🏞｜general-with-images

[elfin obsidian]

Bro where’s the actually ai like where you can make images

[sick tangle]

vampire

[still flax]

is my address gonna be leaked from this van gogh painting? O_O

[sterile kiln]

• All the info should be in the embed.*

?

I don't understand

[rain gazelle]

embedded in the png

[grizzled sage]

I'm playing around with Lexica.art's new model today. holy moly it's easy mode. short simple prompts yield stupid good results. the tags are filtered and blocks things like "fart" but here's "Elsa from frozen on a beach" https://i.imgur.com/gP2Eywt.jpg

delorean mecha... i think that's also the head of the man delorean too

or elain's boss

[grizzled sage]

I really like lexica. I keep diving back to it to just fire off a simple prompt idea with zero engineering. .. boom

[rain gazelle]

evenin'

Didn't know lexica had generation. Is it new?

Very plasticy results

[gritty cedar]

Said by the same artists that couldn’t tell the difference between human and ai

[finite dagger]

Does anyone know a good way to implement hands into stable diffusion? i've been trying to recreate this drawing(but more high quality) but the hands always turn out very weird

[outer plaza]

I feel emotions when creating the prompts and art through AI since I can't draw to save my life.......

[grizzled sage]

noice

[sterile kiln]

There is nothing about in EXIF data. I'm very interesting about the settings you used to do this 🙂

[solid dawn]

prompt ford transit repair

[bitter spruce]

yo yo

[wispy nest]

hellooo

[quasi dagger]

Hi

I have to talk later :((

Byee :((

[wispy nest]

byee

ill wait for screen shots

[wispy nest]

pls ping me when back ty

oh

i've seen these ss on yt

[bitter spruce]

so, what is confusing?

idk what ss is

[wispy nest]

i followed the instructions on yt

screen shot

[bitter spruce]

oh lol

[wispy nest]

and i downloaded the automatic1111

the videos told me to edit the "webui-user.bat"

so i did

[bitter spruce]

where are you downloading your models from and are they ckpt's?

[wispy nest]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/6556

i reported my bug there

[bitter spruce]

don't hold your breath, look at what number you are

[wispy nest]

number?

[bitter spruce]

i'm the kind of guy that solves things and not wait around for someone else to solve them for me, but anyways

yes, you are number 6,556

[wispy nest]

im waiting?? xd

[bitter spruce]

there's a lot of ppl ahead of you

[wispy nest]

:

[bitter spruce]

bro

think from the developers point of view

when you see over 6,000 bug reports, you might just get a little overwhelmed

[wispy nest]

yea..

[bitter spruce]

specially, if you are doing this for free

and so many ppl have such un real expectations, its sickening

and not fair to the guy sharing it in the first place

[wispy nest]

sry..

[bitter spruce]

but anyways

(I just mention that stuff to get ppl back in reality)

[wispy nest]

how old are yoou?

[bitter spruce]

36

[wispy nest]

wow

[bitter spruce]

dick

lol

im not that old 🙁

[wispy nest]

..

[bitter spruce]

ya i kno 😒

So.....

what are you confused about with training?

[wispy nest]

how did you get to that page?

[bitter spruce]

see under INPUT, go through the tabs and fill stuff out

it's my web-ui

what do you mean?

[wispy nest]

i only downloaded the file: https://github.com/AUTOMATIC1111/stable-diffusion-webui

[bitter spruce]

yup

same shit

[wispy nest]

is the next step head to "webui-user.bat"?

[bitter spruce]

but the difference with me is, I go through the folders and read the README files, learn and add shit to make it better

oh, also

I use Linux

[wispy nest]

windows 11

[bitter spruce]

I should still be able to help you

[wispy nest]

can use make too

mac*

[bitter spruce]

oh boy, you really are super noob to all of this, aren't you?

[wispy nest]

yea.

you cant do this on mac?

[bitter spruce]

that doesn't matter, look, if you have specific question hmu, maybe I can help

[wispy nest]

yea, sounds good

[bitter spruce]

but I don't have the patience or time, to walk you through how to use everything

[wispy nest]

do you check dms?

[wispy nest]

thanks for your time

[bitter spruce]

sometimes, but I prefer answering questions in the open, so that other people with the same question can learn from our open discussion

[wispy nest]

question: whats model.ckpt?

immediately runs into problem 😅

[bitter spruce]

use google for simple questions

[wispy nest]

oki

[bitter spruce]

I dont have time to also teach ppl how to use the internet

take baby steps, and use what you have, use critical thinking, and common sense

[wispy nest]

👍, ill dm you when i have real problems

[bitter spruce]

cool cool

i have faith in you.

[wispy nest]

tysm for your time

[bitter spruce]

yw

🙃

[stuck turtle]

Someone is hungry

[still flax]

nice, I like it

working on a terraria model. Its not perfect yet but heres a flying volcano

[sterile kiln]

Starbound? 😄

[still flax]

heheh yeah, terraria and starbound are very similar

btw

thats a good idea actually

starbound model, with planets and stuff. Terraria doesnt have that...

what do you guys think?

a bit rough still. Needs some work

[sterile kiln]

need more shadows, but it's damn similar

[still flax]

yeah, I guess that would make the images better... Thx for the tip

[still flax]

thats really cool actually. I like the fantasy setting

[velvet lichen]

Tweeted by Laion. Seems to be a generation from the Deepfloyd model.

Same for this one.

https://twitter.com/AVShonenkov/status/1612486644761825284

Alex Shonenkov (@AVShonenkov)

A photo of rainbow owl with deer horns in the woods. Portrait photography, photo realism. by Alex Shonenkov

Twitter · Jan 9, 2023, 4:29 PM

Name of the model seems to be: 'DeepFloydif'. As there's already a hashtake used by the team that's creating it.

[velvet lichen]

The resolutions of these pictures is 2048x2048 by the way. That's interesting, but could be the upscaler from Stabiliy at work.

[rain gazelle]

oh oops! Well I'm sure I can dig it up, which one was it?

[sterile kiln]

This one @rain gazelle .
But what would have interested me to know is how you did it. I see on your screen that you first used characters, and that you outlined them with the inpaint tool to have the scenery behind. Do you do this every time?

[rain gazelle]

Ah gotcha

That image was pure txt2img. The widescreen ones are done in a variety of ways -- those in particular, I actually expanded the image in gimp with just empty white sides, and inpainted the sides to draw in the stuff around her.

It was one of the new protogen models, probably 5.8. Just learning how to use 'em.

Here's the deets on it:
Ultra realistic illustration, a (young) carrie fisher as princess leia, white outfit, hair buns, silver belt, (leg). Inside a space station. Highly detailed, digital painting, artstation, concept art, smooth, sharp focus, illustration, art by artgerm and greg rutkowski and alphonse mucha, Negative prompt: ((watermark)), ((signature)), ((logo)), bad anatomy, disfigured, deformed, malformed, mutant, gross, disgusting, out of frame, poorly drawn, extra limbs, extra fingers, missing limbs, blurry, out of focus, body out of frame, cross-eyed, ugly, poorly drawn face, mutation, mutated, ugly face, extra legs, extra arms, blurred, strange colours, boring, sketch, lacklustre, extra_limb, poorly drawn hands, messy drawing, text, malformed hands, malformed face, crop, extra limb, two heads, deformed hands, mutated hands, bad proportions, cloned face, gross proportions, malformed limbs, fused fingers, too many fingers, wip, fanart, crude, rough, amateur, speedpainting, draft, misshapen, defective, weird, broken, disassembled, mangled, unfinished, disproportioned, disproportionate, unappealing, amateurish, basic, beginner, training, grotesque Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 9, Seed: 1900979682, Size: 512x512, Model hash: 13a6777c, Model: ProtoGen_X5.8, ENSD: 31337

[sterile kiln]

Protogen works with Stable Diffusion?

[rain gazelle]

yes of course

does it work with something else? 😛

[sterile kiln]

I had tried several times to do this, to enlarge my image with photoshop and to fill it with inpaint, without great results. But you prove to us that it gives good results.

[acoustic eagle]

LoL

Bored Ape embed on my anime model 😄

"Bored Ape Waifu Club" ?

[bitter spruce]

catgirl realism

If only Elon didn't get picked on at Twitter...

shame, a dam sham

https://tenor.com/view/waynes-world-waynesworlddream-dissolve-dream-gif-20735278

now, bk to reality

[grizzled sage]

SciFi Diffusion 1.0 . Panda space marine.

sorry, Pandaa

[grizzled sage]

https://tenor.com/view/john-travolta-lost-confused-uh-gif-5916341

https://twitter.com/stabilityai

you rascal

[wispy nest]

where do the pics get generated

[left crane]

@grizzled sage

[grizzled sage]

we're on top of it!

[dense tapir]

WHOA!!! Awesome stuff.

[pine briar]

I've been slowly working on Fantasy Diffusion

I need to put it to use to make concept art for a tabletop campaign

[past peak]

oh yea cool thing i found out, i did this as a small test and yes it does work.

[dense tapir]

I am now on Linux with Automatic. Test image

[heady spruce]

Beans

[wispy nest]

#🏞｜general-with-images data centre in an oil paining

[bitter spruce]

https://tenor.com/view/petty-parker-smoke-snoop-dogg-homer-simpson-disappear-gif-17093677

she consumes man

[brittle skiff]

These images were generated from our website ,draft.art / generated by our AI , head to our website to generate yours!

These images were generated from our website ,draft.art / generated by our AI , head to our website to generate yours!

[mellow crypt]

can someone help me create a prompt to generate this kind of image?

or medieval illuminated manuscript image

[vague oasis]

like this?
a painting of a group of people standing in front of a column, by jean fouquet, renaissance, royal - blue, greek columns, inside view, centered

[mellow crypt]

i'm trying to create something inside a medieval manuscript decorative fram

1: an illuminated manuscript with a depiction of a man and a woman, a silk screen by Limbourg brothers, featured on flickr, academic art, academic art, grotesque, chiaroscuro

2:Create an illuminated manuscript illustration, depicting a man and woman, similar to a silk screen painting by Limbourg brothers titled "Illuminated Manuscript" featured on Flickr. Take inspiration from the academic art movement and the use of grotesque, adding a sense of realism and detail to the figures. Use the chiaroscuro technique to create a sense of depth and drama in the scene. The illustration should include decorative borders, patterns and symbols in the medieval style, pay attention to the details in the clothing, jewelry, and the environment. Use of vibrant colors with gold leaf to add a sense of luxury and opulence. Consider including symbolic elements that are relevant to the central image and pay attention to the spatial composition and hierarchy of elements. Text should be written in a calligraphic script, with attention to the balance and harmony of the letters and words. Make sure the final illustration is high resolution and scanned from a parchment paper

this 2 are the prompt

but mmmmh i don't really enjoy it, it's dark and idk why it keep with color palette

[gloomy elbow]

water malone

[pallid ruin]

Lol

[frail isle]

?

why

[frail isle]

how do i fix this?

[dry crow]

How much vram do you have ?

[frail isle]

how do you see that?

4gb

[dry crow]

What graphic card do you use?

[frail isle]

geforce gtx 1050 ti

[dry crow]

Okay so 4gb vram

Do you wanted to generate an image or upscale ?

The error triggers when there is to less vram to perform the task, like generating a highres picture or upscale one

[frail isle]

both

i want to generate and hires.fix it

[dry crow]

I see you want to upscale it by 2

[frail isle]

yea

[dry crow]

I think thats to much in one run

Try generate first and upscale later in a single task

[frail isle]

where i can upscale?

[dry crow]

In the extra tab

[frail isle]

oh i see thx

[dry crow]

Np there it should work

[sudden steeple]

Accidentally made the aftermath of an AI art apocalypse

[thorny zephyr]

do i have the right version? i cant find textual inversion but it does talk about it in the terminal

[grizzled sage]

once in a while i go back an play with av3. it's always a struggle to make it not do ginormous boobs. even with small_breasts and cleavage in the neg, it does it. anwyays.. this is a blend of doom slayer and princess peach. the art style is :chefs kiss:

[weary folio]

genned with midjourney in a few seconds.... @grizzled sage deepfloyd seems way behind

[dense tapir]

@grizzled sage python3 -m xformers.info
xFormers 0.0.15+6cd1b36.d20230110
memory_efficient_attention.cutlassF: available
memory_efficient_attention.cutlassB: available
memory_efficient_attention.flshattF: available
memory_efficient_attention.flshattB: available
memory_efficient_attention.smallkF: available
memory_efficient_attention.smallkB: available
memory_efficient_attention.tritonflashattF: available
memory_efficient_attention.tritonflashattB: available
swiglu.fused.p.cpp: available
is_triton_available: True
is_functorch_available: False
pytorch.version: 1.12.1+cu113
pytorch.cuda: available
gpu.compute_capability: 6.1
gpu.name: NVIDIA GeForce GTX 1060 6GB

I removed xformers and can't get it to lock up my system now when loading other checkpoints. That shows xformers is perfect so something fishy is going on in the pipeline I can't figure out.

[grizzled sage]

no horns. don't @ me. i don't like people who are toxic for amusement.

[weary folio]

I @ who i want

[grizzled sage]

ooo i should run that on mine to diagnose. i keep getting bs cuda crashes. last one that happeend just a moment ago when windows popped up a notification for steam

[dense tapir]

Yeah, in my three days of hell I stumbled upon that. Some of those said unavailable until I finally got it all right

Still, why the total system meltdown? I mean I am not locked up I just can't

do anything

[grizzled sage]

i guess i have to activate the venv to run it maybe

[dense tapir]

yes

linux it is source ./venv/bin/activate

[grizzled sage]

on windows theres some files in scripts for it

Error caught was: No module named 'triton'
xFormers 0.0.14.dev
memory_efficient_attention.flshatt: available - requires GPU with compute capability 7.5+
memory_efficient_attention.cutlass: available
memory_efficient_attention.small_k: available
is_triton_available: False
is_functorch_available: False
pytorch.version: 1.12.1+cu113
pytorch.cuda: available
gpu.compute_capability: 8.9
gpu.name: NVIDIA GeForce RTX 4080

[dense tapir]

In windows it is "./venv/scripts/activate"

yes, triton I had to get as well

[grizzled sage]

two issues. triton and 0.14

/knuckle crack

[dense tapir]

0.14?

[grizzled sage]

version is 0.0.01 less than yours

[dense tapir]

Yes, on Windows I started with that but on Linux and Windows I built them so both are 0.15

[grizzled sage]

i wanted to do the same already and i thought i had. but i guess it didn't take

[dense tapir]

Yeah, for sure

That triton one is a HUGE one

[grizzled sage]

giggity

[dense tapir]

Lucky on linux I can back command and here it is how I got it

pip3 install tritonclient\[all\]

[grizzled sage]

windows terminal has made strides since i last used it and has colors, back commands, TABS, cross ls dir understandings, its amazing kind of

i even had to ./activate.ps1 to use it in powershell. i'm home

[dense tapir]

stay in your venv and do the above command

My back buffer is so full trying to get all this to work to finally be beaten down like this

[grizzled sage]

ew!

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/6601 i found this

some russian guy's build though head scartching

[dense tapir]

Yeah, I had that on windows too. On windows I am unsure which torch I have

[grizzled sage]

sorry i haven't been able to help much with your investigations but i appreciate sharing cause i think this is a lead

[dense tapir]

No worries as all of this is new to me but the problem squarely lies with Xformers

It is a hunk of shit

I reported a few other issues and their devs are bad

taking me 1.5 times longer without xformers. Fun to watch how this almost eats all my vram (I mean it gets so mb close).

[thorny zephyr]

what animal is this?

[dense tapir]

@grizzled sage I just grabbed dreambooth on Linux and damn.

023-01-10 20:14:37.761954: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2023-01-10 20:14:38.551623: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer.so.7'; dlerror: libnvinfer.so.7: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /home/john/stable-diffusion-webui/venv/lib/python3.10/site-packages/cv2/../../lib64:/usr/local/cuda-11.7/lib64:$LD_LIBRARY_PATH
2023-01-10 20:14:38.551700: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer_plugin.so.7'; dlerror: libnvinfer_plugin.so.7: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /home/john/stable-diffusion-webui/venv/lib/python3.10/site-packages/cv2/../../lib64:/usr/local/cuda-11.7/lib64:$LD_LIBRARY_PATH
2023-01-10 20:14:38.551710: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.

[muted moss]

Dang

[boreal crown]

a capybara/sheep hybrid that has attained enlightenment

[dense tapir]

@grizzled sage pip install tensorrt
Collecting tensorrt
Downloading tensorrt-8.5.2.2-cp310-none-manylinux_2_17_x86_64.whl (549.2 MB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 549.2/549.2 MB 3.0 MB/s eta 0:00:00
Collecting nvidia-cublas-cu11
Downloading nvidia_cublas_cu11-11.11.3.6-py3-none-manylinux1_x86_64.whl (417.9 MB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 417.9/417.9 MB 3.5 MB/s eta 0:00:00
Collecting nvidia-cudnn-cu11
Downloading nvidia_cudnn_cu11-8.7.0.84-py3-none-manylinux1_x86_64.whl (728.5 MB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 728.5/728.5 MB 2.4 MB/s eta 0:00:00
Collecting nvidia-cuda-runtime-cu11
Downloading nvidia_cuda_runtime_cu11-11.8.89-py3-none-manylinux1_x86_64.whl (875 kB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 875.6/875.6 kB 7.1 MB/s eta 0:00:00
Installing collected packages: nvidia-cuda-runtime-cu11, nvidia-cublas-cu11, nvidia-cudnn-cu11, tensorrt
Successfully installed nvidia-cublas-cu11-11.11.3.6 nvidia-cuda-runtime-cu11-11.8.89 nvidia-cudnn-cu11-8.7.0.84 tensorrt-8.5.2.2

@grizzled sage 2023-01-10 20:25:03.624607: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2023-01-10 20:25:04.630698: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer.so.7'; dlerror: libnvinfer.so.7: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /home/john/stable-diffusion-webui/venv/lib/python3.10/site-packages/cv2/../../lib64:/usr/local/cuda-11.7/lib64:$LD_LIBRARY_PATH
2023-01-10 20:25:04.630781: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer_plugin.so.7'; dlerror: libnvinfer_plugin.so.7: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /home/john/stable-diffusion-webui/venv/lib/python3.10/site-packages/cv2/../../lib64:/usr/local/cuda-11.7/lib64:$LD_LIBRARY_PATH
2023-01-10 20:25:04.630790: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.

WTF?!?

This is a good thing: This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2 FMA

The rest is bad.

[grizzled sage]

yuupp. looks about right. rebuild one thing, you might as well just go full gentoo

just build the whole os. build it all!

(dont go full gentoo)

i'm actually rebuilding it too myself

Successfully installed xformers-0.0.15+3df785c.d20230110 woot

WARNING: [WinError 127] The specified procedure could not be found
Need to compile C++ extensions to get sparse attention support. Please run python setup.py build develop
A matching Triton is not available, some optimizations will not be enabled.
Error caught was: DLL load failed while importing libtriton: The specified module could not be found.
xFormers 0.0.15+3df785c.d20230110
memory_efficient_attention.cutlassF: unavailable
memory_efficient_attention.cutlassB: unavailable
memory_efficient_attention.flshattF: unavailable
memory_efficient_attention.flshattB: unavailable
memory_efficient_attention.smallkF: unavailable
memory_efficient_attention.smallkB: unavailable
memory_efficient_attention.tritonflashattF: unavailable
memory_efficient_attention.tritonflashattB: unavailable
swiglu.fused.p.cpp: not built
is_triton_available: False 😦
is_functorch_available: ** False** 😦
pytorch.version: 1.12.1+cu113

[dense tapir]

yeah, it is really grinding my gears

what my errors mean is that it will suck ass and is not using CUDA rather the cpu

[grizzled sage]

i think cause i installed the other triton first, this is happening. maybe if i just delete that. if it doesn't work i'll rebuild the whole venv. ez

not compile, i mean like, let python make it again

[dense tapir]

I am about at my end with this shit

[grizzled sage]

https://tenor.com/view/go-on-pulp-fiction-samueel-l-jackson-gif-19148118

me during a build. "say what again!"

[dense tapir]

It is built on a foundation of crackers. Do one thing and it all crumbles to crumbs.

[grizzled sage]

wow i've been so engrossed in command line work all month long i hardly realized my explorer was still set to not show hidden files

[dense tapir]

screw it, as the main reason I went to linux was for dreambooth but the sob refuses to act right.

[grizzled sage]

it do look like there's hope through wsl2 on 10 and 11

i think that docker might tie into the same systems

[dense tapir]

naw, not into wsl. tbh, everything is for ubuntu and that is almost as much of resource hog as Windows.

I can't find any info that is right now so I am lost af. Last string before I just repartion the drive.

[grizzled sage]

i think it's headless mostly

[dense tapir]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/5926

[grizzled sage]

What should have happened?

Not done this

[dense tapir]

btw, that was a great find I just posted because automatic not using tensor is pretty ugly

tensor cores is how the new speed ups are about to happen. Your card doesn't have tensor cores? No speedup for you.

[grizzled sage]

mine do xD

[dense tapir]

Automatic better get on the ball. I don't think mine does

[grizzled sage]

Ada Lovelace name sake tensor cores if i do say

[dense tapir]

yes, yours does

I have 0 TC but I have 1280 Cuda

[grizzled sage]

Daughter of Lord Byron. He was such a rockstar of his time that her mother didn't want her in arts at all and put her education entirely into mathematics

[dense tapir]

TC didn't come until RTX and the 2060 has 32

[grizzled sage]

tc were a smart direction for nvidia. a few neat tricks make them really a hot seller for a while, then when the software and models mature, suddenly they'll bring new power to all their chips

[dense tapir]

Here is a nice one showing how valuable TC is - https://towardsdatascience.com/rtx-2060-vs-gtx-1080ti-in-deep-learning-gpu-benchmarks-cheapest-rtx-vs-most-expensive-gtx-card-cd47cd9931d2

[grizzled sage]

all these machine learned tricks are going to start springing up using TC and nvidia will dominate the market for another decade

[grizzled sage]

i dont want to install cuda 11 maaaan

that sounds dumb

[dense tapir]

The problem with the 20xx is it didn't do fp32 well at all asthey were crippled while the 4xxx series cards are not.

[grizzled sage]

there are special motion vector instruction sets in 4* too that i want to look into using. its how they're utilizing dlss3

i'm wondering if they could aid motion inbetween frames with sd

[dense tapir]

I know

I think 5k, and 6k is when a whole lot of real time stuff is going to hit.

Man, I want to try Lora but I have no idea

[grizzled sage]

this first bit is going to be a taste. they were talking about images per second.

[dense tapir]

I saw the vids and still have no idea

[grizzled sage]

even 3 per second would be like, watching regens as you tweak it

[dense tapir]

Just remember this if they can ever manage 15fps with generations the game is over they win.

[grizzled sage]

fidelity is cool but speed and fun will win the game. thats why hifi audio died and sony walkmans changed everything

[dense tapir]

AMD will have their asses handed to them if they don't jump into this which they are, kinda sorta, trying to do.

Yes

MP3 lacked but won

[grizzled sage]

AMD has carved a niche they can live in. their dgpu days may be coming down

[dense tapir]

They are reaching into ML right now. I forgot what it was but it even has ML in its name. RDNA3 is almost as good at it as Nvidia but it didn't have xformers AND is not really made for ML. RDNA4 is when it will get hot as they see the trends.

[grizzled sage]

yeah rdna3 has ml cores now. and chiplets lets them be a lot more versatile. its been a very functional kit on my deck. fun little device for what it is. powerful stuff

couldn't get SD working on it butttt i wonder. is that a steam os3 thing or a hw compatibility thing

[dense tapir]

steam

[grizzled sage]

i dont think the deck's little igpu has the ML cores

[dense tapir]

SD is what was shown working and it rendered in 6s while the 4090 did it in under a second BUT 4090 had xformers

[grizzled sage]

yeah??

[dense tapir]

yep

[grizzled sage]

mobile sd

on my sd

intriguing

[dense tapir]

that is when I got a woody and threw all my viagra out the window.

[grizzled sage]

i need a voice recognition prompt input thing on it

[dense tapir]

the problem is with AMD their driver division it has been, and still is, and probably always will be utter SHIT!

[grizzled sage]

this is a fantasy scenerario, but john carmack suceeds in making gai and the first thing he does is tasks it to writing a universal hardware driver of maximum optimization, or some

i'll brb i'm hungry

[dense tapir]

k

[grizzled sage]

installing cuda 11 i guesss

[grizzled sage]

whew. almost missed that. cuda 11 almost installed cuda11 era drivers for the gpu

custom install

[dense tapir]

Damn, lora dreambooth has changed so much in two weeks a lot of options are now gone so nerdy rodent's video is obsolete

[grizzled sage]

lol yeah i almost might go to college to learn all this stuff so i don't need to follow tuts on every new version

[drifting oyster]

@broken oracle by low quality deviantart images, I mean I want a model that can create masterpieces like this

[broken oracle]

It wouldn't surprise me if someone did

Though tbh you could probably just spend 20 minutes in MS paint and make something on this level without the robot

[drifting oyster]

i don't really hold the knowledge to figure out how to train a model and all that 😔 (also just doing embedds makes my PC get hot lol)

[drifting oyster]

True!

but it's part of my comedic broken brain

the idea of something that can mass produce that seems hilarious to my immature self lol

[grizzled sage]

o

i'm genning images again with the new venv. back that up now, and start tweaking things again

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/2103

tests are extensive and exhaustive

[broken oracle]

Yo is that the hit 1997 animated comedy tv show Cow and Chicken?

[dense tapir]

@grizzled sageI am done with Linux as it froze up on me again with no way to taskmanager or three finger salute it only the reset button. With no idea what is going on, and kept blind, I am done.

I did not have xformers so it is something in their code causing it.

[grizzled sage]

lol yeah linux kernel can really halt up hard if the correct settings for the driver aren't completely dialed in

[dense tapir]

I already know they do not have any devs that work on linux by the errors I get

No three finger salute is the breaking point

[grizzled sage]

some linux do have it but it's an oil fashioned sysreq thing REISUB

old unix trick

[dense tapir]

I'll just have to forgo training or wait until I get a beefy card and stick with Windows

[grizzled sage]

train on collab is a thing too

i've heard

[dense tapir]

I am on Linux Mint XFCE

I do but dreambooth extension is not allowed

even if I manually add it the thing is blocked

[grizzled sage]

alt+prntscrn(sysrq) + R then E I S U B should on mint

shut it all down

[dense tapir]

I am not going to hit all those buttons

[grizzled sage]

give each command a lil to initiate

its like a prayer for unixers

[dense tapir]

lol

well, I am not happy with the state of Linux today as I was in 1998-2006.

[grizzled sage]

garuda is what i used. the bloatiest of the bloat lol. i wanted to explore what was out there. lots of cool tools

[dense tapir]

At least DF I got to work right in it 2018-2020 but not this

[grizzled sage]

last time i used things it was ubuntu dapper drake era and let me tell you, things have improved

[broken oracle]

As far as I can tell what Hyper network I have selected has no impact on the images the model produces. How do I fix that?

[dense tapir]

Just too many stupid shit things that requires too many stupid shit things that must be those precise stupid shit things to work or it refuses

[grizzled sage]

thats why i like arch as its a rolling release and more flexible to throwing new things at it

[dense tapir]

The other issue, the main one, is too many devs only working on Windows so bad things happen in Linux.

naw, rolling release will not work with CUDA stuff as everything must be the exact versions or forget it. I got all that down but I can't help Automatic1111 and the extensions.

This is not a war I wish to fight as all I want to do is just use the damn shit without issues.

[grizzled sage]

yeah lots of great stuff i know how to use and can rely on working is on windows. i don't justify my switch back at all. i have no loyalty to software lol. i was using linux to get the most of my vega 64, now i use windows 11 to get the most of my 4080

[dense tapir]

yeah, same

only with auto linux version is bad stuff

[grizzled sage]

and while i ws over there i used plasma and made it like windows

[dense tapir]

That I didn't but screw it. Had I 10-16gb vram I would not have touched linux.

I bled again and lost a lot of shit I can't get back too

[dense tapir]

@grizzled sageI tried to do lora but so much is different from two weeks ago it did nothing as I am too lost.

[grizzled sage]

i still have no idea what lora is. it's changed a bunch since i first looked. is it generating small files like embeds?

[dense tapir]

it is embeds of about 160mb so a cross between a full DB ckpt and TI embs

there is an extension we no longer have to merge too

[grizzled sage]

apparantly i'm not building xformers 0.0.15 with cross attention support

hmmm

[dense tapir]

merging was never supposed to be a thing as the files were supposed to be used like an embedding.

so many options are now just gone that I can't make heads, nor tails of this

Returning result: Exception training model: 'max() arg is an empty sequence'.

I went to DB's issues and yeah, they broke it.

[grizzled sage]

nope, that uh, that didn't work. now torch can't load the gpu, which i think is not good

[dense tapir]

that is bad, yeah

[grizzled sage]

it did not like when i rebuilt it that time

[dense tapir]

well, apparently lora will just have to blow away into oblivionas far as I am concerned.

[grizzled sage]

use collab. i've heard people are doing that with lora

[dense tapir]

I tried that but it is really bad.

Unless it got some love it was horrible

[grizzled sage]

well each collab is it's own script set. thye're like codepens

[dense tapir]

I never found but one two weeks ago written by some dude in India and it was just horrible.

Show you how bad auto is I click a link in it and it goes to edge YET my default browser is Vivaldi.

[grizzled sage]

https://github.com/TheLastBen/fast-stable-diffusion this ones regualarly updated

[dense tapir]

That is not Lora

besides that is what I use for TI. HN is broken in it and TI is barely working

[grizzled sage]

its automatic1111 with all that set i thought

[dense tapir]

nope

extensions are mostly verbotten because DB goes out and loads other things. Security risk and colab says no

That notebook is so bad that when you have downloaded everything then hit the cell to start it up that last part takes 10m

wtf is he doing? beats me

PLUS it will lose sync and interrupt it, or it finishes, you MUST restart the cell so 3-10m wait again

Ben says it is everything from 2.1 model is broken (now that I know is BS) to a man on the moon is currently rubbing one out.

At least it does embs but is so annoying due to bugs. Only one we have though.

Total images: 4
Largest prime: 2
Best factors: (2, 2)
Total VRAM: 6
Saved settings.
Custom model name is artefx
Starting Dreambooth training...
Initializing dreambooth training...
Replace CrossAttention.forward to use xformers
Change in precision detected, please restart the webUI entirely to use new precision.
We have 0 sample image(s).
Traceback (most recent call last):
File "D:\stable-diffusion-webui\extensions\sd_dreambooth_extension\scripts\dreambooth.py", line 571, in start_training
latest_file = max(list_of_files, key=os.path.getmtime)
ValueError: max() arg is an empty sequence
Training completed, reloading SD Model.
Restored system models.
Returning result: Exception training model: 'max() arg is an empty sequence'.

[grizzled sage]

wel yeah. its probably chopped right to bits in order to run on colab and is magically stiull running hah. i thought it did lora though. the one thats on the main lora repo seems fine

[dense tapir]

Meh, not for me as it is missing all sorts of optimizations AND it demands (due to a lack of optimizations) HIGH MEM.

[grizzled sage]

i forgot to update subbbbmodules oooo hooo booboohooo rebuilding again then geeze

stupid compiler machhiens. do what i want

[dense tapir]

Ben's is barely able to fit and how he managed was by taking a model then pruning it, even if it says already pruned, then it works but take any 2.x model from HF and BANG

I had to take the 2.1 model and prune it here then upload it there so it could work as the 2.1 I sent it was too much for the ram (not vram).

all of this is getting ridiculous and eventually we will all need 32 gigs of ram and 24 gig vram.

https://civitai.com/models/4086/luisap-tutorial-hypernetwork-monkeypatch-method

[grizzled sage]

i just backed up and nuked the whole folder. i've tweaked the repo so many ways now. will have to copy cudnn files over again, few otehr little tweaks i found over couple months. still working on that xformers 0.15 bit

sick of random gpu crashes that shouldn't be happening. like full black screen wdm resets

[dense tapir]

If I were on Windows with what was happening in Linux I would have had that or a BSOD

btw, to use the lora files without merging you have to use the dreambooth extension.

Oh, and TLB's auto colab pukes all the time and in the root of my directory is a fuckton of files from auto. I mean it is crazy.

171 files it pukes up onto my root folder all the time

What SAI really should have done is make a program for all of this instead of just leaving it up to the community to do. Make it non webui (iow a real program) and leave it up to the community to work with it.

[grizzled sage]

yeah same idea. the kernel is derailing in ring 0

[dense tapir]

OUCH

[grizzled sage]

anytime a driver misbehaves the kernel panics

heh

[dense tapir]

Yep

[sonic vessel]

ok - guess what model this is.

[dense tapir]

She needs to clean her face as it is too oily.

[sonic vessel]

you don't get to tell my art what it needs!

its a hot summer night my friend, sweat glistens

[dry crow]

Seems like a model which contains samdoesart

[sonic vessel]

only if its in SD 1.5

[grizzled sage]

i've got xformers 0.14 going again and so far today i haven't had any gpu crashes. i wonder if it was an extension i was using causing them.

[dense tapir]

Possibly

In linux I had to remove most of them

[sonic vessel]

That's the 3rd worst Morgan Freeman I've ever seen

but if that's what you're going for, that cool, and you're awesome

[sonic vessel]

dear lord

[grizzled sage]

i told chatgpt how to write one prompt, then been asking it for prompts since. its not half bad

[sonic vessel]

uh oh I know where this is going and that's def. banned.

[grizzled sage]

sometimes i correct it, tell it what's wrong and it builds a better prompt

[grizzled sage]

Here i only told it to give me a detailed prompt for a portrait ofa horse in a starfleet uniform, with enougn detail to make the generator produce a great photorealistic portrait"

prompting for a prompt. now i need one to generate prompts for chatgpt to give me prompts

[boreal crown]

"give me 50 image captions for the least popular art on deviantart"

or "alt text for the finalists of worst drawing on the internet"

[grizzled sage]

i could totally awaken my deviant art account, i mean, whats that site

[boreal crown]

gpt is good at inventing characters though. ask it to invent and describe anthropomorphic mascots for really inane companies

[dry cosmos]

inpainted face

[grizzled sage]

yeh not bad not bad

[fervent rune]

Working on a border embedding for a friend

[dense tapir]

Interesting this emb I am doing

Turned the first into the second

Only 100 epochs though

200 epochs

300 it got its appendages back

[jagged fox]

How to get rid off multiple characters when doing landscapes, i tried with prompts "single character" and in the negative ones "multiple subjects"

[floral kelp]

Is there a good way to fix hands?

[hearty wasp]

heya guys newbie here, keep getting disfigured face like this on 1.5 pruned ema only, any ideas to fix this ?

[floral kelp]

Are you using img2img?

[bitter spruce]

@tropic shell

[tropic shell]

Oh! I love the wings! They're so vivid!

[bitter spruce]

and the warior pixie

[tropic shell]

Fantastic outfit, too

Tats are great

[bitter spruce]

you can see all my stuff i think is good enough to share by clicking my profile and visiting the ART link @tropic shell

🙃 and ty

[tropic shell]

I'll check it out!

[bitter spruce]

u r over 13yrs old tho, right? @tropic shell just checking

[tropic shell]

Hahahaha, I'm halfway to 80, we will say

[bitter spruce]

aight, lol i'm half of 65 so

oh and btw, every image in that link, is all the same image, just molded

I never changed, just kept remixing from the one I started with

even this, came from being a pixie, to catgirl, to this

trip out on that

mmmm the angel of death was it's own image, hmmmm so technically i lied, but that's the only out liar img

this 1

it's.... so/so

oh, and the evil snow white was its own image too

so 3 started out as there own, tink, snow, and this guy above

SFW of my Evil Snow White @tropic shell

i have no clue what her hair is doing but it doesn't look bad, just interesting

evil snow white as evil queen

[tropic shell]

Oh, love the gradual values

[bitter spruce]

yea, i would step stuff down sometimes just by 1

[tropic shell]

Its a hair-hood-crown

[bitter spruce]

lol, oh yeah, i see it

huh

weird

[tropic shell]

Dundundun

[bitter spruce]

so I would change the slider by a factor of 1 up and or down, and add or remove 1 prompt at a time

slow, yup, but are my results good, yup

lol

then sometimes i would use in painting to coerce the ai to fix or change things

[hearty wasp]

nope using text2img

[covert marsh]

what makes an embedding "fuzzy"

is it just the still learning vs when you generate an image with img2img or txt2img or is it needing to train more to get more details?

[weary bison]

on a website looking for samples, and had to do a doubletake on the thumbnail descriptions.. almost looks like the BLIP'd them

[frail isle]

how can i make somone face to the background?

[harsh ledge]

add the keyword ",from behind,l"

[frail isle]

thank you

[frail isle]

she is still looking at me ahhaah

[bitter spruce]

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

My Picture gallery is up

XD

[bitter spruce]

cool robots

that made me smile, you rock @fresh hound

[gloomy elbow]

wooden robots? Someone's been playing Autonauts, haven't they?

[tropic shell]

I love it but so spooky!

[still flax]

ikr... heheh, thx!

[hasty dove]

OMG!

[thorny zephyr]

what does it matter if it gets made safe

[dense tapir]

If it isn't OVER 9000! then it is junk.

[boreal crown]

all the same prompt, im really happy with the effect

[wispy nest]

Wtf happend to the site

How do I run multiple images now

Where Is the step bar and gfx

Why is it defaulting to a editor

[foggy knoll]

Wow

Im scared

[boreal crown]

i love this

[boreal crown]

Prompt: greg rutkowski looking at greg rutkowski while facing greg rutkowski, trending on greg rutkowski by greg rutkowski

[tame forge]

mfw I never use greg rutkowski in my prompts cuz I hate oil painting styles

[barren vault]

Hi, sorry I think I asked in the wrong channel. Is this the right channel to ask? If it is then I will ask here.
I have asked this in #🤝｜tech-support . But I still haven't found the answer. Please help me to figure out.
Thank you~

[boreal crown]

Oh the model's hash is its checksum iirc. Since you can change the filename of a model to whatever and blend models together in different amounts it can be difficult to keep track of which are orignals, which are slightly changed etc. Think of the has as its license plate. You'll get like Model Name: WhateverModel-V1, Model hash: 012345ba1

[barren vault]

I see.. Thank you~

But how can I search for Stable Diffusion models on the internet with the given hash?

[still flax]

comfy

[still flax]

thomas is that you?

[dense tapir]

His demented little brother.

[versed furnace]

Love that retro vibe

[versed furnace]

Like the star wars energy there. tried to make a retro-madmax-fashionweek model a few weeks ago. got some good things out of it, but have had other things going on lately.

[dense haven]

Inpaint darkens and desaturates my image. Am I doing anything wrong?

[versed furnace]

I have horrible luck inpainting large portions of images. I need to stick to smaller areas for reliable results. I could just be doing things wrong, though.

Unless it's to completely replace one entire subject/object

[deft phoenix]

From which second it could be considered as "not stolen"
*mine

img2img feedback loop

[wispy nest]

Daughter chrome

[thorny zephyr]

what am i doing wrong? all pics are png 768px. Nothing is happening, not even in the terminal

batch size should be 18 but still doesnt do anything

[wispy nest]

greg rutkowski looking at greg rutkowski while facing greg rutkowski, trending on greg rutkowski by greg rutkowski

promp /greg rutkowski looking at greg rutkowski while facing greg rutkowski, trending on greg rutkowski by greg rutkowski

[versed furnace]

00:00.001

[gloomy elbow]

That depends... whose lawyer are you asking? 😆

[radiant orbit]

There is possible to know some of this prompt? I want to create the nearest one to this

[boreal crown]

https://www.reddit.com/r/midjourney/comments/106z2s0/denizens_of_hyperspace_my_attempts_at_using/ ask this artist, they're exceptionally close in style and have a few prompts in those threads. for future reference, using images.google.com and searching by an image url is a great way to track down styles and things. you can get a chrome extension for it too.

oh interestingly enough, they're training a model for this @radiant orbit

[radiant orbit]

First thank you for your reply. what do you mean by model for this? where?

[boreal crown]

the artist is training a model based on stable diffusion at the moment. as in feeding it in lots of trippy art so its better at outputting trippy art. probably worth following them

[radiant orbit]

Thank you brother!

[gloomy elbow]

starting to get there...

[radiant orbit]

If you will get there please share with me your results 🙂

[gloomy elbow]

I like that yellow cat, very neon light-painting

[boreal crown]

you could try generating the cat head to begin with, then with img to image add the trippy prompt and adjust the denoise level maybe

[radiant orbit]

I glad you like it 🙂

[boreal crown]

but yeah, you're making great progress, congrats!

[radiant orbit]

trippy its a keyword to this style?

[boreal crown]

no i just use "trippy" as shorthand/slang for psychadelic, it might work tho

[radiant orbit]

Ohh ok

[boreal crown]

shamanic would be one of the better words i think, it would encourage the ornate lines and dots

[radiant orbit]

Now I will add the shamanic to the prompt and see

[boreal crown]

also try words like detailed and intricate when talking about the styles. but i think your last pic is pretty much perfect in terms of level of detail in patterns

[radiant orbit]

Thanks! I will catch up with this prompt. I really get nice results just from trying to get to the original

[gloomy elbow]

lol now all I have are normal looking cats.

[radiant orbit]

Now make it transparent and with black background

[boreal crown]

it went from purrfect to catastrophic

[radiant orbit]

The secret is to keep trying

By the way this is the model I use

[gloomy elbow]

@vast cape Using the standard 1.5 model, here's the difference between "photo" , "anime:1" and it looks like the sweet spot is right around 1 for weight, though you'll have to vary it some depending on your prompt/seed. I haven't seen a one-size fits all with my weights.

for example this prompt/seed had a nice little style boost I liked around "anime:1.6"

[vast cape]

Oh, thanks)

[gloomy elbow]

@vast cape , here's the full progression of weights for this prompt: black haired golden eyed girl wearing purple, neutral expression, portrait, detailed face

[vast cape]

And yea, as thanks I would send my best result in niji journey

[gloomy elbow]

Looking great!

[vast cape]

Yea, I like her a lot )

[boreal crown]

related to this, theres a "shift attention" extension/script that lets you put in ranges like photo:1~0

[hasty nova]

What is the difference between all these?

They all seem really similar

[gloomy elbow]

did not know, I'll have to check that out!

thanks!

[vast cape]

fantasy, detailed face, portrait, pixiv fanbox, red haired, neko, kawaii, cat ears, anime:1, stockings, holding wine glass, castle:0,5
Negative prompt: low-res, deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, poorly drawn hands, missing limb

Not bad, but not neko(

I will work on this seed

Little better when I set cat ears on 1.5 but still not nearly a neko

[boreal crown]

it might be this specific composition/pose. if you're telling it to use this seed then its going to try to squeeze the ears into shot

maybe try this image in outpainting?

[vast cape]

This was random seed

I than inputed It's seed and proceeded

Well, third try was bad tho

But ... It made smth like real neko...

How to fix this ?)

Well, If I can't I will make another random seed

[boreal crown]

if you try it at 2:3 aspect ratio it might give you better results. sometimes in square images it smooshes the details together

[vast cape]

I will know this for future, thanks)

[vast cape]

I ran in 512x512

[boreal crown]

itll make something unlike your current pose but try 768 high 512 wide

[gloomy elbow]

680x384 is about 16:9 with about the same pixel count as 512x512, if you want to change aspect ratio but not increase step time.

[gloomy elbow]

I've had okay luck using (cat:girl:1) to get cat ears.

for example, I used (fox:girl:1.2) to generate this fox spirit:

[vast cape]

What script did you use?

[gloomy elbow]

the model was either jamix v1 or anygen v3.7

I was playing with both that day.

[vast cape]

Ah, well, you know

Alternate model can even take a " neko " IMAO

[gloomy elbow]

that's helpful

[vast cape]

I'm chalanging myself to make a good neko art without com models

BTW, how are samplers different, and is there any reason to go out of euler?

[gloomy elbow]

I've heard DPM tends to deliver better textures, but I haven't tested the differences extensively. Euler seems to run faster on my system.

[vast cape]

well, It fixed a little

Like

Really close to greatness

How to fix her head and fingers?

[boreal crown]

try adding in a few negative prompts about badly drawn hands etc

also re samplers: no real reason to venture from Euler honestly, dpm is good with glossy/shiny things but is situational. on the subject, codeformer my goto face restorer, but gfpgan is supposed to be better for cartoons

[vast cape]

This is previous version

I added detailed fingers

And it broke more(

[boreal crown]

yeah its like a cursed monkey paw wish thing with prompts. "draw a perfect hand" puts perfect hand coming out of nose

i think we should all just make an art movement that has mickeymouse style chubby gloved hands with 3 fingers

[vast cape]

Well, but we made progress

She is certainly a neko

xd

[boreal crown]

neo-maus revival

[vast cape]

eeeeeeeehhhh

How is this in the same seed?

Well, ti be fair it is closest to my goal

But proportions

They are out of the window

[sonic vessel]

is this good?

[vast cape]

Yes

[flat dagger]

I'm on the way to being a movie star. 👍 😎

[boreal crown]

try adding "natural lighting" or something to do with lighting to your prompt, should make the subject blend in and might actually clear up the eyes

[shadow lily]

/

[flat dagger]

This was just an outpainting experiment which started from a dreambooth shot. The original pic was just the framed head/shoulders. Everything beyond that is outpainted/ai generated a few chunks at a time.

I wanted to see if I could generate the whole scene a little piece at a time.

[boreal crown]

pretty good stuff

i wonder what the longest outpaint someone has done is? i reckon you could make like a bayeaux tapestry

[flat dagger]

Would have been much easier to just inpaint the body, then inpaint the background in a separate step.

@boreal crown - That's a really cool idea, actually. I might give that a try.

Outpainting with openOutpaint is great, tho. Makes this stuff a breeze.

[boreal crown]

yeah it could be fun and definitely something novel. good luck with it if you run with the idea

[gloomy elbow]

well, now that's ridiculously cute

aaand that's unsettling:

it's an unexpectedly fine line.

[sacred otter]

Hi, I understand with 2.1 we need very precise prompting and -ve prompts, but is it normal that without specifying the art style, with only generic prompting, all image generations tends to be so.... idk, dreamy? It makes me wonder if I even have the right model downloaded....

[dry crow]

hey, your cfg scale is to high, try 7-12.
Also you are using the 768 Model but generate 512x512 images. You should use the 512 model or generate images in 768x768 for the 768 model

[sacred otter]

thanks my man, I see! immediately better result. Can I assume that with more precise prompting we can have higher CFG, but with only generic prompting lower CFG is preferred?

[dry crow]

np, CFG is like hit or miss, but you cant go too high, it will always look blurry and high saturated that way. For better results you can make more Steps.
You can compare your images by setting the same seed and only change the Steps or CFG

[sacred otter]

thanks will try that!

[sacred otter]

yeah I can see CFG is best around 10, sharing in case that can help someone else

[sour matrix]

I'm training a very specific style and wanted to use an existing checkpoint, since I lack enough images to really converge. Anyone perhaps knows of a model/checkpoint that looks somewhat like these images? I already found comic-diffusion, but those comics are quite different to these. Any help appreciated!

[thorny zephyr]

Download a file of the comic and manually scissor tool them out in the right width/height

There are potentially thousands upon thousands of images available with that artstyle if you try hard enough

[sharp phoenix]

souls like, baby dragon, prision

[vast cape]

well, in the morning it generated real neko)

Well, I will delete glass of wine because of artifact, and than I will try to create dark elf art

[dry crow]

hey, nice, did you get SD1.5 or SD2.1 downloaded meanwhile ?

[vast cape]

I use SD1.5 RGN

When I deleted wine eyes got strange a little

But not bad too

[dry crow]

looks pretty good for 1.5

[vast cape]

Thanks)

[fresh hound]

I LOVE the last one due to the lights on the planet

showing life

Last set for now!

[rugged vigil]

beautiful. what's prmopt and model ?

[sonic vessel]

It is my own custom trained model. From scratch on top of 1.5, no merges

[minor glen]

man, i was thinking of submitting a new issue but bruh

[frail isle]

somone can help me why this do the 100% and dont make the img? without one error

[dry crow]

what is your graphics card vram ?

[frail isle]

4

but sometimes work another no

[dry crow]

you should use --xformers and also --medvram in the web-user.bat

[frail isle]

what that do?

like this?

[dry crow]

close the cmd, right click on the web-user.bat and then edit it like here:

[frail isle]

done

what this do?

[dry crow]

save it, then start the webui again

it will install xformers, take maybe 3-5 mins

[frail isle]

okok thx

[dry crow]

np, then you can try again, it will now use less vram and should work

[frail isle]

oh okok

and the quality will be more low right?

[dry crow]

no, the quality will depend on your image size

[frail isle]

ohh okok

[dry crow]

try 512x512 or 512x768 for portrait

[frail isle]

so will take more time

[dry crow]

higher resolution will take more time, also more steps

you can try 50 steps and cfg 7 at 512x768 for example

[frail isle]

oh i see

thank you^^

[dry crow]

np, happy testing! 😄

[frail isle]

im trying with 768*1152 it work one time and stop working on the 2 one

[dry crow]

if the error says no vram then its too much

[frail isle]

dont say any error

[dry crow]

realy, even under the image generation field ?

[frail isle]

yea

like this the 2nd bar are 100%

and dont generate the img

and just stop

and dont say any error

[dry crow]

what is your webui showing ?

under the image field

[frail isle]

nothing

[dry crow]

can you screenshot the webui settings ?

[frail isle]

ofc

this time is that because i lower the scale

[dry crow]

okay but there it shows 512x768, does this still work ?

[frail isle]

yea

[frail isle]

but this one he generate the first one and dont generate more

[dry crow]

did the image field showed error ?

[frail isle]

nop

i will generate for you

[dry crow]

maybe you could open the task manager and look at the GPU Vram its using while generating

[frail isle]

for some reason now work -_-

ahahah

[dry crow]

xD okay perfect

[frail isle]

but will stop soon i know ahah

[dry crow]

yea maybe keep an eye on your vram usage in taskmanager

[boreal crown]

I just crack an egg on my rig, see how long it takes to cook

[frail isle]

see @dry crow

[dry crow]

yea 100% of your vram is full, so it stops

close every other programm that may use the gpu or use a lower resolution

[frail isle]

oh okok

[dry crow]

would recommend to use 512x768 and then use the Extra Tab to Upscale youre image

[frail isle]

yea thx

and i cant interrupt -_-

[gloomy elbow]

I use this spreadsheet to calculate resolution for a desired aspect ratio with a consistent pixel count, it makes upscaling easier because I don't have to crop the composition.
https://docs.google.com/spreadsheets/d/17LXM_GgHib7W9dj88j7pqgenKsohNht-9f-CaV-yp9w/edit?usp=sharing

keeping the pixel count the same reduces memory errors

[dry crow]

ouh very helpfull thanks! gonna try it with highres fix

[fluid palm]

Hi, if I use brackets () like this it means that i want to emphasize this specific word? Can i use brackets in negative prompts?

[cosmic halo]

yes, and weights. (ugly:1.3) will have more of an effect than (ugly) or (ugly:1.2) in either prompt

[fluid palm]

does this double bracket ((ugly)) make sens?

[cosmic halo]

I think double bracket is the same as (ugly:1.2)

[still flax]

what do you guys think?

[lucid kelp]

I tried the Ultimate SD upscale (from image size to scale 2.6) and the output turned out to be quite different. Is this because of denoise level? Also the auto1111 console got "stuck" to
Total progress: 67%|██████████████████████████████████████████ | 750/1122 [15:31<07:41, 1.24s/it] but webui said it's done

[vast cape]

Well, I got 1 good dark elf

I am going to 2.1 now

Then to community models

[lucid kelp]

play that G-string

[gloomy elbow]

ngl, I'd rather go to the bottom rock concert.

[gloomy elbow]

sometimes I tweak a prompt for hours getting the weighting right. sometimes I just type in 'greebles' and get this:

[rain gazelle]

Any new movies out worth watchin'?

[ebon oracle]

James and the Slightly Above Average Peach

[rain gazelle]

hmm, is that on netflix?

[still flax]

really cool is that the openjourney model?

[wispy nest]

Some of these are epic. Just wondering how long it took to kind of work out the prompts that work best. Installed locally so that I could try to get some small additional images to use as backgrounds, wallpapers or add to educational material.
And, some have came out amazing, others...

Does anyone have some of their first sets, testing prompts, and ones that just... Didn't quite get it. I'm wondering how I can adjust mine, just playing around.

[boreal crown]

if you look in #1019361238234443776 some people post comments that explain what cause certain things to happen. also it can get really funny

[wispy nest]

OMG, thank you. XD That's perfect, and great learning opportunities. Definitely taking a look now.

[fierce crest]

did the bot

just get lazy

💀

[keen crest]

anyone know what this specific hair-do is? it's not in my prompt, the ai randomly generated it, but now i want it

[hallow totem]

10 Images sobre guerras de edad media

[boreal crown]

I'd call it a flicked bob

[keen crest]

nice, thanks

[wispy nest]

I was kind of thinking Jane Jetson bob

[edgy wren]

Emad, the artists are probably going to sue you.

[still flax]

that does not make sense... You cant say an AI gen image belongs to an artist with that style

[edgy wren]

I don't know, but they're suing anyway

https://investigations.stopconsumerharm.com/ai-image-model

[still flax]

well, I dont care 😐 will gen whatever I want

they cant sue ppl for that

like

[edgy wren]

they'll try

DMCA $2,500 minimum penalty by law

[still flax]

per violation of copyright. But you cant copyright a style

im not afraid

also, ppl here dont care about that so...

[boreal crown]

someone should ask greg rutowski to dmca them for his art on their stupid website

[grizzled sage]

freshest off the inpsiration generation script press. Evgeny Lushpin style

[still flax]

gotta use more Greg Rutkowskii in my prompts 🙂

[boreal crown]

also https://www.stopconsumerharm.com/all-consumer-harm-reports/ lol you can write your own investigations basically. i wouldn't even call this an ambulance chaser, looks more of a janky third party lead generator

$15 a pop. I feel like sending them some bollocks for a laugh like how cheez whiz doesn't even contain whiz or that the difficulty of using an EZ bake oven is understated

[grizzled sage]

there's lots of artists who do magic card art, dungeons and dragons art, an all the same style that greg does. It's more of a house style than his style. Games Workshop produces oodles of that stuff.

Greg isn't in 2x anymore, but like i said, so many other artists do the same sort of fantastical concept art that greg is able to

[still flax]

ok thx thats nice to know... To you know any names I can use?

[grizzled sage]

https://en.wikipedia.org/wiki/List_of_role-playing_game_artists i intend to see what the model knows of these artists sooner than later

Interesting that greg isn't on that list

[boreal crown]

do you need to use peoples names really?

[grizzled sage]

It helps

[still flax]

O_O

I dont NEED to I just like to use sometimes for the style

[grizzled sage]

its not like whoever is named will have any copyright claim on the generated material

[still flax]

yep, that was my point

[boreal crown]

what im saying is, you can describe things without having to just copy other peoples styles

[grizzled sage]

i dont think people can hold claim to a style personally

that's over reaching of entitlements

morally, i think belieivng you have ownership over a broad idea is sinister af

[boreal crown]

thats not what im saying lol. im saying using an artists name and typing "X with boobs by Y artist" is just a bit of a crutch 🙂

[still flax]

yeah, thx for the tip... But sometimes its kinda hard to replicate a style... Im got a very gud prompter yet

[still flax]

100% agree

[grizzled sage]

that's a little condescending for no apparant reason. Gate keeping basically. puts a gross taste in my mouth

[boreal crown]

and who is saying that?

[grizzled sage]

It's no more a crutch than any other prompt technique

[boreal crown]

you're fighting some strawman here

[still flax]

g-guys

._.

[grizzled sage]

You want to insult people by calling them disabled and in need of a crutch, while also acting like you're not insutling people... damn i hate passive aggressive dicks.

[boreal crown]

"red sideboob, by mike wazowski"

[grizzled sage]

Why are people this way?

[still flax]

ok ok how would you improve a prompt

like

how would you describe a style?

[boreal crown]

this is bizarre. im saying you can get fantasy styles without putting in magic artist name words and now im attacking the disabled.

[grizzled sage]

What do you think a crutch implies

[boreal crown]

describe the palette, the medium, setting composition etc?

[boreal crown]

absolutely not disability when used in context, grow up lol

[grizzled sage]

you chose the metaphor. honest liars. seriously.

[still flax]

can you pls show an example like of something you did?

[still flax]

hey its ok I guess... Maybe hes right like you could get better results, more precise stuff describing everything

[grizzled sage]

don't let people trample on your creative process ever dude. it's yours. do with it how you will. excell with it.

people saying "you can't do that" are wrong. do it

[boreal crown]

when did i say you werent allowed to?

[still flax]

ok, got it. Thx bro, also thx Wit for the tips

[grizzled sage]

Now you suddenly don't understand metaphors

[still flax]

I think I gotta improve my vocabulary maybe

[boreal crown]

heres one. its a doodle, i just happen to like the result blending these things. "porcelain ecorchè style anatomical sculpture of the female form in front of velvet curtain background, (thorns:0.6), rose petals floating, (crazed eyes, grinning with long sharp teeth) rococo and baroque intricate details, cinematic lighting,8k"

escorche + intricate details is fun as hell it just does its own thing really.

[still flax]

ok, thx, thats a good prompt... I know rococo and baroque but never heard of ecorche? How do you discover new words or styles?

[boreal crown]

none of it is "original". porcelain escorche stuff is pretty common

you know those half skin on half skin off human body dolls/statues from biology classes?

thats escorche

[still flax]

hmm I see

how to you know that many words

[boreal crown]

i read a lot i guess? also https://www.onelook.com/thesaurus/?s=descriptive words&viz=1 this website is great

https://www.onelook.com/thesaurus/?s=%3Abig words&sortby=ly1&res=res_77 sesquipedalian lol

[wraith wigeon]

anyone get this error b4

[boreal crown]

but like, fancy words are fine, but actual words used in photography/art are better as they'll be the ones people know when describing images initially

[still flax]

cool! byzantine, implex, fiddly, this is gonna help me improve my english actually. Thx you bro

[boreal crown]

yeah i really love it. use ":" in front of like an idea and itll give you the closest thing to a word for it

[still flax]

sorry, maybe you can try #🤝｜tech-support I can try to help you there but are you up-to-date? Can you describe what you were doing?

[still flax]

noice I hope to improve my prompt game

[hasty nova]

thought this was funny

[boreal crown]

those two could honestly pull anything off

[hasty nova]

ikr

I'm using a princess embedding (I sadly don't remember where I found it, probably civitai?)

Called Style-Princess

[boreal crown]

oh i thought that just stuck a tiara on heads or something, didnt realise it was so well fleshed out

[hasty nova]

yeah my prompts are super basic

and it works very well

[boreal crown]

thats dope

[hasty nova]

those two images were with the elldreths mix I think

it does try to turn things into women though

But those images above were first-try

[wet blade]

in case anyone ever gets bored or wants a little prompt inspiration, I started working on this random character prompt script

https://github.com/526christian/SD-Prompt-Generator

[unique scaffold]

#LunarNewYearAIArtChallenge starts on Instagram today, for anyone interested!