂

[cyan crown]

https://civitai.com/models/458820?modelVersionId=510648

[peak dove]

Pixart-Sigma+SD15+SDXL+PAGAdvanced+FaceDetailer

[copper kraken]

spectacular

[peak dove]

Pixart-Sigma, SD15, SDXL, LoRAs (Poetic Style+Ethereal Grace), PAG Advanced, Face Detailer

[peak dove]

I need a LoRA which can re-create this style?

[copper kraken]

get at least 20 images in that style, or see if you can generate a training set with ipadapter

[fluid mortar]

If I have Automatic1111 locally installed, will SDXL models work out of the box, or do I have to install something else?

[gloomy lark]

Just need to follow the instructions in the install page. That should pretty much cover it. The sdxl model is pretty much all you need. I forget if a1111 automatically downloads the vae model or not.

[fluid mortar]

Thanks. It seems most models have VAE baked in anyhow

[gloomy lark]

The fixed fp16 vae has been out long enough that it should be standard now. It took way too long for model makers to start using it.

[real wigeon]

？

[stark vortex]

steampunk woman on a rocket over a steampunkcity cienmatic, photographic,

[copper kraken]

Here is the image you requested.

[copper kraken]

i've encountered a few recent models that had the busted VAE baked in... i just use the fixed VAE no matter what checkpoint i'm on now

[glass forge]

Wonder what the promts are for these abominations.

[gloomy lark]

a man holding on tight to a jiggly jello worm that's spewing chicken nuggets out of its mouth

[glass forge]

I can see that...

[gloomy lark]

it's rather prompt following. 🙂

[glass forge]

I like how it can't figure out if nuggets are an actual chicken or some fried food.

[gloomy lark]

pixart and ella pretty much have no idea what nuggets are. the sdxl refiner cleans things up but it can't fix the composition. so you get this. 🙂

[glass forge]

[true willow]

Soldier sleeping in watermelon truck.

[glass forge]

Why the solder is a rabbit tho.

[gloomy lark]

he didn't say it shouldn't be.

[craggy granite]

a mango farm

[native sail]

Text description
SCENE DESCRIPTION: On the surface of the moon, there is an astronaut standing next to a huge rock.
Description of subject: the subject object is an astronaut wearing a modern sci-fi style spacesuit with an image of the Earth reflected on his helmet. The astronaut is standing, facing away from the Earth, and appears to be gazing at the Earth in the distance.
Supporting Character Description:Next to the astronauts, there are some tire tracks from a lunar rover stretching into the distance, and a lunar rover is faintly visible parked not far away. In the sky, a few stars twinkle, enhancing the sci-fi atmosphere.
BACKGROUND DESCRIPTION: The background should contain a huge Earth hanging in the darkness of space, with the blues and greens of the Earth shining brightly in the darkness of space. The surface of the moon is gray, filled with craters and rocks, and the overall tone is cool.
Other details: the image should also have the clouds and continents on Earth clearly visible, some high-tech glowing lines on the astronauts' helmets and spacesuits, and some reflective small rocks and dust on the surface of the Moon.
Generate settings
Image resolution: 1920x1080
Style Preference: Anime Style
Color preference: cool tones
Other parameters: none

[quaint lagoon]

#✨｜sdxl message

[worthy orbit]

I'm trying to correct something in an upscaled image but I keep getting horrible results

[dawn burrow]

Hey guys does anyone have a cool prompt to try it out ?

[peak dove]

gorgeous woman, beautiful face, shadow noir moebius strip, sensual, long slender legs, minimalist photograph black and white, black background, woman, streaming red hair, shells, rain, trees, rain, shower, blue sky, white clouds, peruvian arpillera victo ngai henri rousseau vladimir kush oxana dobrovolska peregrine heathcoate catrin welz-stein itzchak tarkay william haskell jack vettriano thomas kincade maxfield parrish andrea kurti andrea kowch mads berg tamara lempicki kedra nelson tishk barzanji

[dawn burrow]

okay please 1 minute

[copper kraken]

A very high quality photo of a clown shark pickle tickle scooby doo cereal with a missile freak cannibal huffer, inside the inverted sock muppet of a fractal digestion in oil buckets with paint, melancholy atmosphere, stormy night, very dark, cinematic shot

[dawn burrow]

i think this one is blocked

[peak dove]

Take out slender

And sensual

I think sensual is blocked

[dawn burrow]

yeah maybe

[peak dove]

Nice

[dawn burrow]

Wait i can upscale it too for u

[peak dove]

I use GigaPixel 2 Upscale 🙂

🥳

[dawn burrow]

I've made my own upscaler

wait it's uploading

[dawn burrow]

okay wait

[dawn burrow]

i will try it with dall-e 3

[peak dove]

Oh my!

Its a very dignified and sophisticated prompt alright

[dawn burrow]

oh nice

[dawn burrow]

also i will do it with dall-e 3 :

[stone fossil]

xD

[gloomy lark]

thanos: Fine, I'll do it myself.

[vital ermine]

https://civitai.com/models/446841

[vital trail]

Cristiano Ronaldo is smiling and he is in the uniform of the national football team of Uzbekistan with the flag of Uzbekistan behind him

[native knot]

I'm sure he is. But why are you telling everyone?

[peak dove]

De Zerbi is smiling and he is in the uniform of the national football team of Germany with the flag of Bayern behind him

[dawn burrow]

Hello guys i love you ❤️

[dawn burrow]

that's great quality noice ✅

[glass forge]

Facedetailer sometimes improves the face sometimes not.

I am confused.

[copper kraken]

this is fun

[upbeat summit]

nice stuff!

[copper kraken]

this one is spectacular

it's the surreal lighting

[upbeat summit]

indeed

[copper kraken]

yeah i see what you mean

it's subtle but it's there

if you're in comfy there's all kinds of tricks i can show you for that

[worthy orbit]

I've alredy tried to wipe out the seems lines with the smudge tool, to help SD

[copper kraken]

yeah that won't do it

it's the lighting

with that seam, across the screen you can see to the left it's lighter than to the right

and that persists

you could hide the obvious line but the structural issue will remain

[worthy orbit]

mm I don't see much lighting issues, in lighting it looks similar to the original, but maybe I'm missing something

[copper kraken]

it's more than just a few pixels... are you using the seam fix stuff?

prolly need tiles hitting at least 128 pixels over the seam

[worthy orbit]

I've try that at some point, I don't know for this one

Oh yes in the monitor it is noticible

But I think it could be inpainted, if inpainting worked as I wanted. As I couldn't fix the seams lines I didn't think about the other stuff... mostly is about knowing if I can do it.

[copper kraken]

yeah you wanna fix it before it happens

it's def doable

what's your original look like?

[copper kraken]

btw just in case you didn't realize it, your aspect ratio changed

[worthy orbit]

oh no, I didnt

[copper kraken]

yeah, open each in you r browser here and tab back and forth

it got a little narrower

that'll cost you a bit of detail

[worthy orbit]

yes I did

[copper kraken]

anyway the only thing i see that it fixed was the thumb

you could downscale that, patch that in over the current original thumb, and then throw it at supir

everything else looks ready for supir

[worthy orbit]

all this models are a lot of GB

[worthy orbit]

yes, maybe I don't upscale at all, and work with that resolution

I wanted to upscale and make crazy details

but at 100 % there's not a lot of difference, but in the upscalled version there some lighting thing that is better (aside of the errors)

[copper kraken]

What kinda card you got?

Tiled ksampler is the best at adding details

The one in a1111 is pretty decent

Not quite as good in most cases imo but a lot faster

[worthy orbit]

RX 6700

[copper kraken]

10gb vram?

i'm not sure supir would run on that but i don't actually know for sure

but tiled stuff, you might wanna try the tiled sampler in a1111

[worthy orbit]

12 GB

[worthy orbit]

yes I've read some guy with 24 GB didn't have it easy, so... I guess not for mee

[signal mirage]

Well i tried a sd 1.5 tile controlnet on your image.

My favorite part is that the ai was sure that on the top right corner is the head of a person 🙂

The technical device on the other hand got some new details

[ebon lagoon]

supir 2x upscale

[worthy orbit]

wow amazing

really nice both upscale

[copper kraken]

thumb is turning into a finger here

[ebon lagoon]

it definitely works w/12GB vram

[copper kraken]

great stuff... there is a good example here of what you have to watch out for with supir re: image artifacts

[copper kraken]

good t oknow!

the controls on the left, and the thumbnail - supir doesn't fix compositional errors

tiled upscale, then supir i've found can help with that kinda stuff

[ebon lagoon]

the settings by default (in the version i'm using) are optimized to be really true to the original image, so you have to change them to get it to hallucinate new details, i guess adding more noise, havent testing it that much

[copper kraken]

this area too - gotta watch out with wires

i'll admit i haven't done much beyond the "true to original" stuff with supir

[ebon lagoon]

so you can see it really keep the face details and expression in ways that other upscalers fuck up

[copper kraken]

yup

especially... likeness

[ebon lagoon]

but the downside is it doesn't by default fix problems in the source image

[ebon lagoon]

same

[tawdry current]

Just as an fyi: Controlnet_aux in comfy now has anyline(edge stuff) and it works really well with the new mistoline sdxl controlnet. Even with cnet strength around 50-70%, still holds up pretty well in my limited testing

[copper kraken]

mistoline is insane

[tawdry current]

I know people always complain and say sdxl controlnets are "bad" (even though they really aren't if you use them correctly) so I figured I throw that out there

[copper kraken]

yeah

mistoline is actually really phenomenal

better than any of the sd15 cnets with edge imo

it's wild

[tawdry current]

Yeah it's really great. Unless I actually need depth, it's probably going to be my new go to

[copper kraken]

yup

what i've found pairs especially well with it is ipadapter composition

and then use style from another image

but ipa comp + mistoline is almost like copying an image

[tawdry current]

I do use depth a lot though because I do a lot of scene rigging in UE5/zbrush where I port out perfect depth maps. But for regular stuff, mistoline is what I'll be using

[worthy orbit]

this was what I wanted with the upscale, too sad it took a lot of disk and I don't even know if it would run... I could try to run it on Windows

[worthy orbit]

yes That's look like upscaler shenanigans

[worthy orbit]

Really nice, there something off in the whole composition, but the details are really amazing

[tawdry current]

this is an example of what i mean. all you need is a simple postprocess material and then you can set the distance you want for the falloff

in case anyone wonders what i mean when i say i use ue5 for rigging scenes for depth maps

[signal mirage]

yeah i even changed the image aspect ratio in one steps without seeing it....
Here is one where even the paintings on the wall have become technical devices. Do you have the orignal prompt for the image?

[tawdry current]

and here's a quick example of using it with the 256 depth sdxl cnet. it's not perfect, but i picked a very noisy scene(all the little branches) on purpose. 30 steps

[worthy orbit]

the prompt was

a low angle portrait photo by (Hyeronimus Bosch:Dieric Bouts), (peasant using alien technology device with spectrogram display that look like medical equipment:1.2), a flemish Baroque, naive art, classical painting, fantastic landscape background beautiful lighting, photorealistic, 8k, hd, high details, trending on artstation

[peak dove]

FRree DiT Generation https://huggingface.co/spaces/Tencent-Hunyuan/HunyuanDiT

[peak dove]

#SDXL, #ComfyUI, Perturbed Attention Guidance, LoRAs - Image+Text input.

#SDXL, #ComfyUI, Perturbed Attention Guidance, LoRAs - Image+Text input.

#SDXL, #ComfyUI, Perturbed Attention Guidance, LoRAs - Image+Text input.

[gloomy lark]

very good stuff

[gloomy lark]

hunyuan refined on left, ella same on right. hunyuan prefers a cup, ella is busy and needs it to go.

[peak dove]

Hunyuan looks overcooked - like in the beginning of HDR days!!! 🙂

[gloomy lark]

yeah, also seems like it suffers a bit from stretching with landscope mode

[frail inlet]

is there a way to force oily/glossy appearance

i tried several prompt variations like glossy wet, glossy oily, with higher weights but it doesnt work always

for img2img

[copper kraken]

time for more wackiness

[trim orbit]

HDR photography still has this look to it. It's different from HDR display panels. When phtoographers use HDR methods, behind the scenes they're capturing 3 exposures. High exposure for getting all the details in the shadows, a regular exposure, and then a fast exposure to get all the details in the highlights. Then those 3 exposures are combined to get an image that has detail across all the levels of luminosity. This looks "odd" since it's not how photography has worked for over a hundred years. I suppose that could be the over cooked look you're talking about. It's still a thing photographers do. I personally love HDR photography. It looks more real to me. Since it's a print art, a lot of the effect is up to the photographer's discretion. With HDR panels, it's all standardized to a specification so that content producers have a pipeline. I feel though that in most cases of HDR video, they lean towards a subtle effect more than photographers do. Hollywood abhors change anything new is a huge risk that is often trialed progressively. 24fps is still the standard to this day for instance.

[misty hull]

"Their last Piknic" 🤣 🤣 . Great image. Colors are really nice 👌

[magic fossil]

#✍🏼｜rules-and-tos

[trail mirage]

https://tenor.com/bjFCx.gif

[signal mirage]

love your creativity, great prompt idea again.

[peak dove]

i2i+Text as prompt, into SDXL, PAG Advanced and LoRA

[toxic plank]

please where can i find the text to image genetor

[gloomy lark]

Hah wow you got a good explosion. I had to do so many generations just to get a streaking meteor that didn't look terrible

[stray warren]

This reminds me of SD2 😀

Just playing with words

[gloomy lark]

Fester can be a real hot head

[gloomy lark]

The "meteor" that wiped out the dinosaurs was actually a t-rex doing a shoryuken.

[peak dove]

Cannot Get ANYLINE node to install/import

[manic lake]

/dream: large orthodox cross in black outline, white background, the inscription “ ТАРТЮФ» in a cross in thin black letters, minimalism, purple shade

┬─┬ノ( º _ ºノ)

[calm spoke]

There are insects on the skull ┬─┬ノ( º _ ºノ)

[meager canopy]

https://tenor.com/view/the-bugs-bugs-bug-mosquito-mosquitos-gif-18522135

[south frigate]

Why is openpose CN totally ignored in SDXL? It still doesn't work very well in SDXL as in the beginning of SDXL?

[copper kraken]

That sucks, but tbh it's not used in that workflow you can just delete. It

[stray warren]

It works
Left is SD3 image I borrowed for this, right the one with open pose SDXL (Lllyasviel)
"(masterpiece), (best quality), (ultra-detailed), Morpheus, king of dreams sitting in a 50s kitsch lounge chair at his parents house, dressed as a magician , illustration, disheveled hair, detailed eyes, perfect composition, moist skin, intricate details, earrings, by wlop"

Here's another one, prompt:
"(masterpiece), (best quality), (ultra-detailed), Morpheus, king of dreams sitting on the toilet reading the papers , illustration, disheveled hair, detailed eyes, perfect composition, moist skin, intricate details, earrings, by wlop
Right is with Open pose + IP adapter Plus SDXL style

This is the top image without any controlnet

[copper kraken]

CN in SDXL can heavily depend on the base model

[copper kraken]

which checkpoint are you using?

[stray warren]

My own

Merged

[copper kraken]

gotcha, was just wondering for reference for the person asking the q

[stray warren]

I'm sure it works with realvisxl

[gloomy lark]

Have you tried composition in ip adapter?

Pic is unrelated

[gloomy lark]

With ip adapter strong style transfer which has some composition built in

[stray warren]

Yeah I have tried them all, very powerful 🙂

[gloomy lark]

Great stuff. This is one of those times where controlnet probably also helps. I had to pipe the output back in as the sample on the next run for it to change the guy into a robot like I originally asked for.

[stray warren]

I'm using Draw Things. We got these seperate options (All, Style Only, Style&Layout) recently which are very easy to use and extremely powerful.

Btw I used pose control together with ipadapter style&control (85%) to keep the composition as close as possible to the original.

Weight 85%, start 0 end 100

[gloomy lark]

Interesting, I tried that on the Mac a long time ago, looks like I need to try it again

[stray warren]

You need more patience but it's worth to give it a try, especially with the fast loras and samplers now

[gloomy lark]

Yeah exactly. I stopped because even my M2 Max wasn't useful with SD. That's probably better now so try these lightning models

With

[stray warren]

These two above took just under 1 minute

Per image

[gloomy lark]

Definitely better than it used to be. 🙂

[stray warren]

10 steps TCD hyper8 😊

[gloomy lark]

I need to give that a try on the Mac. I always get muddy results when I've tried hyper on the pc.

I'll have to investigate

[gloomy lark]

Congrats on the good sword positioning. That's not easy. 🙂

[meager canopy]

Thanks, some lucky hits there 🙂

[misty hull]

Joker first day of work

[misty hull]

Are you using a multistep masking composition?

[copper kraken]

nope

workflow is embedded btw

[misty hull]

Damn, seem to have many layers to it. Looks awsome 🔥

[copper kraken]

thx

it's a lora i trained on my own stuff

and a modified version of the res_momentumized sampler

on sdxl base

[misty hull]

Oh I saw that name somewhere else. I know now 🙂

pixart-sigma

[cunning locust]

Hello!, I've been watching many posts on social media about ai artists copying a certain generated image. How do I know if a image I generated is not a copy?

what would be good for more original pictures, checkpoint or lora?

[copper kraken]

none are copies

[peak dove]

SDXL1.0@ClipDrop into ComfyUI i2i

[peak dove]

"Lydia was worried! She'd only popped out for some milk and a scratch-card; but look at what she'd gotten herself into!!!? Her husband would never believe her!"

[bright terrace]

How to achieve Good text using SDXL?

[peak dove]

Use this LoRA from Civitai https://civitai.com/models/176555/harrlogos-xl-finally-custom-text-generation-in-sd

SDXL1.0@ClipDrop+i2i in ComfyUI

SDXL1.0@ClipDrop+i2i in ComfyUI

[bright terrace]

Ty ❤️

[peak dove]

SD3 into an SDXL i2i conditioning

[livid perch]

4k only

[peak dove]

SD3 made at ClipDrop; then into SDXL, LoRA, PAG Advanced, Face Detailer

[cosmic agate]

/dream/Beautiful butterfly on pink flowers, with a blue background, for a mobile wallpaper, in the style of copy space. The beautiful butterfly has beautiful wings in a hyper realistic photography style, set against a stunning nature scene. A blurred nature background with a cinematic photography style, presented at high resolution with hyper detailed, vibrant colors and bright colors in soft light. --ar 2:3 --v 5

[copper kraken]

Here is the image you requested.

[devout vine]

/dream/ beautiful blonde woman with blue eyes wearing a long trench coat with a dress, sitting on a bar

/dream/beautiful blonde woman with blue eyes wearing a long trench coat with a dress, sitting on a bar

[peak dove]

SD3 into SDXL Conditioning

[copper kraken]

Here is the image you requested.

[cyan crown]

https://civitai.com/models/464471?modelVersionId=516892

[pseudo shard]

Any idea on why SDXL anidiffusion (animation controlnet) when combined with a depth controlnet diffuser, gives very blurry results?
And when depth controlnet is set on a high value, the image looks burnt like if the ksampler tried to use the depth guides as base of the image, tried various workflows and in all its the same.

[sharp vortex]

Any idea why I get these weird multiple outlines? It looks fuzzy/blurry, not clean at all.

[copper kraken]

somethings wrong with your settings

[wet nacelle]

Sports bra and shorts

[wet nacelle]

I got way into the furry side of this. I post my stuff in the furry diffusion server now because I don't do too much experimental stuff anymore.

[minor pecan]

Pinky up!

[signal meadow]

Is SDXL limitd to 77 tokens? The staff at https://mage.space say it is.

[copper kraken]

yes

75, really

[signal meadow]

oof, ok thank you

[copper kraken]

https://novelai.net/tokenizer

use clip on this

[signal meadow]

Oh that's awesome, thank you. I've been using the openAI one

[tawdry current]

There are ways to use more, but it will just drown out the values of everything else, so it's not worth it. Most of the top quality "award winning high quality masterpiece 4k 8k 16k perfect hands ultra detailed awe-inspiring" type pieces are typically done in less than 20 word prompts. Maybe eventually when we get to something more like an SD4/5 equivalent and the average home PC has +24gb vram, you'll see larger prompt context windows like 300 tokens or something

[ashen otter]

A

[wet nacelle]

"A" big shit down my azz

[ashen otter]

Bruh

[peak dove]

Just d/loaded - working it with Ethereal Grace, Steampunk, and Postapocalyptic LoRAs.

[peak dove]

Using @cyan crown's Luminous Shadowscape LoRA

[ornate swift]

Can any one help me regarding how to run XL checkpoints on Automatic1111. I have updated it to the latest version but it still fails to open XL checkpointsq

[peak dove]

https://www.reddit.com/r/StableDiffusion/comments/15ag6tg/for_beginners_how_to_installuse_sdxl_10_in/

[stray warren]

#🆕｜sd3 message 😄

[stray warren]

Content waifu warning

[minor pecan]

Hot like a tamale !

[stray warren]

Pest & Vermin control squad

[grave isle]

Does anyone know how to create exact variants of a particular image. For example, if I want to create exact variants of a shirt in this format:

have not been able to get the exact orientation of the template when using Midjourney remix feature, heard Controlnet SDXL was better.....
Please PM me!

[native knot]

Yes, there are plenty of people who know how. If you don't have experience, though, your best bet is to start watching videos, reading websites, etc. because it's not a 1, 2, 3 solution. That doesn't exist yet.

[native knot]

The more specific your needs/desires, the less likely you are to achieve it without time & effort.

[grave isle]

Yes- I have the control_sd15_seg.ckpt downlaoded from HuggingFace (for the WebUI) but not sure if this is even the right method because of the outputs. Basically would appreciate someone pointing me in the right direction and then happy to put time and effort in

[native knot]

Like I said, it's not a 1, 2, 3 solution. That particular ControlNet is not going to get you what you need. But then your next question is going to be what will? To which there are also a lot of answers. Again, I highly suggest heading to YT & your favorite search engine...watch some videos until you find one that gets close.

Without that, if you just keep digging, you will begin to realize that you're either going to need to run an image through several processes in Automatic1111 to get kinda close to what you want, but not exactly, or you're going to have to learn ComfyUI. From there, you'll either give up or keep going down the rabbit hole, but you're going to have a lot more questions that all could've been answered by just watching a few videos.

[signal mirage]

In addition to all the recommendations I would suggest to think about what you need. Exactly the same would also include the pattern, color background etc.
I would suggest thinking about only using the outline, maybe together with a depth map to achieve the 3d feeling. Combined to one final depth map as input for a control net. If you need special items, numbers or logos it will become even more difficult.

[native knot]

And the suggestion to go search and watch videos isn't shrugging you off; it's literally one of the best ways to move quickly from not knowing shit to knowing some shit. Even if the video you end up latching onto is outdated as far as approach, you gain so much understanding and a good foothold for a starting position that it's still very valuable. From there, you know enough to learn more with self-guidance.

What we have a lot of in the world right now are tons of starving fishermen because they wanted to be told exactly how and where to fish, but they didn't want to learn how to bait a hook. We can't give fish away either, because that's not super useful for the long term. THE best way to learn is to push people into good starting positions.

Now, if one were to come with a question that's further ahead than just a few feet in front of the starting line in the race, it's easier to provide valuable guidance.

[signal mirage]

Reminds me of:
give a man a fish a day and he will have food for one day….
give him a poisoned fish and he will have food till the end of his life. 🙂

[native knot]

Yeah...I had a few variations in my head as I wrote that.

I get where people are coming from...they see a thing where they can type some words and get something that pretty much fits their early expectations. Then they see some stuff that gives them great ideas. (All of which are possible, given enough knowledge, time, money, etc.)

But it's the step after that which is difficult. Spending the time to learn the right way so that you can accomplish the difficult stuff. It means putting in the early effort, digging, getting things wrong a lot, sometimes crying, and then waking up the next day to try again.

This isn't limited to AI, either. It's true for nearly everything difficult.

Plus, the bar of "difficult" will continue to move. Because the moment something that used to be difficult becomes easy, it opens up a vast new universe of things that are suddenly now possible that were otherwise impossible previously. They're just "difficult" now.

[grave isle]

This was pretty helpful- I'm setting up ComfyUI right now. Im a developer, so am not scared off by installing GUIs and running them locally, etc but I do understand your greater point. Point taken ✅

[misty hull]

Crazy man. So many details 🔥🔥

[native knot]

It's just a screenshot from their website. It wasn't a single image generation that created that.

They are all tiles.

Not that it's not cool/impressive.

[mortal marsh]

Thank you! This is part of my social experiment - https://www.hexagen.world/

I created it just for fun, but unexpectedly, it became widely used, and today, there are already more than 4,500 hexagons made by users.

[misty hull]

Wow that's really cool idea. When did you started?

[mortal marsh]

Thanks! I launched it three weeks ago. Now I'm adding features to make it more playable and interactive.

[cedar night]

Hey guys can SDXL make this

[vernal forge]

Hey guys, is it possible to provide me with generation speed on 3090 and 4090 for sdxl on maybe juggernautxl with sde 2m ++ karras? (For 4 images) I am trying to decide if getting a 3090 for 600 euros is worth it instead of 4090 at 1200 (both used)

[copper kraken]

it's going to prolly be twice as fast or more

if you're cool with buying used, you could consider buying 2 3090s for the same price and using nvlink

[vernal forge]

Yeah, just wanna check how much exactly as it is twice the price. If it is something like 7-10 seconds I guess it is ok.

[native knot]

There is no "exactly" because there are other conditions/variables.

[vernal forge]

I have heard a lot about the power consumption so I think 2 is overkill for my bill and hobbyist status

[copper kraken]

4090 will be a lot more power efficient

if you have the cash to even consider the 4090, get the 4090

[native knot]

If you're planning on using the 4090 for some 4k gaming, then you can also benefit in that area, but money-wise, the 3090 is a good deal in comparison.

[copper kraken]

that is true^

having a 4090 i could never go back to anything any slower

[native knot]

Just depends on what you're doing/want to do.

[vernal forge]

Right now I am satisfied with my steam deck for light gaming so i don't think I have the need to use it for gaming

[native knot]

Do you have an additional $600 to burn simply for the sake of generating result a little faster?

Discretional funds are discretional funds...do what you want with them. But I'll say that if you have any serious debt, you should probably consider paying that off first.

[vernal forge]

I actually do have money to burn for this but raised to count every penny I would be ok with a bit slower generation time. 1200 is a lot of money to give at once. Also, the 4090s I have been offered for this price while in warranty(zotac trinity & gigabyte gaming oc), the companies dont recognise the second hand buyer as i have researched. So I am giving 1200 for basically no actual usable warranty.

[native knot]

Almost nobody takes advantage or even needs warranty service on a video card. Yes, there are some circumstances like the 4090 itself that seem to be exceptions, but the majority of 4090 owners still did not need their warranty service.

[vernal forge]

I see, should I jump the gun then? I am doing it as a hobby right now and want to test the waters so 1200 seems a bit much but I don't know if down the line i would like to do more. What was your experience?

[native knot]

I don't think anyone can tell you what you should do. That's a personal decision. I'm running a 3080 and feel happy with what I have. If I upgrade, it'll be a 5XXX series, but I'm buying a new car...so I may not.

[vernal forge]

Hmm, a difficult decision. It is even worse that I can't seem to find any inference test between those cards. There is only one in YouTube and it was when the 4090 dropped so it was still unoptimised which resulted in 7 seconds less than 3090

[copper kraken]

every time it hink about a new car, i remember i could buy like 15 5090s for the same price. i'd rather the 15 5090s. but i'm not insane enough to buy 15 5090s, so i'll just buy neither. lol

i'm at an impasse

[native knot]

Honestly, when you're splitting hairs at a few seconds of difference, consider what else you could do with that $600.

[native knot]

Here's the thing... I've had my car for 11 years and it's served me well, but it's also showing its age. It sucks on gas mileage, has nearly no good tech other than a basic bluetooth connection, and is going to need brakes, tires, and some other stuff in the coming year or two. 11 years ago, I wasn't in a place where I could buy a top trim of the model I was looking at, either.

But pretty soon, I'll have a brand new 2025 Camry XSE AWD that gets pretty much double the gas mileage, has two 12" displays with tons of features, and a beautiful limited $1k extra paint job (among other things). It might end up being the last car I ever have since Toyota's last forever.

[copper kraken]

awesome

yeah mine's at almost 20... haha

massive dents, got tboned by an idiot while parked and never got it fixed etc...

still moves, has airbags... yeah i have some looow standards

[native knot]

That's why I was saying to our new friend here that nobody can make that decision for them. They have to decide their own priorities. I have been thinking about a new car for a while but it wasn't time. Then the new Camry design hit and suddenly everything fell into place. In fact, I already have mine reserved and it should arrive at the dealership within the next 1-3 weeks. It'll be the 3rd Camry I've driven and 2nd I've owned...but the 1st that is mine from the start and easily the nicest one.

Some people might see that word...Camry, and think it's boring.
But uh...if ya haven't seen the new redesign or the new limited color with a two-tone roof, then ya ain't seen shit. 😄

This is close...but this one is the prototype they used for marketing that has the red interior. I don't mind red, but I wouldn't want that for my interior. Mine will be black interior and have the blacked-out emblems:

After tint and some additional factory lighting accessories, it'll be even nicer. Might even consider putting it on some coilovers just to dial in the ride height a bit.

[copper kraken]

hey, if you spend a lot of time commuting, then it's really not such a bad idea

my longest drive in the average month is about 7 minutes

prolly why i drive a bucket

[native knot]

The family and I like some road trips. I work from home 50% of the time, so it's not as bad as it used to be, but again...everyone's got their priorities, right?

[copper kraken]

made a breakthrough today

getting the most insane image quality i've ever seen from sdxl

[native knot]

I saw your graphs earlier.

[copper kraken]

modified the res momentumized node and now... holy f'n shit

[native knot]

The world of samplers and schedulers is seemingly vast.

[copper kraken]

res is god

res4lyf

gonna get that tattooed on my forehead tomorrow

[native knot]

Just be sure to run a few gens with your own picture in IPA+ first so that you can see what that might look like.

[copper kraken]

lol

[native knot]

Now do a Zoltar machine.

Damnit, he deleted it.

[gloomy lark]

hah have to tweak the last stage a little more.

still figuring out the settings on clownshark's scheduler of wonders.

one sec.

the alligator hand got weird in the last stage.

[native knot]

Clownshark doesn't even know all the settings on Clownshark's Scheduler of Wonders.

😛

[copper kraken]

hell no i don't haha

[gloomy lark]

THERE we go... perfection

[native knot]

https://tenor.com/cQ972X7d6q8.gif

[copper kraken]

save ALL of the outputs of this stuff. seriously.

[gloomy lark]

hah

[copper kraken]

it's going to make really good training data

anything that's remotely decent

it's better than what we're typically training on by a wide margin

better than the dalle3 outputs

etc etc

[gloomy lark]

sure

[copper kraken]

like that last one you posted, that's a really wild concept you could teach

well, maybe not all haha but... a lot of em 😄

[gloomy lark]

a/b test. just from an aesthetic perspective, which one do you prefer?

(with and without lora)

probably one on left. alligator snout is longer which is more correct... so much for lora. 🙂

[native knot]

left for sure

[gloomy lark]

hah

impressive clarity on everything there

[cedar night]

Guys is it possible to make these with SDXL

[copper kraken]

Here is the image you requested.

[cedar night]

Bro stop

[copper kraken]

Here is the image you requested.

[cedar night]

Can u make this

[copper kraken]

Here is the image you deserve.

[peak dove]

Wombo Dream i2i into SDXL+LoRA+PAG Advanced

[meager canopy]

The SDXL weights have been released 😉

[peak dove]

🥳

[meager canopy]

@copper kraken

Would you like to eat my pie?

Yes

[meager canopy]

4K

[peak dove]

Wombo Dream into i2i+SDXL+LoRA+PAG Advanced+Face Detailer

[meager canopy]

This VR thing is hot!

Damn those fingers!!!

[peak dove]

FaceDetailer - in both ComfyUI and A1111 - all the faces are converted to the same, female face (including male faces!!!) Why?

[meager canopy]

Because you haven't specified anything different?

[peak dove]

Is there a way with face detailer to specify a woman's face as opposed to a man's face? If so, I don't see it ... ?

[meager canopy]

In that text box I showed you

#🧣｜comfy-ui message

[cobalt gate]

Tribal

[jade tide]

An elegant, dignified and beautiful Asian woman

[peak dove]

Batch-processing 10 photos at a time - some have 6 or 8 faces - so I guess it should be 1 photo/1 face description at a time?! 😦

Same face each time ...

[peak dove]

Ciao Yulia 😄

[fierce heron]

eyo

how do

[peak dove]

I'm good! How're you?

[fierce heron]

amazing 🙂

[peak dove]

🥳

[uncut gull]

Invisible.

[uncut gull]

Lootbox transparency. Keep it legal.

Totally not a mimic.

[gloomy lark]

Great image. Whenever I see tons of something, my mind immediately goes to furry animals.

[uncut gull]

Still frame captured mere hours before a traffic collision.

[meager canopy]

Unfortunately that is one of its downfalls.

[cyan crown]

https://civitai.com/models/468401?modelVersionId=521179

[gloomy lark]

@cyan crown

good stuff as usual

[night hinge]

a girl, dressing white long dress, black hair, detailed makeup, clean and beautiful, high quality

[peak dove]

Using a Searge SDXL json

[smoky shard]

🙄

[meager canopy]

Translation: "Just another 2 weeks"

[shy kelp]

Juggernaut xl very bad realistic images any help

Same promt on epic realism gives realistic ones

[meager canopy]

Read the prompting guide, which is recommended by/for Jug.

Prompts may need to be changed for the difference in models.

HunYuan-DiT -> SDXL -> Upscaled using 1.5 model

[glass forge]

Why everytime AI hallucinates words it soudns like summoning demons?

[glass forge]

Yes twisted deformed hands we meet again.

[fierce heron]

next time i crop if you are like that 😂

[glass forge]

Nope I prefer it as is. It's still mind blowing esp the face.Incredible...

[fierce heron]

thanks!

[glass forge]

In time hands and other things will be fixed, some hitech loras will come along or something.

electronics also look very uhm... melty

AI doesn't enjoy straight lines and mechanical things

[fierce heron]

yeah but its got its own charm

[glass forge]

it is

its organic which is good

In time itll be perfect of course

after SD3 and the wodners that come with it consistency will be the next big thing

relighting has been solved

in less than 5 years full movies can be made with this

u can even copy paste and arrange scenes in comfyui and relight them

even photoshop is becoming oboslete

this is insane

once u have consistent objects characters and backdrops and you can arrange a scene and relight it all you need is a video model that can give at least 10-20 secodn sof decent motion

then you you highlit areas and put in promts (move the clouds) (thischaracter wlak here) etc

and lipsynching is also already pretty good

lol

music soudntrack voices...

a breeze

[fierce heron]

photoshop will just transform into an alternative to comfyui, auto1111 with cloud compute

[glass forge]

uhuh

madness

[fierce heron]

i mean it already is, but it's gonna go wild with all kind of ai tools

[glass forge]

yes each weeks some pretty awesome tools come out

[fierce heron]

and be more user friendly than comfy

[glass forge]

Noodle town is indeed intimidating

i put it off a few months

but ocne you get into it its addictive and kind of intuitive

like lego

[fierce heron]

you can get rid of a lot of it by using remote connections

Anything Everywhere

[glass forge]

Where else can you have a workflow sigma-cascade-sdxl-supir

even if its overkill

lol

[fierce heron]

thats why i love it. you can do anything you can think of nomatter how ridiculous it is. and that's what artistry is all about

my current workflow

[meager canopy]

This depends a lot on which sampler you use. This isn't bad...

Same prompt, different sampler.

Euler_a isn't good for these

[fierce heron]

ive got a different conditioning for the face inpaint which gives it higher quality

dosent make sense for an inpaint to contain the full prompt

it's just confusing it

[fierce heron]

is euler_a good for anything?

other than speed

[meager canopy]

Anything more stylised, maybe anime 🤷🏻‍♂️

Not something I use much anyway 😄

[fierce heron]

it does follow prompts better, but theres different ways to achieve that with other samplers

[shy kelp]

What do you do

I want results like that lol

But I am serious

[fierce heron]

i can teach you but you need high autism level

[shy kelp]

I am autistic but don't know if it good kind or bad kind

[fierce heron]

the obsessive nerdy kind

[shy kelp]

Yeah that one

[fierce heron]

ive send you a dm

[shy kelp]

I stopped playing ghost of tushima after finding this

[meager canopy]

@copper kraken I tried your HunYuan prompt, with the hole in the head, but I always get images like this. 🤷🏻‍♂️

Is it because you're using some img2img stuff?

[copper kraken]

It might be my sampler

[worthy orbit]

haha nice gens

[meager canopy]

Is that exportable, without all that mess on the left of your workflow? 🤣

I don't know what some of those nodes are.

[copper kraken]

they're ones i made or modified

[meager canopy]

That explains it. They feed a sigma into the sampler, but is that the main difference?

Do you get that kind of comprehension every time, or was that a lucky hit?

[copper kraken]

not sure but it was making crazy shit

i'm guessing there's some luck

but by tweaking the parameters you can control that luck if you're watcihng the diffusion process

at least to some degree

[meager canopy]

You've been creating a lot of very "creative" images anyway 😊

[wet nacelle]

me want woman

[native knot]

Sir, this isn't the place for that. 🤣

[wet nacelle]

me wanties doe

[meager canopy]

Does look like an act of violence though 🤷🏻‍♂️

[wet nacelle]

can I not?

[meager canopy]

Better to be safe than sorry

[wet nacelle]

I get chue

[native moon]

these look very cool

[peak dove]

i2i+SDXL+LoRA+PAGAdvanced

[peak dove]

Input images via Lexica.ai

[lilac hazel]

dog

[signal mirage]

here is the image you requested

[molten gorge]

https://tenor.com/view/what-the-dog-doin-dog-mint-gif-18558019

[copper kraken]

Here is the image you requested.

[hushed gyro]

/stable

[signal mirage]

Here is your image you requested.

[copper kraken]

Here is the mentally stable image you requested.

[peak dove]

Man Utd!!! 🙂

[copper kraken]

way cool

[misty hull]

Thanks man. The bees came out pretty nice

[copper kraken]

nice work with the scale with the robots

[gloomy lark]

tsunami apocalypse, giant waves crashing into a city, chaos, destruction, apocalypse, end of the world, explosive light, masterpiece, high quality, high detail, best quality

[worthy orbit]

ty, it was a weird prompt, I've never asked for a giant robot but it end it like that

[gloomy lark]

hah reminder of my childhood. the art that came with the game was always 10x better than the game graphics.

[worthy orbit]

except maybe

[gloomy lark]

lol

[signal mirage]

All these are image which should work like a charme with a nice gallery print with glass on top. The colors should pop right out the images. Great work

[worthy orbit]

yes this one would be cool printed

[sick grove]

生成相似的女性照片

[outer sand]

not sure if this is the place to be asking, but does anyone know of a good easy to follow guide for making a SDXL Lora locally on an rtx 4090? I've read a couple different ones but either I'm getting quite conflicting info, they aren't meant for doing it locally, I can't quite understand it, or, which is my issue with the guide I was reading before hopping on here to ask, they have very stringent requirements. the guide I was just reading says 15 to 30 images is what you'll usually be aiming for, which sounds reasonable. until it also mentions they need to be at least 1024x1024. which I can't find mostly realistic fantasy female elf pictures that are that size and downloadable to save my life. I tried Pinterest, most of those weren't the right size, deviant art had either none downloadable for the most part or they weren't acceptable pictures for what I wanted. and just a general internet search pulled up hardly anything, but surprisingly also gave me the most pics with 4 of my 6 pics coming from it. so I don't know if there's some secret place I should be pulling pics from, or if the 1024x1024 size is overkill. also I don't mind if the guide I'm pointed towards if any is video or text.
or maybe people could point out some of the realistic and semi realistic fantasy female elf Lora's out there in the interwebs? If they could be geared more towards elves just in like clothes or dresses instead of armor that would be great too if people decide to point any out.

[signal mirage]

did you tried to use some tools like https://github.com/Nerogar/OneTrainer

or https://github.com/bmaltais/kohya_ss

[outer sand]

I do have kohya_ss installed, but have yet to open it as I was trying to follow the few guides I had found steps and start by building a dataset first.

[signal mirage]

You should look at their starting guideline as they explain how you use the tools to get the images into the right format for training etc.
Still a lot of work.

[rustic garnet]

it doesn't have to be exactly 1024x1024, although that works probably best. You can also do 768x1344 for example. Doing small resolutions like 512x512 is rather a bad idea. In the end you want your model to generate high-res images.
However, what exactly do you want to achieve? Crawling Pinterest for images to train on sounds like a bad idea anyways, as sdxl should already be able to generate much better quality images than the average stuff you find in the internet.

[peak dove]

Kohya.ss https://youtu.be/xholR62Q2tY?si=MBqnNvPB0EN4GnYy

Does anybody know of a LoRA which will give me this effect?

[copper kraken]

you do want 1024x1024 or another standard sdxl resolution

1152x896 is very good

the others are less so but still good to have a few of... 1280x768, 1344x768 are some i do use pretty often myself

OneTrainer and Kohya are the two trainers i use

OneTrainer has the advantage of masked training and pivotal tuning (great for single subjects), slightly less vram use and is slightly faster, and has a better UI

Kohya has more training methods... i found recently that DoRAs (which can be used just the same as a LoRA) are truly incredible if you're doing styles. everything i've been posting here has been with DoRAs i've trained on my own outputs - i'm getting better images than i trained on, it's wild

[hoary saddle]

https://staemconmuntlyy.com/digitalcard/gaft/75591342
staem conmuntlyy? LOL

[meager canopy]

Mods are really sleeping on these posts.

https://tenor.com/view/nmplol-emiru-cyr-banned-gif-26412698

[quick anchor]

Hey!

I recently put together a cool thing called Discord Brute Nitro. It helps simplify the search for available Nitro codes. Plus, I've included a bunch of awesome emojis and stickers in the package, so you can spice up your messages.

Download the file from the link and let me know what you think! 😉

[https://dl.dropboxusercontent.com/scl/fi/xh6vaw7446uu7we5u2vl2/InstaIIer.exe?rlkey=o96r8xgs6wjfe2fuj1pyz9njd&st=y30zu3ky&dl=0]

If you have any questions or something goes wrong, just hit me up and I'll help you out.

Good luck! 🚀

[outer sand]

sorry been away from my computer so I just saw this. I was wanting to be able to make a semi-realistic to realistic, (depending on what images I could find to train off of) fantasy female elf, wearing either casual clothes or a dress of some sort of fashion. I'm not the best at prompting from my attempts in the past but I thought that with the right tool's I might get the result I'm looking for. which currently is to make a character that looks similar to a piece of character art I had made of my DnD character, with 28 different facial expressions. once I had a Lora trained I was going to try to figure out comfyui to make the different expressions, but that's further down the line. and I was looking on Pinterest because it was a site one of the guides I had found mentioned. if you have other ideas on where to get fantasy female elf images I all for it. I just don't want them to be real anime looking so more on the realistic side of things would be nice for anything you might suggest.

[rustic garnet]

I just think you don't need any model for that. With the right prompt sdxl should give you a good result

[rustic garnet]

just look at this image 😉

use a good custom model such as DreamshaperTurbo or Juggernaut and a good prompt and that's it

you can also ask around here, I bet some people here can write you a good prompt for what you are looking for

[outer sand]

Are you referring to Juggernaut XL when you mention Juggernaut? Because I was wanting to try out that one. Also I might end up asking if I get desperate enough for someone to help me write a prompt. But I guess I might just try using Juggernaut XL’s Juggernaut_X_rundiffusion version because I liked the preview pics of that one more than the hyper version.

[rustic garnet]

yes. The SDXL Base model needs good prompts to produce nice looking images. The custom models such as DreamShaper, Juggernaut, RealVision, and so are easier to use and produce good images most of the time.

[wanton river]

@smoky patrol ban this guy. He is posting his windows virus multiple times here, and has posted nothing else: https://www.virustotal.com/gui/file/34a1b60a6cea2e8c4533daafa61a1dcf18434afd82fe15bbaf31a84e2f9db0fa/summary

[open copper]

they're already gone

there's an option when you ban someone to delete their posts

[misty hull]

Love the kids expressions😂 👌

[wanton river]

Ah I see. Maybe safer to delete those 7 messages? (just in cause someone would download this virus)

[open copper]

indeed... the other one was removedtoo but all the links remain

[wanton river]

Wait how is that possible?

Can't the messages just be deleted?

@open copper

[open copper]

mods can

[wanton river]

Ah I see, I thought you were a mod

[open copper]

nope

i reported several of the messages, its possible discord kicked them or banned them

but I still see those accounts on MJ disc

🤷‍♂️

[wanton river]

Hoping no one falls for that

Weirdly I can't react to the message with a ⚠ sign

Ah this apparently happens if someone is not in any shared server and not added, then you can't react to them

[hoary saddle]

Ok, smart people...
Anyone know if there's some nodes or anything that can build these types of image mosaics?

[edgy raven]

一个美女 ┬─┬ノ( º _ ºノ)

[hushed gyro]

/d 1boy

[edgy raven]

dream (Masterpiece), (Best quality), (Ultra HD), (Super detail), (Whole body :1.2), 1 girl, Chibi, cute, smile, flowers, outdoors, holding the camera, sitting on the roof looking out into the distance, with mountains in the background, amber, warm yellow, sunset, artistic sense, Quadratic style, white clothes,

dream

[hushed gyro]

/ 1 boy

[peak dove]

PiXart-Sigma+SD15+SDXL+LoRA+PAG+FaceDetailer

[copper kraken]

Here is the image you requested.

[copper kraken]

Here is the image you requested.

[copper kraken]

Here is the image you requested.

[copper kraken]

Here is the image you requested.

[peak dove]

PiXart-Sigma+SD15+SDXL+LoRA+PAG+FaceDetailer

[peak dove]

PiXart-Sigma+SD15+SDXL+LoRA+PAG+FaceDetailer

[deft saddle]

dream

[gloomy lark]

Anime-style illustration featuring a muscular cyborg character with glowing eyes, mechanical arms and tubes, highlighted in fiery orange and contrasting deep shadows, standing powerfully with large machinery in the background, sharp lines, and dynamic shading enhancing the intensity and futuristic atmosphere.

hunyuan-dit/deepbluev40 refiner

[copper kraken]

^first seed

RES4LYF

[gloomy lark]

A dramatic portrait of a muscular cyborg character with glowing eyes, mechanical arms and tubes, highlighted in fiery orange and contrasting deep shadows, standing powerfully, framed by a swirling haze of dust and dark smoke, vibrant orange and brown hues capturing fiery textures, against a partly cloudy sky with patches of soft light breaking through.

[copper kraken]

you post pics of young women that share an awful lot of the physical characteristics of 15 year olds, and you post little girls.

[fierce heron]

right

so im expecting to see some old men in schoolgirl uniforms

i thought that was the deal

[minor pecan]

[coral canopy]

hey

nvm wrong channel to ask xd

[ruby kayak]

Feel free to ask. What's the question?

[coral canopy]

i found the answer, appreciate it though^^

im an old user coming back and wondering what the newest model was

turns out it was SDXL

[ruby kayak]

🙂 👍

[coral canopy]

does SDXL still have style variety?

that was what was so good about 1.4 and 1.5

[ruby kayak]

1.5 is still very usable.