📝｜prompting-help

[chilly karma]

it's pretty old for this stuff 😄

[tired vigil]

I have a RX 580

Yours can't be worse

[chilly karma]

Mind if I get a couple of these, I'd like to try copy the prompts on mine 😄

mine is 1060

not much, but it's been pretty good to since the beginning, so I won't complain about it

mind if I can get these?

these have exactly the same style I'm trying to make

[tired vigil]

I'll send you in dm to not spam too many images here

[chilly karma]

OK thanks a lot

[somber charm]

Hey all, trying to make an album cover, and cant get stable diffusion to generate people in corpse paint makeup. any help on prompting it? I found and SDXL LoRA but am on automatic1111 so cant use SDXL

[silver valley]

Hey, Automatic 1111 can use sdxl

It depends more on your gpu vram

[slim axle]

I'd like the ai to center the car in the middle of the image but give it a bit of whitespace all around. Any Ideas how to tell SD that?

[open basin]

Okay so I know with mid journey if I tell it to give me an image of "broken glass and rusty nails rolled up like sushi". Several different models I've tried, some food based some object-based some generalized. They all just give me images of sushi completely seemingly ignoring the variation on the idea.

With 1.5 at least I never actually tried with XL.

But I guess my question is how does one prompt to make something that is not directly of the training images but novel from a combination of what is within the models training data

Rusty nails rolled up to look like sushi,

[elfin crater]

Hi I am trying to generate 2 flags (canada, australia) on the office table. but it is not coming along with the following prompt. please help "professional handsome man sitting inside luxury office with canada australia flag, wearing premium suite and a watch, looking towards viewer, D-ID friendly style:cinematic width:1152 height:896 aspect:5:4"

[fickle sparrow]

I'm trying to create a top-down old fashioned drawn map of a small mountain region. but it keeps making me islands and way larger scale than I'm looking for.

current prompt:
mystical++ realistic close up old++ fantasy map+ from 16th century of a valleys in a mountain range. ultra-detailed. winning tabletop roleplaying map. 2D flat view.

even negative prompts of seas and oceans yet that appears.

first time trying to make maps

[orchid ore]

cant be better use rather (realistic close up old:2) Then ++. I met with +++ for second time, just not sure if it works. But should apparently. Just () seems to me be more precise.

[crimson patio]

guys, best embe or lora to fix hands right now?

i just cant, like using loras and adetailer for hands - it still bad...

[silver valley]

@teal inlet here is an quick example of tiled diffusion
i made 3 squares, 1 background and 2 foreground, each person i described differently and put it inside its description

[orchid ore]

@silver valley it is only for SD or for SDXL as well?

[compact maple]

Hoi, what promps will help to tell SD to make the eagles feathers be rifles and guns?

[silver valley]

didnt tried it yet, but should work i guess

[orchid ore]

i got error, not sure if related

[boreal pecan]

Hello guys

which is the best model for making like, weird things that cant have a refernece

por exapmple

"Plantas made of metal wire, "

Plants* lol

[tired vigil]

@boreal pecan Do you have access to SDXL ? xl understands that concept well I believe

[urban ibex]

Hey question: I am trying to create a character with complex features (Object built into armor, Items in hand, detail etc) SDXL can handle the first couple of prompts but loses the idea and misses the finer details. (For example I can get the armor but none of the details I want with it) What is the best way about building the detail into the image? Should I try to add more details off the same seed or is ther a technique or process someone can point me to please?

[tired vigil]

@tranquil totem I am trying to get that style, can you give me more information of the image, like artist or source

And my computer crashes

[tired vigil]

Look up how to use regional prompter

I believe that can do it, it will separate regions to generate for each thing

Like
Girl
BREAK wearing armor
BREAK bird inside armor

And so on

No idea how to actually use it :P

Looks like that when people do

[urban ibex]

Ok awesome! Thanks so much for the reply, I will do some research.

[tranquil totem]

can i dm?

[tired vigil]

Yes

[trail zinc]

Please please please do your best to help me because I might lose my job.

I trying to create one color logo or icon of random robots like images of this post

above logos or icons have only 1 color, they are simple and straightforward, they don't specify the robot's size, shape, or other features.

Now I want to use ControlNet Stable Diffusion to create random icon and logos like above.

In ControlNet Stable Diffusion:

1. what checkpoint I must use for this?

2. what SD VAE I must use for this?

3. what ControlNet model I must use for this?

4. what prompt I must use for this?

[rigid wind]

what are you getting? Also what model are you using? I would say that most likely what you're going to want to do is to inpaint to fix it up but hard to know without seeing your results

I am really struggling to get this pipe wrench in his hand to look right. For some reason it really doesn't want to put it in his hand it really wants to be behind his hand, or the pipe wrench disappears all together.

I've been trying to inpaint it using a myriad of different settings but nothing seems to quite work. I have gotten some settings that seem like they will work but then I run 10s of generations of it and this is the best I've been able to get it.

[vivid plaza]

When browsing community feeds, i see a lot of different ways of writing prompts. Specifically how people seperate the commands. Some just write (thing, thing, thing, ...) and some do (thing. thing .thing. ...). Is there an advantage of using period over comma or vice versa? Is it same?

[tired vigil]

You can do whatever you want
There's no real difference from what I see, I like using commas and a space after to make things well organized

Anything you type, even a space in the prompt, will make a difference

Might be very small but they are all accounted for :p

[vivid plaza]

ok thx!

[little granite]

why is it no longer possible to upload images to generate from?

[slender pasture]

Im making a image with my anime girl- But I can't get it to make a background AT ALL.

I've put "(scenic), outdoors, sunny, pumpkin patch, arctutecture "

and the backgrounds still empty

I put "empty background" and "no background" in the undesired and it did nothing

Also- Everytime I give her FANGS it makes her has some kind of f***ing hair accessory and no matter what I do I cant get it to remove them even with "hair accessory" or "up hairdo" in the undesired prompt

[slender pasture]

https://tenor.com/view/family-guy-stewie-griffin-gone-crazy-insane-letter-l-gif-17135412

[silver valley]

what do you mean with empty`? literaly blank ?

[slender pasture]

example: (ill post it when the image is complete)

I mean it's not making the background I ask it to make what-so-ever

the other images had even less in the background- like floating rose petals or something

Which Im so fucking confused because the settings are the same as I've always had them-

and I've made some pretty fking amazing images before with dynamic backgrounds and the codes were MUUUUUUUUUCH more complicated that THAT

[silver valley]

i got this with:
1girl, (masterpiece, best quality:1.4), vampire girl, open mouth, fangs:1.2, teeth, in pumpkin patch, sunny, architecture, intricate,

[slender pasture]

does it matter the model thats used?

[silver valley]

yes

[slender pasture]

😐

[silver valley]

some models can do stuff better than others

[slender pasture]

I mean Ive been using the same model so i dont see why it'd just stop making backgorunds out of nowhere

These are all from the same model Im currently struggling with >:[

[silver valley]

maybe because you used the silhouette lora?

[slender pasture]

👀

i used it on those images too

and the barbie one

and a couple others 💀

[silver valley]

hmm

[slender pasture]

(easynegative:1.1), (badhandv4:1.5), worst quality, low quality, normal quality, lowres, bad anatomy, extra toes, extra feet, missing toes, missing feet, extra legs, missing legs, extra hands, missing hands, bad hands, text, error, missing fingers, extra digit, fewer digits, (cropped fingers), jpeg artifacts, signature, watermark, username, blurry, quadstar, empty background, no background

this is my negitive prompt

[silver valley]

remove the background tags

[slender pasture]

i just tacked those on like 5 mins ago in a last ditch effort to try and fix it

ill remove them now and try again

ok its improving

maybe I need to like add ( ) so some things 😐

[silver valley]

pumpkins are there

[slender pasture]

yeee its a bit

[silver valley]

but yea its still empty

[slender pasture]

Its very frustrating cuz youd think "sceneic" and "outdoor" shoule make it be like "ah yes- I need to make trees and a sun!) and pumpkin patch you'd think might actually look like one too but 😐

oh well i guess ill figure it ou

🤭

[silver valley]

yea its maybe only one word that triggers that background xD

[slender pasture]

🤔

cant say thats 100% as Ive said some of my more complicated codes seem to understand better than this 😐 LMFAO

💀

This one is tagged in my server under #Top Codes 🤭

✨ boop ✨

somtimes the more complicated ones actually turn out more like what I WANT

anyways im back to it ❤️ thanks for the lil help~

[silver valley]

yea sometimes they work good or if you just throw some random stuff together ^^

[silver valley]

np 🙂

[slender pasture]

Why is it when Im watching the image be made, it will go from like

idk it looks like a great image but then like in a second it changes the WHOLE THING

nad now she has like 5 arms and a 5 mile long neck 😐

2 heads 💀

ooomgg

(think i know why im just being silly tbh) its so weird how fast it can go from 👍 to 😬

Guess this is best in #1019361238234443776

[silver valley]

Hires fix didn't fixed it xD

[slender pasture]

NO IT DID NOOOOTT D:

😂

I just decided to go back to one of my old propts that I know works* 🤷‍♀️ Hopefully I dont fk this one up 🤡 Im just glad I have my bases saved n_n

[trail zinc]

nobody can help me?

[slender pasture]

Yeah so- It's still fucking doing some weird shit- I restarted the whole program and it's not doin' any better 💀

I mean I got backgrounds now! That's cool and all 😐

[trail zinc]

I want to create random robots face with exactly this logo design (one color blue logo and flat design). please suggest some Stable Diffusion checkpoints for this

[low lodge]

Hello, how to make the same character on 1 image, both from direct straight face and profile view ?

[little granite]

why is it no longer possible to upload images to generate from?

[spring tiger]

does anyone have any tips on how to make my generations 'simpler' in terms of clothing and background, but keep the character detail?

my model seems insistent on adding wings, and halos and crazy background

[slender pasture]

makes me think putting fantasy in the undesired prompt might help- but then again it might make it do that to the whole image, and that may not be wanted either 🤔

Mines doing the same when I give my model fangs 💀

[spring tiger]

I'll try that

[slender pasture]

🤷‍♀️ a shot in the dark tbf

why not hope it helps

[spring tiger]

I still seem to be just getting wild backgrounds and overly elaborate clothing

I just want like simple tshirts and stuff haha

[slender pasture]

intricate clothing in the undesired*??

again i have no idea- just a guess

orrr do you mean like

casual clothing?

[spring tiger]

that one may have helped a bit

I guess I just keep adding more and more stuff to the negative lol

maybe we'll get there one day

maybe I just switch out of comfy and use a111 for reference only to get a base

[slender pasture]

Best of luck n_n

Im currently trying to figure out how to get mine to quit making these:

'detached sleeves'

[spring tiger]

(((((((Sleeves)))))) in the negative xd

weird thing is if I switch to a111 it generates stuff way more in line with what Im looking for even without reference only. So strange

oh maybe because I didnt have loras turned on... wonder if its a lora causing the issues in comfy then

[orchid ore]

too much ((( causing unpredictable with prompt issues.

[slender pasture]

yeah i wont go over like 2-3 ((s

This is what I got it giving me + my settings

I can't get rid of the sleeves to f'in save my life lmfao

[orchid ore]

568 is weird resolution for as well 832
576 x 848 <- try this. Or classics 512x768 seems more safe

[slender pasture]

Or classics 512x768 seems more safe is what its set to?

OH NVM

WRONG SCREENSHOT HOLD ON

[orchid ore]

i see 568x832 🙂

o.k. then

[slender pasture]

Here we go- this is what it's at- I changed it since then- Sorry about that ❤️

[orchid ore]

you are as well using it seems older version of A1111, if you have trouble, try to upgrade to nowadays 1.6.0 Or update extensions

[tired vigil]

"showing (body part here)" works well for me

[tired vigil]

Cursed 1.6

my GPU can't run 1.6 for some reason

[orchid ore]

yes but if issues, what to do then. o.k.

[tired vigil]

Reduce upscaler denoise to like 0.5

And I'm not sure but you could try another upscaler, some might increase the randomness of the final image

I love the ultrasharp ones ( can't really use it all the time it's slow my GPU bad)

[slender pasture]

oh huh i never thought about that tbh I am using XL but my pc cant really use many XL models so im just using the regular ones

upscaled from 2 to 2.5 and my pcs not having it

🤡

[silver valley]

What's your GPU ? With the 1.6.0 version SDXL runs better

[slender pasture]

uh

these are questions for my husband 🤡

oh thats his geforce thing

its a geforce rtx 2080

[silver valley]

Ah well then it can run any sdxl model if your on 1.6.0
My GTX1080 can it too

With --xformers --medvram-sdxl --no-half-vae

[slender pasture]

huh O-o im just pc illiterate and dont know what im doing then LMFAO

Either way I have it making images now that I semi like so I guess I can't complain

Yall have been very helpful

I appreciate it alot

x

[silver valley]

if you want to update the webui in the future feel free to hit me up in #🤝｜tech-support

[toxic dragon]

This is an annoyingly simple question but in ver 1.6, wher did the save prompt disk file icon go?

How do you save your prompts now?

[charred ore]

is there any way to quicksave my current profile setting?

so when i reopen it i can quickly get back to my preset

nvm found it

[fluid solstice]

What would be the best way to achieve things made of materials they are not usually made of?
For example, a human made of veggies, a turtle made of bottles.

[boreal pecan]

Idk I tried with a plant made of metal and I faild

But I heard sdxl is good for that

@fluid solstice but I didn't tried yet

And also I wanted to ask here guys how do you properly put negative promt to avoid people or silhouette's or hahaha weird things that look like a person. I tried lots of different convinations and I can't found one that works always.

,robots,silhouette,men,man,woman,women,android,person,lady,adorable_girl,cyborg,human,humanoid robot,

example

[bitter hornet]

Hi guys, does anyone knows what happened to the controlnet option? Why can't we upload pics anymore?

[slender pasture]

Was in here yesterday gripeing about not being able to get a decent background-

I saw someone do something with their code- and decided to try it 👀

It works, not complaining.

So- What I did is I went to ChatGPT and asked it to write a "dark romantic poem describing the following scenes" (with a list of like 5 scenes i want)

My code for stable diffusion was set up as follows:

Quality tags-
Model (1girl, x hair, x eyes, ect)
clothing tags
lora tags
loras
paste the short poem

I need to keep testing it with different short poems, but its been giving me very beautiful results ❤️

[tired vigil]

What are you trying to get

It also depends on the model

If it's heavily trained to make people then find another one

I was trying that the other day and literally just describe the image as much as you can and it didn't make people all the time

[tired vigil]

Doesn't it have an option to allow you to upload an image? "Upload independent image" or something

Also it defaults to using the same your img2img has

[boreal pecan]

so im making an animation with SD CN Animation and u know mainly there is no ppl in the images of the animation (frames) but sometime there is one

that ones are good,

but sometimes:

and im using Gosthmix because is the one that give me best results with the animation results

the other ones is just weird stuff as resulty

[tired vigil]

[tired vigil]

I dunno much about animation

Try some big negatives like (people:1.4)

Go high on it

[boreal pecan]

is ok, u are a kitty, I understand it, u.u

I will try and let u know

if it worked

[tired vigil]

Stick to mostly humans

Or is it actually making robots

Lmao

[boreal pecan]

haha it was making robots

or like just silhouettes

like not humans at all but like weird

[tired vigil]

Try "using a computer"

[boreal pecan]

oki

[grizzled gull]

Hello i have been using comfy for 3 weeks now, question is: How can i enable the progress of the ai that is shown in the ksampler node. Thank you

[upbeat phoenix]

Hey, I found that the only way to have reasonable times with image generating (despite having quite the decent pc I guess?) is by doing 512x generation and upscaling it to 1024
the image though comes out quite lowres
anyone knows how to make them look better?

like the excact moment you zoom in its very blurry

[finite pendant]

How do I get a "low angle"? I want it so the character "stands over" the viewer and looks down, but standing over viewer, looking at viewer, looking down doesn't work.

[brave fulcrum]

hello I have a small question, is it normal that inpainting gives extremely blurry results?

like in the zones I put the mask on?

[tired vigil]

@upbeat phoenix what is your GPU ?

[tired vigil]

looking at viewer, from below

or view from below

That gets you there, depending on how close you want, try adding "close up" and insert body parts that you want to be in the image like "waist"

[upbeat phoenix]

3070

[tired vigil]

Hmmm

Something seems wrong

What times are you getting on image gen

like iterations per second

for those 512x512 or so images

[finite pendant]

Thanks. What do I use to get it further away when full body doesn't show the whole body or cuts off the legs/feet/head?

[upbeat phoenix]

wdym

[tired vigil]

full body is the exact prompt to show the whole body, are using a realistic model ? most anime models do it fine

[tired vigil]

like try a 512x512 whatever prompt image, use hires fix to upscale it by 2x, it should take you like 14 seconds for that image

[finite pendant]

I'm zig-zagging through models, just trying to find ones I'm comfy with.

[tired vigil]

@upbeat phoenix do they take longer than that ? 3070 is very good for SD

[upbeat phoenix]

yeah that's more or less what it takes
but ther esult looks like what i sent

maybe i should change sampling method? dont evne know what htey are ;-;

[tired vigil]

hmmm

Okay I took a look at your image info, looks fine, wait a second

@upbeat phoenix try this https://civitai.com/models/116225/4x-ultrasharp

download this model, put it in your path to your sd\stable-diffusion-webui-directml\models\ESRGAN

And then try generating something similar with hires 0.5 denoise

and that model instead of latent

also

those negatives, do something better, its not very good to have things like (extra arms) and stuff, they don't really do much

@upbeat phoenix download these https://civitai.com/models/7808/easynegative
And https://huggingface.co/SweetLuna/Aurora/tree/main/AuroraEmbeddings

put them in your SomethingSomething\webui\stable-diffusion-webui-directml\embeddings
and do a simple negative with them like this one I like using: negative_hand-neg, (KHFB, AuroraNegative:1.2), (Worst Quality, Low Quality:1.2), border, grayscale, (watermark, multiple views:1.2)

you can probably remove multiple views my model just really wants to do multiple views

the positives you have look fine, see if the images come out better for you with those

[tired vigil]

What is your gpu

is it amd by chance

and show the settings you use on inpaint

[brave fulcrum]

Ill send it tomorrow but if I remember right it started to happen already with default but changed to only affect the masked part and latent somethimg instead of original

And yeah amd, rx 6600 with 8 gb

[tired vigil]

oh no

You might be with the same problem I have haha

[brave fulcrum]

Could that be behind some other issues I been encounterimg?

[tired vigil]

With inpaint you should only change the setting to so it inpaints masked only and keeps original

if that isn't working you might need the no-half argument

which is my problem

my images either don't change or become a blurry mess

But controlnet inpaint still works, so its not all bad

[brave fulcrum]

I have the second happen with inpaint and the first one with adetailed or however is called

[brave fulcrum]

Is that an extension?

[tired vigil]

yes

lots of useful things controlnet can do

[brave fulcrum]

Ill install tomorrow, rn in bed

Mind me asking what can it do?

[tired vigil]

It does a lot of things...

Like inpaint is one example
it can tile so you upscale images by however much you want and not losing quality
openpose you can get a pose image to get your generation in that same pose
depth map, canny, softline models can guide your generation, imagine you want a moon and a mountain in a certain place, you can draw a black and white picture (or tell controlnet to extract it from another image) before genning something

you can use reference to try and generate the same person or style 2 image, to get the style of an image you send to controlnet

animations...... lots of things depends on what you want

[brave fulcrum]

Oh, dang, thats a lot, thank you ill check it out tomorroe

[tired vigil]

my inpaint does this without --no-half (and with that argument basically nothing works for me, my gpu is very slow no-half makes it even slower )

either nothing changes (tried prompting for blue eyes) or does the blurry

controlnet inpaint works for me

[upbeat phoenix]

i have the first, what's the second?
Now I tried generating something with the gacha splash art lora
but the images size required are so big that it takes 30 mins for a single pic

[tired vigil]

@upbeat phoenix What do you have in your webui-user arguments ?

@upbeat phoenix the second link is for embeddings too, just different site

they go in the same folder

[upbeat phoenix]

what do you mean

[tired vigil]

here, what do you have in that

[upbeat phoenix]

in the bat file right?

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=

call webui.bat

[tired vigil]

you need xformers at least, try with
set COMMANDLINE_ARGS=--xformers --no-half-vae

I'm not sure if no half vae is good/better/necessary, saw some people recommending it for 3070 on #🤝｜tech-support

Go in there and ask what webui arguments is better for you

[upbeat phoenix]

dafuq its faster now

[tired vigil]

yes its for nvidia cards, some optmization stuff very good

[tired vigil]

my friend who has a 3070 doesn't use no-half-vae so it should be fine to remove it, are images looking better now ?

[upbeat phoenix]

well yes cause they're like double the size i was uising, thanks!

[wraith tulip]

how do i make it seem like someone is behind glass? Is there a lora for it? I'm struggling to get a girl in a containment tube, and I need some help.

I can get her in front of the tube, but I can't get her properly in it, so I'm imagining I need to put a "behind glass" or similar tag in

[tired vigil]

@wraith tulip I've seen people try that and its pretty hard
Looking through chat history someone used this prompt to get a woman behind glass
"in a wall, transparent window of glass reflecting sky, (woman covered by translucent reflections:0.2)"

also you might wanna checkout some loras on civitai if you just search glass ((and enable the +18 filter you'll find some that are not particularly the context you might want but maybe at a lower strength it works for generic purposes))

[wraith tulip]

Ty! 💜 I'll definitely try some loras and that prompt suggestion

Damn! I found the perfect lora but it's SDXL!

[tired vigil]

Take images of other people using it and make a 1.5 lora

Big brain

Also try those that I mentioned

Which are for people doing naughty things behind glass but try it with lower strength, it might give you the result you want

Good Loras are flexible :b

[brave fulcrum]

that is exactly what happened to me

Should I be able to find controlnet in extensions by searching for it or should I install from url?

[upbeat phoenix]

if i make a rough mspaint sketch of what i want which function do i have to use on sd
inpaint, inpaint skethc or img2img?

[brave fulcrum]

Any recomendation to get good faces?

[mighty vigil]

I'm trying to learn how to use stable diffusion and copying some workflows from civtai to understand how everything works, but for some reason most of the time my end result is completely different from what I'm seeing at civtai. Take this one for example: https://civitai.com/images/2784714?period=AllTime&sort=Newest&view=feed&collectionId=66689

I'm using the same model, loras, prompt, negative and everything else

Does anyone know why?

[orchid ore]

for sure i believe you have wrong resolution? Who would generate for civitai in 544x960? Going to check it

616x768 @mighty vigil

He resize it later to 540x960 with hires fix

[silver valley]

upscaling or inpainting

[mighty vigil]

how can i see the original resolution?

[silver valley]

you have to click the Copy Generation Data Button. There are All the hidden informations stored. Then paste that into your Positive Prompt line and then press the blue arrow on the right. That will apply all the settings automaticly

[mighty vigil]

ooooh nice

didn't know that

thanks

[silver valley]

no problem. you also need to download a vae file for color correction

[mighty vigil]

how do i select a vae when generating images?

[silver valley]

Go into Settings > User Interface > Quick Settings there add sd_vae.
Then hit apply and reload ui button

[mighty vigil]

nice

thanks

[silver valley]

np, you then have a dropdown to select different vaes. put them into models/vae
take the 8400-mse for example

[bitter hornet]

not in discord, I wanted to generate img2img through stable diffusion on discord channel

[azure timber]

I'm trying to replicate something like this(see link below) in Automatic 1111 but am struggling to do so.
https://github.com/sail-sg/EditAnything

I was thinking that I can do an inpaint on the target image while using the clothes as a a control net reference.
However, I'm just getting jibberish... can anyone give some pointers?

[tired vigil]

hello haha
Can you show what you're getting as result

is it either nothing changes or a blur by chance

[azure timber]

yea! basically the area that I select to become inpainted becomes something similar to this.
similar to this discussion on reddit
https://www.reddit.com/r/StableDiffusionInfo/comments/149tx00/comment/jo7rpwj/?utm_source=share&utm_medium=web2x&context=3

[tired vigil]

my gpu is too slow to use these controlnet models

are you on amd ?

[azure timber]

i'm on a M1 macbook actually!

[tired vigil]

are you using cpu only or does it have a gpu

in my case and some others I need to add the argument --no-half

maybe that helps, inpaint only works for me with that

[hallow pelican]

im getting this when its supposed to be 2 girls sitting on a bench in the forest, along w/ other random monstrosities

[tired vigil]

put the first region as background and the others are foreground (however many you'll use) if they arent

[hallow pelican]

It is the background

[silver valley]

Can you show the settings you tried ?

[hallow pelican]

And in the regular t2i promt is just 2girls

No negative because I forgot

[silver valley]

What's your image size ?

Because if you want 2 characters in one image you should use at least a resolution of 768x512
Then make the squares for each character bigger

[hallow pelican]

they were 960 x 640 and hires is turned on

[silver valley]

After changing resolution you have to click create txt2img canvas again. Because your image looks like a sqaure

So it used the wrong resolution

[rigid wind]

If you want to change the colour of something but not the shape of something you still need a high denoising strength right?

[upbeat phoenix]

before I lose the picture
I generated a pic but it's not quite right
I tried to go into inpainting but instead of changing the face it generates another picture inside the face
what can i do?

[rigid wind]

what are your settings? Also you haven't lost anything. you can find all your generations in your output folder

[upbeat phoenix]

what settings

[rigid wind]

for your inpainting

should look something like this

[upbeat phoenix]

inside the hand there's now a smaller girl instead of a hand

[rigid wind]

the reason for that is probably going to be because "Only masked, padding pixels" is set to 32. That means it will only consider 32 pixels around your mask when creating the generation. You also have a high denoising strength so it's going to be replacing 75% of what's there. If you just want to touch up a face try like 0.2-0.4

[upbeat phoenix]

face is done, slightly lighter than the rest of the skin but it's cool

whole picture is sprouting abomination hands and they're b/w
wat

[rigid wind]

Uhhh what?

[upbeat phoenix]

look
the face is now fine
but the hands? what's goin on

she got the power of tv static

[rigid wind]

Oh wow I really like the colours

[upbeat phoenix]

yeah that's why i really wanna get it fixed cause the pic is super nice

[rigid wind]

So to fix the hands you're probably going to want to use a different prompt and put things like "bad hands, extra fingers, bad anatomy, etc" in the negative prompt, you can also explicitly say "4 fingers, 1 thumb" in the positive prompt. Hands can be an absolute bitch sometimes

[rigid wind]

so it's just gonna be a matter of running more inpainted gens on the hand, use whole picture with a high denoising strength (probably in the 0.7-0.9 range) and run it until you get hands with the structure you want then when you have ones that are structurally correct go back at it with masked only, with a high padding value so that it continues to consider a large amount of the picture for context and a mid denoising strength like 0.4-0.6 (depending on how much it needs to change) and then when you find one that you are pretty happy with run a lower padding size and an even lower denoising strength to add in the detail

It's important to try to keep enough context considered that it doesn't start mixing up left and right hands, I have unfortunately gone through the process just to realize that I gave them 2 left hands xD

[upbeat phoenix]

i'm not understanding the color issue though
it's like going black and white in the inpainted area

[rigid wind]

are you still using the same prompt?

and what is your padding area? Also you're not accidentally in inpaint sketch are you?

[upbeat phoenix]

nono i'm in inpaint sketch
same prompt i just added good hands
i was using whole picture, trying with masked area now cause whole thing wasn't making any progress

ok masked area only gives a more accurate result, still weird abominations but they're beter

i've read around to get a hand structure than img2img it but there's no hand tow ork with here

[real echo]

anyone knows any lora for this gesture? cant find any

[fluid solstice]

What is the current best way to prompt 2 different characters in the same image?
Just region extensions or is there any other way?

[hallow pelican]

now its creating images in the correct resolution but it's still ignoring the region promts entirely

im trying to mimic your example but with two girls sitting in a forest

[rigid wind]

sorry I had a mettering. The reason you're having the problems is because you're in inpaint sketch meaning that that white mask you're putting is in affecting the colour, go to just normal inpaint

[rigid wind]

control net, or loras

[silver valley]

hmm can you show the whole settings used ?

[upbeat phoenix]

I swear to god i'm in inpaint

[silver valley]

can you try without hires first

[rigid wind]

lol then you typed your message wrong you said "nono i'm in inpaint sketch" xD

[upbeat phoenix]

woops

anyways i made a little progress, now I have one weird hand but at least recognisable hand

the other hand got turned into a face though

[hallow pelican]

i got this

[rigid wind]

you could also try to use control net with an openpose for hands but idk how well that would go in an already generated image. You also might want to use a different resolution for the hands 768x512 is too big for a SD 1.5 model which is probably why you got the second person in the original generation, but it will also make it hard to get good hands with that big of a resolution in a 1.5 model

[silver valley]

thats really strange

[hallow pelican]

yessirrr

[silver valley]

okay try first with only 1 character

if this works you can add the second

[hallow pelican]

didnt work, just a random anime girl not the character lora

[silver valley]

are you and amd or nvidia graphics?

[hallow pelican]

nvidia

[silver valley]

okay is your webui updated and the extensions?

[hallow pelican]

yea

oh wait

lemme see

aw man i gotta reapply on the settings XD

it didnt work sadge

i think the background one is functioning though

but i can't straight up tell if its doing random bgs or not

oh yea no it seems to be doing random things

[tired vigil]

Not for the extension but cut those negatives a bit, keep the textual inversions and do simple negatives with them like this one: (Worst Quality, Low Quality:1.2), border, grayscale, (watermark, multiple views:1.2)

Also when I tried bad prompt version 2 it altered the image a lot

But it you like it it's fine

Also try regional prompter

Might work for you instead

A bit harder to use but way better I feel

[hallow pelican]

I'll try

I can't tell if it's working
Probably cause I have no idea how it works
I'll try more with it after I get home

[upbeat phoenix]

I made some progress by photoshopping two png hands over where the hands should've been but they are of a different color, is there a feature that helps me with making color more omogenous?

[tired vigil]

you can do img2img

But hmm that might change the overall image a bit even at like 0.2

ínpaint should work better, cause its basically img2img but you target the area

[upbeat phoenix]

but that will remake the hands won't it?

[tired vigil]

it should keep them similar to your "better hands"

and homogenize the colors a bit

you can also do controlnet canny/depth to keep the same shape on things

[upbeat phoenix]

i'll try

meanwhile im doin it with photoshop

[fluid solstice]

I mean, but how exactly do you do it? Just adding 2 character loras and using controlnet over an image with 2 characters? Wont it try to mix both?

[tired vigil]

Getting a lot of gens where my subject is cropped from the middle of the head up. I'd like for them to be more centered. Any advice for prompting?

[silver valley]

a protrait resolution can help

like 512x768

[tired vigil]

That's actually what I'm using currently

[silver valley]

oh okay

and the head is cut ?

you can try portrait as tag

or modelshoot

these usually never crops

[tired vigil]

oo, modelshoot is new. Will try both. Thank you

[silver valley]

no problem 🙂

[tired vigil]

Hi everyone, is it possible to include a reference image in our prompt ? i need to do some images with a particular product in it

[cold coral]

Can someone help me please
I'm trying to submit an image to change the style or add on.
Where can I do thay

[hallow pelican]

Anybody aware of what | does, ive been seeing it in a few prompts on civit

[twin yoke]

They were likely using the Prompt Matrix script which allows you to put in several different terms and have multiple versions created based on the words between the pipes.

[tired vigil]

Isn't there a thing like {cat|dog|eldritch horror} ?

if you can help me remember what it is

[twin yoke]

That's what I'm talking about with the Prompt Matrix script. I'm sure there are other things that use the pipe symbol as well. But that's the one that always comes to mind for me with A1111.

[tired vigil]

I wanted something like that to randomly generate one of the words in my prompt, I need to take a look at the available extremely, maybe it's one there

[hallow pelican]

Ah, so if I copied and used it
It wouldn't do much of anything for me since I don't have the Promt Matrix thing myself?

[tired vigil]

prompt matrix is a script that you can enable at the bottom of the page

it will make images with each of those and compare side by side on a grid

i prompted this a woman in {white|red|black} evening gown, {blonde hair|red hair|brown hair}, {blue eyes|yellow eyes|red eyes} and its generating images with all those attributes one after the other

[tired vigil]

oh so it works by default I'm not crazy ?

its working as regular prompt

Like it made a woman in white evening gown, blonde hair, blue eyes first ?

the patterns seems to be random, i have it on generate forever

hmmm :D I need to try then

[hallow pelican]

Oh so I can use it for like
A ton of alternative outfits

That's p cool

[tired vigil]

yeah so far i see the images remain in those pre defined combinations

[hallow pelican]

Sick, thanks

[tired vigil]

lol np, im trying this but its pretty random

[tired vigil]

used a simpler prompt 1girl, {blonde|brunette|red head} dont think its picking up on the variations

possibly it would require a script

[old dove]

hi, any tips to improve the lines so they are more straight and less curvy and wobbly, this is with SDXL but I think the lines are to "unstable"

[tired vigil]

https://youtu.be/s-1L6MCVh-E

[tired vigil]

basically have to enable these options in the image then prompt something like.... a woman with {blonde|brunette|red} hair, wearing casual {orange|white|black} dress to generate images of a woman first with blonde hair and all three colors for the outfits then then brunette and so on ...

[smoky vapor]

I'm not sure I understand how styles work with the SDXL bot. Surely it just appends a user's input prompt inside a pre-made style prompt, right? I've seen things like this and wonder how StableDreamer bot works its style button because of it - https://github.com/Douleb/SDXL-A1111-Styles

Say I want to generate images with @delicate kernel with the prompt "Test" and choose the Anime style. Does my input prompt "Test" get changed to "anime artwork {Test} . anime style, key visual, vibrant, studio anime, highly detailed" with a negative prompt of "photo, deformed, black and white, realism, disfigured, low contrast" secretly without the bot explicitly stating that? How would this work with tokens sequence limit and would long prompts or even the style itself be truncated from an input prompt?

[silver valley]

By default you can mix between 2 things like.
Animal, [cat|fox], cute,
it would then switch every step between cat and fox

No extension needed for that

[hallow pelican]

I see I see, also I just came across
" Differential Regional Prompter " in the script section, any use for it?

[raw nexus]

how can I make this image sharper? Ive tried using r esrgan but I only get black squares

ive used the same setting to generate its previous image which was wayy sharper

[silver valley]

What settings did you used in txt2img ?

[silver valley]

Don't know that script

[raw nexus]

i dont really remember since i have tinkered with it to find out the problem. however the first pic was upscaled with r esrgan and the second one was upscaled with maybe euler or esrgan 4x

[silver valley]

Euler is a sampler not an upscaler

You can put the image into the PNG info tab to see all the settings you used

[raw nexus]

im so lost

oh

wait lemme see

can I dm you the settings?

too big

[astral isle]

is there a way i could change her right hand to face down from this point or would i have to remake the hand again?

[silver valley]

Sure

[raw nexus]

sent

[tired vigil]

Inpaint that area

[astral isle]

yes but how would i get it to completely rotate down?

[tired vigil]

Easiest way I can think of is just making a quick sketch of what you want and inpainting it to match the whole image

SD is pretty bad with directions but I guess you already tried "hand, back hand, hands facing down" and stuff

[astral isle]

yea thats what i did for it in the first place just wondering if i had to redo it again for a better pose unfortunately, ty tho

ill try backhand but most likely will just redo the base sketch

[tired vigil]

If you still want the whole arm in that position, you can just crop things together after a img2img and a bad paint sketch :p

Since it's out of the frame of the body

And then inpainting

[astral isle]

haha yeah thats fair enough lol. im just trying to get this inpainting thing down so im practicing on this hand

[tired vigil]

You'll have a high detail sketch of what you want from img2img xd

[astral isle]

im just making the hand in gimp rn

[tired vigil]

Also I'm sure controlnet can do it with open pose or something

Maybe combine open pose and inpaint

[astral isle]

whats open pose

[tired vigil]

A controlnet thing

You can tell exactly how you want a character

Using ""coordinates"" of sort

Look it up, you'll see

Very fun to use

[astral isle]

ok ill look into it ina bit

[wintry urchin]

Does anybody know how to use Dynamic Prompts to sometimes choose a word? e.g. I want the character to sometimes be sitting on a couch. I tried using {.|sitting on a couch} but, unless the RNG gods are against me, it doesn't seem to be taking using . ever

[maiden geyser]

Sign up now! And sorry for the advertisement, it is for an educational service for beginner and intermediate SD users. Thanks!

[maiden geyser]

Try with something like xyg instead of a symbol, make sure Combinational Generation is on

[maiden geyser]

You could first use an extension like Minipaint or use Photoshop to repaint the hand roughtly, then inpaint, it makes it much easier.

[maiden geyser]

I would try Ultra-Sharp x4, here check this site out
https://openmodeldb.info/

[maiden geyser]

I found styles odd in XL, and discovered something in testing - if you want to prompt an artist, you have to leave out other descriptors like detailed, 8k, intricate, etc to get the real style. When I would prompt the artist with any of those, the style was messed, but without those, I could even combine artists

These came out once I removed the other descriptors

[flint radish]

Wondering if there's a good way of using "BREAK" and assigning colors to things?
Should the actual color name be used, OR can I specify that it should be a random
color?

Example:
I'd like to create colorful lights OR lighting someplace; the end result often bleeds
into other parts of the scene or subject, (ex. hair or eyes of subject, clothes)

[potent hawk]

Hello 👋

Can anyone help me, how to give existing picture reference and ask to modify in it?

[smoky vapor]

This may work for artist names, but how about art styles and mediums? Have you went that deep into extensive testing or are there resources you could link me to?

[upbeat phoenix]

anyone has tips on not making everyone so muscular? masculine bodies fuckin shredded while feminine have some differentiation despite being mostly 90-60-90

[flint radish]

You could try "Muscle Slider - LoRA" and give it a negative value and see if that will generate more appropriate results for you

[upbeat phoenix]

also i'm having a hard time with making male fairies
it defaults to girls despite me saying no women

[flint radish]

@upbeat phoenix ¨
If you already haven't you could also put all these things you don't want in the negative prompt and also make it extra strong by adding weights to it
negative prompt
(woman:1.3), (female)

[tired vigil]

I never tried with guys but abdominal muscles in negative do the trick for me

[tired vigil]

Regional prompting would be perfect for that :p
Regional prompter is the name of the extension

[upbeat phoenix]

I did :(

the muscle lora seems to be working, i've also added gender slider from the same author

[flint radish]

@upbeat phoenix nice! hope it turns out well for you - you could also try the suggestions "cat" above gave 🙂 and see what works best, maybe combine

[upbeat phoenix]

idk if my pc is getting tired (i've been doin this all day) but i've goen from 30s/it to barely 10)

[tired vigil]

What gpu do you have

[upbeat phoenix]

yep, it works
i can work with this!

[tired vigil]

If you have one

[upbeat phoenix]

3070

i meant 30s/it tokens for whatever the thing it to 10

[tired vigil]

Wait

[tired vigil]

Is this sdxl

[upbeat phoenix]

dreamshaper 8

sdxl styles is an extension i'm using if that's your question

set on fairy tale

[tired vigil]

No I mean are you using SDXL or SD 1.5

[upbeat phoenix]

sd1.5

i just googled automatic1111

[tired vigil]

Then how are you getting 10s/it

Or do you mean per image

[tired vigil]

What resolution are these

[upbeat phoenix]

768x512 upscaled to 1536x1024

[tired vigil]

Something is wrong

Like it shouldn't be this slow

[upbeat phoenix]

yeah, this morning it was 3x faster

[tired vigil]

My friend has a 3070 and images come out in 16 seconds

On that same resolution

Go ask in #🤝｜tech-support

[upbeat phoenix]

i will ty

[pine inlet]

Added this to my Patreon (https://www.patreon.com/posts/sota-very-best-90744385). 1 click install and 1 click run. Batch process images and auto save captions. Also properly deloads model. I think the very best image captioning tool. Supports 90+ Clip Vision models and 6 caption models

Can be used to get amazing prompt ideas

This image captioned with Clip Vision model of Stable Diffusion XL (SDXL)

[quasi flax]

I have a seemingly simple question but I’m having trouble getting to the bottom of… what does putting a number less than one do when in parenthesis? Like (small dog:0.5), does that actually lower the “weight” of the words?

Put another way, is (small dog:0.5) actually similar to using [small dog:1.5]? They both lower the weight?

[tired vigil]

yes it lowers the weight of the prompt, I believe "(small:1.3) dog" would work better, I need to test it more :d

not sure if [anything] works in a1111 at the moment, its used to lower the weight of the prompt, but I just do (anything:0.X)

I reduce the weight of many things like smile, view angle, poses...

[orchid ore]

@upbeat phoenix have you not open hires fix? I have 3070 and 1s/it i got when in comfyUI strikes buggy AITemplate.

[lyric goblet]

Any advice on how to prompt Vedalkens from DND like this?

[upbeat phoenix]

I do use hiresfix mostly yeh

[orchid ore]

so your speeds are probably o.k.

[boreal pecan]

Hey people, Im making some animations with AI , is for a presentation that will be showed on schools about the danger of drug consumtion, they hired me to make some AI animation but more like backgrounds and stuff and now they asked me if I can make some icons or symbols of diferent "substances".... Everything is a big FAIL, is there any model or LORA that is good for objects and like, weird things?

If I dont archive this, I will just other technic that is not AI but I want to keep trying

[noble sable]

Bro how the hell do y’all get nice pretty out painted pics? I ALWAYS get a seam/border in mine. Whether I use OpenOutpaint, a photoshop extension, or poor man’s out painting. What’s the trick? I mainly want to use OpenOutpaint, and I’m trying using SDXL Inpainting model, and had the same issue with 1.5 inpainting models

[grizzled otter]

tbh I found MJ to be much better at specific subjects like this, if you can get around the censorship

[boreal pecan]

yeah well sure I can get around I just descrive it with other words, but IDK anything about free ways for MD, like well they pay me so I could be able of just buying credits but mmmhmhmhmh nah, is there any free way? With collabs or somthing?

[maiden geyser]

Inpaint the border at denoise 0.15 to smooth it out, resend, redo

[maiden geyser]

This still leaves it in the prompt, but less that normal. Remember, weights are all relative. So if you have 0.5 and 1 and 1.5, its the same as 1, 2, and 3 for everything. I often use 0.5 weight on LORAs, as well as things like (fit:0.5) to just give a slightly in shape look

[maiden geyser]

certain keywords, although offensive to some, work. Try overweight or fat or soft, or put something, or put (athletic:1.3) in the neg prompt

If anyone needs more prompting help, feel free to PM me, here is some of my work #📝｜prompting-help message and I guess my ad was removed, I am an educator and prompt engineer.

[upbeat phoenix]

I downloaded the weight slider Lora and it works wonders

[maiden geyser]

Oh I DLed that but did not try

nice, will try tonight!

I am not sure why admins removed my link, as I see links to Paetrons and such, but I am running a seminar for training prompt engineers. PM for details. If not allowed, just remove the post admins, sorry I am not clear on the ad rules. I see other links above, and this seminar is educational, and tbh is being put on because I have been teaching for 30 years and want others to know what I experienced and needed to do at my prompt engineer job. Thanks.

[wooden ore]

Hi, I am new in the world of stable diffusion and I am looking for advice. I would like from a video recreate the face of the person but on fire. And I can’t do it on deforum. Have you some advice:)

[supple torrent]

is it possible to write a prompt that will generate 360° sky (cloudy, sunny, cloudy sunset, etc.) seamless panorama without any problems?

[snow herald]

can some one help me make a promt that will give me this kinda of result,

or just give me some tips bcs i havent used stable diffusion before i only used midjorney

[upbeat phoenix]

has anyone found a good way to generate symbols? cause so far i'm doing characters on SD but logos and icons on bing

[noble sable]

Just use the same prompt for the inpainting? Now I’m having an issue where it extends the design really well but the color is always off or washed out. Idgi

[tall prism]

Is there a way where I can gain complete control over the output being generated? Like, I want to specify where each object in the scene goes. For instance, I want to make a comic page, so I want to be able to specify where each panel is created and what goes in them. Is that possible?

[tired vigil]

yes, very possible

I believe on comfyUI you can do some black magic for that stuff to come out better but either regional prompter or multi diffusion region control should be able to no problem

[tall prism]

Oh. I am using a1111 hosted on Docker and called through RunPod

Sending prompt through a Python Script which is a Google Cloud Function

[tired vigil]

so you can't use extensions ?

if you can't do with extentions then... I dunno

generate them at the resolution that would fit each section of the page

[tall prism]

I actually can, but ComfyUI is not an extension right? It's like a platform like WebUI

[tired vigil]

ye

Regional prompter and multi diffusion are both extensions

You can still try those

[tall prism]

Do they require actual input from user like inpaint? Like, I mark which area needs to be changed after generation etc?

[tired vigil]

https://github.com/hako-mikan/sd-webui-regional-prompter/#how-to-use-via-api maybe this is what you're trying to do ?

I am honestly a bit clueless on how your thing works

normally you just enable some stuff and through prompt you tell what goes where

no need to send images or paint anything like inpaint

[tall prism]

Yess exactly like thatt, thank you so much! will look into this.

I have tried using Adetailer extension but for some reason, it isn't working. Like normally, it throws an error saying ADetailer is missing although it clearly is there but I need to fix that anyways so I will look more into this Extension and try to use it too!

[tired vigil]

Good luck :D

[tall prism]

Thank you!

I want to generate two characters through one prompt. My prompt is like this:

Full Body Shot, Two People, Standing, Character 1: (male: 1.5), 4 year old, (child: 1.5), (brown coloured skin: 1.3), (blue eyes: 1.3), (mullet hairstyle: 1.3), (red coloured hair: 1.3), (Wears rounded styled glasses: 1.4), (brown coloured clothes: 1.3), Character 2: (female: 1.5), 6 year old, (child: 1.5), (white coloured skin: 1.3), (red eyes: 1.3), (wolfcut hairstyle: 1.3), (pink coloured hair: 1.3), (yellow coloured clothes: 1.3)

But it is only generating one character by combining both the characters as shown in the image below. How do I ensure that it generates two separate characters?

[tired vigil]

Were you able to use the extension ?

It also helps with getting multiple people in one prompt with distinct features on each

also "character 2" isn't gonna help I believe, try "multiple people"

When I tried that without extension it was pretty low success rate without but I did: "prompt for the 1th person and all the stuff BREAK prompt for the second one"

It helped separate them with getting eye colors and hair right but low chance of it all working, I imagine it would be really hard on your case, you'll need the extension to do its thing

[tall prism]

I haven't used it yet since I would need to recreate the Dockerr image

[tall prism]

Yeah 😔

[tall prism]

The extension is working! Thank you! I am yet to figure out how to properly use it since the Documentation is not really useful

[tired vigil]

https://tenor.com/view/cat-cats-pet-cat-cat-pet-cute-cat-gif-24810247

[maiden geyser]

Not always. Say the border that is fuzzy is grey, use grey background, smooth as the prompt

[charred ore]

if im looking to transfer the colors from image left to image right is there a good method

this is photoshop not sure if theres an easy way to achieve with SD

[hardy tulip]

how do is end prompt to get images

do i send

[tired vigil]

Have you tried style to image with controlnet ?

It does something like that, see if you can get it working
example

To get the exact image but with another style I believe you can do it with other models together like canny/img2img/depth... the possibilities are endless good luck

[solid forge]

how do I set aspect ratio to 9:16?

[zealous bobcat]

How can I achieve this illustrated look? I need to keep the car's fine details though.
So, the image below has been hand-illustrated / traced over a real photo.
I need to create several more images in this style, but of different cars.

I've tried in SD, and used ControlNet, but its all very new to me and I can't nail the look exactly.

I need to replicate the graphic-novel look, but keep the finer details of the grille/headlights etc... img2img I guess? But then what?
Thank you

[charred ore]

i havent yet i will look that up thank you for suggestion

[noble sable]

why are the colors always washed out when I try OpenOutpaint? It seems to outpaint the patterns just fine, but with the same exact prompt as the original image, the colors are off. I can use the same prompt 100x in t2i and get the same color scheme every time. Here are my settings and initial image. This was outpainting, not i2i, but the right half of the init image is colored manually because I was trying to brute force with i2i, but it didn't work. didn't denoise enough no matter what I did when I tried i2i.

I just wanna expand this damn image 😫

I've tried including up to like 90% of the pic in the outpaint bounding box but same results

[tired vigil]

@meager tangle

[meager tangle]

thank you! ill look into A1111

[tired vigil]

https://www.greataiprompts.com/guide/install-stable-diffusion-in-windows/?utm_content=expand_article

[molten moat]

help

help

Please help, after installing the github link(a1111) I try to run the file to start it and I get the following error in the console:
Creating venv in directory C:\Users\TITIR\a1111\stable-diffusion-webui\venv using python "C:\msys64\mingw64\bin\python.exe"
venv "C:\Users\TITIR\a1111\stable-diffusion-webui\venv\Scripts\Python.exe"
The system can not find the path specified.

[tired vigil]

Ask in #🤝｜tech-support if you couldn't figure it out yet

👍

[queen urchin]

hello guys, can you help me here

im struggle using charturner

i got this result double hips 😦

[tired vigil]

Triple!

Try using a more "normal" resolution

Models like 512x768

[queen urchin]

oh yeah triple sorry 😄

thank you so much, i will try that

[molten moat]

I got a question

If you can help me in the topic of nsfw

If a download stable diffusion with a1111 it will be automatic able to do nsfw content?

[tired vigil]

@molten moat yes I guess >.>

[robust lion]

Hey guys so check this out, this website is designed for prompt engineering and offers a search feature for millions of AI art images generated by models like Stable Diffusion, Midjourney, and more. You can upload an image and receive a prompt in return, which can be used for further experiments or creative projects.
https://prompthero.com/

[molten moat]

When I try to make an image I get the following error:

OutOfMemoryError: CUDA out of memory. Tried to allocate 26.00 MiB (GPU 0; 4.00 GiB total capacity; 3.36 GiB already allocated; 0 bytes free; 3.51 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

[molten moat]

heeeeeeeelp

[tired vigil]

Try to use the --medvram or --lowvram flags. --xformers can also help!

[rain tendon]

Trying to get images in Stablediffusion to do something like this. I can't seem to get it with Open Pose, and common key words like high angle shot, view from above, Overhead shot, birds eye view, etc. This was generated with Bing's Image generator, but I'd like to do it on SD. any tips?

[queen urchin]

hello guys, is there any tutorial how train lora for clothes only

[thin creek]

Hey how do i use the XYZ script for different prompts

[thin creek]

sry i have no idea : (

[tired vigil]

Hello everyone, what would be a good prompt for base SDXL for a close-up of a face?

[tired vigil]

Hey, could you tell me your offer?

[maiden geyser]

@tired vigil I will dm you

[tired vigil]

{|sitting on a couch} should work.

[worn copper]

Um, should I use articles - a an the - in prompts or rather avoid them in sentences for SD?

[orchid ore]

i think you would as there is classic prompt "A photograph of an astronaut riding a horse"

[uncut glade]

How can I stop or change a LoRa in the middle of the generation process?

[stiff bison]

When I promt "a girl with pink hair,..." I always get pink influence in buildings, furniture and so on.
Can I use the extention "regional promter" with the BREAK commands somehow to avoid that?
In only get it to work with "regions", but i dont want a specifit region, i just want overall distinction from the haircolor and the rest of the photo

[silver valley]

If your using an anime model try the cutoff extension

[tired vigil]

thats exactly how regional prompter can work

You instead of selecting regions you go on the prompt settings (Theres prompt-ex or something, its one of the modes inside the extension)

it does regions automatically, it will make a mask for each thing you do a break for
You just need to adjust some treshold settings so it works for your prompts (Im not good with it good luck )

One for the girl in question and the rest wouldn't have her prompt

[glad thistle]

Anyone got a prompt that creates a clean visual representation of a neural network? Or else a Lora or some embeddings? I'm lost and can't create one in SD (Automatic1111 webUI),.. any help appreciated, thanks. I tried Img2img and it just copies the existing one I got from the interwebz,.. hardly original or accurate.

[hybrid thunder]

What went wrong ?

what is this xdd

why is there a head on the helmet

Can someone help ?

This is CMD

[tired vigil]

Try hires 2x upscale, 0.5 denoise and 10 steps

And you don't need a refiner

[tall prism]

How do I get different characters in different regions using Regional Prompter?

masterpiece, best quality, child, kid, ADDCOMM Kid 1 in astronaut costume ADDCOL Kid 2 in Sailor costume ADDCOL Kid 3 in Race car driver costume

This is how I am getting it now, but this is not what I am expecting

[hybrid thunder]

pretty blurry and bad background

[tired vigil]

Can you try another checkpoint ?

[hybrid thunder]

what is checkpoint

I only have this option

[tired vigil]

https://civitai.com/models/4384/dreamshaper download this one

[hybrid thunder]

wait

I tried another generation

Put this in img to img with exact same prompts 20 sampling steps

pretty decent tbh but couild be better background

background is not accurate at all to the prompts

[tired vigil]

thats what it should look like normally

I think its something with your hires

[hybrid thunder]

what are hires ?

[tired vigil]

Can you disable hires

Its one of the options in txt2img

just close it, its a little tab called hires under your prompt

[hybrid thunder]

Tried again, this good ?

A bit weird body composition

How to make it better

[tired vigil]

that looks like a normal generation yes

try the dreamshaper model

download and put it in your webui\stable-diffusion-webui-directml\models\Stable-diffusion

[hybrid thunder]

and how to make it more armor

for example

more spikes on armor etc

just (( )) the armor prompts ?

[tired vigil]

You could try spiked armor

in the prompt

I don't know, try stuff :D

to increase the strength of the prompt you can do (spiked armor:1.3)

[hybrid thunder]

what do those numbers mean

[tired vigil]

that number will be the strength

[hybrid thunder]

I saw others do prompts with numbers like

[tired vigil]

1 is the normal value

[hybrid thunder]

negatives too

like bad quality:1.5

[tired vigil]

that also works in negatives yes

If you want something but want it lower you can do 0.5 for example

[hybrid thunder]

is it 1.5x the strength that way ?

0.5 would be half strength ?

[tired vigil]

You can think like that, it will break the image if you go too high

[tired vigil]

yes

[hybrid thunder]

And what are

((( ))) ?

(( ))

etc

[tired vigil]

usually you won't need more than 1.5 for something

() is 1.1
(()) is like 1.22

Its the same thing as using numbers

I don't know the values exactly, but think of it as increasing 10%

its easier to do numbers, you can select the prompt and press CTRL + Up arrow

[hybrid thunder]

so if you use numbers and (( )) at same time

[tired vigil]

it will add the (:1)

[hybrid thunder]

it will multiply the numbers by percentage ?

[tired vigil]

no

[hybrid thunder]

So (spiked armor:1.3) is same as spiked armor:1.4 ?

[tired vigil]

you need the () to make the numbers work

[hybrid thunder]

Oh its basically choosing which part exactly should be strengthened up ?

like if I have spiky shoulder plates,

if (shoulder plates) then the shoulder plate armor gets more detailed, bigger or similar ?

if (spiky:1.5) shoulder plates, they get more spikes ?

if (spiky shoulder plates:1.5) everything ?

[tired vigil]

@hybrid thunder I believe that works

I tried it for a bit only

I tried with "destroyed city background"

"(destroyed:1.3) city background" would make it more destroyed

[hybrid thunder]

show

[tired vigil]

Oh its been a while

[hybrid thunder]

how long have you been using

stable diffusion

[tired vigil]

XD I will never find it in thousands of pictures

few months

[hybrid thunder]

nice

and do you have practical use for it as well

or just experimenting

[tired vigil]

Just experiment

It also depends on what model you're using

some are better with prompts or different

some work better with very minimal prompting

[hybrid thunder]

which models are good for very long detailed prompts ?

[tired vigil]

I have no idea I found a few I liked and have been using them

Dreamshaper is a very good model, I recommend that one

also instead of using those negatives, use textual inversion https://huggingface.co/SweetLuna/Aurora/tree/main/AuroraEmbeddings download these 2

[hybrid thunder]

what is textual inversion

[tired vigil]

and this one https://civitai.com/models/7808/easynegative

[tired vigil]

These are like trained negatives (trained using images and AI magic)

they are more effective

[hybrid thunder]

AI magic ?

is that some model

[tired vigil]

No

I just don't know how they do it

Which is why I say magic

lmao

[hybrid thunder]

xD

[tired vigil]

you put them inside the embeddings folder \webui\stable-diffusion-webui-directml\embeddings

[hybrid thunder]

Maybe I need to call Gandalf to help out

[tired vigil]

@hybrid thunder instead of those negatives you're using start with this only: negative_hand-neg, (KHFB, AuroraNegative:1.2), (Worst Quality, Low Quality:1.2), border, grayscale, (watermark, multiple views:1.2)

[hybrid thunder]

wait but checkpoints are the models, right ?

[tired vigil]

yes

[hybrid thunder]

I put these at the beginning of the normal prompts ?

[tired vigil]

in the negative

after you place the files inside the folder

they will work

[hybrid thunder]

cannot find

[hybrid thunder]

cannot find embeddings either

which webui ?

python or batch?

or shell script

I have folders like this

where is webui and directml folder

I found something like this, put it below the txt doc or put something in the txt doc ?

here ?

how can I put safetensors file into txt doc

and pt file

[silver valley]

Hey

You don't put them into txt files

You put them into folders

[hybrid thunder]

which folders

I have these

[silver valley]

You put the Dreamshaper model into the stable-diffusion-webui/models/stable-diffusion folder

[silver valley]

Here

And the other ones go into stable-diffusion-webui/embeddings

[hybrid thunder]

what I dont understand is

Why are there those text documents with put checkpoints here etc

when you put it into the folder

not into the txt docs

[silver valley]

Because its just a notice/comment to put the models in the same folder as the txt file

[hybrid thunder]

ohhh

like those "readme.txt" stuff

lol

Now I get it thx

I am sadly negative IQ when it comes to stuff like these

and now I just boot ?

the webui batch file

and try it out on diff model and negative prompts

[silver valley]

Yes exactly

[hybrid thunder]

what is diff between

webui.user batch

and normal webui batch

and the shell script, python script webui ?

[silver valley]

You always need to run the webui-user.bat
As this also starts additional settings for SD optimisations

In fact the webui-user.bat starts the webui.bat after its done

[hybrid thunder]

after the optimizations are done ?

like this

this looks good right ?

[silver valley]

The webui-user.bat gives additional commands like yours to the webui.bat

So then it can use these

Xformers is for the performance optimisation

[hybrid thunder]

What is the advantage of running stable diffusion on my computer compared to something like NovelAI image generation? (If you know that one) or any other paid image generators? (Other than that its free)

[silver valley]

The biggest advantage is that you can run any Model you find on civitai.com
Also the community made a lot of extensions for different purposes that help you to get what you want. For example img2video or GIF maker extensions.
Also your local SD webui is completely uncensored.

You also have better control over all the Settings. Some websites only give you the basics

[hybrid thunder]

uncensored?

Thats why some people have negative prompts like these?

xD

I was thinking wtf do they need these prompts if they have nothing of that sort in their positive prompts anyway

[silver valley]

xD some models have special tendencies.

But mostly your fine without them in the negatives

[hybrid thunder]

Until one day one model decides to think otherwise, and my parents walk in

[tired vigil]

@hybrid thunder Sorry I was playing a game here, got focused lmao

I believe cs1o explained the stuff

Whenever you can try generating those images and show if the result is good :D

[hybrid thunder]

Sure

I mostly try to generate images from my mmorpg game characters

[hybrid thunder]

what game were you playing ?

[tired vigil]

backpack battles lmao

just found about this today

[hybrid thunder]

First try with the new model

It has 2 ears

xD

[tired vigil]

now try with hires

see if it works

restart is a good sampler

it can fix some details that others might now

but its a bit slower

[hybrid thunder]

how to disable/activate hires

these are my settings rn

when I expand it

[tired vigil]

@hybrid thunder that is how you activate it

you expand

[hybrid thunder]

what

if I dont expand it its inactive ?

[tired vigil]

yes

[hybrid thunder]

lol

so now I try it out again while its expanded ?

[tired vigil]

yes

[hybrid thunder]

settings fine this way?

[tired vigil]

try lower denoise

0.5

and it will copy your steps in the other setting

so set it to 10 or so

[hybrid thunder]

why

[tired vigil]

denoise is like

how much it will change the image

too much and it might break everything

[hybrid thunder]

this good ?

[tired vigil]

too little and it might look low res still

yeah thats about what everyone uses

[hybrid thunder]

so denoise is basically sharpening the details ?

[tired vigil]

no

think like

you gen one image 512x768

that imagine is then used on hires, it will retain 50% of it to generate a higher resolution one

[hybrid thunder]

First try

[tired vigil]

yeah looks nice

[hybrid thunder]

More armor would be good

[tired vigil]

try full armor

put stuff you don't want to see in negatives

[hybrid thunder]

(full armor:1.5) ?

[tired vigil]

try it >.>

its all about try and error :p

[hybrid thunder]

The background, pose and setup was better in the first one

Maybe I should just image to image the first one with armor prompt ?

(full armor:1.5) doesnt work there

and one horn + ear is missing lol

also what is that in the beer xd

Should I set sampling steps to 10 instead of 20 on img to img ?

[tired vigil]

What is your prompt again

the positive

[hybrid thunder]

Armor on upper body would be good

[tired vigil]

instead of exposed skin on thighs

Just type thighs

the ai kinda of doesn't understand some things like that, it goes by rough words xd

at least thats how I use it

[hybrid thunder]

Like a child basically? xd

[tired vigil]

so instead of "Blood elf from world of warcraft" I would just do "Blood elf, world of warcraft"

or not even describe the game

just what you want to see

[hybrid thunder]

Bloodelf is something out of the game though

It was invented by the game basically

[tired vigil]

"Girl, pointy ears, tavern, holding a glass of beer, blonde hair, full armor, thighs"

So then you describe what a bloodelf is

[hybrid thunder]

this fine ?

[tired vigil]

I never played wow so I don't really know what a bloodelf should be

Your images just look like a normal elf to me xd

[hybrid thunder]

yeah thats the problem

xD

This one is fine

But I typed "exposed skin on thights" to make it look like the armor is some kind of dress

Like a skirt

skirt armor

that does not entirely cover the legs

[tired vigil]

@hybrid thunder https://civitai.com/models/111802/blood-elves

is that what you want

there are loras for blood elves

they help the AI understand what the hell you're talking about

:p

[hybrid thunder]

what are loras btw

[tired vigil]

a mini-model on top of your model

idk its AI magic again

to use you download it, put in your webui\stable-diffusion-webui-directml\models\Lora

and then in the web page you should see it

in the lora tab

under your negatives

[hybrid thunder]

when I make any changes to the folders, I should exit the CMD file and SD on the web first right ?

and then restart

once its done

[tired vigil]

the author put this in the description:
Activator: bloodelf
helpers: blonde hair, green eyes, armor
additional: glowing eyes

so you put bloodelf in the prompt to activate the lora, blonde hair, green eyes and armor will help it more if you need

[tired vigil]

Only some of the time

theres a blue button near the checkpoint selection

it will update the webui

[hybrid thunder]

for Loras I dont need it ?

just press the blue update button

lol

[tired vigil]

did you close the cmd ?

[hybrid thunder]

no

its on the civitai site

when I download

[tired vigil]

oh

is the link broken

[hybrid thunder]

Now it works

just needed to spam reload page

[tired vigil]

ah

[hybrid thunder]

nothing is broken right ?

just server thing

I just put it into the Lora folder here, right ?