🏞｜general-with-images

[wanton nimbus]

every time i generated

[limpid lichen]

weird. if you followed the steps and didn't change things like default vae, that shouldn't happen

doesn't seem lke you asked for help with it though. i was just curious what went wrong where.

[short hamlet]

Can you send an image showing what you mean?

[wanton nimbus]

nah i deleted it

ill redownload it if you want @short hamlet

[ember mirage]

I would like to use an open-source AI for image processing, which can be installed locally. I have an RTX 3090, and I've achieved good results in replacing objects and people in images. I would like to replace the skateboard from photo 1 with the one from photo 2. Additionally, I want to replace the entire person. For example, a middle-aged male with blonde hair, wearing a necklace around the neck, a white T-shirt, and baggy gray pants. Now, this described person should replace the 'old' skater along with the skateboard. I am working with Python on a Linux system.

[runic ivy]

https://cdn.discordapp.com/attachments/446305615808167936/1157438025081557032/b4e052a8-81b1-4f06-90ac-370e386bf829.jpg?ex=65189ba9&is=65174a29&hm=66102ed9b5d0fbc6d493b095ff5e95e7c07b43f12649092cb17ea1cb48f6c4f2&

waltuh

[nimble scroll]

poster worthy?

[fleet goblet]

So I just realized I had access to Dall-E 3 through bing create.

It's too good.

Like, a level of good I cannot describe.

[nimble scroll]

i like to make alternate coverarts with dalle3

[fleet goblet]

No AI is this good.

I love SDXL, and the fact that it's locally-run.

And I have it always-ready through WSL

But you just can't get this level of coherency without REALLY being specific on prompting.

[nimble scroll]

agreed

[heady anchor]

now try with anime style

[fleet goblet]

Sure, anime style time.

[heady anchor]

waltuh rly be cookin

[past pelican]

That's not meth, that's um... blue Himalayan salt.

[fringe spindle]

Is there a problem with "Interrogate", or am I doing it wrong? I drop a image in the Img2Img part of A1111, click either of the "Interrogate" buttons, the spinner appears, but nothing else. Spinner just spins forever.

Or I get "Error - connection timed out".

[fleet goblet]

How the hell does Dall-E 3 do this?

It's almost nailed Super Mario Bros. on the screen.

But the console and controller's funny lookin

[runic ivy]

it does nail mk10

[wanton nimbus]

has to be the most cursed car picture ive seen

[wanton nimbus]

crazy

[runic ivy]

and another stupid design xdd

[wanton nimbus]

aerodynamics

[wispy nest]

This is too funny

I have not seen dalle3 looks insane

[runic ivy]

can also do memes

[short hamlet]

It's okay, you don't have to

[wanton nimbus]

its kinda too late

i kinda downloaded it already and i kinda fixed it

[short hamlet]

well thats good

[heady anchor]

Unsafe image content detected
Your image generations are not displayed because we detected unsafe content in the images based on our content policy. Please try creating again with another prompt.

[runic ivy]

i guess cs go in pixar style was too much lol

[wispy nest]

looks cool tho

valorant ahh

[wispy nest]

if only mk had the levels of crossover fortnite gets

[mild tusk]

what are the prompts?

[runic ivy]

cant find it now but it was something about csgo in pixar style but nothing to specific

[mild tusk]

im so trying this

pov work as a prompts it nice

[fleet goblet]

Idk man this is kinda nuts.

[mild tusk]

layer

[heady anchor]

shrek knows what u deleted

[fleet goblet]

almost

[mild tusk]

shrek but thanos attempt?

[mild tusk]

what is your model?

[runic ivy]

well dalle 3 seems to be available on some web browsers

[mild tusk]

is there a prompts for text being good?

[mild tusk]

that what i was trying doing

is that spongebob background?

[runic ivy]

yes

[runic ivy]

just some random prompting

shrek as thanos in a spongebob episode,absolute spongebob style,detailed

[mild tusk]

nice

im using dreamshaper as model

omg amazing

we having so mcuh fun

[mild tusk]

real people behind

[wanton nimbus]

dall e 3 is great at making misinformation

[wispy nest]

this will happen 76 years from now

[serene flint]

https://cdn.discordapp.com/attachments/1037799583784370202/1157461073650262107/harrowed_None_89213987-fb25-4d99-948b-d8ac1148e735.png?ex=6518b120&is=65175fa0&hm=9bbceec717804842ce7892286bdb98fa72a979581bb9d0c17b104148b5b222b2&

[wispy nest]

Correct prompt to create something like this?

[serene flint]

with that specific person's face in it? Ain't no promptin' that my guy

[runic ivy]

xdxd

[serene flint]

wanted to find out what shrek + spongebob =

now I know

And if you take SpongeShrek and mix THEM with Thanos....

[kind quartz]

@wispy nest install controlnet and qrmonster and upload face and then something like in storm.

[wanton nimbus]

can someone help me with bing image creator

[boreal falcon]

what do you need?

[wanton nimbus]

its taking forever to generate

it says 15 sec wait but ive waited for 20 minutes and it still hasnt generated

[wanton nimbus]

nvmn

fixed

[fleet goblet]

I'd play this.

[fringe spindle]

@keen bloom Still running - could it be a problem with the image?

Is anyone else having a problem with Interrogate CLIP not working?

[wanton nimbus]

spongebob catching lambos?????

[torpid rampart]

guuuys im getting very bad quality in my stuff, some advise?

[nocturne oak]

What VAE are you using?

[torpid rampart]

a little new here, with vae you mean sampling method?

model?

like checkpoint?

[nocturne oak]

No, literally what VAE do you have selected.

[torpid rampart]

Mh

I think thats the problem

[nocturne oak]

In order to decode the latent information properly into an image that has the right colors, you need a VAE. SDXL requires the SDXL VAE.

[torpid rampart]

probably I didnt set up that

yet

[past pelican]

Some vaes are baked in though.

[nocturne oak]

Some, but not all.

https://huggingface.co/stabilityai/sdxl-vae

[torpid rampart]

Ok

And how I set it up?

Ok I see

I need to run the command

ok I will try to do it thanks

So you mean sdxl_vae?

[nocturne oak]

yes

put it in your vae folder

[torpid rampart]

yeah done

I think is working better

do someone has experience here with infinite zoom?

hehehe im also strugguling with that

[mild tusk]

question is there a guide for stable diffusion? ik how to use it btu i want to improve with settings so ye

[wispy nest]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#attentionemphasis

Hardest thing on stable diffusion is knowing what you want to gen

This page has a bunch of technical stuff

[mild tusk]

thx

[wild sorrel]

https://stable-diffusion-art.com/tutorials/

[mild tusk]

thx

[sterile kiln]

Not my prompt at beginning

[wanton nimbus]

Bro I fucking hate mobile data

[serene flint]

@sterile kiln I love merging two images together, with no other variables involved. No prompt, LoRA, etc. Most of the time I combine people's images with one I made of a jellyfish.

[serene flint]

Here's this one with Mario

[wanton nimbus]

ferrari f8 covered in jelly

[light fable]

DALLE3... what have thes closed source guys figured out like this is just too good

look at those fingers

by now i thought SD would be caught up

[obtuse palm]

I just want something that stays in the same quality across different version.. lost my faith on auto1111 already. give me an a better alternative

i want the checkpoints and everything to deliver their optimal performance without the ui versions affecting anything.. is that possible?

[obtuse palm]

nvm i think i found the mother of most of my issues.. it was this culprit extension which was trashing up everything

if anyone else using this extension then i bet you're getting less than what you deserve.. for this hilarious extension

it may look great and handy in it's appearance and functionality but that's where it goes wrong.. it does way more than what you see, changes the whole structure and makes generations different than usual and bad to be honest

here the 1.6.0 and 1.5.2 were generated while this extension was active but on 1.3.1 it wasn't there so i got the usual output with the 1.3.1
but still yeah.. auto1111 version changed generation across different versions
1.6.0 and 1.5.2 looks pretty similar compared to 1.3.1

[obtuse palm]

i think i'll trust the old version auto1111 for sd generation than new updates because quality matters the most over speed or anything

for sdxl i think i'll move to another ui like comfyui or something better.. i need yours suggestion

[wild sorrel]

Imo all 3 images are kinda meh

• if you want to compare something you need alot more examples than 1 with different settings and models too

[obtuse palm]

I'm trying with different models and setting and always version 1.3.1 came with the best outputs (realistic or anime)

Most importantly the hands and fingers are generated much better on version 1.3.1

That was the most of my regrets lol

Things become unstable with version upgrades imo

[quiet current]

Keep it tasteful please.

#✍🏼｜rules-and-tos message (Rule #4)

[prisma stone]

I need a little help here plz

[delicate pilot]

Hey guys my new workflow is now to start with dalle3 and then bring those creations in to unified canvas on invoke-ai and then continue working. Anybody else that is doing that as well?

[delicate pilot]

Is your new workflow now also to start with dalle3 and then bring the generation in to a unified canvas running a SD model and then continue working on them?

[limpid lichen]

requires a server farm though. huge limitation of it. the host owns all the rights to the images created and only licenses it to people for personal use. That disqualifies it from being impressive at all. It's not a tool anyone can use for anything other than an extremely limited use case.

[limpid lichen]

if you intend to accept money for doing anything with dall-e 3, you're putting yourself at huge legal risk

[glass crescent]

What setting i need to change to have a result more like this ??

Instead of this

[bold mist]

Surprised at how clean the BG gradient is, and how accurate Bugs is.

[kind quartz]

@glass crescent you have result source and source result order.
You need qrmonster for SD Controlnet

[glass crescent]

@kind quartz Do u have a video that explain qrmonster ? I already downloaded so much thing i don't understand anything

[kind quartz]

i have not video sorry. Just make sure you have high denoise strength. Here are some images by it.
If you want text, you have to do it before in krita or inkscape or photoshop if you have something like this.
qr monster for controlnet for 1

for 1,5 model is on huggingface, it has 750MB and yaml file you need as well.

you can send me image and all i can do for you is one transformation as you wish. Thats all i can do. I bet there are videos on youtube how to.

[golden oriole]

trying my hand at photorealistic concert photography thru ai

[lapis quarry]

How can I get the colors normal? Am I missing something?

[obtuse estuary]

my own a1111 theme, how is it?

(the focused input field gets highlighted in blue like shown in the pic)

[sinful urchin]

Bing AI regularly produces this quality or better. What models, checkpoints, or whatever do I need to match that on SD?

[wispy nest]

https://tenor.com/view/copium-cat-gif-27161395

[wanton nimbus]

i have never seen this message lmao

[wild sorrel]

I'd give fields \ interactive elements and some text a little bit more contrast.
hover looks fine, negative barely visible, I'd make inputs a bit lighter.

[serene flint]

One Punch Jam

[wispy nest]

https://th.bing.com/th/id/OIG.Pmjwb3S_vmUu5OIZZ_nu?pid=ImgGn Dall-E 3 generated my most beautiful image of all time

[foggy frost]

how did you get dalle 3

[wispy nest]

I think it's 3, bing account.

[foggy frost]

ah

yeah i love bing images

i wish SD would be as good

[muted vault]

Anyone here GOOD at art? DM if yes - and I mean GOOD,. professionally good. Esp' helpful if you can emulate the styles of/either Luis Royo, Frank Frazetta or of course Boris Vallejo.

Failing that,.. if you can complete a render that was too closely zoomed in and ControlNet refused to work with the Checkpoint (ZavyChromax v12) I created it with.

[muted vault]

No one?

[sterile kiln]

😂 stop, he will think that SD is responsible of this style xD, he will run away

[wispy nest]

it's just Dalle-3 has like 15x the amount of parameters

it's not anything to do with SD being poorly trained, SD is just made to run on slower computers

I'd imagine in order to run Dalle locally and not through bing's server rigs, you'd need some kind of commercial GPU

[sterile kiln]

@wispy nest
Because you think that Dall-E 3 will offer people models and generations so greedy that they would need Nvidia A100s for each generation? Their models are not necessarily bigger and more qualitative than Stable Diffusion.

[wispy nest]

no I prefer fine tuning sd 1.5 myself

I've made loras, lycoris, stuff like that

generate a few thousand images, toss like 40 of them into my core and tag them with txt files using kohya then hand edit the tags in the txt files myself

train at 10 epochs at cosine with AdamW (not AdamW8bit) then use a network dropout of 0.2 or so across the board

[sterile kiln]

I also created a pletorium. I didn't see the link

[wispy nest]

their models are bigger

but bigger doesn't have to be better

it's about how you use it 😏

[sterile kiln]

https://www.youtube.com/watch?v=k5JfV6rqajg
Ok i got out...

[wispy nest]

lol

oh shit this is actually a vibe

https://tenor.com/view/mood-dance-russiankiddance-club-dancevibe-gif-21421102

IT'S IN THE WAY THAT YOU UUUUSEEE IT

[sterile kiln]

dude, its one of my favorite from EC

[rustic kraken]

Does anyone know when is impainting and outpainting going to be enabled via API?

[foggy frost]

how can i bet better results

i want it to still look like the input image

[short hamlet]

You could try regenerating with the same seed

That'd ensure similar results

[foggy frost]

wdym

the image on the left is real

just playing around but like what if i wanted it but as a pencil sketch

[sleek needle]

Anyone know why my images are looking like this in ComfyUI? Just started with it, using a custom XL model that works/looks fine in auto

[serene flint]

What VAE

[serene flint]

Adding a little Italian plumber works well

[sleek needle]

None, also lold

[serene flint]

😂 maybe try?

[tidal iris]

When i promt "..girl with pink hair.." I get also pink chairs, or buildings... How can I avoid that?

[wild sorrel]

yea, that's SD problem...if 1 color specified on any given item - this color will be everywhere SD can put it, it's not limited to a word or phrase with comma or anything...

specifying more things on the image, giving it more details helps a little, no actual solution as far as I know

[heady ice]

the controlnet?

[kind quartz]

@foggy frost try play with strength as it is most important. Higher value higher change, lower value less change. Also try keep AR as posible to original say 768x512

try pink color, rainbow theme. It should put in picture more colors. @tidal iris

yes light pink drink

[sterile kiln]

Can anyone tell me more about --opt-split-attention?

[wild sorrel]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Optimizations

[sterile kiln]

I meant, more than here. Because the explanations on A1111 are not very precise.

[wispy nest]

hey! Dall-e 3 supports controlnet too. Just drag an image to the chat window in windows 11 and type the prompt

https://www.bing.com/th?id=OUG.E582543CAAB50DD1AB8EED81AD2563C3&w=236&h=320&c=7&rs=1&qlt=90&bgcl=ececec&o=6&pid=PersonalBing

https://www.bing.com/th?id=OUG.A531A3BD54DBB40E727BE1C4CEC2E2CA&w=236&h=320&c=7&rs=1&qlt=90&bgcl=ececec&o=6&pid=PersonalBing

the second was generated by using the first as input and the same prompt

[nocturne oak]

Looks more like IPAdapter or BLIP than controlnet.

[wild sorrel]

idk what those smart words are, but looks like img2img

[nocturne oak]

The angle of the cube is not the same and it's not a manipulation of the existing image.

BLIP and IPA look at the context of an image, similar to how a human might look at it, and then take that into account when a new image is generated.

It's not manipulating the image given to it itself like img2img is.

controlnet would be looking at something like the depth, the outline, etc. and fitting a new image into those specific parameters. (It could be done with a weak adherence, but controlnet just doesn't seem likely compared to BLIP or IPA.)

unCLIP is a similar approach and it could be that too.

[limpid lichen]

dall-e 3 doesn't even support in painting. there's a big reason stable diffusion people keep rolling their eyes at it.

[timid yew]

NEGATIVE PROMPT

[wispy nest]

something looks like it's too high of a strength as well

if you're using loras you want the weight to be 0.6-0.8

[boreal falcon]

they will likely add it

but for part of the users its not even needed because DALL-E 3 might not even be the main AI image generator for them

or the all in one tool at least

and SD users imo would be anyway less likely to drop SD for D3

[limpid lichen]

its not there now. so...

you don't own rights to the images out of dalle-3.

if i were using them, i sure wouldn't brag about using them

[boreal falcon]

i wouldnt know why anyone would brag about using any of the image generators or in general

[limpid lichen]

anytime you announce that an image is made with dalle3 , you're declaring that you don't own rights to it

[boreal falcon]

well you dont have to ask OpenAI for a "ok" to sell what you generate etc.

you can more or less do what you want with them

as of copyright, ofc nothing different than other generators

[limpid lichen]

https://www.bing.com/new/termsofuse?FORM=GENTOS

dalle 3 terms on openai site haven't been talked about yet. i doubt they'll allow commercial usage on this version

these limitations mean one thing. you have a license, not ownership

[boreal falcon]

https://help.openai.com/en/articles/6425277-can-i-sell-images-i-create-with-dall-e

[limpid lichen]

that's for 2

[boreal falcon]

uinless

unless*

yeah

but why would they not allow it for 3

[limpid lichen]

3 is only available through bing

[boreal falcon]

its rolled out soon via ChatGPT

for some it already is

they handle Bing differently probably

makes also sense

[limpid lichen]

more lawyers are looking at the new dataset and microsoft is heavily invested now, using this tech to make bing good

its only a matter of time before MS acquires openAI

altman might even be inline for MS ceo tbh

[boreal falcon]

if MS acquires OpenAI i dont care as long as it brings more good than bad tbh

[limpid lichen]

i like the moves they're making. but honestly, the image stuff isn't relevant to people using open models

[boreal falcon]

well kinda like with Adobe if you ask me

[limpid lichen]

Adobe have always done really great stuff with computer vision and ai tech

i'm impressed at how fast they got firefly rolled out. photoshop has had the healing brush for a long while though

[boreal falcon]

i mean in a sense that people of SD are much less likely to use Firefly for example and it doesnt even really target them primarily or at all

[limpid lichen]

so much awesome content aware stuff in photoshop. the scaling tech they came out with a decade ago, oo 🤌

[boreal falcon]

they have yet to bring stuff out they were exploring and some of showcased

[limpid lichen]

firefly also provides a legal boon for studios looking to bootstrap generative pipelines today

[boreal falcon]

i wait for the next wednesday, thursday and friday

their MAX event

[limpid lichen]

adobe is a giant shield against possible lawsuits in that regard

[boreal falcon]

but as said that doesnt matter much or at all for majority of folks of the SD community and i understand why imo

[limpid lichen]

i think most people doing generative work in adobe are using a plugin with sd

[boreal falcon]

yes, although it has a price (literally and symbolically)

[boreal falcon]

from what i got to witness, talk to etc. not really. But there were some that did, also the other one what was it called

Alpaca

yeah, Alpaca

Alpaca has SD in it tho as well

at least one of the models

[limpid lichen]

https://github.com/AbdullahAlfaraj/Auto-Photoshop-StableDiffusion-Plugin i mean this monster of a project

[boreal falcon]

yeah, i even tried it xD

it was buggy when i tried it the first time

but was updated later apparently

[limpid lichen]

i'm not sure why the past changed the address to phoooshop

i didn't edit it at all lol

[boreal falcon]

A111 got that plugin now which is supposed to handle contrast, color etc. of the image

basically alternative to Photoshop or parts of it

[limpid lichen]

i'm weirded out by how my clipboard just lost a "t" randomly

[boreal falcon]

Phooshoop 😄

man now i forgot how the plugin for A111 was called

[limpid lichen]

lol thats phooshopped

[boreal falcon]

do you use SD only?

basically for the whole process of an image

i might have asked you that before lmao

because i always like to hear from people what and how they use stuff

[limpid lichen]

i have a few different creative processes. sometimes i start on paper. I don't like to limit myself to one tool for a job.

[boreal falcon]

Oh okay

Me neither

With exceptions

[heady ice]

what happend with the controlnet?

[boreal cedar]

Something happened to Dream Studio that completely ruined my results. If I reuse a prompt with same seed and all the settings the same, the result is so wildly different that it seems broken. I have tried all the engines and they all look terrible compared to the original.

has anyone else had this happen?

[livid ermine]

is it just me or does sdxl seem overy polished? we could try a SDXL vs SD 1.5 to see which is better. my bets are on the latter.

[earnest solar]

How did it get the americas almost perfect yet screw up the thumb so bad? X.X

[sterile kiln]

Stable.Art

use inpainting with stable.art

[limpid lichen]

i'm not sure why you're replying to me with that comment. where's the context?

It's just you. prompting and process issues

[sterile kiln]

Hey man,
did one with my face long time ago x)

[sterile kiln]

Turing is missing

[kind quartz]

but arc and 3060 which were issue are included 🙂
Pascal as well, and is working somehow.

[rain gazelle]

haha nice! I like how it blended the styles.

[sterile kiln]

I used a LoRA from my face, a COntrolNet tab with a t2i_style, another with a Depth. The 2 ControlNet were inspired by this coin from google.

The same settings with a recent model (Dreamshaper 8)

[rain gazelle]

right on

[craggy wolf]

Greetings friend, excellent work, I would like to know how you managed to work with ControlNet and Lora in this discord, I need to change the facial expression of a face but I can't get the ControlNet option, thank you very much

[sterile kiln]

ControlNet is an extension. Go on your Extension Tab, in sub tab, go in Available, clic on Load from, then search sd-webui-controlnet

For the LoRA that I used, I did it myself with 15 photos of myself. If you have an Nvidia card with at least 8 GB of Vram, you will also be able to do some quite quickly.

I need to change the facial expression of a face but I can't get the ControlNet option, thank you very much

I'm not sure you know what ControlNet is for.

[rain gazelle]

I wrote another image browser, this time for quick metadata lookup. Would appreciate any feedback.

V1.0, Open Source and binaries available: https://github.com/GarlicCookie/SD-Quick-View

[mighty orbit]

I want to create random robots face with exactly this logo design (one color blue logo and flat design). which tool can help me?

[mighty orbit]

😭 no any forum could help me for this - please help me

[rain gazelle]

you could maybe try img2img?

[mighty orbit]

no no - this generated images are very similar to my sample - I want to use only design of that logo not face of robot.

[rain gazelle]

oh

you should be able to do that with some prompting

there's my garbage attempt 😄

[mighty orbit]

good - but still not simple - I want to use only blue and white color like that sample but tools use extra colors like black color 😒

[rain gazelle]

i like this one

[rain gazelle]

yeah you'd have to play with the prompt some, I'm not the best prompter

[mighty orbit]

yeah and my problem is that I don't know what prompt and checkpoint must use for this. are there any way to convert my sample to prompt list?

[rain gazelle]

I think just about any prompt can do that

err... checkpoint

just use "vector art, 2d, logo, flat design, blue and white"

[mighty orbit]

How can I prevent the use of colors other than blue and white?

[rain gazelle]

well, you could do it in post, that's really easy

otherwise, just prompt best you can. "two colors" maybe? play with it

[mighty orbit]

my pc GPU is very old - are there any discord robot to use for generating image from sample and prompt?

[rain gazelle]

you can do stable diffusion on colab online - check for a youtube tutorial on setting that up.

[mighty orbit]

please tell me - what checkpoint and VAE you use for this one? and what ControlNet model you use for it?

[rain gazelle]

i think it was RealCartoon. Just the regular SD VAE. No controlnet.

[mighty orbit]

this mean you set VAE on automatic?

[rain gazelle]

vector image, , flat color, 2d, circle logo, blue and white smiling robot, blue and white only, white background Negative prompt: Frame, framed, ((text)), grainy, low-res, titles, text, error, cropped, worst quality, low quality, jpeg artifacts, duplicate, morbid, mutilated, out of frame, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, ((extra limbs)), cloned face, disfigured, gross proportions, (malformed limbs), missing arms, missing legs, extra arms, (extra legs), fused fingers, too many fingers, long neck, ((ugly)), fat, obese, chubby, (((deformed))), [blurry], bad anatomy, disfigured, poorly drawn face, mutation, mutated, (extra limb), (poorly drawn hands), messy drawing Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 4258414840, Size: 512x512, Model hash: 75d32966d0, Model: realcartoonAnime_v5, VAE hash: f88e1cfb7c, VAE: v1-5-pruned-emaonly.vae.pt, Clip skip: 2, Version: v1.6.0

there's the full settings

[mighty orbit]

tnq - i will try this again

[shell sleet]

Did this with a junji ito lora and a sally lora from nightmare before christmas! It's kinda neat seeing her in a creepier way.

[glossy herald]

Meow fellow discord users

bad cosplay man as a cat

bad cosplay man as a cat singing karaoke
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1168721058, Size: 512x512, Model hash: e65c57a0a6, Model: DB_fun_bad cosplay man
model link

same params,

bad cosplay man as anime girl, crazy yandere
Negative prompt: text meme

[past pelican]

We posting cats now?

[glossy herald]

when aren't we ?

let's cat-party!!!!

[glossy herald]

(I'm cheating, using my "good generations" folder, I'm not generating 100 styles on the go)

[past pelican]

I'm sort of cheating. This is just a very good model 🙂

[glossy herald]

ho the cat sushi chef xD

[glossy herald]

beautiful color palette

really nice model

[rain gazelle]

Nice!

[sinful urchin]

Bing AI seems to shadow black prompts with "monster" now

[hasty tundra]

Hi everyone, I've been trying to find a good method to help me achieve this kind of effect: https://www.youtube.com/watch?v=gZprUqsLOyQ&ab_channel=uisato
But can't seem to make it work... I am using stable diffusion with control net enabled, I've also tried img2img, it's been driving me crazy. I've tired 5 different models and some Lora's as well but my effects are nothing like the ones in the video. Does anyone have any idea on how to do this? Or any good models which would help me achieve a similar style etc. I've attached below the reference image and the style which I'm trying to obtain.
Thanks.

[wispy nest]

https://www.bing.com/images/create/five-wizards-on-a-circle-evenly-distributed-on-the/6521b789252b4553b796d12beddba2a7?id=K547b2LLeMjG0imW8MaYQQ%3D%3D&view=detailv2&idpp=genimg&idpclose=1&FORM=SYDBIC too bad that SD cannot generate this. The sampe prompt, when I call the discord bot, generates one wizard and some arid landscape from the side, like some portrait

[glossy herald]

controlnet + tweeter logo

never thought the bird was facing us

[past pelican]

Low budget Mercy cosplay?

[glossy herald]

exactly

"bad cosplay man as mercy"

[tepid lance]

Sans

[dry blaze]

Friendly reminder to not do absurdly large generations with automatic1111, it is not efficient at fuckin all with memory and just eats everything

[past pelican]

https://tenor.com/view/what-shocked-surprised-why-how-gif-24191425

[wicked sparrow]

1

[celest halo]

award winning photography, 1woman, swimming underwater, coral reef, portrait photo, realistic

[wanton nimbus]

skill issue

[dry blaze]

Lack of video memory/automatic is too damn ram greedy issue

[dry blaze]

As i pruned lots of my models, they have no longer any metadata from civit, so i regen alot of the images as well as harvesting loras and whatnot to recreate the images in auto1111, but one of the gens were fuckin 1500x2048 native with x2 upscale on top lol

Sadly it's not as efficient as comfy

[torpid rampart]

Can you recommend me a good model that was trained with only landscape and urban scenes or stufffff
im tired I cant avoid having people or "people" in my animations and I add all the possible negative prompts
and no luck

all that 4 have some kind of person

[sterile temple]

did you try 'no humans' in the positive prompt?

[torpid rampart]

im afraid

[sterile temple]

it's a common booru tag

[torpid rampart]

that it confuse it

but lieraly bro

etf

wtf

hahaha

promot:

digital drawing of pink backrooms, by Beeple and Jean Giraud,

negative:

(people:1.5), robots, silhouette, men, man,(woman:1.5) women, android, (person:1.5), lady, cyborg, human, humanoid robot, human shape, human shadow, girl, boy, kids, old people, young people, humans, frames, border, edges, borderline, text, character, duplicate, error, out of frame, watermark, low quality, ugly, deformed, blur, bad-artist, people walking, people standing,(girl:1.5)

lol

[sterile temple]

are you using 1.5 or SDXL?

[torpid rampart]

1.5

because is animations

so I cant in SDXL make some animations I think

[sterile temple]

https://civitai.com/models/5041/cheese-daddys-landscapes-mix

this was one of my favs from 1.5

[torpid rampart]

ok ok

even for like urban landscapes

Im making a lot of cyberpunk stuff but I have few loras for that

[sterile temple]

I used it for everything

but it can be good to mix with another model too, I used to mix it up a lot

[torpid rampart]

mmmmmmm

I think I still didnt learn how to mix them

do I need to have an extension for that?

[sterile temple]

no I think A1111 has a basic mixer

[torpid rampart]

mmm ok ok I will check it up

I think anyway I found the reason of my problem

I used by Syd Mead and Beeple and both artist have lots of art stuff with silouethes

[rain gazelle]

hysterical. 🙂 https://www.reddit.com/r/StableDiffusion/comments/172p4in/i_wanna_make_a_fullbody_image_based_on_this_one/

[rain river]

https://stablediffusionweb.com/#demo

OK does this website just not like drug use or what?

At this point I'm content to just get an anime girl doing cocaine.

[obtuse palm]

how do i make 4x ultrasharp upscaler not oversharpen the face? like .. is there a strength setting or something that will fix sharpening specific realistic shadows on face?

the sharpening feels a bit more and if you know any setting that can reduce the sharpening or any other upscale that has less sharpening with clean result then suggest me such one

[kind quartz]

You can try SwinIR. But what do you expect from model which is called ultrasharp?

[obtuse palm]

i think something can be done in here.. just need to figure out the ideal combo

[obtuse palm]

GPFGAN visibility causing color banding issue like this.. any known fix for it?

[kind quartz]

i dont see reason why use gpfgan there, her face seems to me be o.k.

[limpid lichen]

manual touchups

[obtuse palm]

how is that?

but i kinda think codeformer does better in keeping color uniformity in eyes

GPFGAN makes each pupil color different

[limpid lichen]

I'd take it into Photoshop, adjustment layer, and mask it in with a feathered edge

[obtuse palm]

Ahh Photoshop.. well, yeah

[mild jay]

there is also gimp

[obtuse palm]

Do you think GFPGAN gives the best result for face compared to Codeformer?

More natural appearances?

[wise field]

so um what did I do wrong I installed python like it told me too

[frigid oar]

picturesque nature scene with happy house pets, where the pets are enjoying the beauty of the outdoors. The scene should evoke a sense of tranquility and harmony between the pets and their natural surroundings."

[harsh marten]

Howdy y'all.. question, does anyone have a good suggestion for which model to use to get good/consistent die-cut sticker style images? For example I made this in DALL-E 3, and have installed ComfyUI locally with SDXL, but am wondering if there are other models (variations of 1.0, etc) that would be better for this kind of thing, and/or if anyone can recomment the right prompt/settings to get something this good? I just got this setup locally and my results so far are very underwhelming and I'm sure that's got a lot to do with my ignorance.

[plain tinsel]

Hi everyone

I've developed a web app called Lookr https://lookr.fyi that utilizes stable diffusion for image-to-image transformations. I'd greatly appreciate any feedback or suggestions on how to enhance it. Thanks in advance!

[plain tinsel]

https://tenor.com/view/thank-you-gif-25358104

[sterile kiln]

a biker friend who usually rides very fast, whose face I learned, I'm quite proud of myself xD
(his face is very similar)

[rain gazelle]

anyone buy anything from prime day?

[jolly scaffold]

only thing I am considering buying right now is buy an A770, wait for a 7600 XT, or maybe get a 3060 12GB

[rain gazelle]

get 'em all

[jolly scaffold]

hmm?

have any of you made any really good, moody, outdoor shots before?.. like night time stuff?

[wispy nest]

Definitely 3060 12gb if you really want to do SD

[jolly scaffold]

if RAM matters that much the A770 has 16

[wispy nest]

No like Nvidia has much better support

I am not sure where a 7600 would slot in this

[jolly scaffold]

that was 10 months ago

[wispy nest]

But you'll definitely want Linux for amd

[jolly scaffold]

i got rid of windows a while ago

someone needs to do a more updated test

[ebon zealot]

winblows suck

only use is gaming

[hoary tartan]

If you are just generating and not training my favorite is a 2080 ti

If you want to train sdxl you need 24gb so go witha 3090, 4090 or 7900 xtx

I get pretty much the same it/s for generations on my 2080 ti as my 3090 hybrid

[stoic pivot]

Hyper advance prompt smithing: "cat with hypnotic eyes".

[jolly scaffold]

I use Linux for gaming, have done so successfully, yet to find a single game that doesn't work, in the past 18 months of using Linux

[stoic pivot]

And the end result.

[jolly scaffold]

What do you guys think coming off of an RX590 GPU, SDXL, 1344x768... took 6 minutes though but... it works

[undone ruin]

Thumbnail for triceps pullback for reels skeleton muscle from back side

[jolly scaffold]

I'm looking for good gore/horror/holloween type stuff

and perhaps some kind of dark, moody, or sci fi ideas for building up some new desktop wallpapers

[jolly scaffold]

hmm, Looks like the A770 has had major improvements for stable diffusion, and it's almost tied with the RTX 4060 TI in 512x512 rendering, with the higher VRAM it might even surpass it in SDXL 1024x1024 renders

[dry crow]

codeformer is better for realism

[wispy spindle]

please adhere to the #✍🏼｜rules-and-tos @outer sandal

[desert estuary]

@wild sorrel here's my stable diffusion folder as well

[wild sorrel]

I guess?

[wild sorrel]

I have nvidia, no idea what people have to go through with other GPU's , try asking in#🤝｜tech-support , there are smarter people

[desert estuary]

alr!

[glossy herald]

skyrim logo

[wild sorrel]

I wanna see Baldur's gate 3 logo

[past pelican]

Looking more like Diablo than baldurs gate...

[glossy herald]

I went off from this :

problem is mainly the 3 in latin numerals

it gets merged and goes full on demon

hard to get the tentacles

[past pelican]

You using controlnet canny?

I suppose even with controlnet the model will do what it wants to do.

[glossy herald]

yeah it was canny, 100% strengh

but I'm not at my best

2AM30, I'll go to bed

[rain gazelle]

A classic scary movie poster for a movie called "Pumpkin Duck"

[rain gazelle]

quiet...

[jolly scaffold]

yeah, it does seem like that, and I'm going to bed.. only have a couple more days to decide if I really want to buy the A770 or not

[rain gazelle]

sounds expensive

[wispy nest]

intel gpus are really not

[rain gazelle]

i wouldn't buy an Intel gpu, personally

still a bit early in their development

[jolly scaffold]

Intel is intentionally trying to undercut AMD and NVidia in order to establish themselves in the market

[rain gazelle]

yep

that's true

[jolly scaffold]

which means they're selling stuff at prices they wouldn't otherwise be selling them at, which makes them the best bargain option at the moment

they're already just under the performance of the RTX4070, which is like $600, double the price, at least in terms of SD

[obtuse palm]

By realism you meant realistic or not? I think those aren't the same

GFPGAN has better understanding on facial features from what I've seen.. eyes, lips, teeth etc whereas Codeformer needs visibility and strength adjustments to see different different results but results are mostly unpredictable.

[dry crow]

I mean for realistic images yes. Codeformer seems to do them better. Gfpgan is more trained on Asian faces so there its better

[kind quartz]

Don't forget about drivers, which is very important part in GPU world. It is not easy. Also to fully utilized arc you have to have some intel cpu's? This is another problem. And not support for some dx i believe as well. So plenty of things to consider to go with intel gpu.
I personally would love if they are successful, others should be forces to go with prices down.

[jolly scaffold]

@kind quartz dx?.. directx?

[kind quartz]

yes

[jolly scaffold]

I'm on Linux, anything DX gets converted to Vulkan first

[kind quartz]

old one but still missing it. Majority is on windows, playing on windows

[jolly scaffold]

looks like it was DX9

[kind quartz]

yes it is, why was?

they solved it somehow?

[jolly scaffold]

ooooh.. that's what Intel is doing

it'll play DirectX9 stuff just fine, it's just doing it differently

[kind quartz]

hm anyway me personaly have not intel gpu, so definitely not way for me to go with it.

[jolly scaffold]

it's focusing on an open source D3D9On12 layer, which basically up converts DX9 stuff to DX12... that might actually give better compatibility and performance with old PC games, not less

if you play old PC games like from the mid 2000s, a lot of that is DirectX8, and won't run well at all on modern GPUs, so a common tool people use is DX8to9, which is usually a drag and drop DLL mod that converts those games to directx 9 for compatibility, so D3D9On12 sounds very similar

DXVK does the same thing but converts DX7 to DX11 to Vulkan

[kind quartz]

Better always must be native support. It is my believe.

[jolly scaffold]

and it's why retro PC games are MORE compatible on Linux than Windows

having the compatibility layer lets them patch in enhancements not originally capable on directx9, like higher resolutions, wider resolutions, multi monitors, and likely even shader improvements

[kind quartz]

and what about non intel cpu users, which is quite large group?

[jolly scaffold]

yeah, that doesn't seem to be an issue, from what I'm reading there's a feature called "Resizable BAR" within CPUs that their GPUs benefit from, Intel CPUs have that, but also, AMD CPUs do too in the Ryzen 3000 series and later

it's a feature started with NVidia, so it's not really any different than running Nvidia cards with a Ryzen

it seems like Intel is actually supporting DXVK right on their GPUs

that might actually be huge for Linux gaming

[kind quartz]

o.k. i hope many ppl find way to arc and force nvidia gpus prices down

[jolly scaffold]

yeah, screw it, I'm going to buy an Arc, either an A750 or A770, which one, I don't know, the A750 just dropped to $190 for what is essentially RX 6600 XT performance in gaming, and closer to 3060 performance for AI

[kind quartz]

dont forget as well you need good drivers to it run properly, it is nothing easy to do.

[jolly scaffold]

actually, according to a chart I posted recently, the A750 is benchmarking between the 4060 and 4060 TI for AI

the differences for both AI and gaming are minor between the two except for 8gb vs 16gb VRAM

[kind quartz]

go for it definitely. if they solved one thing, another things has to be solved same way. I wish you best with arc!!!

[jolly scaffold]

it seems like Linux will benefit from it the most.. I just need to decide if I want the extra VRAM, or something that'll physically fit better in a mini ITX build

[kind quartz]

4060 TDP is 115 W

arc twice more

[jolly scaffold]

and I have an 800w PSU

[kind quartz]

just it is difficult to compare 2 cards performing same when 1 can use 115W and other 225W.

[jolly scaffold]

that's not really relevant to me though

[kind quartz]

mini-itx you should care. about temps in case. Not sure how about cpu for mini-itx, but efficiency should be critical thing with your case format.
It is up to you. You will see.
If you have electricity for free, there is not point to care maybe temps you have to get out of case, we have quite expensive electricity. So it matter here.

[jolly scaffold]

I'll be designing my next PC case myself, and 3D printing the parts I need to build it around the components in a way that will optimize airflow, and keep maintenance fairly simple

[last jacinth]

NVlink wont work with just one bridge and both your cards need to match 100% (Just a heads up to anyone playing with the big cards)

[jolly scaffold]

nice.. finally learning how to weight out prompts 😄

[fleet goblet]

Dall-E 3 is still mind-bendingly good to me.

Despite them content blocking celebs and certain show characters, which is to be expected from a closed-source model.

[boreal falcon]

that doesnt matter (too much) for a lot of people

[fleet goblet]

Yep.

It's still an amazing tool with infinite creativity.

[boreal falcon]

it fits me kinda well as a supplement for Firefly or other pieces of software

plus vision feature of ChatGPT is amazing

[olive harbor]

can some one help me make a picture like this im new to stable diffusion

[jolly scaffold]

you could take a photo of yourself in that pose and use controlnet... you might need to get help from a plastic surgeon for the face though

[rain gazelle]

nice

[rain gazelle]

Meet the iCheese

[tough cobalt]

You can tell it's AI art with the polydactyly

[mild tusk]

behold the Lava Knight

[limpid lichen]

pride jesus

sdxl pride jesus

[wild sorrel]

wtf

[limpid lichen]

J was ace so he's part of the spectrum. he's in the +

[jolly scaffold]

what's the point of it?

by that i mean, making images that you know will only insult a group of people, then posting it amongst a group of people

[ripe bluff]

trying DALL E 3

[rustic marlin]

how ı text prompt

[boreal falcon]

Just wait till someone generates Muhammad

[crimson dawn]

it's pride month in Canada and many lgbt+ people are religious. no one can't prevent you from being insulted about rainbows and i don't think rainbows are inherently offensive

[dawn seal]

Someone give Pride Jesus a hand!

[crimson dawn]

wait till you see korean jesus

[nocturne oak]

Unit.

[crimson dawn]

chadus

buddy jesus might've offended a whole lot of people too. should kevin smith not have created dogma and instead considered people's feels?

https://tenor.com/view/buddy-christ-dogma-george-carlin-thumbs-up-gif-3486405

[lapis quarry]

I am extremely new to Automatic1111, how do I use this to make a full character turn around of a demon loli?

[crimson dawn]

isntall controlnet extension since A1111 doesn't come with it by default

[lapis quarry]

I did that. Now i don't know exactly what to do next.

[glossy herald]

I'm not running it right now so no screenshots but,
you would need to :

• activate the extension controlnet in the txt2img tab
• put this image inside the controlnet window
• select the "openpose" model, no preprocessor
• prompt what you want

you may need a special model trained on turarounds though, because this is quite specific... something like this https://civitai.com/models/17012/character-turnaround-openpose

[lapis quarry]

That is actually the one i have

[glossy herald]

then this model is the one you need to select in the "model" section of the controlnet window

check the model description

it explains how to use it in detail

[lapis quarry]

Ok, the only issue is that when i download it it gives me a zip with a folder that has 3 pngs in it

where do I put the folder or pngs?

[glossy herald]

this is not a model, it's just 3 poses. You extract the poses and use them like the instruction says. it's the step "Upload the OpenPose template to ControlNet"

[rancid bloom]

https://youtu.be/SacK9tMVNUA?si=FgG6CLlTUWDQHaij

[stable flax]

Control net pose thingy

Then use a model of your choice

And bam type in your prompt and it’ll generate

[lapis quarry]

ooooooh

Apologies, my brain not braining correctly

[glossy herald]

no problem mate

it wasn't clear to me either

[pine briar]

Hey everyone! It's been a bit. What are everyone's go-to settings at the moment?

Specifically wondering about samplers, I don't recognize a lot of these

[wispy nest]

Impossible question

Restart is very good

DPM ++ 2m Karras, that DPM 3 SDE exponential something, Euler a classic

Haven't tried them all

[jolly scaffold]

which model are you using for this?

[bold river]

[versed void]

riskme.app

[glossy herald]

https://civitai.com/models/44412/perpetual-diffusion-10

[wild sorrel]

That one is nice, I'd outpaint if I wasn't lazy

[jolly scaffold]

this is the kind of stuff I've been trying to make without much luck

and I should consider trying more SD2.1 models

or does SDXL sort of succeed 2.1?... it's confusing, they seem to be separate directions

[sterile kiln]

@lapis quarry
Skeleton images like this one are already "preprocessed", or made by hand by someone, but if you select "Openpose" or "full" in preprocess, you can add real photos of characters to have a rendering in skeleton.

My android 18 gen 🙂

[soft orchid]

Logo for Nail art called ivaz

[sterile kiln]

Hey @rain gazelle , still working on png-sd-info-viewer?

[stoic pivot]

I have downloaded so many models and Loras, well over 100 Gb, but I often come back to the Stable Diffusion XL 1.0 without any addons and bling bling, and I have started to as for images rather than demand them 🙂

The prompt was: "Make a cool image of a realistic Hello Kitty like character in a buzzling digital metaverse and make her wear clothes that you think fit the teme, then frame the image in a cinematic way to make her look like a cool space hero. Try also to add some film grain, lens flares and blown highlights with inspiration from the movie TRON"

[heady meadow]

it works with a full sentence?!

i thought u needa make it point form

[stoic pivot]

There is so much "magic words" in AI that folk use just for they seen other use them, "ultra realistic", "best quality", "8K HDR PMS", "RAW", those seldom do what folk want them to do. You can ask AI, use it in point form and also make it like a script with "Subject = cat" and so on, AI often find a way to do something of whatever you throw at it. Just play around and find your way.

[heady meadow]

ohh

[soft ginkgo]

Prompting is a bit of a cargo cult, CLIP can understand a simple description fine. everyone just uses booru tags/copy's someone else prompt so stuff ends up looking like a mess

https://media.discordapp.net/attachments/930499731451428926/1033538575330971708/Untitled.gif

[heady meadow]

what do u guys think? i think the background is kinda awkward :(

[heady meadow]

icic

[stoic pivot]

"draw an sketch of where the old pirate treasure was hidden, I think it was on some tropical sand island with palm trees or so."

[nocturne oak]

Modified your prompt a bit and put it through some different models to get some interesting variations.

[stoic pivot]

Neat.

[rain gazelle]

Hi, it' s still active. Version 5.7 was released in August: https://github.com/GarlicCookie/PNG-SD-Info-Viewer

I also released another image program with a slightly different purpose. It loads thumbnails so you can quickly mouseover and view prompts. (And copy to clipboard for quick use) - https://github.com/GarlicCookie/SD-Quick-View

PNG-SD-Info-Viewer is a little more feature rich. Not sure what the last version you used was. I think the last thing added was drag & drop.

[jolly scaffold]

that could be a fun little exercise... having everyone use the same prompt, any other settings or models, and see who can come up with the best, and how

[jolly scaffold]

hmm, now I wonder if there's any metadata plugins for comfy

there is 😄

[sterile kiln]

Nice,
I updated to 5.7 recently. I check often 🙂
I wanted to ask you.
We often search for images via the file name, it would be nice if there was this function, like in Explorer. Example : if I have pictures of Kirby, I type Kirby in the search field and it brings them up for me. This would avoid repeated Alt+Tabs!

[limpid lichen]

is something like ip adapter possible for LLMs? like, could we have a conversation with a LLM that would plug into the diffusion without tokenization?

[rain gazelle]

Not sure I really want to replicate all of the search features when they're already integrated to the OS. I think Explorer is going to handle that better, anyways, since it ties in with Windows search indexing.

Searching through the open folder for prompts with a certain word could be interesting....

[limpid lichen]

Sonic and Mario, high five, nintendo, mushroom kingdom, Super Mario neg: dead, deformed, corpsed, demented, broken, trash, jpeg low quality, compression artifacts, lossy
hjahhahahaha

okay dalle gets this one slam dunk.

[jolly scaffold]

check this out.. automatic masking of the foreground character, with automatic reconstruction of the background 😄

and the reconstructed background takes into consideration the size, shape, and features of the masked off foreground character

[wild sorrel]

idea is good, result tho...

[frozen nebula]

Been trying to find a good place to buy a used 3090, tried ebay got burned on a bad card (return accepted luckily so didn't lose any money but still) are there any reliable ways to buy a good used card or is it just take your chances on ebay and keep trying till you find a good one?

[jolly scaffold]

@wild sorrelreferring to my post?

[wild sorrel]

yea

[jolly scaffold]

oh, I just wanted to do something generic so I could do two distinctive enough looking backgrounds to see how well it built the second background around the character

[jolly scaffold]

i don't know why, but I really like visual scripting languages which is sort of what Comfy is

[serene flint]

amazon

just bought one from there myself

am verry happy

[jolly scaffold]

i think I'll be fine with the A770 I got, it'll still do SDXL faster than what I have now will do SD1.5

how much VRAM?.. I've been doing SDXL stuff on 8

8gb ram or vram?

comfyui has lower VRAM requirements than automatic1111

[wild sorrel]

isn't SDXL even on auto1111 is ~8gb vram?

[jolly scaffold]

it'd be hard to do on automatic1111, but I've gotten SDXL to work fairly well on 8GB VRAM using Comfy

it does take me 6 1/2 minutes per image though but that's because i'm using an RX590

i bet it'd work pretty well on something like a 3060

i even thought about just upgrading to a better 8gb card for SDXL, but, there's other stuff VRAM can be beneficial to, like larger batch sizes

[serene flint]

from the day i started using AI tools until a couple of weeks ago now, I used exclusively a 2060 6GB

[jolly scaffold]

AMD decided not to support ROCm on my GPU so I have to use an older version of ROCm and torch.. i hear more updated versions of some of that stuff has some noticeable speed improvements too

[weak hollow]

what is batch size

[weary scarab]

hi everyone, i'm trying to ID the model for the animation i took this frame from, thanks in advance!

[silk tendon]

Amazing what people create without censorship

[hearty karma]

Flaming skeleton on a motorcycle go brr

[jolly scaffold]

for sure... I asked bing to show me an image of a girl without eyes, all body horror like, and actually got my account suspended for that

[honest talon]

goes unbelievably hard

[silk tendon]

skill issue

it focused too much on the sky and made the kissing couple a blob

didn't figure a way to make both high quality

[frozen nebula]

detail inpainting

[silk tendon]

there had to be a compromise (close up)

[honest talon]

pablo

[stoic lotus]

hello everyone,
on the verge of releasing CCXL-Photo
so please help vote on the version you prefer with 1️⃣ or 2️⃣

[hearty karma]

Planes :)

[glossy herald]

Cow 🙂

cowplane

[jolly scaffold]

Finally started cleaning up my workspace

the lines leading down are headed to an inpainting specific section

[eternal lance]

having an organized workflow is the work of the devil

[jolly scaffold]

i learned it from my time workign with Unreal Engine, using their visual scripting language, without organization visual scripting is a nightmare

my inpainting section

[eternal lance]

blueprints? I can't imagine being any more organized there than I am in comfy

[jolly scaffold]

in blueprints for Unreal, I can basically save one of these entire sectioned off groups into a single node that I can plug in places

[jolly scaffold]

but yeah, inpainting is actually pretty easy with Comfy.. debating if I want to use the plug-in from the main checkpoint loader, or if I should consider some inpainting specific models

something I really like about comfy.. let's say there's a project I'm working on, but I want to take a break from it and move on to other stuff... well, I just clone the text node and stick it off to the side somewhere, prompt saved

[jolly scaffold]

oh... I just realizd my wyvern isn't exactly a wyvern, it's a dragon, how rude

[eternal lance]

you can do that in Comfy as well

Nested nodes builder. It has some quirks

[glossy herald]

the biggest problem I had with comfy (don't think it has been solved) is that you can't import/copy paste nodes from a workflow to another

it's not a problem if you want nodes from only 1 of your workflows

but it prevents you from building a library of base logics to put here and there

[jolly scaffold]

not sure, I'm fine using just one screen for now, and sectioning things off that I can disable and enable if I need to do different tasks

[glossy herald]

like, I spent some time making a randomizer for the prompt based on dictionaries. took dozens of nodes. I would have loved to use that system in lots of workflows to replace the basic "prompt"

[jolly scaffold]

i basically made an automatic character mask

I've been trying to find a good way of doing gore and horror stuff

[glossy herald]

mojo tree got old

[eternal lance]

What? That's like the primary feature of comfy

You drag and drop a png with metadata from another comfy user and it automatically imports their workflow and settings...

Additionally, you can easily save workflows as .json

[glossy herald]

yes, but if you take a workflow of someone else for example, and want to take some nodes from one of your workflows to put in there, like some post processing logic or anything

dragging your workflow inside will just load your workflow

[eternal lance]

just copy and paste your nodes

ctrl click all your nodes, copy and paste

[glossy herald]

it's now possible ?

[eternal lance]

it's always been possible

[glossy herald]

well, I missed it last year when I discovered it

thanks !

[eternal lance]

np, enjoy

[rain gazelle]

Opening up a new large convenience store.

[sturdy pewter]

https://youtube.com/shorts/h24nAtOlHdQ

[crude sable]

Img2Img + GIMP to stitch together elements from different gens and also to make the text (which I ran through canny to get the texture)

This is the original txt2img

[tropic remnant]

beby

[weak hollow]

Where does the beer stand on lol

[dusky swan]

Yeah.. I’ll never be able to understand the dumb ass wizardry that goes into these prompts

[cosmic lion]

#🏞｜general-with-images nami

[naive rain]

[mild tusk]

yes

[weak hollow]

what is this stable diffusion XL ?

[weak hollow]

Nice colors

[mild tusk]

thx

[lofty moon]

Nice.

[serene flint]

First person to reply to this message gets a free high quality logo of their choice

[wispy spindle]

i win

jk, i'll let someone else take it

[serene flint]

Too late

[serene flint]

Give me a theme or style

(I'll give someone else one too 😉)

[wispy spindle]

fruit made of pastel metals

soft colours

[serene flint]

Here's some examples

Roger that

Stand by

[raven mirage]

I'm really liking this style. Are you using ComfyUi or Automatic?

[serene flint]

Okay I made 2 without text

And 2 with

[wispy spindle]

CUTEEEEEE

i love the hearts

[serene flint]

I know they came out so cool

[wispy spindle]

i'm gonna make those folder icons on my desktop so cool and cute

ty!!!

[serene flint]

Awww, the dream! You're welcome I'm so glad you enjoy them

If anybody else wants one just gimme a ping

[raven mirage]

Love the look. That metallic on the apple looks quite fantastic.

[serene flint]

Yeah that was actually a really neat idea

[zinc maple]

I did something pointless for fun

[silk tendon]

I'm using dall-e 3
Just don't tell Microsoft I'm making this

I'll dm you the prompt

[raven mirage]

Thank you!!

[weak hollow]

Your Bing account or Midjourney account ?

Does bing have image generation ?

[jolly scaffold]

bing does have image generation

[weak hollow]

Got your whole Bing account suspended or does it have an extra account for that image generation ?

[silk tendon]

I challenge anyone for a more over the top image than this

😵‍💫

[boreal falcon]

whats the prompt actually?

[silk tendon]

Check dm

[wispy nest]

What model

[silk tendon]

The proprietary one

[wispy nest]

Dalle 3

[silk tendon]

[silk tendon]

Someone should send infiltration specialists

For science

[wispy nest]

Yes

[boreal falcon]

With censorship they kinda lost some customers

I guess at least

[limpid lichen]

dalle, midjourney, all the others, they've all censored the entire time, but are huge

no other AI service even competes with MJ. They've carved out such a huge market segment that is basically untouchable

the idea that censorship is holding them back is just dumb. honestly. dumb.

[jolly scaffold]

it was temporary, and some of the prompts I used successfully for horror shots in the past are restricted now, so, screw Bing

[boreal falcon]

Honestly, they dont have to

[austere leaf]

Can I get some quick help with a LoRA model? I'm incompetent when it comes to a lot of technical things so I think less than five minutes of looking could give me the information I need...

I have a LoRA model trained and created with CivitAI, linked below I have attempted to run it on TensorArt for multiple hours now, with no results. I am repeatedly getting the message when I try to run it in TensorArt that "the format of the model file is wrong, needs to be re-uploaded." I've checked everything and have uploaded both the zip file containing the training data and the LoRA model itself - it seems they go together to get uploaded. I don't know where to go or what to do to get this to work. It seems to be a formatting issue but that's the extent of my knowledge. Please help.

https://tensor.art/models/650010538509815933

https://civitai.com/models/166731/ellie

[boreal falcon]

OpenAI? (Chat)GPT is what is the main "money maker" for them. Google? Gemini soon and right now they have not to worry about their image generators might not beat the competitors. Adobe? They have a specific target audience and even then its not even nearly the main money bringer and product of them although they heavily focus on it right now

Stable Diffusion has other products as well and not just Stable Diffusion

[jolly scaffold]

it annoys me they use the word "open" in the name of their company/project

[boreal falcon]

Well they used to be once upon time

[austere leaf]

Then they realized they could profit enormously off of it and put it behind a paywall...

[jolly scaffold]

at this point it's just false advertisement

[boreal falcon]

It paid off

[kind quartz]

why not use local installation of llm. Lots of fun i must say. Yesterday i laughed very. It is big fun.

[weak hollow]

Can you install Dall-e 3 model on SD

[limpid lichen]

100% no

[limpid lichen]

openAI is basically owned by microsoft. their tech is built into tons of microsoft services now and runs on microsoft datacenters. it's only a matter of time before MS acquires them wholly.

it's like youtube before google bought them

[boreal falcon]

Maybe but whether or not, DALL-E isnt thair major product

[limpid lichen]

dalle is restricted to personal use only. at this point, microsoft is treating it like solitaire. solitaire was only included free with windows to teach people how to use the mouse. thats why bing and windows are going to provide openai's services, to teach people how to interact with ML systems

could call it a halo product

https://www.investopedia.com/terms/h/halo-effect.asp

[glossy herald]

bacon smoke

[boreal falcon]

Actually you can use them commercially

[limpid lichen]

actually???? https://www.bing.com/new/termsofuse?FORM=GENTOS

and unless you're uploading your pics to a non profit or personally hosted website, it's out of scope. uploading to facebook, google, discord, any commercialized service gives the image commercial purpose. Microsoft could DMCA each one if they wanted

[boreal falcon]

Not Bing, ChatGPT one

Directly from OpenAI

I expected the Bing one to be limited anyway

And they are

They are now at 15 prompts per day lol

[limpid lichen]

i use my microsoft account that i've had forever and it was up to 1000s of credits last i looked. some accounts i guess get more

i used a ton

[boreal falcon]

They limited it hard now, what was the last time you looked up?

Maybe you spend those reward points as well

For me technically those 15 credits per day are enough

For what i need them

[limpid lichen]

dalle3 isn't offered through openai yet. no indicators showing that it'll allow commercial use. they caught a ton of shit for that with dalle 2 services and it's no surprise they haven't rolled out 3 yet

[boreal falcon]

Wdym its not offered through OpenAI yet, i used it today via ChatGPT

Yes, im in EU

[limpid lichen]

you were using dalle 2

their website says it's releasing premium in october, but no terms are up about it yet. microsoft is sticking to their guns that they're not even touching commercial licensing. i think they've got issue with where the data set came from for it. the rights holders might've shown up saying "well hold on here one second" and they're doing their best to cooperate with rights holders going forward

[boreal falcon]

No im using D3

Even in ChatGPT app

It may be not rolled out for everyone yet tho

[nocturne oak]

ooh...new toy

[boreal falcon]

And vision one is on GPT 4

[nocturne oak]

I have it too

[boreal falcon]

Its good possible that i will unsub again tho

[rain gazelle]

Dalle

Crazy stuff

[limpid lichen]

yeh it is crazy. just wish the weights or any of the research behind it even was published. character meetup i did had that white split style too

[marsh wedge]

anyone knows how can i generate a reference sheet from the front view with sd? i only need a side view and if possible the back too

[dry wagon]

Trying to create a Halloween background with img2img from the bare 16:9 room. The two Halloween versions are XL but squished from playgroundai. Does anyone know good img2img settings to change room enough for Halloween but keep the layout close? Or if you're a legend and help take a crack at the bare 16:9 room, all I need is to change the purple lighting to a green with Halloween decorations. My 1080 gets crippled with XL, so any help would be great. This is for my YouTube videos. This was a quick test for a talking head format I'm exploring and using AI images in my work. https://www.youtube.com/watch?v=40PUF-gG_SkThank The videos I'm working on aren't political or in politics that was just the popular topic when I made it. Working on a younger generations vs scams and a film review video that's why I need a Halloween background. This became long to read, yikes, if you read this thank you and any help would be great!

[glossy herald]

who will find the trapped pizza slice ?

[wispy nest]

you could train a lora

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

or use this i guess

https://www.youtube.com/watch?v=EW2MAaNhZJA

just dont do it like this guy with a static background

or else the lora will also add that

[jolly scaffold]

I think I am getting closer to what I want

that one needs work... but still 1152x896 SDXL doing 100 steps in about the same time as it took me to do a single 512x768 with only 25 steps before

[rich coral]

#1010934719455707218 🏁｜start-here

[empty wasp]

Do you think a 4k image is possible for a monitor wallpaper?

[dawn seal]

Ok, so I'm showing dalle-3 when I look at my options, but it won't just let me give a picture prompt. How do you access it?

[serene flint]

you have it selected? the just ask GPT to make whatever you want

It basically acts as the text encoder

@native ginkgo

[dawn seal]

Ok, I didn't realize I had to select it.

[serene flint]

It's a phenomenal model, enjoy

[gray falcon]

Is GPT4-V available right now?

[serene flint]

What is that

[warm needle]

Why doesn't it work when I'm playing Diablo 4?

[gray falcon]

this might help: https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions

[gaunt heron]

... because Diablo4 uses your GPU

[glossy herald]

lol that was a funny question

[teal zephyr]

Hey, I made this mech photo and I was wondering if theres a way to easily turn it into something even cooler with SD, like change the background but keep the mech same?

[glossy herald]

sure

(starting SD to try)

let's go epic

[teal zephyr]

I'm new to this so I don't even know the possibilities.

Nor know if its easy

[glossy herald]

well, there are multiples, here are my first thoughts of "how to do it"

you can use img2img, a way to modify a picture, and draw a mask on top of the robot so the AI just changes the background. This is not that great usually in terms of results, but with some good mask, you can do great

the second option is to use the classic txt2img, and a "controlnet" on top. The controlnet will help extract some parts of the picture, like the robot, and force the new image to be generated folowing those extracted features

I'm going to try the second one. it may change the colors on the robot, but it should render cool. if not, I'll try a mix of both

[teal zephyr]

Hmm interesting, sounds trickier than I though.

I saw some background removal tools online so my first idea was to use that go get the robot on a transparent bg, but then use SD to generate a background of any kind - like big explosions - and then just paste it on.