🏞｜general-with-images

[teal zephyr]

That said, the result would probably look tacked on.

[glossy herald]

first try

[teal zephyr]

: O

[glossy herald]

not 100% the same robot

but epic

[teal zephyr]

Wow definitely cool!

[teal zephyr]

I love the heat sinks and energy details it added... man I wish I could keep the original photo but with those enhancements and an AI generated BG. Seems like I'll have to do the overall result I'm looking for manually.

[glossy herald]

I'm trying other things. I got further away but wow

[teal zephyr]

hot damn thats awesome

[glossy herald]

this works great with children's drawings too

they are amazed by the results

staying closer from the source picture

going into other styles

[strange nova]

Make it a Pakistan cricket team fan realistic cartoon

[frank egret]

https://tenor.com/bDEmn.gif

[eager dragon]

guys ive just started trying to learn how to use image generation and ive run into a problem where it says in queue 1/1 but wont generate. any ideas why or help?

[jolly scaffold]

trying out revision

hmm

[kind quartz]

Those red things on end of prompts are saying i think it is not ready. Can you post what is in cmd window? @eager dragon

[south lion]

Make this model a building photografy

[frank swallow]

https://tenor.com/view/i'm-sorry-sad-cat-love-cat-cats-cat-gif-3997044575451409500

[silk tendon]

I'm now

A pony enjoyer

[wispy nest]

Go to pony jail

[silk tendon]

Can't stop the pony poster

[wispy nest]

I battle your pony

With cat

[silk tendon]

So you want to go that route

[jolly scaffold]

ya know... I was just thinking... someone above posted a few blocks, and you could use that with something like controlnet to texture something of that shape. It's be a flat texture though, but, it seems like that could be used to make background images for a Resident Evil style game with pre-rendered backgrounds

I think I might do that this weekend, see if I could slap together the basic layout of a room and let SD do the rest

[wispy nest]

Illuminate thy soul

[wispy nest]

I declare pony a crime

[silk tendon]

I declare you dead.

[silk tendon]

Won't be of much help

[wispy nest]

Okay then

I activate cheats

Fluffy

[silk tendon]

So you want to play the gods game

Are you sure about that

[wispy nest]

The hell

Lmao

[silk tendon]

Pony ate the cat

[wispy nest]

I'm laughing

Lmao

[silk tendon]

Gonna beat you up

[wispy nest]

Peace time

[silk tendon]

I'm not even a pony

Total pony death

[cinder trellis]

Is this sdxl

How your ponies look so detailed and realisitic

😡

Teach

[silk tendon]

I'll teach

At one expense

Send me your collection

[cinder trellis]

uhhhh

It's all nsfw

[silk tendon]

Did I stutter?

[cinder trellis]

I CANT JUST SEND IT HERE RIGHT

[cinder trellis]

Here is a

NonNsfw one

[silk tendon]

I said send it not send it here

[cinder trellis]

nah bro what if u reported my account and it got terminated 😡 😡

[silk tendon]

And, that's not a pony

[cinder trellis]

That's a

That's a cat

Do u not like cat

[silk tendon]

Whatever

No pony secrets for you

[cinder trellis]

@silk tendon WILL THIS WORK

👉 👈

[cinder trellis]

Give me them pony secrets😡 😡

[silk tendon]

MODS HE'S POSTING HARMFUL CONTENT

[cinder trellis]

WHAT WHAT

NOOO

It's just a baby fighting a

a dog/

🥺 🥺

[silk tendon]

Explain that to the police

[cinder trellis]

Officer you gotta believe me that baby is safe

[glossy herald]

he got a sword, fair fight

[cinder trellis]

Yes bro

Fair and square

What can go wrong

[glossy herald]

don't go into baby violence for real for sure, but this seems not harmful lol

[cinder trellis]

Officer

I would never do something so sinfull like that

Please don't fine me.

[glossy herald]

that's not reassuring the least but nice x)

[cinder trellis]

phew

[glossy herald]

you trained a poney model by the way ? or is that just a base model

[glossy herald]

time

[cinder trellis]

What does training a model mean
Is it like going through various controllnets with different modes?

[glossy herald]

training a model is completly different from making pictures. it's using other tools, in order to make a new .ckpt or new .safetensors for example. You can train on new things, like your own face or your dog, to be able to prompt for those for example
when a model does really good for a specific subject, like the ponies you did, it usually comes from someone training a model specifically for ponies

[cinder trellis]

ow

[glossy herald]

like, I trained lots of models on stupid things, or specific subject

[cinder trellis]

gotta research bout that

[glossy herald]

it requires more VRAM usually, just so you know. but depending on your setup, you can find different tools fited for your needs

check my guide if you want, some months old, but still holds up on the theory about all this

[cinder trellis]

don't worry Google Collab Cloud Vram can handle anything

Trust

[glossy herald]

just know, this is not easy

https://github.com/Guizmus/sd-training-intro

[cinder trellis]

O

I will sure check it out after I'm done with my maths homework

[glossy herald]

have fun 🙂

[cinder trellis]

One more question
What is Revision

Is it like the same thing as Sd

[glossy herald]

no idea what you're talking about there

[cinder trellis]

oh no I think i sounded dumb

Nvm nvm

[glossy herald]

a link to Revision maybe

cause google doesn't land any AI related results to me

[eager dragon]

its not happening anymore so far

[silk tendon]

Look ma'am, I'm just a simple dall e enjoyer is all, no training

Stable diffusion is beyond me, I don't own 2 4090s

[earnest solar]

I'm doing this on a 2060 filled with porn, dust, and viruses.

[silk tendon]

He just took that image from 4chan most likely, they have posted dozens of these, the idea of it is to push for hate against pitbulls as they harm children unpredictably.

[silk tendon]

won't fall for that

[carmine blade]

I love my gpu.

[serene flint]

I built an entire career on a 2060 6GB

[jolly scaffold]

I'm quite pleased with my new GPU, having the power to put pretty much anything I want into a prompt is cool

lmao, good chance I'm going to lock it up now though pushing my limits

100 steps of 3440x1440 SDXL just to see what happens

[carmine blade]

my 2080s is thermothrottling from 1900MHz to 700MHz, but hey it works

[jolly scaffold]

I'm using an Intel Arc for this

[barren jackal]

How do I generate ts?

[jolly scaffold]

ts?

tiny squares?

[weak cipher]

ai progeammer

[jolly scaffold]

What do you think?

[jolly scaffold]

you have a model that makes things into Lego pieces?

[rain gazelle]

A couple from dall-e 🙂

[rain gazelle]

naw just dall-e 3

[jolly scaffold]

it'll be nice when there are open source tools that work as well

[serene flint]

I managed to render a 1200 frame animation with SDXL tonight

and nothing even melted or exploded or caught fire, or disintegrated at all

[jolly scaffold]

you using Comfy or Automatic?

[cinder trellis]

Do u use vid2vid

Or manually press generate 1200 times

with each video frame

[serene flint]

me? I dont know of any way to animate with SDXL on comfy, im using a11

[serene flint]

3D animation

#🎥｜animation message

i just uploaded it here

[jolly scaffold]

You can do anything on comfy that you can do on other stuff, but for advanced stuff you build your own workflow for. For example, I pieced together a system to automatically isolate foreground characters and generate new background using a new positive prompt

[serene flint]

Not Deforum though

[cinder trellis]

Sheesh bro it looks good

but how

[serene flint]

SDXL

[silk tendon]

I'll NTR your gpu

[silk tendon]

Omg just like no man's sky

[silk tendon]

The gradient is so smooth, no artifacts

Impressive

[wispy nest]

filósofo chino, Wang Chong, carboncillo estilo zen

[modest marsh]

how can i get this exact painting but in another style https://upload.wikimedia.org/wikipedia/commons/thumb/3/3d/Fallen_Angel_(Alexandre_Cabanel)_crop.jpg/450px-Fallen_Angel_(Alexandre_Cabanel)_crop.jpg

Fallen Angel by Alexandre Cabanel

[raven mirage]

https://stable-diffusion-art.com/stylize-images/ Here’s a good start for you.

[raven mirage]

https://youtu.be/1-26FDZM-Zg?si=WWT7A8fg9SFmbW6_ Closest result I think

[stoic pivot]

Random watermelon cat.

[cinder trellis]

Why the watermelon look so scary

[glossy herald]

High watermelon five

[wispy nest]

https://tenor.com/view/cat-watermelon-watermelon-cat-cat-stunned-heonism-gif-23758788

[cinder trellis]

idk if u guys get it or not

[stoic pivot]

My friend drew the first image and the second image is when it was used as a guide for the AI, the third one is me adding some prompts and tweaking the settings.

[cinder trellis]

Make it more spooky

Not cute

[stoic pivot]

Trying but failing

I tried too hard, so I call it a day now.

I lied for I sent wrong image.

[soft orchid]

#🏁｜🏞️

#🏞️ | logo game for YouTube channel with name mad game at

#🏞️ | logo game for YouTube channel with name mad gamebaz

[jolly scaffold]

Heh, stuck the main entrance of the RE1 mansion into controlnet to see how accurately SD could rebuild the interior of the mansion, given the backgrounds in that game are made up entirely out of 2D textures...

[jolly scaffold]

getting closer

[lapis quarry]

Is that supposed to be that youtuber who climbed the bridge live on twitch or something?

[jolly scaffold]

I'm kind of curious if anybody can get closer to a 1280x960 rendering of the inside of the RE1 mansion

[cinder trellis]

It's PewDiePie's bridge incident

[raven mirage]

something like that….

[cyan snow]

Captain Sponge

[gaunt heron]

With real pieces of sponge?

[hidden mortar]

prompt?

hey guys, I'm sure you recognize those funny memes - movie posters from disney. Are they SD? How to make one? I saw a lot of them and they are extremely similar. What is the model, prompt, settings etc.

[blazing gate]

Viking woman intricate detailed (skeleton face) plus many other added details...

[lapis quarry]

A classic. Me and my ma use to watch this every Saturday night.

[prisma stone]

How to stop making a head out of the frame?

It's impossible.

[wispy nest]

Do 512x768

Taller images

[tall shadow]

@slate bridge

[slate bridge]

Yes? You were saying...?

[wise wraith]

Can Anyone make anything similar to this? Worth a try !?

[slate bridge]

Looks like Midjourney. Similar to any of these.

[slate bridge]

https://replicate.com/methexis-inc/img2prompt for image to prompt utility. There are others on HuggingFace but they are heavily loaded with wait times.

[silk tendon]

Someone knows this guy?

[sleek flax]

a book cover for "O Labirinto de Cristal", featuring 4 children (2 boys and 2 girls) in a colorful classroom filled with educational elements. There is a blonde psychologist woman participating in the adventure. The cover is whimsical and captivating, with a touch of mystery, in a modern and imaginative art style, UHD.

A captivating book cover for "O Labirinto de Cristal" featuring a colorful classroom filled with educational elements. The cover depicts four children, two boys and two girls, embarking on an adventure. The classroom is vibrant and whimsical, with books, globes, and other educational objects scattered around. There is a blonde psychologist woman who plays a significant role in the story, adding a touch of mystery and intrigue. The art style is modern and imaginative, with vibrant colors and intricate details. The cover design is in ultra-high definition, showcasing the stunning visuals and inviting readers into a world of wonder and excitement.

【反向提示】Blurry, low-resolution image with pixelation. The classroom appears dull and uninteresting, lacking any educational elements. The children are poorly represented, with awkward poses and unappealing features. The blonde psychologist woman is barely noticeable and does not contribute to the overall composition. The art style is uninspired and lacks creativity. The cover design is mundane and fails to capture the essence of mystery and whimsy.

[marble flame]

画一个身穿瑜伽裤的美女

[crude schooner]

Why there image been blurred? Is there any censorship on the image produced?

[silk tendon]

Interesting

Wonder what they're using for this

[crude schooner]

Ya...a few of my image generated is blurred...the prompt is normal, the result pretty much a SFW...but looking at the shape of the body maybe either way... I don't know. The developer may know

[silk tendon]

Perhaps now female bad

[scenic umbra]

the blur filter works on images it considers nsfw, that one looks like it's the breasts, are you prompting for it?

[dreamy bough]

Keep having black images I installed rev animated. The problem is with 3D models. It creates a video but after first image other ones are only black images. Any suggestion how to solve this problem ?

[wispy nest]

I'm trying to nail the concept of a black spell that destroys all creatures of a chosen color

https://th.bing.com/th/id/OIG.CRhooHUB55SbvpkxQ7T1?pid=ImgGn bing made this

[crude schooner]

Nope...just normal prompt

[wispy nest]

Hello, I'm new to Discord.

I really liked this image generated on Bing.

Furthermore, I want to generate it, maintaining the shape and clothing.

[wild sorrel]

This might be a way to start
https://stable-diffusion-art.com/controlnet/

or just search for "controlnet" guides anywhere you want, there are plenty of articles and videos

[wispy nest]

Women in war

[cinder trellis]

will this work

if i manually draw openpose

[wispy nest]

https://th.bing.com/th/id/OIG.ymv4AS9TaDNYhqU9i1jO?pid=ImgGn

bing is able to understand what I want

https://cdn.discordapp.com/attachments/1148388684836638820/1166757545080270918/a_plains_with_higher_ground_closer_to_the_viewer_where_multiple_characters_despair_down_to_their_knees_and_screaming_in_pain__Full_moon_on_the_sky_wit_style-Fantasy_Art_seed-0ts-1698247048_idx-0.png?ex=654ba6a4&is=653931a4&hm=dd2cea6c8210c9e0e6692cd552d52bfec6095db134ee110007784c7f6432de4c&

sdxl fails to understand what is a skull in front of the moon. "skull" alone, not a whole character.... damn

[wispy nest]

Bing is cheating

:P

[golden tree]

cozy

[wispy nest]

it seems that if I want the shadow of star on the ground, the AI inferes that I want a star on the sky or a dark star on the sky

I write "shadow of a skull on the ground" and I get a skull on the sky.

[pine briar]

Bath time

[small slate]

as far as I know, it shouldn't be 70% for this purpose, there's too much water? instead, it should be more like 99%?

[wispy nest]

200%

Thing evaporates before touching the card

[heady anchor]

u can even use water as long as you dry it properly

https://www.youtube.com/watch?v=_ruNPYSIKtY

[pine briar]

As long as you dry it sufficiently afterwards it doesn’t matter at all

[quiet moat]

Does anyone else have their webui configured to save images under /modelname/00001-prompt.png? How do you browse your previous stuff when looking for a mood to regen?

I can't find an image tool to conveniently browse my models and their outputs. I end up browsing in File Explorer, then switching to webui and pasting the path of a file I like. Not very nice. Can anyone recommend something? I tried:

1. Infinite Image Browsing (extension): kinda ghetto. You can configure it to look in subdirectories, but often the directory doesn't refresh even if you click Refresh. Gotta restart to see new gens. Also the lack of polish worries me, it has some ghetto because it wants to access the whole filesystem.
2. Image Browser (extension): badly developed when it comes to using subdirectories. Every action seems to scan all your subdirs, taking several seconds. No caching at all.
3. Breadboard (desktop app): just ghetto. Doesn't support subdirectories.

The requirements I'm looking for is:
-be able to see most recently-generated images (even if they're in different subdirs)
-assign a score to images, find high-score images
-browse a specific model (subdirectory)
-nice to have: a "random model" browser

[wild sorrel]

Does anyone else have their webui configured to save images under /modelname/00001-prompt.png? How do you browse your previous stuff when looking for a mood to regen?
I'm just saving images, seperating them in folders based on different themes...
About copy-pasting - not really needed.

You just need to keep 1 image with prompt you want to keep and drop it into png info. From there you can send prompt and settings into txt2img or whatever you need.

Don't remember if it swaps model from there, that might be one of the things you'll need to change. Seed is another one, set it back to random

[quiet moat]

so you're doing what I'm doing, manual image management from File Explorer, + PNG Info

I prefer to save them based on model (instead of theme) to be able to quickly inspect a model's capabilities.

[kind quartz]

i think you can in settings do that. What format of save you want. But i can be confused using more programs for image generations, but A1111 should be able save based on models. let me check

[wild sorrel]

makes sense if you swap models often

yea there was a naming scheme

[kind quartz]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Custom-Images-Filename-Name-and-Subdirectory
Check settings and this page

[quiet moat]

that's what I already do. It's just 2 chatlines above.

it's just that the existing image browsing extensions suck when you use that option

[kind quartz]

So make it more detailed, not just model

add date

[quiet moat]

you misunderstand what I want. Do you use any image browsing extensions?

do you see the advantage they bring over browsing from the filesystem only?

[quiet moat]

of those 4 requirements, #1 cannot be done by filesystem browsing, or #2. It does #3 great.

[kind quartz]

you are lucky, i keep some images in special format and having not any preview. But one has to cattegorize images anyway. I think using full capabilities of Xnview could help you

[wild sorrel]

that's not bad idea, I haven't seen anything like that tho

that's alot of work

[quiet moat]

OK, I'll download xnview. Never tried it before.

[kind quartz]

xnview is great, you can rate and many more. It supports wide range of formats, batch converting included subdirectories and many more. But it cache things i think probably can be switched off/on dont know.

[quiet moat]

cache is needed because I have 16k images in 100 models so if it has to "scan" everytime like that image browser extension, it's DOA

[kind quartz]

Let us know if any problem with it. @quiet moat

[kind quartz]

There is in settings automatically select new files, may be difficult to find.

[quiet moat]

@kind quartz UI is kind of rough. For example if I assign a 5-star rating to an image, when I go to my txt2img root, I can't filter by 5-star images. It's only for the immediate folder. I need to go within one of my 100 model subfolders for ratings to work.

[kind quartz]

yes it looks outdated but it is realy good sw. There must be way. search in subfolders try

i got it in my native language so it is for you pointless.

[kind quartz]

Search all - including subfolders - and then sort by rating?

[steel abyss]

Depth render is all you need 😁

Mixing SDFs, fractals, AI, all powered by nodes... 🙂

[slate bridge]

you might try infranview. https://www.irfanview.com/ It supports thumbnails and subfolders, but you have to press F9 ? to refresh and they don't seem to be cached, although they do refresh in the background. Open an PNG file and press T for thumbnail page. I also use TweakPng to view tEXt block of PNGs http://entropymine.com/jason/tweakpng if SD isn't running. Also, I asked ChatGPT to write a python script to extract all of the tEXt blocks from my PNG files and save them as png_filename.TXT and also had it rename all of my files based on comments in my prompts. If you can explain what you want, ChatGPT can write it for you. ( It could fix all your old filenames based on a new pattern you prefer going forward, like add the model name so you could reorganize your folder names into ratings instead of models for example)

[tropic remnant]

8X，3D

[wispy nest]

i love this artstyle

[wild sorrel]

interesting how just a few words can often change style from somewhat realistic

to this lol

not even related to style words, just few random words have different effects on image, not even sure which words yet

[wild sorrel]

nvm I'm dumb, wrong model 🫠

[tight whale]

8k, 3d

/create Rainy night

/create

/create

[kind quartz]

@brave wraith first two images, removed beautiful, and empty negative prompt.

[fleet goblet]

I would say this fits here very well.

https://www.youtube.com/watch?v=7jgVgCI-R8k

The image is DALL-E 3 generated, music by me.

[obtuse crystal]

Why did the text to image API only call three times and prompt 429 Too Many Requests?

Is it a restriction on newly registered users?

[obtuse crystal]

I understand that the balance is insufficient.

May I ask if the purchased credits have a usage time limit？

[river isle]

These are not SD but I had fun making them

Dall-E 3 + Pika Labs (LeiaPix for the toy one)

[silk tendon]

I'd like to file a complaint
@scenic umbra

[glossy herald]

I reported it to mod channels, we got some of those currently, I got one too earlier

[silk tendon]

honestly discord should be the one to do the moderation work, you're doing it for discord, for free

[gritty vigil]

bit of a action scene

[lapis quarry]

What’s the difference between the first two?

Is the inpainting one only for inpainting? Or can it generate the same stuff as the regular one?

[tulip yew]

Why do I get an error when using the reactor extension?

[past pelican]

Inpainting models work better with keeping images coherent, but it's not as good at creating good one shot images.

[lusty otter]

Is there any easy place to find white outline & black background images for use in ControlNet for the Canny option? I understand I can make my own but was wondering if someone has some I can take to mess around with easily?

[wild sorrel]

basically you want to use inpainting model for inpainting and outpainting

[sterile kiln]

You should try Multidiffusion

An a1111 extension

[latent sleet]

Trying to get game/anime characters to real life version. Can't figure how to fix hands yet 😦

[tulip yew]

If I set a 4GB RTX 3050, will it pull the SDXL in comfyUI?

[misty bane]

no shot theyre selling 30 series with 4GB VRAM

ight ight, this one's 🔥

[spark orchid]

no shot theyre selling 30 series with 4GB VRAM

[lusty otter]

That 3.3GHz is for the speed of the processor. Reading that, the laptop has no dedicated graphics card which would make running Stable Diffusion hella slow.

Additional, I think at least 6 to 8 GB of Vram is recommended/ needed for SDXL.

You can try using RuinedFooocus UI which works well on lower end systems.

[wispy nest]

That's the laptop version

Even still, Nvidia is being cheap x3

[limpid lichen]

Its time for David S Pumpkins baby

[limpid lichen]

Any Questions??

[limpid lichen]

theory. papa pumpkins is the papa of David S pumpkin

the resemblance is unfathomably canny

i ressed my case

[quartz ravine]

The concept of the word "violated"... I am not clear what SDXL thinks it means

More violated

[quartz ravine]

infanincubator - just the two concepts "infan" and "incubator". Two tokens

[limpid lichen]

papa pumpkin incubating pumpkins

the breakthrough moment

[hoary summit]

Many students flocked to elect the head of the organization at school where there were 3 candidates for head of the organization in front and many students at the back were holding the election and made the atmosphere very dramatic and futuristic.

[quartz ravine]

mingoaur

seems to make various beasties

[pure flower]

Curious what people think of stable diffusion combined with data viz

[trail lantern]

why does this happen when using inpaint am i doing something wrong

[limpid lichen]

when properly used, why not. no different than any other artistic touches to data vis. long as it doesn't hallucinate details and is guided visualization, do it

[prime flame]

Been working on a new variant of controlnet if anyone's interested 🙂
(Flip the mountain range upside down)

[prime flame]

I have a gradio link I can share to let you guys use it easily - @ me or DM me!

[main kiln]

I asked for Mickey Mouse.

[wild sorrel]

It does what it's saying, "latent nothing"

[pure flower]

Could I .... Make a channel around it to promote the concept? I think it's another potential vein of creation

[limpid lichen]

I think a channel can be made around any concept if you've got good production value and heart. I watched a deep dive on murder she wrote because the content creator had heart and enthusiasm. I dont even like murder she wrote that much. it was my mom's favorite show when i was a kid

[quartz ravine]

gophon grylem

[trail lantern]

even if i change it to latent noise or fill its the same output every time

[tiny siren]

can anyone identify what model is being used to create these?

been tryna figure out the mode for a hot minute

some guidance would be appreciated

[wild sorrel]

resize+fill, fill, higher masked paddings, more steps, lower denoise

[minor wraith]

If I wanted to add additional muzzle flares, shooting fx to the guns, what’s the best way to do that? Do I photobash a shooting effect plug-in and img2img or is there an inpainting method(this is an sdxl image).
Also want to put some defenders in the castle 2nd floor

[heady anchor]

photobashin is faster,with inpaiting you rely on rng

[minor wraith]

For the extra guys, would you maybe cut and copy one of the guys in the foreground and paste him in the second floor of the castle?

I would have to turn him around somehow, could get lucky with inpainting him after I added the guys with photoshop

[pine briar]

When doing architecture, how do I avoid this sort of thing

Like the non-parallel lines and distortion

Using SDXL

[heady anchor]

just find random pictures of the soldiers online,and photobash them into the pic then img2img

[trail lantern]

Thanks for the suggestions but i think the problem is elsewhere because in the render preview is see some realy nice faces but the output looks like that as it would skip the last step and not apply the face.

[wild sorrel]

yea that's weird

[trail lantern]

maybe because im using an AMD GPU

Because as i sad in the previev it looks good but in the output on original it does that and on fill it just does nothing

Im using those commandline args maybe there is the problem? set COMMANDLINE_ARGS= --opt-sub-quad-attention --medvram

[kind quartz]

Try lower Denoise strength. also if in last step, probably vae issue? Cant be?

[trail lantern]

did sadly not work and i also disabled the vae

[kind quartz]

disabling VAE is bad idea. Do you have enabled settings, show all, QUICK SETTINGS, sd vae <- or how exactly it is called?

you may have in use bad vae, not for model are you using

[trail lantern]

could be i use chilloutmix and VAE set to autoamitc and i dindt install any VAE´s manualy

[kind quartz]

try some model having vae baked inside, there are as well inpainting models, probably you can try them.

[trail lantern]

i have teh SD1.5 inpainting model same results

[kind quartz]

just for curiosity, try this one
https://civitai.com/models/4384?modelVersionId=131004

[trail lantern]

same thing

[kind quartz]

Try ask in prompting help or tech support probably. There are skilled people can helps you probably. One thing is for sure --medvram shouldnt change quality for sure.

[trail lantern]

will do thanks for the help should i try maybe highvram but i only have 8gb so idk

[kind quartz]

no, i think amd got its special settings.

you probably should have more arguments

[trail lantern]

also i already asked for help in tech supoort no answer yet

[trail lantern]

in the prompt or the command line

[kind quartz]

i mean in command line args

[trail lantern]

i now used the argumants from the github instalation page --opt-sub-quad-attention --lowvram --disable-nan-check i will try again

[kind quartz]

--lowvram --no-half-vae --opt-sub-quad-attention --opt-split-attention-v1
found this 🙂

disable nan check is solving nothing

[trail lantern]

the weird thing is in the render preview it looks good until it finishs and than it lokks bad

[kind quartz]

yes it is usualy when VAE is missing or wrong

[trail lantern]

Will try these thank you very much

[kind quartz]

no problem

yw

[trail lantern]

i downloaded this VAE https://civitai.com/models/88390/difconsistency-raw-vae-pack maybe it will work now

used two diffrent VAE´s now no chnage preview looks fine output not so much

[kind quartz]

Vae should be for exact model. Realy weird. Have you tried lowvram?

[trail lantern]

yes tried Lowvram

could it be somting with clip skip

[trail lantern]

but for dreamshaper i dont find any specific VAE for that

[kind quartz]

if you dont change it it shouldnt. There are not vae in inpaint model but baked vae is in classical dreamshaper models. Just not in inpaint. At least not mentioned.

[trail lantern]

i changed it to two

[kind quartz]

Cant it have something with padding 64px? if whole image is 512x512 cant be 64px too much? But on other hand, it shouldn't look in preview well then ....

[tulip yew]

Hello, does anyone know why I get this error when trying to install insightface?

[trail lantern]

i just read that the padding pixel could be the issue and you have to turn it higher because default is 32 and its the same result maybe i increse the resoluten and look what happens

[trail lantern]

do you have an amd gpu maybe you installed the wrong version not suitable for AMD

[tulip yew]

How do I find the version for amd?

[trail lantern]

just google i reacon. do you wanna install SD1.5 or something else?

[tulip yew]

Everything works fine for me but I can't use reactor to faceswap

[trail lantern]

oh okay i dont know much about that sorry

[wild sorrel]

Try default.
vae-ft-mse-840000-ema-pruned.ckpt
or
vae-ft-ema-560000-ema-pruned.ckpt

If there's none with model - it's either baked in model itself or just "use default"

[wild sorrel]

64 should be fine, more doesn't make it worse anyway, it just gives more "to look at"...like more details outside to base result off

[wild sorrel]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs

[silk tendon]

I didn't even prompt for this

The "painting by a feminist" is to help ease up on filters

[silk tendon]

Oh god I butchered the spelling

Poor asuka

No wonder it freaked out

[pine briar]

Asking this again. How do I make more geometric shapes less distorted/botched?

[limpid lichen]

the fnaf movie has no right being as well made as it is

i went in expecting a horrible film i could laugh at but it was actually legitimately a good horror

[grave flax]

wheres a good place i can use img2img on? i have a potato pc so i cant run ai locally

[frail tendon]

is there a way to take an image like the first one and use img2img and get an image something like the second one. The shirt could even be in the same pose as the original image it does not have to be laying flat like in the second image. Basically take the shirt from the first image and remove the person and background. Is this even possible? I have been trying in the img2img sandbox, but have not been coming even close.

[cinder trellis]

Can anyone tell me why does T2iA do this to me

[silver latch]

I need some help with training a lora

What are some good base models?

for sd 1.5

I have a mix of 2d and 3d for a cartoonish video game charater

Currently not happy with the results

If anyone also wants to help test the lora currently just dm me

[slate bridge]

KJV, this is my trained voice.

[rain river]

Anyone know why the website is just slicing off the rest of my images on it?

It shows the whole thing before I generate the image, but it slices it in half when I go to generate?

[kind quartz]

what website? @rain river

[azure gull]

Hello everyone who can explain to me why I have this problem, I will be very grateful ❤️

[lapis quarry]

Could someone help me figure this out? I’m trying to turn this photo into an anime style photo. Basically anime-fy it.

But so far I haven’t had any luck.

[past pelican]

Which model are you using? You probably have a resolution set too high for the model.

[azure gull]

1.5

[past pelican]

That model struggles if you try to make images much larger than 512x512. Even 768x768 can cause that duplication effect. You'll want to do a high res upscale on the lower resolution image.

[azure gull]

How can I do that?

and where

[past pelican]

Well for starters, what are you using to create images?

[azure gull]

realisticvision

[past pelican]

Not the model, are you running this locally on your machine and if so what program?

[wild sorrel]

probably cause you're trying to generate too high resolution your model isn't trained on and not using hires fix

oh...didn't see it's kinda old, my bad

[past pelican]

Yeah I mentioned it. I need to know what program they are using.

[wild sorrel]

can someone share like a ...definitive, but compressed guide on comfyUI for people who kinda know things, but no idea wot those new shiny buttons are

and how you put together things

or just a basic workflow lol

[past pelican]

Actually, if you have auto1111, you can drag and drop images into comfyui and get the equivalent workflow fyi.

It helped me figure out hires fix.

[wild sorrel]

oh

[azure gull]

automatic1111

[wild sorrel]

yea that actually works it seems

[past pelican]

Just below the negative prompt window, there's a drop-down called "hires fix". Start by generating images at the proper 512x512 resolution and if you see one you like, try generating it with the default hires fix settings.

You can grab the previous seed you used for the image you like by clicking that recycle symbol next to the seed number.

[wild sorrel]

is there a way to like...separate prompt into different nodes in comfy?

Like 1 for "general" things, 1 for subject, 1 for everything around...

[past pelican]

That's one of the default template images that comes with comfy iirc.

[lapis quarry]

Could someone please help me out? Or at least tell me how I can change this photo into an anime style with prompts. (Ie: what would be good prompts, etc)

[wild sorrel]

AI had hard times figuring out where's rain and where's not, but that actually made image look interesting
(had to delete bottom, it got a bit more nude than it should've)

[wispy nest]

Why you posting that cat

Where is dark images

[past pelican]

I don't have a great answer for you. You could try controlnet but I'd try my luck in #📝｜prompting-help

Ken Sugomori is the Pokemon guy if you want to try that

[wild sorrel]

I don't understand what UI it is, too little info to figure out issue too

[past pelican]

Looks like they are just doing img2img stuff.

[wild sorrel]

The only thing that catches my eye is (:-1.5) weight on prompt , never seen that...not sure if it works

((:x)) inside double quotes might lead to weirdness too

other than that should be just fine after chosing proper anime model and playing with denoise strength

[lapis quarry]

It’s the draw thing app for IPhone. It runs stable diffusion locally

[lapis quarry]

Some Lora’s on Civitai have that sometimes

[past pelican]

I'm guessing the -1.5 thing is just negative prompting.

[wild sorrel]

yea I thought so too, but idk if it actually works

[fair void]

@wild sorrel

[wild sorrel]

fill masked content

[fair void]

creates a blurry spot on image

[fair void]

^

[wild sorrel]

huh

swap to resize and fill

[fair void]

same thing

[wild sorrel]

is original image 512x512 too?

[fair void]

yes

[wild sorrel]

try increading padding to 64 or 128 (altought 128 might be a bit too much for 512x512) and random seed

or "only masked" , but it should be fine on whole image too, I'm missing something

dumb question...is anything masked on the image? 😄

[fair void]

yes

[wild sorrel]

how big mask is compared to image?
It might be blurry cause it's too small and mask blur eats everything...

[fair void]

like a third of image

[wild sorrel]

yea okay I give up

[fair void]

i used to use easy diffusion

but i moved to automatic 1111,

easy diffusion inpainting worked, but i had to use cpu cause i have an AMD card

[wild sorrel]

either I'm missing something obvious or something is broken...

are you using inpainting model?

[fair void]

no

I get an error when pressing generate

one sec

[wild sorrel]

There's an installation for amd cards too
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs

[fair void]

I did that

[wild sorrel]

you need inpainting model to inpaint and outpaint, usually you can find one seperate model on https://civitai.com/ for most models for both gen and inpainting

[wild sorrel]

no idea what's that =\

[past pelican]

You didn't randomize the seed as well.

[wild sorrel]

maybe #🤝｜tech-support is a better help for that error

[fair void]

no differet

[wild sorrel]

keep it on fill when you'll attempt to test inpaint, original just keeps original image content

latent noise - just adds noise, nothing - fills it with color iirc

[fair void]

ok

fixed the error

but it just makes a blur again

[wild sorrel]

well...you either need inpaint version of your model or idk

[hushed flame]

great work everyone

[inner totem]

And the eror code is -->
Traceback (most recent call last):
File "G:\stable-diffusion-webui\venv\lib\site-packages\gradio\routes.py", line 488, in run_predict
output = await app.get_blocks().process_api(
File "G:\stable-diffusion-webui\venv\lib\site-packages\gradio\blocks.py", line 1431, in process_api
result = await self.call_function(
File "G:\stable-diffusion-webui\venv\lib\site-packages\gradio\blocks.py", line 1117, in call_function
prediction = await utils.async_iteration(iterator)
File "G:\stable-diffusion-webui\venv\lib\site-packages\gradio\utils.py", line 350, in async_iteration
return await iterator.anext()
File "G:\stable-diffusion-webui\venv\lib\site-packages\gradio\utils.py", line 343, in anext
return await anyio.to_thread.run_sync(
File "G:\stable-diffusion-webui\venv\lib\site-packages\anyio\to_thread.py", line 33, in run_sync
return await get_asynclib().run_sync_in_worker_thread(
File "G:\stable-diffusion-webui\venv\lib\site-packages\anyio_backends_asyncio.py", line 877, in run_sync_in_worker_thread
return await future
File "G:\stable-diffusion-webui\venv\lib\site-packages\anyio_backends_asyncio.py", line 807, in run
result = context.run(func, *args)
File "G:\stable-diffusion-webui\venv\lib\site-packages\gradio\utils.py", line 326, in run_sync_iterator_async
return next(iterator)
File "G:\stable-diffusion-webui\venv\lib\site-packages\gradio\utils.py", line 695, in gen_wrapper
yield from f(*args, **kwargs)
File "G:\stable-diffusion-webui\extensions\sd-webui-inpaint-anything\ia_threading.py", line 119, in yield_wrapper
yield from func(*args, **kwargs)
File "G:\stable-diffusion-webui\extensions\sd-webui-inpaint-anything\scripts\inpaint_anything.py", line 782, in run_webui_inpaint
for count in range(int(webui_iteration_count)):
TypeError: int() argument must be a string, a bytes-like object or a real number, not 'NoneType'

İ am trying to use inpainting webui but not working

How can i solve this problem ?

[fickle matrix]

Ay yo, anybody wanna buy an NFT 😂 😆

[glossy herald]

Macronator

[tulip yew]

Any help to solve this reactor error?

[wild sorrel]

looks like movie poster

[wild sorrel]

Can someone explain me what's the difference between those things in comfy UI Lora settings?

I'm guessing model - is actual model strength, while clip Lora strength or wot...?

and also what's the difference between those samplers

[rustic mantle]

I'm a newbie so please don't crucify me >.<. but can someone tell me how to AI upscale this image? the resolution seems too low for printing

please and ty

[wild sorrel]

okaaaay....

[steel osprey]

is it possible to use your own face images in foooocus or soemthing similar locally (easy to use) to create ai images?

[solid mesa]

hey guys ever since I came back from automatic 1111 after an updated all the generations became like this
I used same prompts models and lora before but it showed a lot of really good generations. However, it's been a week of trying different models and loras with experimenting with the steps. All I got was a really ugly images similar to this.

[wild sorrel]

stable diffusion "lora" - google-able.

[stoic lotus]

i'm having a preliminary training in 90 minutes for those interested in getting a glimpse on the Stable Diffusion Training Package where i will explain in details the impact of training parameters and how they affect image generation and when they should be used.
Example: cosine vs cosine with restarts vs polynomial vs linear; what is snr_gamma and when to use, max vs min timestep, lora dims and alphas, optimizers such as prodigy and adafactor, why use repeats and epochs ? what is regularization ?
All the advanced training methods and the impact they have on your training.
Link: #1095477236074827933 message

i will be showing you the exact impact of each training parameter, optimizer and scheduler

you may use the link in my profile if you are not part of the discord server yet

[dry blaze]

Insane to realize that this is SD 1.5 lol. Sadly i can't do my mad lora's as it's tensorRT accelerated

Took 3 min for a 2048 upscale with pretty much 200 steps from 768 to 2048 with 4x_ultrasharp

[frozen nebula]

Finally got some decent base generations working with comfy and sdxl. It's taken a while to get used to but I think its clicking now

[lapis wyvern]

I can’t use Adetailer! With SDXL model — I am getting errors - do I need to install something else?

[unborn heart]

No, it should work with the same files as SD1.5

[lapis wyvern]

Ahh.. depends of the Model then

I was trying with Hephaistoss10 SDXL

[unborn heart]

face_yolov8n works with SD1.5 and SDXL

[wild fractal]

Hi, I need a software or a feature or a script that does this... creating a black white mask image where the face is. I wrote a small python script that works with OpenCV but it's a disaster, I did this manually in PS.
I have a bunch of photos, say 300, I need to run an inpaint batch to replace the face, trough trial and error I am happy with face swap lab, however I am bumping the denoising to 0.4-0.6 and wanna do img2img batch on a bigger image like 2k, so I have to have an area mask for every image because i'm running 512x512 res for the faces

I'm at a dead end at atm

anyone tried doing batch inpainting ?

the "segmented mask" feature in faceswaplab is not so good, it gets buggy alot, I'd hate to manually inpaint 😦

[dry crow]

@elder oracle

[elder oracle]

omg thanks ill do that rn !!

@dry crow

[dry crow]

Its not in there

[elder oracle]

oh haha

[elder oracle]

i went into model and i saw a config folder in the model

[dry crow]

There it is:
Extra_model_path.yaml.example

Remove the .example

Then edit it with a text editor

[elder oracle]

ohh will do !

[covert harbor]

regenerate

make full c

[brave pebble]

can anyone please help me with this 🥲

[raven mirage]

It looks like you don’t have python installed locally. At least that’s what I’m getting from the error message

[brave pebble]

already installed

[raven mirage]

“You need to install and add python to your system environment path, assuming you are on windows.”
https://docs.python.org/3/using/windows.html

[raven mirage]

Take a look at this forum post. Should be able to help. https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/1423

[brave pebble]

i already enabled add python to path. it still shows the same error

[raven mirage]

I’d say uninstall and reinstall python then, if you haven’t already…

[brave pebble]

Thanks you one of the post in GitHub helped me rectify it

[raven mirage]

Nice, no problem

[wispy nest]

https://cdn.discordapp.com/attachments/1148388684836638820/1171224232114524221/a_stuffy_doll_inside_a_stony_chamber_with_arms_and_legs_attached_to_glowing_violet_chains__The_chains_are_pulling_the_arms_and_legs_to_the_point_of_tear_style-Fantasy_Art_seed-0ts-1699311993_idx-0.png?ex=655be691&is=65497191&hm=854a754cb1bff284b609f253abd39ab49729099365b0729e65fdda3ce2a081be& I want to see a stuffy doll being teared apart by chains being pulled

[edgy shuttle]

What's the difference question mark The normal models have in painting function you scribble around and it fills in the blank right?

[wild sorrel]

idk what's like...actual technical deep difference, but...inpainting models just work better for inpainting \ outpainting, normal models kinda sucks

[edgy shuttle]

I'm coming over from midjourney so learning a lot of the equivalents is pretty confusing as a first timer!

Is there an extension that lets you use more than one image reference? That's been invaluable for tweaking a style after I nail a composition

[wild sorrel]

one image referense for what?

[edgy shuttle]

Well you can img to IMG but past that further tweaks seem to base off the one image or it's direct results

[wild sorrel]

img2img? You can use controlnet with more then 1 img as input

[edgy shuttle]

Oh shit I should really dig into control net I installed it

I thought it was more for poses and stuff

[wild sorrel]

Yea, that's...a bit complex topic, you can use it in a different ways , get different kinds of information from image, I'd recommend looking for some videos

[wild sorrel]

nah, there are lots of different controlnet models...you can take pose, lines, style, colors...bunch of stuff

[edgy shuttle]

Where would you point a beginner? The web UI is really not self explanatory I've just been figuring it by screwing about

[wild sorrel]

idk, I was just learning from random videos youtube suggested to me 😄

[edgy shuttle]

I wish there was like a graphic that has overlay explaining the settings...or a mouse over text...

Between Lora, checkpoints, models, after adjustments, scripts it is much more complex than midjourney ui

[wild sorrel]

There is for basic things, like cfg, img2img settings and stuff...but not for extentions, usually they have their own docs.
Controlnet should have it's own docs, different models for it made by different people...so it's kinda all over the place...

[wild sorrel]

mj doesn't even have those 😄

[edgy shuttle]

Would be nice if there was some meta guide linking it all together in a way that's not a rabbit hole of disjointed guides

[wild sorrel]

https://stable-diffusion-art.com/tutorials/
Like this?

[edgy shuttle]

Does that apply to auto1111

[wild sorrel]

That's about everything related to SD

a1111, comfy, other things

[edgy shuttle]

It's definitely a lot to crack into. I can't remember the last time there were so many jargon I didn't understand not even working in linux

[wild sorrel]

lol

[edgy shuttle]

I tried to get gpt to explain how it work but it is very limited

[wild sorrel]

some tutorials there might be outdated tho...or just ...debatable...youtube might be better tbh

[edgy shuttle]

Maybe once it has the new feature I could fill with tons of documentation make a SD gpt

My thing with YouTube is often out of date info mixed with...people dragging things out to take more time

[wild sorrel]

You can try phind https://www.phind.com/ and feed it link to SD forums, docs and tutorials and ask it whatever you want, probably will work better than gpt

[edgy shuttle]

Oh hey nice thanks for the info

[wild sorrel]

It's kinda...gpt but with it's own model (trained mostly on coding), but can look for general info too...
Main it's thing is - you can give it any links to go through and prioritize it's search on those links when giving answer

[edgy shuttle]

Oh nice that's useful for website documentation

Or gaming Wikia for instance

[wild sorrel]

idk...I'm mostly using it when I need to find somethign quick and don't want to look through tons of links...or even make it code something simple

I won't let it document or make wiki...it's still AI that makes tons of mistakes

[edgy shuttle]

It does seem useful wonder if it will take a full domain like Reddit or something

I wish discord guides were searchable by gpt

Tons of them it's overwhelming

[wild sorrel]

it doesn't read whole thing, it just searching for things you are asking and trying to find something related , put it in a sentence that makes sense

[edgy shuttle]

Still sounds useful for tech help

[wild sorrel]

about reddit...there was AI that goes through social media garbage too

This
https://www.perplexity.ai/

idk how good it is, I kinda...tried few things and gave up on it

[edgy shuttle]

Useful for not having to visit reddit as much for stuff

[warm vine]

Prompt : a ghost mixed with a tree, glowing
Sdxl fine tuned to create Midjourney like images

Playground : https://ionium.web.app/sdxl

[gray plank]

prompt: Shrek2

https://aiphoria.world

[fleet goblet]

He sees you.

[obtuse hull]

this dindt go as expected

[mellow orchid]

Does anyone know what might be wrong with my computer to generate these photos?

[past pelican]

Cfg scale looks like it's way too high.

That, or you aren't using enough steps.

[wild sorrel]

send screenshot with your settings

[mellow orchid]

Cfg is set to 12, and sampling steps 30.

[past pelican]

What sampler?

Those settings can cause burn in if you are using an older sampler.

[crimson dawn]

could it be a bad vae? it does look like poor sampling settings though. whether the cfg or sampler itself. 12 is high but not SO high to look like that

[warm vine]

What model are you using? this thing happened with me also

[past pelican]

I mean I really can't help them out if they don't respond back after 2 hours.

[mellow orchid]

I'm using ComfyUI in my PC. through Stability Matrix Application.

It should be looking like this as they do in tensor.art.

But they look like these, in my local machine.

Try more and more, I seem to find it relavant to my Negative prompts more than VAE or others.

Unthinkable, right?

The Prompt is :

(1girl:1.4),(front body:1.6),solo, (a woman sitting at a desk use quill writing on a piece of paper),melanchonic soft light, wet dripping long hair, still from a live action movie, streaming on twitch,elegant, sadan vague, white facepaint), ink painting, ((moon behind:1)), (chinese style:1.2),(concretism:1.2), (masterpiece, top quality, best quality, beautiful 1.2), extreme detailed,theater dance scene,(hypermaximalistic:1.5),((Traditional Chinese table lamp)), (Glowing ambiance, enchanting radiance, luminous lighting, ethereal atmosphere, mesmerizing glow, evocative hues, captivating coloration, dramatic lighting, enchanting aura),

lora:anxiang:0.55

The Negative Prompt is :

(worst quality, low quality:2), monochrome, zombie, overexposure, watermark, text, bad anatomy, bad hand, extra hands, ex tra fingers, too many fingers, fused fingers, bad arm, distorted arm, extra arms, fused arms, extra legs, missing leg, disembodied leg, extra nipples, detached arm, liquid hand, inverted hand, disembodied limb, small breasts, loli, oversized head, extra body, completely nude, extra navel, easynegative, (hair between eyes), sketch, duplicate, ugly, huge eyes, text, logo, worst face, (bad and mutated hands:1.3), (blurry:2.0), horror, geometry, bad_prompt, (bad hands), (missing fingers), multiple limbs, bad anatomy, (interlocked fingers:1.2), Ugly Fingers, (extra digit and hands and fingers and legs and arms:1.4), ((2girl)), (deformed fingers:1.2), (long fingers:1.2), (bad-artist-anime), bad-artist, bad hand, extra legs , (ng_deepnegative_v1_75t)

I find when I use another Negative Prompt, it would be okay:

worst quality, normal quality, low quality, low res, blurry, text, watermark, logo, banner, extra digits, cropped, jpeg artifacts, signature, username, error, sketch ,duplicate, ugly, monochrome, horror, geometry, mutation, disgusting,
bad anatomy, bad hands, missing fingers, extra fingers, three hands, three legs, bad arms, missing legs, missing arms, poorly drawn face, bad face, fused face, cloned face, three crus, fused feet, fused thigh, extra crus, ugly fingers, horn, cartoon, cg, 3d, unreal, animate, amputation, disconnected limbs,

When I say "Okay" it means it doesn't look like totally shit, but they are still far away from good, they are no where near the quality as they are in tensor.art. It might be the lack of Hires Fix or upscale I suppose, but the color fell short and the atonomy and facial expressions aren't anywhere close.

[mellow orchid]

xxmix9realistic_v40

[mellow orchid]

Euler Ancestral

[mellow orchid]

I tried different VAE, including vaeFtMse849999Ema_v100, they seem to reproduce

[past pelican]

Ok so while the CFG is on the high side it shouldn't be causing an issue considering you are using Euler ancestral and 30 steps. My other thought is it could be vae related but you might just consider reinstalling your program and see if that fixes the problem.

[mellow orchid]

Well.... I would really try to avoid reinstalling everything. Until I get very desperate.

[muted vault]

Does anyone know where I can get a REALLY comprehensive list of artists? No, not music. I tried wikipedia but they're scattered all over the place and trying to sort them is a real mess.

[heady anchor]

theres a new extension for webui with a lot of artist styles

[wild sorrel]

I'd pin on high cfg + high weight on prompt and negative prompt...looks overcooked

2.0 is extremely high...I'd avoid even going 1.5+

[muted vault]

Oh? A rough idea of how many? Hundreds? Thousands? I got 2k today but that's not a particularly good list tbh.

[heady anchor]

go to webui extensions>available>load from and its the one on top

i just checked the page and its 600+

[muted vault]

600,.. Not good enough,.. but,.. Hey,. if I can add them to my list and they're not ones I have already, I'll have it. Gonna add it to my dynamic prompts. 🙂 Thanks. Any others anywhere else I can look into? Does anyone else know? 🙂

[mellow orchid]

humm.... yeah, I didn't look into the prompts very carefully, since it works on tensor.art. It might be the issue, I'm not sure.

[mellow orchid]

Art Station? https://www.artstation.com/

[muted vault]

Hm, does anyone here have the paid-for version of ChatGPT,. if yes, would you do me a big favour please? I have a large (almost 4k) list of artists for SDXL, but,. they're on a website,. could I ask if you might feed the link into CGPT and ask it to collate all the names into a .txt file in alphabetical order please? Pretty please? 🙂

I'd rather use deceased artists if possible, it's not exactly fair to use living ones too often IMHO.

I like to blend them (2 or more in one render to make them a bit more of my own).

[mellow orchid]

What website...

[muted vault]

https://stablediffusion.fr/artist-style

[mellow orchid]

That's a great list!

[muted vault]

Aye,. but I am struggling to get the list only. Am trying to run a script to search the text but I can't do it, I don't understand it at all. I'm not a coder.

ChatGPT says I can run a python script, but I'm having a headache getting it to launch.

[muted vault]

You got the first page? o_0

[mellow orchid]

I think it contians from A to Z

3906 lines of names

[muted vault]

Sweeeeet!!

[mellow orchid]

No sweat, I'm an engineer. And ..... it's not even a secret. It's written in the code. It's our lucky day.

[muted vault]

Thank you! I hope this can be of use to other users too of course!

Maybe SD could pin the text so others can use it?

[mellow orchid]

I'm not sure, I guess the website does a good job already. Actually, I'm not sure what you're up to with those names... unless you are to randomly automize it with artiest styles.

[muted vault]

I was getting fed up with an image I was trying to create,.. but just could not, for the life of me, work out what the style needed to be and was getting sick of trying to enter names two or three at a time,.. so, decided to try Dynamic Prompts,.. so ultimately figured a massive list of artists would be useful.

Had been at it for 3 days without any luck. So, yeah, dynamic prompting.

[mellow orchid]

Dynamic prompting.... I'm still new in this area. Good luck with it!

[muted vault]

This is the list again without punctuation, just names. Thanks. Fingers crossed. I have one render I'm working now, inpainting and tidying up the 'wallpaper' in the scene so it better suits the finished image. 🙂

I'd show you but it's (tasteful) NSFW.

[mellow orchid]

Humm.... I'm not sure if it's safe for display at this channel . . . .

[muted vault]

No, of course. 🙂 Wouldn't dare to post here. *Despite its tameness.

[rose hinge]

美女

[trim scaffold]

/imagin

[twilit verge]

I just started using dynamic prompts with wildcards earlier this week. It cut down a series of generations from about 3-ish hours (with interruptions) to about an hour. Of course, I also didn't have to be there typing in the names each time, which is another bonus.

[muted vault]

Tidied the list still further, and a handful more names too.

[muted vault]

I got a really cool image earlier, have been tiredly (if that's a word) working it since. A dancer's legs, ballet shoes,. bare mid-riff, dark contrasted rim-lighting. A powerful image.

Damn, another render of it has ggone awry. Suddenly she's got thongs on. 😕

[waxen zealot]

Hey guys, sorry if this sound dumb but please, what exactly does this button do

[stable hatch]

Hello i'm newbie, any guide how to make my own animation(jpeg sequence) with new style?

[stable hatch]

make all settings from last generated pic(promt, bad promt, size and etc)

[limpid lichen]

hellstrom got it but also, if you paste meta data from other images into the positive prompt box and hit that arrow, it propagates all the settings from that pasted text into the gui

[waxen zealot]

@stable hatch @limpid lichen Thanks a lot 🙂

[cunning thunder]

there's an error when running webui-user.bat

[dry blaze]

144k res grid anyone?

3 and a half hour 768x1024 100 steps per image to test every model i have with certain prompts to test not only what can do what, but also see what each focuses most on.

[pallid ruin]

Lmao

What happens if you try opening that? 🤣

[pallid ruin]

What´s the error that you got?

[dry blaze]

I can then zoom in and find out what each model's result is and which isn't suited for high res without upscale :P

And whadda ya know, refiner doesn't do great as the generator here

[prime flame]

Hey guys! If you have any time could you please check out this user study I made for CVPR? I'm trying to get some human-evaluation results for our new dataset, and we're doing a last-minute user study by the submission deadline Friday lol https://forms.gle/wKPFtTm8178adL1r6

It's a new massive image generation dataset specifically made for generating transparent images - we're hoping to release it to the public soon

Right now I'm hoping to get at least 20 people to do this study...it's not going to be very large so every person that fills it out helps tremendously

If anyone wants access to the dataset early please let me know!

[devout gust]

how to create pictures in stable diffusion

[limpid lichen]

refiner sucks when used as a straight generator ahhaha. i still don't understand the refiner architecture. it's completely not needed especially if you're using a refined model or a lora. add detail leco lora especially!

i suppose it gets very minute details better is the idea? theory behind that seems odd

[exotic glade]

Just enter a prompt of what you want your image to be,the style of how you want it to look, example: photographic or anime style...and if you want it to be a image or short video... pretty simple....

[dry blaze]

Leco?

And yep. Several XL models that doesn't need a refiner

[limpid lichen]

leco loras are the slider ones, where as you crank it up to power of 3 it does more of the concept, and 0 is none of the concept

https://civitai.com/models/122359/detail-tweaker-xl

you train them without images and instead subtract concepts or something

[fringe gulch]

So ask your ppl posting questions understand, Admin and devs do not answer any questions in this server or offer help.

[dry blaze]

Shit.. This is hard

Trying to make this screencap a proper 3d/realistic version, but anything cfg above stock and he turns crisp

[wild sorrel]

Check your prompt, usually that happens cause something has too high weight

Not sure what it is, but looks interesting

[oak yoke]

looks really good

[wild sorrel]

i'd argue on good, but I just love weirdness sometimes

[oak yoke]

this is the best image i have created so far of learning

[wild sorrel]

oh...I tried dragons or dragon-like creatures so many times when I started...that's oooof...not easy to get good looking thing

[oak yoke]

im trying more like a boat and sunset rn

[wild sorrel]

there are probably models that are better at this by now

[wild sorrel]

I just do random weird shit that comes to my mind 😄

Here's another thing from today

[oak yoke]

i mixed open journy v4 and dreamlike diffusion for my checkpoint

[wild sorrel]

broken too 😄

[oak yoke]

wow

[wild sorrel]

lemme find something actually not broken lol

Here ^^

Making people hold swords was challenge for a long time too lol

[oak yoke]

wow

i just got this one broken but eh

[wild sorrel]

not bad, even tho by brain refuses to understand what's happening 😄

oh it's half shadow half boat lol

[oak yoke]

tim to generate 20 images and go to school

nvm more like 8

[glacial bane]

is there anybody could teach me how to use this to create images ?

i cannot understand

mygod

[royal hound]

Hi, I was looking for references on SDXL training dataset and couldn't find anywhere. Do you have a source for this? Or maybe you were involved in the training? Thanks

[oak yoke]

How would i make these more realistic

[dry blaze]

Hoi, any idea why controlnet won't generate image previews?

[oak yoke]

I love this

[oak yoke]

HOW DO YOU MAKE IT SO GOOD AND PERFECT

[slim dirge]

tbh its a nsfw prompt but some of them comes up like this

its dalle3 though not stable diffusion

[oak yoke]

Ah

I didnt want to pay so im trying to learn stable diffusion

[slim dirge]

im using it for free tho

from bing

[oak yoke]

Huh

Is there a limit on how much you can use it?

[slim dirge]

15 free weekly booster credits, after those 15 u can still generate but it takes long (for me in the morning it takes jist like 30 seconds, in the night it can take like 30 minutes though xD)

[oak yoke]

I waited an hour for 1 image of stable diffusion

And it looked horrible

[slim dirge]

xD

so bing is better for u

[oak yoke]

Prob

[slim dirge]

also u can just create more bing accounts

[oak yoke]

Smart

[tawny drift]

Actually no, it's from ChatGPT, it also uses DALL-E. But you need to carefully make request prompt if you want something forbidden. You have unlimited amount of attempts if you have monthly subscription

[oak yoke]

Wait why bing?

[slim dirge]

https://bing.com/create
their image creator uses dalle3

[oak yoke]

Ah

[tawny drift]

ChatGPT as interface for dalle is better, because you can make desired image iteratively adding details step by step, with this method you can avoid censorship. Also chatgpt has weaker AI post-gen censorship analysis (that means you have higher chances to see your image)

[slim dirge]

true but it costs money though

[tawny drift]

well, yes, 20$

[oak yoke]

and that is the reason im going to use bing now

cuz i dont got $20

and stable diffusion takes to long for my low quality images

[tawny drift]

actually you can get such quality and impressive results with SD, but you have to spend some time on messing up with models.

[oak yoke]

yeah idk much about it and i dont really know good places to learn the different ones

[slim dirge]

im actually so proud of this image lol

so good

[oak yoke]

this is my best with stable diffusion

[slim dirge]

is semi nsfw ok to send here

[oak yoke]

idk

[slim dirge]

idk if some of it counts as nsfw

[tawny drift]

Made this recently with SD, looks good on phone background

[oak yoke]

HUH THATS SO GOOD

how long did it take to generate?

[tawny drift]

40 seconds, i have 4090

[oak yoke]

ah

i got a 3060TI but even my images take over an hour

i mean ig a 3060ti isnt that good

[tawny drift]

3060 Ti is not bad at all, it will take like 2-5 minutes

[oak yoke]

tf

1 sec

over an hour for these images i believe

i wonder if my stuff is getting limited

[trail canopy]

I think it can be generated normally on a 3060ti as well. You can do it quickly if you use something like xformers.

[oak yoke]

whats xformers?

[trail canopy]

Techniques to optimize memory

[oak yoke]

ah

see it uses like 2-5gb of ram

i wish i could allow more for it to run better as i got 32gb

[tawny drift]

i used 2070 Super for high quality images one year ago, it wasn't speedy, but it can handle something good
you better don't try something higher then 512x768 or 768x512 (in original resolution)

[slim dirge]

https://cdn.discordapp.com/attachments/1159160808203366400/1166483950844260362/image0.jpg?ex=655d1cd6&is=654aa7d6&hm=b4c60129f65aed63429cf86f5e09266c8a9930022e45d6fbc0f2173f73945595&

thats a good one i sent, should be sfw ig

[tawny drift]

and just applying upscaling with factor 1.3-1.5 you can get pretty good results

[tawny drift]

looks nice

[trail canopy]

Oh, and due to technical constraints, I think the 4000 series has the highest accuracy and speed.

[tawny drift]

\

[oak yoke]

im trying another image real quick and ima see how it runs

[tawny drift]

i love bypassing dalle censorship, lmao

[oak yoke]

1 hour and 40 minutes

is my sampling and cfg just that high that it takes that long

nvm its now 2 hours

[tawny drift]

50 steps, bro even i don't use such a high values

[oak yoke]

whats a good step amount

[tawny drift]

it's like 24-32

[oak yoke]

ima try 28

[tawny drift]

27 is good anough in most cases

[tawny drift]

and don't generate 768x768 this is a waste of resources
if you want rectangle use 512x512 this is anough

[oak yoke]

my thing is set for 768

i downloaded 2.1 for 768

[tawny drift]

also high gcf scale causes more artifacts

[trail canopy]

If the SD version is 1.5, it will be 512, if it is 2.1, it will be 768, and if it is XL, it can be generated in a larger size.

[trail canopy]

that's right. cfg should not be too high

[oak yoke]

is 11.5 high?

[tawny drift]

Sure it can, but if you have limited hardware capabilities it will be better to set lower original resolution

[trail canopy]

Normally something like 7 should be fine.

[tawny drift]

7 will be anough

[oak yoke]

oh

so maybe thats why it takes forever

[trail canopy]

can be generated very quickly

[oak yoke]

6 gpus....

[trail canopy]

No, it's seven GPUs.

[oak yoke]

my bad

[tawny drift]

nice

never used multiple gpus

[oak yoke]

oh now it generates more quick

[trail canopy]

But 1111webui doesn't have that much of an advantage.

[oak yoke]

dont look that good

seems a bit blurry

[trail canopy]

What model are you using?

[oak yoke]

idk

[trail canopy]

fmm

[oak yoke]

i combined open journy and dreamlike

[tawny drift]

By his screenshot it looks like it custom merged model

[trail canopy]

It's a model that you merged yourself.

[oak yoke]

yea

[slim dirge]

@oak yoke do u want me to generate images with ur prompt

[tawny drift]

I don't recommend you merge models until you definitely know what are you doing

[oak yoke]

nah i think i got it

i mean 20 minutes i can wait

[oak yoke]

i watched a yt vid and a guy reccommended this merge

[trail canopy]

🙂

[oak yoke]

A Black dragon with a purple aura, dark background, Attached body parts, breathing a purple colored flame, 4legs, 1 pair of wings, 1 head, 2 eyes, proportional, 3d rendered, mystical, magical, Euphoria this is the prompt im trying to use

[slim dirge]

aight

[oak yoke]

also where is a good place to learn all the different sampling methods

is that with bing?

[slim dirge]

lemme send u some good ones

[oak yoke]

i can already tell my image is gonna be messed up

😦

[oak yoke]

how can i get better with like image duplication like how its shown in the ss

[slim dirge]

use bing

[oak yoke]

isnt it from bing most likely to just be online

[slim dirge]

wdym

[oak yoke]

like other people can also get the images

[slim dirge]

no

your recent photos are saved in ur account
only the last few photos are displayed though

[oak yoke]

ah

[slim dirge]

so like u wont be able to get ur first image u made if u made like another 100

[oak yoke]

ah

i got a discord server where i just post my images in there to save

so i got that covered

oh wow bing does work better

[slim dirge]

good

[slim dirge]

nice

[oak yoke]

dang

[oak yoke]

why is this going to take an hour and a hlaf

the steps and cfg arnt even that high

nvm its going down

1 hour now

[rigid fractal]

Second test with Animate Diff. Holy craaap it's so cohesive! no post processing at all! Running the same thing with roop right now to see how that works haha

[sinful urchin]

so good, but that's not how you hold a sword

[vernal hollow]

hello cummunity i need a little help know some one a lora or a checkpoint ? where i can doo picture like this?

[quiet moat]

I dont keep up with animation news since its pointless with my 6GB card. What is the input for those animations? A real video + a prompt?

[rigid fractal]

yes. and you can do custom assignments per frame. ie 15 "man smiling" kind of thing.

I have a 12GB card (3060) and like... it probably hates me but it churns it out.

works fine with roop as well. Just did myself.

[quiet moat]

how long did your animation take to make?

[rigid fractal]

I didnt time it but I would say an hour, though it's all about fine-tuning the quality vs cost settings.

[quiet moat]

ouch

[rigid fractal]

why ouch? that is honestly super fast for video.

I did animations with blender and a few frames would take all night

[quiet moat]

well, SD is not great at accuracy, so with pics at least, you gotta do many gens and pick out the best. I imagine for video it's the same.

for something good enough to show off

And 1h is beyond the threshold for me to wait for a dopamine hit

[rigid fractal]

you could lower the settings for quicker results. I did 768x1024 here, you could do it in 15 minutes @360x512

I am much more concerned with quality - so I let it burn hotter than I should

[quiet moat]

if you do it in 360x512, and you're happy with the result, can you then switch to 768x1024 and get a guaranteed great result? or is it kinda random?

[rigid fractal]

does that work with normal images? I don't think it does.

[quiet moat]

Hmm, I guess not. Only the upscale part does.

[wind iris]

Hey, when i run diffuser on colab, it runs the entire code every time, is there a way just to run the code i added recently?

[sinful obsidian]

/animate

[cyan gulch]

Мем про Барбі

barbie meme

[rocky quail]

POV: you are a peasant, it has been 1 day past the 90% yearly tax deadline which you cannot pay since you just broke your arm, unable to work and currently malnourished. Your benevolent lord baron send this guy. what do you do

what the knight sees

[manic olive]

I hate nvidia for this! My card is 16xx series. Barely minimum vram, old pc on it's last legs, and I wanted to upgrade to 40 series and am4 socket pc. But I can't justify the price for the vram.

[hearty karma]

I understand ya man. My computer had a 1660 gtx card so I understand how painful it is

[rugged token]

i bought a 4060ti for 599, has 16gb. sold my old 2070 for 200. so not too bad

not quite 24gb, but more then 12

[sharp sail]

Ola gris

[quiet moat]

what's the term for the camera angle where it's not a portrait (body facing viewer), not from the side (body facing left/right), but somewhere in-between, at 45 degrees?

[past pelican]

Three quarter pose I believe.

[quiet moat]

thanks, I'll try

[past pelican]

Wait, that's a head and upper body shot.