🏞｜general-with-images

[modest niche]

@burnt flower

some new tries

[burnt flower]

looks amazing!

mind generating some space related stuff when you’ve got some time?

while I set up the model

[modest niche]

sure!

[still flax]

lawsuit incoming

[vague oasis]

so no lawsuit against OpenAI, probably cause he know he will lose against them

"collage tool"

[grizzled sage]

github pilot doesn't use a diffusion model for creating code, and it often uses code blocks from trained data verbatim. that lawsuit actually has a lot of teeth and microsoft stepped in a beehive there

it's unfortunate because people will think that one sets precedence for all AI systems

[wispy nest]

"I'm a little teapot Short and stout. Here is my handle, Here is my snout."

[sharp hamlet]

What prompts did you use to make the lego ones?

[sterile kiln]

https://www.youtube.com/watch?v=NmJSQn0dui4

[pine briar]

I wonder if SD is considered “remix culture”

[worthy stag]

ai

[sharp hamlet]

There must be at least a few people who are salty and jealous about Stable Diffusion’s success 😂

[pine briar]

I think the dislike is totally understandable and I would probably be commissioning art for this project I’m doing if

a) I wasn’t a broke high schooler

b) There weren’t about 100 images I needed.

[fiery carbon]

Hi. Newbie here. I am training a dreambooth model with rendered images from an 3D engine.

[boreal crown]

thats like saying a paintbrush is a collage tool connected to your brain.

[pine briar]

Good luck!

[boreal crown]

recite all the states in order, and don't be collaging them from memory of this copyrighted map.

[sharp hamlet]

You might not be the only one who sees price and quantity as benefits of ai art

[sharp hamlet]

Looks good

Are there any easy-to-use mobile apps on iOS based on Stable Diffusion technology?

[fiery carbon]

The idea that comes to my mind is if SD can be used across the network from a browser maybe on iPad.
I mean, the IPad doesn't have the power to run SD, but it could connect to your computer using the web browser. But i don't know if that is possible.

Another favorite so far

Getting so many nice training images, that is just amazing

[hard wagon]

New embedding glitch is out now on civitai and hugging face.

https://civitai.com/models/4511/glitch-for-sd-2x

[dry crow]

yes, its called Draw Things, here is the link:
https://draw.nnc.ai/
Just click on on the mini link (take me to Appstore) no need for phone number

[sharp hamlet]

Thanks

[hard wagon]

https://huggingface.co/joachimsallstrom/Glitch-Embedding

[fiery carbon]

I think she is beautiful

[silver solar]

3d dart

[wispy nest]

Is there a place I could find some well engineered prompts I could give as example to ChatGPT?

[hasty nova]

something like lexica would work, or I just found this cool script on reddit and it made a really good prompt that I made an amazing image with https://github.com/526christian/SD-Prompt-Generator/blob/main/promptgen.py

hope that helps!

[wispy nest]

Oh wow that's exactly what I needed

❤️

Much Appreciated

[hasty nova]

No problem! It's much faster than Chat GPT, and really customizeable. I barely know python but by the looks of it I could edit it really easily (though I do know how to code)

Very simplistic though.

[wispy nest]

I really needed the list of all the styles and etc. I'll try to force ChatGPT to combine them in an Epic way

[hasty nova]

Ahhh

I see

[wispy nest]

Gosh I love Ai 😄

[hasty nova]

I hope it works out for you!

[wispy nest]

Never ending excitement

Thank you ❤️ have a wonderful time!

[solar ember]

whats the best tool to use ?

[hasty nova]

tool to use for what? SD in general?

[solar ember]

yeah

[hasty nova]

Most people use Automatic1111's WebUI, but there are others

I assume you mean to run locally

[still flax]

WIP

[boreal crown]

looks neat

[dense tapir]

https://discord.com/channels/1002292111942635562/1064045745432444999

[crimson flower]

Accidently generated something cool with Protogen.

[gritty trellis]

Well... hello words.

[grizzled sage]

geeze oh geeze!

GEEZE

[crimson flower]

Whoops, accidently found Google's supercomputer!

[tawdry geode]

somewhat got decent eyes, and the start of a facial expression. Maybe ill work on it someday

or just face mask everything for my sanity

[velvet vessel]

Look at what these guys are doing now https://twitter.com/JonLamArt/status/1614394690572517378

Jon Lam #CreateDontScrape (@JonLamArt)

Today is a good day, Artists. We came together against being unethically scrapped and exploited.

The article below is from the Joseph Saveri Law firm that has filed a formal complaint in the Lawsuit against Stable Diffusion, Midjourney and Deviant Art!!
https://t.co/od2Bd6mVv8

Likes

10076

Retweets

2882

Twitter · Jan 14, 2023, 10:51 PM

[thorny zephyr]

The actual amount of influence a individual artwork has on the input would probably be worth like 0,001 cent

[sharp hamlet]

What’s that?

[open sun]

we're discussing that in general chat, just dropped it here as a showcase.

[left stone]

Hey folks, I am fairly new to Stable Diffusion but not AI art. I am struggling to get something similar to a "Drider" from dnd. If anyone has any tips for promptcraft or model suggestions i'd be keen to hear it. (Pic attached for what i'm going for.)

[steep zodiac]

The lawsuit is riddled with inaccuracies and straight up lies lol

[hard pelican]

I got an update this morning and now the generate button doesn't reset when done with an image or when I interrupt or skip. Images also don't preview when being generated. Anyone else getting this or know how to fix?

[vast cape]

I didn't update cause I'm lazy)

[rugged pendant]

new mix

[rugged pendant]

1.5 still stronk

[vast cape]

Well, Yea)

[violet saddle]

https://www.saverilawfirm.com/ai-art-generators-copyright-litigation

I guess it's going to take a while

[static tusk]

https://www.artslaw.com.au/article/i-like-your-style-part-i-copyright-infringement-or-not/

from 2002. says it all

[steep zodiac]

Has anyone dug into the Joseph Saveri Law Firm's past yet? They to have multiple lawsuits against tech companies going on right now, and they were the ones that the GME meme stock crowd used to sue Robinhood for blocking them from buying the stock

They seem to do a lot of anti trust stuff

https://www.saverilawfirm.com/our-cases/

[vast cape]

Well, WLOP module works)

[violet saddle]

That doesn't cover the use of artworks for training an AI, I assume it's the angle they are going to take

[static tusk]

yeah the AI factor is all new so who knows

but in the end its still use of a style, which technically isnt copyright

[violet saddle]

I don't think it's going to be about the images generated by the model but how the model was trained with copyrighted works without permission

And I don't see it as an easy problem, it's a new concept altogether

[feral stratus]

so, i have been messing with the Easy Diffusion webgui on my gaming computer for weeks... all my pics still suck. First time trying midjourney, and the pic looks 10,000x better than anything i have been able to create with easy diffusion - and on the FIRST try. What am i doing wrong?

[still flax]

using easy diffusion, probably a normal model and basic prompts maybe?

[hasty nova]

My first embedding just finished training! Now I have to try out the 4 different ones at different step counts to see which is best, I hope it turned out well.

[still flax]

WIP

[still flax]

noice

[feral stratus]

Is there something you can guide me with so I can see and realize the potential I’m missing? Here’s my werewolf. The MJ ones look like pure artwork. This looks horrible.

[still flax]

what model, sampling method, steps and CFG scale are you using?

Im gonna try your prompt

[feral stratus]

That info is under my image.

[feral stratus]

Thank you!

[still flax]

idk something like this? Im using a cartoon model

kinda out of view wait

Guidance scale: 50

bruh

Ok so Id recommend using the WebUI and read about the options

[feral stratus]

Ok. Apparently I need to download/use new models and study my settings

I am using the webgui. I like it.

I just bookmarked the settings explanation earlier as well

[hasty nova]

welp it failed miserably, oh well.

It's at least nice knowing I can train locally now.

[still flax]

something less cartoonish

I could work to get better teeth

😄

[violet saddle]

First try, not too bad but lacks details around it imo

Second one

[feral stratus]

Damn. I really have to learn the tweak the settings. Everything you guys are sharing is epic.

[violet saddle]

highly detailed photograph of a menacing werewolf showing teeth, realistic portrait, realism, Canon
Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 1430176969, Size: 512x704, Model hash: 0aecbcfa2c, Model: dreamlike-diffusion-1.0, Batch size: 8, Batch pos: 0

That's for the first image in the first block of 8, the others are the seven next seeds

highly detailed photograph at night of a menacing werewolf showing teeth, realistic portrait, realism, Canon
Negative prompt: 3D, glossy, fake
Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 3407230357, Size: 512x704, Model hash: 0aecbcfa2c, Model: dreamlike-diffusion-1.0, Batch size: 8, Batch pos: 0
And that's the first image of the second block, keep incrementing the seed for the others

The last block of 4 is from img2img based on a similar prompt

[feral stratus]

Wow! Are any of those models ones that you had to download separately, or are they all included with the web gui already?

[violet saddle]

Downloaded separately, I also have bad results with the vanilla sd models

The one I use is this one https://huggingface.co/dreamlike-art/dreamlike-diffusion-1.0/tree/main

[grizzled sage]

some of these merges aren't that bad. here's a good one of the standard deloreon mecha test i do

dreamshaper its called

[grizzled sage]

another i coaxed out of it.

[boreal crown]

That's kinda dope

[worthy stag]

🤓

[wispy nest]

If I merge checkpoints do I need a yaml file for that new checkpoint?

[dry crow]

Yes if the checkpoints are based on 2.0/1 and have a .yaml file

[boreal crown]

oh god dont merge 2.x and 1.x

[wispy nest]

I'm merging 2 and 2 but don't know how yaml files form

and gotta go to sleep 😄

[boreal crown]

dis u?

[dense tapir]

Not me

[outer spear]

Why does my output seem to get darker whenever i use a low denoising strength? (Im using a loop back here so its just progressively getting darker)

[dry crow]

Are you using a vae?

[outer spear]

i think so

i think its built into the model or something?

[outer spear]

Okay setting the vae option to none still has the same effect

[dry crow]

Which model are you using ?

Normaly you need the right vae for the model (only if the model needs a vae)

[outer spear]

anythingv3 + animevae

[dry crow]

You should use the anythingv3 vae

[outer spear]

thats animevae

[dry crow]

Oh okay xD

I will try it later when im at home (loopback)

[outer spear]

well ive noticed in regular img2img aswell, its just more visible on loopback because its happening repeatedly

[errant yoke]

ultra widescreen generations are fun

[hasty nova]

I used soem stable diffusion to help make an enemy for one of my games, what do you think?

I used the pxl8 model to create the base image which looked like this

Which it then turned into this

Which I then turned into the first image.

[violet saddle]

Maybe add texture to the torso but otherwise that's pretty cool

[hasty nova]

The torso has some texture, the little lines that are slightly lighter.

[still flax]

really cool... Are you using the allinonepixel model?

[hasty nova]

no, the one that comes with the pxl8 plugin. It's not great tbh, but what the things turn into is really nice.

I am using img to img from some enemeis already in the game to keep the same colors and style mostly.

[grizzled sage]

That new pxl8 model looks so powerful for pixel art. I am going to get it set up today to play with. cool that you can so easily create new game assets now. Production ready really! nearly anyways

[hasty nova]

It's nice, but the model itself doesn't make great pixel art, rather the plugin does. It's still a nice model to have I suppose.

Look at the difference between the two.

[grizzled sage]

whats the plugin? is that for an editor

[hasty nova]

It's the pxl8 extension for Auto1111

My bad, I meant extension/script not plugin

I'm just dragging the pixelated stuff into aseprite and downscaling it

Though it can still do some nice stuff

[grizzled sage]

AH i haven't started playing with it yet so i didn't realize it has other tools in the ui. i know i had to add a script

Reminds me of old color cycle animations. these things demo'd on this page http://www.effectgames.com/demos/canvascycle/

[hasty nova]

oo those are pretty

With a little work you could probably create something like those with this tool

[grizzled sage]

i learned the basics of how to create color cycling animations back when my school had amigas. it's a neat very technical art form

[hasty nova]

I feel like this looks too noisy...

Eh it's probably fine

[grizzled sage]

i knw there's some pixel artists who absolutley hate pixel filters, and i'm sure this model is going to face a ton of their critiques too. It does seem like it just adds random noise for details much like the last generation filters would

[hasty nova]

the random noise is probably an artifact from the color pallette picker thing, if I were to add more colors it woulndn't be as noticeable.

That's probably fine

I think it will be the boss for the game, so it will like summon stuff I guess.

I guess the best part of using this tool is getting a good natural shape for things, and fixing up the detailes yourself.

[grizzled sage]

Here's my first neat result. i used "reference pose" in the prompt and it did this.

[hasty nova]

Oh that's neat

I got these with a little trial and error

[grizzled sage]

very niiice

[hasty nova]

thanks!

Imma try to get a lighthouse

idk why

[grizzled sage]

it does scenes very well

[pseudo aspen]

Generated with intel HD graphics 4000

[still flax]

O_O

i-integrated g-graphics...

noice

[pseudo aspen]

Yes. Torture

[still flax]

poor thing is doing its best tho

[pseudo aspen]

My ThinkPad screams in agony with this. But it's fun

[pseudo aspen]

Yeah

[still flax]

cool Ive a thinkpad also

old t420

[pseudo aspen]

I have 2012 era E530

[grizzled sage]

nope. It was generated with the cpu. the igpu i don't think has a ml library to execute the model with

[still flax]

thats what matters... Having fun

[grizzled sage]

cpu only generation mode is still super valid. i dont mean to bring it down. excited for you. the future is now

[still flax]

yeah I was thinking that also... The main page for the WebUI says only AMD/Nvidia + Intel HD graphics dont have enough vram

[open sun]

wait 1 year and this technology will be as smooth as running blender, looking back to first versions of 3DS max, comparison is quite close 😄

[still flax]

did you consider colab?

[pseudo aspen]

Wym

[still flax]

using google colab like

https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast_stable_diffusion_AUTOMATIC1111.ipynb

[open sun]

runpod is also quite cheap, i train around 50+- embeddings with 10$

[still flax]

yeah its like SD WebUI in the Cloud, for free

youd get much better speed

[pseudo aspen]

No, I will check it out. Thx

[still flax]

👍

[red plover]

When I launch Automatic1111 (Win 10) now it gives me this error...it's been working fine lately... ideas how to fix? I did a reinstall (Git cloned, as I usually do to update it) and it still happens... 😦

[still flax]

what error

btw you can just click the .bat files

[red plover]

That is what is causing this. I just launch webui.bat and get this instead of it working like it used to work...not sure why.... (see my screenshot)

[still flax]

get what theres no error in the screenshot

[red plover]

It won't open. That IS the error. (not working), but I think I fixed it. somehow I had a git clone command in the webui.bat after my set COMMANDLINE_ARGS=--xformers line before my @ Echo lline.

It's installing/cloning stuff now so it should work.

[still flax]

tbh Im not sure what youre doing. Parameters for the webui should go in the webui-user file... No need to keep everytime parameters. Then just double click the webui file. I think call starts the program in parallel so maybe thats why no terminal output

[bitter spruce]

man, i'm higher than bird pussy. Chillin, making art n listening to Palmer Squares

https://tenor.com/view/bob-ross-pointing-nod-gif-13127635

How you all doin?

[red plover]

I think that was an older webui.bat I had saved as a backup before updating, and it was one that had that line in it from a YouTube video to auto-update the installation every time you launched it...apparently that doesn't work....I'll move the parameters (which is only the --xformers line to that other one, but it seems to work in either.

[bitter spruce]

I'm using invokeai's web-ui. works great.

[still flax]

hmm, do you think its better than auto1111?

[still flax]

fine... Training a model

[bitter spruce]

yes

It also fully works on AMD hardware

https://media.discordapp.net/attachments/1048866961670549566/1063706790719660052/drug-anime.gif

give it a try

[still flax]

hmm noice... Im checking their repo

thx for the info

[bitter spruce]

i censored the picture

but yeah, shit works great

remember, there are no mistakes in art. just happy little accidents 🙃

[still flax]

heheh ok bob ross

the questions is

which one is better?

auto 1111

or

invokeAi

[bitter spruce]

My answer is bias

I have AMD hardware

InvokeAI

just made her

[still flax]

but you cant use invoke+AMD on Windows, the main page says you need ROCm to use AMD. ROCm only works on Linux... So you must be on Linux and therefore can also use 1111

[bitter spruce]

I don't use windows

[still flax]

thats makes sense

but you can also use the 1111

[bitter spruce]

last time I touched that OS was win3.1, I still hate it, XD

I use lots of things, stick man

relax

[still flax]

O_O

[still flax]

but like I think this would imply you cant use 1111 with AMD

[bitter spruce]

Correct

[still flax]

yeah but thats what Im says, you can

[bitter spruce]

I can't, others might, idk

[still flax]

card?

[bitter spruce]

no, I can with InvokeAI

6700XT

Ryzen 5 3500G

[still flax]

for sure you can

Ive a 6600 XT

just use the nightly pytorch

[grizzled sage]

linux was hardly functional at the time of windows 95 through xp. but i'll believe you still. there was apple os still i suppose. and amigas.

[still flax]

instead of the 5.1.1 one

[bitter spruce]

when I try 1111, it's stuck in cpu mode

when i use invoke, everything works

[grizzled sage]

i used a1111 with a vega 64 for months on garuda linux

[bitter spruce]

but that's just me

[still flax]

yeah ikr I used garuda before

[grizzled sage]

it took a little tinkering

[bitter spruce]

everyone's PC is different

that's why I view tutorials and the like, as suggestions

and I ad lib alot of shit

but I'm also confident and not afraid to edit code

it's just code

if I fuck it up, delete it, and start again XD

I don't like garuda

There is no reason to hook timeshift into a btrfs fs

btrfs already has built in snapshot tools

so..... if timeshift fails, so will the entire backup functionality of garuda

(been there) even got in there forum to tell the dev this

they werent intrested

¯_(ツ)_/¯

[grizzled sage]

its the bloatiest of the arch i've heard. and i wanted to explore all the cool new systems available for nix. i found lots of good tools in their bloated kit. btrfs was one of those discoveries.

I'm pretty sure the garuda backups tie into btrfs' own snapshot abilities

[bitter spruce]

oh hey wassup man

i didn't realize it was you

[grizzled sage]

its always me

from my pov anyways

[bitter spruce]

tbh i didn't even look who i was talking too lol

[grizzled sage]

just chatroom life. even more if you run discord in irc mode

[bitter spruce]

or just irc from the terminal

https://youtu.be/daILBURBIac?list=RDEMWLlK8AdgSmJoMz8zcYJEhQ

Here's some word art, for your ears

Enjoy

https://media.discordapp.net/attachments/860627583870566422/914291256530714694/temple-os-terry.gif

https://media.discordapp.net/attachments/1048866961670549566/1063458534416732210/puff.gif

https://youtu.be/XLbiZSbCBVE?list=RDEMWLlK8AdgSmJoMz8zcYJEhQ

Hope you all are having a good day

<3

mmmmm, yeah. I like using InvokeAI Xp

nothing I did looked this good or worked this fast when using automatic1111

[grizzled sage]

i couldn't get it to install. all it's requirements were failing. the install wiki's instructions didn't match the files in the folders. i'm sure i could've wrestled the install into submission but i had it working on another instance already and went there. when i finally migrate to docker, maybe i'll try to make a docker image for that too

i read something about it having 2x model support now so i'll try again sooner than later

[bitter spruce]

yeah, that happens with fresh software, there's prob been updates n shit, it's github stuff so

an easy way to update if you use the terminal and git clone $URL is to cd into the $dir you cloned it to and run git pull --ff . Now everything is updated

no with invokeai, all you do is run ./inokeai.sh then just press enter on every question

and it works

If you need help the actual developers of it will help you.

on there discord

so. If you can't get it to work, It means you didn't try and you realy probably don't care if it works or not.

Or I'm totally wrong about everything I say, idk, ya know.

¯_(ツ)_/¯

and even better I made a desktop launcher

if I want it to always launch the browser i could add echo "2" | /usr/bin/bash /home/mrgfy/invokeai/invoke.sh --no-nsfw_checker--safety_checker --no-patchmatch --embeddings (the echo part, so it will auto answer #2)

oh and & at the end to turn it into a background process

then ; exit at the end of that to close the terminal

🙃

hmmm, no

works best just like this /usr/bin/bash /home/mrgfy/invokeai/invoke.sh --no-nsfw_checker--safety_checker --no-patchmatch --embeddings

[bitter spruce]

so it's taking me about 30secs at 5 images with 300 steps each, image strength at 0.73, k_euler_a, random seed, noise 0.3, perlin noise 0.05, cfg 7.5, 512x512
config using: v1-m1-finetune.yaml

[violet saddle]

300 steps is 5 to 10 times the amount of steps I use

[bitter spruce]

this is all i can show

but I get great results

in a bit i'll upload my new images to my gallery site (found in my profile)

i'll still post the sfw ones that get made

need to do a lil more work on fixing that smirk

unless you like that derpy look

https://tenor.com/view/new-girl-zooey-deschanel-uncomfortable-awkard-what-is-going-on-gif-5250300

[subtle ruin]

"An extraterrestrial invasion has taken over New York City, and the streets are filled with chaos and destruction. The aliens, who are technologically advanced, have brought with them advanced weapons, and the city is on the brink of collapse. The citizens of New York are fighting for survival, and the military is struggling to contain the extraterrestrial threat. The image should be highly detailed, realistic and in high definition, with a focus on the destruction and fear caused by the invasion. The image should also convey a sense of chaos and despair, as the city is overrun by the extraterrestrial forces."

[bitter spruce]

Too bad that wasn't really going down

https://tenor.com/view/star-lord-guardians-of-the-galaxy-dude-omg-ガーディアンズオブギャラクシー-gif-8527541

[grizzled sage]

i run win 11 now so i'd use their bat fiile and have to select all the files that configure it for nvidia on windows

[bitter spruce]

oh yeah, windows .... (shivers in disgust)

man, too bad i'm banned from art servers that DO allow nsfw

I'm making some kick ass art right now

ayyy they are starting to turn more pixie like.. nice

omg, the bot wont even allow an uncovered BACK

who the fuck gets aroused from a back?

really?

lame

[red plover]

I'm trying to find "Additional Networks" in the Automatic1111 GUI to use some LoRa models... I feel like an idiot and can't find that anywhere, least of all "a panel from the left bottom of WebUI...

Can someone screenshot this and share it? Thank you.

[grizzled sage]

that's discord's nsfw filter. it's not so much detecting a back, just patterns. maybe her shoulder blade is too cleavagy

server owners don't have control over the filter's thresholds last i checked. its either on or off

[bitter spruce]

i just tried to dm u the pic

and even your shit is blocked from seeing a womans back

😂

are you all like .. Mormons or something?

[grizzled sage]

its discords own filters bruh...

[bitter spruce]

nah dude

i can get what ever i want sent to me

weird

discord put child locks on you adults

😂

i mean, u be u

[grizzled sage]

but one thing i do do (heh doodoo) is immediately block people who act like anything less than extreme hedonism is religious devotion.. i'm going to go ahead and not do that for you though since nsfw seems like your whole thing to begin with

[bitter spruce]

lol

i had to crop it like this

🤣

[red plover]

Anyone?

[bitter spruce]

OOooOoOo so sexy

hahahahahaha

i actually lol'd

man, guess I'll have to make my own disboard discord server

So the Adults can post there art

Don't want to get all you ppl all sexed up for more pictures of bare "Backs" 😂

I'm just teasing you guys

[violet saddle]

Looks like the tutorial is for sd-webui, just try to place the files in your stable-diffusion-webui/models/ folder and it should work (after restarting automatic)

[bitter spruce]

Sampling with k_dpm_2_ancestral starting at step 75 of 300 (225 new sampling steps) 🙃

[grizzled sage]

unicorn going super saiyan

[violet saddle]

I love the hairs 😄

Did you use the latent mirror extension?

[dry crow]

its at the bottom of txt2img

@red plover

[hasty nova]

Alright, my second attempt at making a textual inversion embedding just finished. It's of a person this time, not a style.

I hope it works

Well it works better than my previous one, but it has a lot of flaws

The last one looks the best, right?

But see, in this one the first one looks the best

I think I am going to go with 2000 steps being the best. I don't think it is overfitted, would you say it could be not enough steps? Do I need to train more?

[hasty nova]

I am going to train some that are 100-400 steps

[grizzled sage]

nope thats just dream builder model. it really nailed the symmetry there though

[violet saddle]

definitely

[hardy flume]

A friend who was curious about AI challenged to make a picture like he wanted, here is the result:

[tame forge]

This is something I still can't wrap my head around. If it's this easy why dont all pytorch programs just also tap into RAM too?

[bitter spruce]

inovkeai's tool uses both @tame forge

vram and ram

[tame forge]

well unless its a new feature its not working for me lol and I have 32gb ram 24gb vram

[bitter spruce]

are you using invoke or 1111?

did you build your conda env correctly

i have 32Gb ram and 12Gb vram

cd invokeai
conda init --all --user
conda env create -f environments-and-requirements/environment-lin-amd.yml -n inokeai
conda activate invokeai
source .venv/bin/activate
and your venv and conda are now setup

@tame forge

make sure you also installed all the correct ROCm packages

https://youtu.be/fNFzfwLM72c

🙃

we're going no where, somebody help me, ya

well..... I thought it was funny 😒

[bitter spruce]

hi

i'm using v1-m1-finetune.yml

great config

I get great results

[wispy nest]

thank you

[bitter spruce]

i run at 300 steps

[tame forge]

yo thx for the tip. I will definitely try this. I usually only use invoke for inpainting

[tame forge]

thats 300 steps? ur wasting GPU power

[bitter spruce]

I use inpaint and img2img

https://tenor.com/view/ugh-as-if-gross-grossed-out-cher-gif-3547950

i'm not wasting shit lol

4 image(s) generated in 162.27s
Max VRAM used for this generation: 3.40G. Current VRAM utilization: 2.25G
Max VRAM used since script start: 3.40G

@tame forge

[tame forge]

Let me see what 300 steps can do

[bitter spruce]

barley even touching the 12Gb of vram

[tame forge]

I assume only invoke lets u do that many

steps dont use up vram

[bitter spruce]

regardless

i'm using like 3.4 of 12gb of vram

at 300 steps

per image

still think i'm wasting vram, please enlighten me on a good step number to use @tame forge

[tame forge]

Ugh 40 steps? Lol? Depends on the model

[bitter spruce]

model is elldrethsLucidMix_v10

now running at 40

[tame forge]

Yea do a 40 vs 300 I'm curious

because I've never head of anyone using beyond 100 steps outside of joking purposes

[bitter spruce]

i won't be able to post here

[tame forge]

why

[bitter spruce]

nevermind, it made sfw

ur in luck

oh nevermind

[tame forge]

Then do a sfw prompt

[bitter spruce]

this fucking server even thinks my sfw is nsfw

[tame forge]

dayum

[bitter spruce]

yeah, that's why i bitch so much about it

[tame forge]

just add "wearing a burqa" and that will cover her up

[tame forge]

I just want to see the quality dif

[bitter spruce]

wow, this one worked

well fuck, guess i got lucky 😒

[tame forge]

so thats 40 steps right

[bitter spruce]

vs 300 steps

sooooooooooooooooooooo

[tame forge]

da fudge?

[bitter spruce]

ya

[tame forge]

Its not the same resolution

or aspect ratio

do the same seed

[bitter spruce]

ooooh ya lol

doesn't work if its randomized, my bad

prompt
(tattooed body), (tinkerbell with long hair over her shoulders and smirk smile)+, photorealistic, full body portrait, shadow depth, intricate details, (large breasts), trending on artstation, intricate details, highly detailed, (fantasy)++ (nude), (realistic), (photorealistic), (NSFW), [poorly drawn face], [poorly drawn hands], [poorly drawn fingers], [poorly drawn breasts], [poorly drawn body], [disfigured], [ugly]

running

(don't judge my prompts, i'm a single male in a town of 2000 ppl, and no single women my age)

[tame forge]

nah its kool I only do chix as well

[bitter spruce]

now i will run at 300

omfg

damn it

[tame forge]

I told u to cover them up

[bitter spruce]

my art is SFW fuck you discord

[tame forge]

put a shirt on em whatever

[bitter spruce]

they are clothed

i assure you, she is wearing clothing

but this server is re tarded

[tame forge]

Ok so 300?

[bitter spruce]

sorry not sorry

no

this is 40

running 300 now

running at 4 images per so give me just a min

[tame forge]

it has to be the same seed to make a fair comparison

[bitter spruce]

then i choose 1 of the 4 to post

yes

i kno

nags a lot lol jk

[tame forge]

if its the same seed then the 4 of them will look exactly the same

if not, something is wrong, I suspect its not actually steps that setting is changing

[tame forge]

If it really were 300 steps, your img would be garbled and overcolored

[bitter spruce]

if wut

really wut?

huh

lol

[tame forge]

Why seed 0

[bitter spruce]

Xp

how many steps

what seed number would you like

[tame forge]

Any normal number lmfao

[bitter spruce]

same

at 300

[tame forge]

and these are uncompressed? Thats how they look like in ur pc?

[bitter spruce]

im not trying to fuck with you chick or dude

lol

so yes

[tame forge]

That is bizarre. There is probably some cap that is setting a limit on steps. Like, look at this one

[bitter spruce]

k

that 1 came out nice tho, right?

[tame forge]

Nice, yes, but it is not 300 steps

I can guarantee you

[bitter spruce]

yes it IS

im not trying to impress you

or lie to you

[tame forge]

This is the detail 40 steps does

It is either 40 steps or 70 steps, I can do some digging

Its a failed generation but it is just an example of why u dont need 300 steps

[bitter spruce]

u really think im lying to you huh?

[tame forge]

If u want to I can do the max 100 steps in the regular webui and show u why 150 steps would look awful

[tame forge]

no, invokeAI is lying to you

It is probably capping your steps at like 70

[bitter spruce]

{'prompt': 'black leather lingerie, (tattoos), (7ink3Я8311 7uЯn3d in70 4 53xy 5uccu8u5 wi7# 4 10ng 810nd3 p0ny74i1 0v3Я #3Я 5#0u1d3Я5 & f4iЯy wing5 wi7# 4 51ig#7 5mi13), photorealistic, shadow depth, intricate details, (large breasts), (sitting on the forest floor with her legs spread wide)+, (facing the viewer), (realistic)+, (photorealistic), [disfigured], [p00Я1y dЯ4wn f4c3], [poorly drawn hands], [p00Я1y dЯ4wn fing3Я5], [poorly drawn breasts], [p00Я1y dЯ4wn 80dy], [poorly drawn pussy]++', 'iterations': 4, 'steps': 300, 'cfg_scale': 5, 'threshold': 0.3, 'perlin': 0.05, 'height': 512, 'width': 512, 'sampler_name': 'k_euler_a', 'seed': 2428777163, 'progress_images': False, 'progress_latents': True, 'save_intermediates': 5, 'generation_mode': 'img2img', 'init_mask': '', 'seamless': True, 'hires_fix': False, 'init_img': 'outputs/00137.png', 'strength': 0.75, 'fit': True, 'variation_amount': 0}

300 steps

{'prompt': 'black leather lingerie, (tattoos), (7ink3Я8311 7uЯn3d in70 4 53xy 5uccu8u5 wi7# 4 10ng 810nd3 p0ny74i1 0v3Я #3Я 5#0u1d3Я5 & f4iЯy wing5 wi7# 4 51ig#7 5mi13), photorealistic, shadow depth, intricate details, (large breasts), (sitting on the forest floor with her legs spread wide)+, (facing the viewer), (realistic)+, (photorealistic), [disfigured], [p00Я1y dЯ4wn f4c3], [poorly drawn hands], [p00Я1y dЯ4wn fing3Я5], [poorly drawn breasts], [p00Я1y dЯ4wn 80dy], [poorly drawn pussy]++', 'iterations': 4, 'steps': 40, 'cfg_scale': 5, 'threshold': 0.3, 'perlin': 0.05, 'height': 512, 'width': 512, 'sampler_name': 'k_euler_a', 'seed': 4029563337, 'progress_images': False, 'progress_latents': True, 'save_intermediates': 5, 'generation_mode': 'img2img', 'init_mask': '', 'seamless': True, 'hires_fix': False, 'init_img': 'outputs/00137.png', 'strength': 0.75, 'fit': True, 'variation_amount': 0}

40 steps

[tame forge]

I believe you that you are entering 300 steps, but I know it is not doing 300 steps, let's get a second opinion. I'ma wait for some expert to come in here and pitch their opinion

What you can do is do an X/Y prompt and do 40 steps, 60 steps, 80 steps, etc.

You will see after some point they will all look exactly the same

[bitter spruce]

on it

i'm changing nothing but the steps

since i did 40

i'm upping in incriments of 20

[tame forge]

Aight I'll be back since I gotta finish troubleshooting my card. But I am curious to see that joint and prove to u it is not doing what u think it is lol

[bitter spruce]

kk

running 60 now, same everying as in those posts above with the {}

sorry if I seem agro, I have really bad ptsd, and I can't help it @tame forge

so please take my attitude as with a grain of salt

[tame forge]

nah its kool, also pick a 8-digit or so seed and stick with it

Im not fighting u I'm just pointing out what I think is wrong, which I am sure it is lol

[bitter spruce]

no, i mean in general

i have no control over it

and i can't work because of it, and prob why i'm single

60 steps

now running 80

i fucking hate living on the edge of fight or flight, i really do

my family has to knock on the door of any room im before they enter in so i dont jump up

it sucks

(Fuck you discord bot)

[red plover]

Not on mine, for some reason...

[bitter spruce]

Hey admin! I'm trying to post a picture of a clothed pixie, so wtf!?!?!?

[red plover]

Trying this....

[bitter spruce]

same as the pic above just better detail

fucking art nazi's

😒

now it looks even more naughty

but i guess thats how they like it here

running at 100 now

man... i see how my mouth combined with ptsd gets me into shit ... smh

srry

running at 120 now

(I hope the admins understand my ptsd is bad, and why i can't work)

i hope they don't take what it writes/types as me typing it, cause well, it is, but it's not

most the time, i don't have this quick clearity

as i do rn

because i'm focused on it rn, so i'm able to catch myself as soon as it happens

running at 140 now

running at 160 now

(personally they are improving, even if it is, ever so slightly)

and i'm not really noticing the 1 or 2 seconds longer they are taking

i'm not addicted to instant gratification

running at 180

and notice, her hair bun

it's also getting better at trying to make my image match my prompt

"tinkerbell"

and, i am ONLY changing the step

running 200

then we will compare 40 to 200

(i really hate living on edge 24/7, you all have no idea)

fin

idk what happend to step 40, but here's step 60 to compare

STEP: 200 above, 60 below

i ran 4 images at the same time, others i can't share so you will just have to believe me when I say, there is a different

and if you look at the OG SD wiki, all of there tests range from about 40 - 200/300 steps

also if you look, some of the info on the models at civitai.com say they run at 200 steps

which is why I usually run at 200 steps

on the wiki page look at the grid showing the types of SAMPLER's

you'll notice they are doing what i'm saying

showing example images of each type of sampler ranging from 40 - 200 steps per example image

if you do, you'll also notice the change happens around the 200 step mark

please correct me if I am wrong.

(I could be wrong, but I don't think I am)

200

80

300

60

👀

see the diff

@tame forge

i choose 300

imo, step 300 beats all of them

300 <--------------------------------> 60

what do you all think?

I choose the one one the left.

with proof i did no fuckery

(I just want people to know, i'm on the ball, and not trying to fool anyone)

changed prompt, image and seed, starting at 40

moving up to 80

moving to 150

and finally 200

last 2 im doing

then you all can judge them for yourselves

2 sets of images only thing changed are the steps

imo step 200 is the breaking point when shit gets better

but I did this so you all can do own assessment and make your own informed discussion

[tame forge]

I believe this is due to the model ur using

I think its kinda like midjourney

Its hard to make something look like shit with their models

Sadly this troubleshooting has taken me all day but tomorrow I'ma post an X/Y plot with the protogen model

[bitter spruce]

https://civitai.com/models/1366/elldreths-lucid-mix

[tame forge]

and ima tag u with it and I'ma show u wassup

[bitter spruce]

the model i'm using ☝️

[tame forge]

Yea his models are pretty good

I think protogen is mixed with that model

[bitter spruce]

this is the config i am using as well

@tame forge so do you concur with my hypothesis that step 200 is when the major change happens?

or are you still fighting (possibly bad choice of word) that steps 40 - 80 are all you "really" need

honest and sincerely

🙃

[tame forge]

I have seen an image suddenly transform itself after a random amount of steps occassionally

[bitter spruce]

and @tame forge ty for being able to handle my energy, it means alot

[tame forge]

But I am telling you, 300 steps is not the norm, not even 100 steps is the norm

Something is wacky with ur model or with invokeAI and ima show u why tomorrow

My head hurts from all this bs I'm doing with python rn

[bitter spruce]

norm, is boring and idk what norm anything is fren

[tame forge]

What I mean is that any other model would look deepfried after 150 steps

[bitter spruce]

ugh, py, don't get me started lol

i fkng hate python

[tame forge]

Its not pythons fault, is its dependencies especially SD that uses a ton of moving parts

[bitter spruce]

like, if your prog lang breaks the versions its replacing, you might as well call it an Apple product

lol, am i right

lol, burn

(reffering to the devastating python2.7 to 3 version upgrade and the BS it still causes today)

when software devs have to introduce, virtual environments or version managers, is when you know, something, somewhere, someone fucked up 😂

and they ran with it.

🤣

because, programming hard sometime

@tame forge mind you as well, everyones pc is diff, so steps for you compared to JimBob and I, will never be equal

and that's ok

we have nothing to "prove" to anyone

(but it is fun to try)

lol

we all love the gratification of validation through other people

it feels good, though it just shows we're all kinda broken inside

[tame forge]

wrong. images can be reproduced exactly between computers if all parameters are the same, steps included, they are no different

[bitter spruce]

so what have you found

maybe, i'm not computer scientist

idk

but ok, lets roll with it

are you still gathering your data to show me?

(I was showing my data as I was working in real time [ he mumbles quitely to himself])

[tame forge]

No I will do that tomorrow. I'm exhausted

I just wanna chill and play some OW2 then hit the sack

tomorrow I'll tag u and post my chart

[bitter spruce]

ok

fun talking with you

i enjoyed it

peace @tame forge

[sonic vessel]

training models is so much fun

going to hang out in voice 3 and stream my new model using InvokeAI.

[bitter spruce]

https://voidofempathy.com/public/uploads/original/f9/64/bfc3a5cd237674936e6d2ffad22a.gif

Updated my Gallery

Link in profile

[fierce crest]

so

nijijourney vs waifu diffusion

which ones better

in terms of coherency at least

[fierce crest]

I CLICKED ON THAT GALLERY

THINKING IT WOULD BE NORMAL THINGS

I SAW NOT SO NORMAL THINGS LMAO

also does mrgfy stand for mr. Go fuck yourself?

nice name man

[bitter spruce]

yes

I'm just experimenting with all the models from https://civitai.com/

@fierce crest my handle come from my tattoo, Go Fuck Yourself - across my throat

-- one day, I just got tired of peoples bullshit

[fierce crest]

sounds legit

[fierce crest]

damn free nsfw

do you just download a model

or?

[bitter spruce]

yup

[fierce crest]

and then

[bitter spruce]

and then what?

and then I make stuff

https://tenor.com/view/sherlock-holmes-now-firm-grasp-obvious-gif-25345453

[fierce crest]

i was using getimg.ai

[bitter spruce]

I'm self hosting

using inokeai, on my pc not a website

the gallery is one of many of my websites

[wispy nest]

that's a lot of steps to create a image :O

[bitter spruce]

I got made it quickly randomly one day because I got tired of every single "art" place online, has an issue with using the femal body in art

is it

[wispy nest]

Might be different in the workflow, but using steps over 60 seem to cause the image to become too blurry in my experience. :/

[bitter spruce]

one the left, 300 steps
on the right 60 steps

I like the one on the left

[wispy nest]

yeah, the important part is the outcome. I just haven't seen people use those high steps in a long while :)

[bitter spruce]

¯_(ツ)_/¯

if you look at SD's wiki on explaining samplers, it shows a grid in the range of steps, 40 - 200

with progression images

and you can see that around 200 steps, is the change between ok, to wow happens

[wispy nest]

I'm sure they are good as well, I have a couple of my own which I tested when 1.5 was released

[bitter spruce]

I don't think many ppl even look at the wiki on github tbh

[wispy nest]

I've made a couple thousand images using over, well 1000 steps, but I found that the ones I liked the most was around 20-44 steps in general, there never is a set number, I like some at 50 steps and others at 9, etc :D

[bitter spruce]

sucks for them ya kno

[wispy nest]

most wiki's an tutorials are behind as the technology and discovery happens too fast, which is a good thing in a way as it means we get lots of new stuff all the time! :D

I'm just longing for a time when it's easier to get shaper/crisper images. Blur is my nemesis! :P

[dry crow]

The trick is to use a higher resolution or an upscaler xD

[bitter spruce]

seed resized

seed not resized

[wispy nest]

yeah, but that's cheating, I want to be able to do it using only txt2txt and img2img :D

[bitter spruce]

crazy how even the simple shit can entirely change the final image

[dry crow]

I dont think its cheating when you just make a 1024x1024 xD

[wispy nest]

too bad the wiki don't show the other samplers, Euler a and DDIM are the only one I sometimes use of those :/

[bitter spruce]

and honestly, if I were getting the results i want at 80 steps, i would be using 80 steps

[wispy nest]

yeah, I don't really think so either, I just want to strive for it :D

[dry crow]

Try DPM++ 2m Karras, its realy good too

[bitter spruce]

but so far 4 images at 300 steps per 1 is only taking 30secs a piece

which is fine

i'm in no rush

[wispy nest]

I use DPM++ SDE Karras the most, not because I know it's better at X or Y, but because it's the longest worded one :P

[dry crow]

Would suggest always to go for 512x768 for portrait

[bitter spruce]

i use mostly k-euler-a or k-dpm2-a

mostly just euler-a

[dry crow]

Same mostly Euler a

[bitter spruce]

switching between those 3 are always gonna be ok

then change to ddim for inpainting

[wispy nest]

I haven't had time to learn inpainting yet, I'm still learning how to "perfect" what I want using the txt2img, so much to learn, and it's so much fun. Was a long time since I was this set on a goal! :D

[bitter spruce]

once I get a txt2img I like, i use it in img2img

getting 1 really good txt2img can be an uphill fight

[wispy nest]

It all depends on what each person want in the end. :)
I'm happy with the results straight out of the txt2img most of the time

[bitter spruce]

I started to make a README file with each models trigger words, and tag words, so it helps me get an image i want

[wispy nest]

auto1111 has an extension that does that as well, but I'm too lazy for it, I just rename the model file to include the keywords inside a ()

[dry crow]

For models i rename them to Inkpunk(nvinkpunk) for the triggerword

[fierce crest]

can u send all ur websites

[bitter spruce]

i use inokeai, i dont use auto1111

[fierce crest]

to my dms

[bitter spruce]

https://www.cursedcode.com all my sites end up here

or begin here

[wispy nest]

that's also my favourite model, how'd you know?! ;P

[bitter spruce]

basedlabs isn't rendering correctly so, ignore that page

[dry crow]

Haha its a realy good one ! 😄

[bitter spruce]

it worked before, just fine, then idk wtf happend, maybe jekyll got an update or who knows, i'v recompiled that site with ruby gems idk how many times and it still wont fucking work like it used too

😒

so i been slowly building a new website for that 1 page

its slow going cus i'm not all that motivated to do it

like imagine being on the production dev team, you finish and upper managment comes over n tells you were switching to a new framework, trash whatever you just finished

non nonnegotiable

🤮 🤬

https://tenor.com/view/looney-tunes-bugs-bunny-tic-distressed-nervous-gif-16888492

its not even like jekyll is hard but idk, i have no clue what's different

you make a _config.yml and a Gemfile, bundle install, then bundle exec jekyll serve -d public

it compiles and serves the site, then you make your edits n what have you.

easy, but ..... maybe its just the theme im using that's broken, idk

[dry crow]

How much vram do you have ?

[livid crystal]

i have 8gb

i have no clue what is going on

I am positive a 3060ti can render this

[dry crow]

Yea easily

Youre on a laptop ?

Also which guide did you follow?

[open sun]

@livid crystal

[livid crystal]

no

[vast cape]

protogen + anything Vae

[dry crow]

Would suggest using the WD1.4 vae for proto

[vast cape]

Close every window on your pc, check task meneger

[livid crystal]

I fixed it

[vast cape]

It happens to me from time to time

[livid crystal]

Well sort of

The solution was to just load it on Linux

[vast cape]

Well, thanks for advice) But I just wanted to experiment , I had anythind v3 vae for wlop model, and I changed to protogen so I thought that leaving anything vae active would be fun

[dry crow]

Yea try to experiment with it. I figured out that using anything vae on non vae needing models can cause The image to get pale(desaturated)

[vast cape]

well, this is pretty good result tho)

[dry crow]

Yea its good! Just sharing tricks ^^

[vast cape]

xd

[dry crow]

Bloody bullet holes wtf

[vast cape]

Yea, but it's kinda cool

BTW, what are good anime models excluding anything vae?

[dry crow]

@vast cape here, i have compared them very quick: ProtgenX2.2. See how anything vae makes it pale xD.

[vast cape]

Hmm, I see)

[dry crow]

AnythingV4, ACertainModel, also Eimis Anime Diffusion maybe

[vast cape]

Is anything v4.5 better than anything v4 / anything v3?

[dry crow]

its different, depends on what you like

the creator said in dc higher number = better 😛

[vast cape]

Btw, what means prunned ?

[dry crow]

compressed

to get a smaller file size

use pruned if you can

[vast cape]

Thanks)

[night sonnet]

@wispy nest again

I've seen this pic once

[vague oasis]

@amber leaf

[wispy nest]

Sometimes life be like

but then its also like

[finite igloo]

When is STable Diffusion going to have their own version of Chat GPT?

[pseudo aspen]

Ah yes, another creation of mine

[dry crow]

very nice! but you should use a vae, if its anythingv3 then the av3 vae

[pseudo aspen]

Yeah it's anythingv3

But I already have the vae

[dry crow]

then its not using it correctly

[pseudo aspen]

Hmm

Ah I see

[dry crow]

make sure its in the folder and looks like this:

[pseudo aspen]

I have the wrong vae

Gonna fix it

[dry crow]

okay 👍

[pseudo aspen]

I guess it's right now

And now I wanna see the results

[dry crow]

yea!

make sure the vae selector is on automatic

then it will choose the right vae for the model

[pseudo aspen]

Loading weights [67a115286b] from E:\stable-diffusion-webui-master\models\Stable-diffusion\Anything-V3.0-pruned-fp32.ckpt
Loading VAE weights found near the checkpoint: E:\stable-diffusion-webui-master\models\Stable-diffusion\Anything-V3.0-pruned-fp32.vae.pt
Applying cross attention optimization (InvokeAI).
Weights loaded in 110.1s.

It said that so it's probably choosing automatically

[dry crow]

yes, ah youre using invoke ok

[pseudo aspen]

Yeah I noticed. Is there a way to change that

[dry crow]

change what?

[pseudo aspen]

To other AI's

I want to experiment with that

[dry crow]

ahh, its not other AIs, they are just webui's

[pseudo aspen]

Ah, got it.

In that case, do you know any better?

[dry crow]

There is Automatic1111 (the most used with the most features), InvokeAI (best outpainting and ui) and SD 2.0 (didnt tested it yet)

[pseudo aspen]

Right. I'll check it out. Thanks

[dry crow]

no problem, automatic1111 webui is not as easy to install like invoke, but message me i can send you a tutorial if you want or just look up Aitrepreneur on Youtube

[pseudo aspen]

I'll try my best ig

[outer spear]

So fp16 is just less precise version of fp32, but what is the regular one?

[dry crow]

@outer spear fp32 is the "normal" one you should use

also you need the vae for anything

[outer spear]

whats just pruned then?

thats also fp32?

[dry crow]

@outer spear yea pruned is compressed for a smaller filesize, always use them

[outer spear]

im asking what is the difference between these two

[dry crow]

the file size, oh weit

[outer spear]

no its same

[dry crow]

yea i see, he uploaded the 32 and 16 two days ago, so use either the non fp32 or the fp32

it shouldnt realy matter in this case

keep in mind that you also need a vae file for 4.5 if you dont manualy select them

[pseudo aspen]

Whoa

That vae file made a change

[dry crow]

yea its like night and day with anythingv3

[pseudo aspen]

Yeah. The colors are much warmer

[outer spear]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/6858

anyone know whats causing this

[tame forge]

Vote

[vague oasis]

GFPGAN is the most accurate, both face restorers smooth too much the skin for my liking tho

[earnest solar]

mofu and mofu

[willow chasm]

its something in your prompt or negative prompt. do you have anything like overexposed, underexposed, contrast, high pass filter, hdr, cinematic lighting, etc

[grizzled sage]

this dreamshaper model keeps impressing me with how well it is doing abstract prompts i'm throwing at it. i think i'm going to keep on this one a bit, pausing my 2x exploration, so that i can get a good set of desktop wallpapers

[earnest solar]

More mofu mofus

[lunar sluice]

Hi ya, I am noticing a bit of a trend on pisstagram. I can tell they are using film stills with their prompts. I would love someone to shed light on this. What's the diagram of operation, you think? Is there specific model/module they run their prompts to achieve these? Is it even Stable Diffusion?

[grizzled sage]

theres lots of cinematic style embeds and models

[boreal crown]

poolsuite is a model trailed on kind of late 80s 90s cheesy magazine lifestyle glamour.

[hasty nova]

https://github.com/Firework-Games-AI-Division/dmt-meshes Anyone else checked this out yet? I am downloading it now, supposedly it works really well.
(It's a Blender plugin to generate 3D meshes from text)

I'm trying it out right now in blender

"A corgi" at 15000 steps...

It was really fast tho

An alien spaceship

Alright, one more txt2model, then I wanna try img2model

A windmill

Definitely not

[gloomy elbow]

It works by vibration without traditional blades.

very advanced.

[hasty nova]

Ofc

How could I not think of that

[boreal crown]

https://imgur.com/EXX4R5f

[velvet vessel]

https://twitter.com/80Level/status/1615571743363530753 😒

80 LEVEL (@80Level)

Online image library Getty Images has filed a lawsuit against Stability AI, accusing the Stable Diffusion developer of infringing its intellectual property rights.

Learn more: https://t.co/D71TNlqB4A

#stabilityAI #stablediffusion #AI #ArtificialIntelligence #gettyimages

Twitter · Jan 18, 2023, 4:48 AM

[teal knot]

Anybody know if you can blend two images like in Midjourney ?

[bitter spruce]

what's up ppl

had an issue, reinstalled invokeai, bk up n running

[boreal crown]

It is Getty Images’ position that Stability AI unlawfully copied and processed millions of images protected by copyright and the associated metadata owned or represented by Getty Images absent a license to benefit Stability AI’s commercial interests and to the detriment of the content creators. <--- see THIS is an actually well worded filing. Because getty absolutely own the shit out of their images inside and out and took pains in doing so.

[bitter spruce]

for some reason invokeai doesn't like the the textual inversion files

[boreal crown]

source: https://newsroom.gettyimages.com/en/getty-images/getty-images-statement

[bitter spruce]

(embedding files)

nice halo sun flare

[boreal crown]

sorry but by looking at this halo flare you now owe getty 100 million dollars.

[bitter spruce]

😂