📝｜prompting-help

[neat dew]

@undone bronze happy

[restive socket]

i couldnt make it work

[silver valley]

do you want the same image just with an other expression?

[restive socket]

i tried to inpaint the face, then I set "shy face expression" as a prompt. it worked but with someone else's face

[restive socket]

yes

[restive socket]

I just can't do it..

[raw zephyr]

Hello anyone are freepik contributor here?

Or Adobe stock?

[agile ferry]

try the baddream and unrealisticdream negative embeddings from civitai

[silver valley]

anime or real person?

[restive socket]

real

[silver valley]

ok let me test

[restive socket]

thank you sm

[silver valley]

so first i tried inpainting

[restive socket]

whats your denoising setting

the denoising strength

[silver valley]

Inpaint area "only masked"
Denois 0.5

inpaint resolution 512x512

[restive socket]

okay let me try the same right quick

main face

result: "sad, masterpiece, best quality" prompts

0.5 denoising

for realistic images its slightly changing the face

[silver valley]

yea let me try with an asian face

whats the model?

[restive socket]

juggernaut XL

[radiant nest]

#1047610792226340935
Prompt: prompt:prompt:Vulcan, star Trek, a female Vulcan, tanned skin, elf-like vulcan ears, arched eyebrows, skinny, slender, femenine body, sensual pink sci-fi dress, in a vulcan white alien mountain, Vulcan robe, fisheye camera, fisheye effect, retro, selfie, summer, transparent dress, sexy dress, 90's, long hair, very long hair, beautiful, exotic, sunkissed
Negative: long dress, covered dress, modest dress, ugly

[restive socket]

btw can I send you my main face I use for consistent images through your DMs?

so you can try it on that

its 1024x1024

[silver valley]

sure

[restive socket]

thanks a lot

[restive socket]

https://civitai.com/models/112552/weight-slider-lora I encountered this while I was scrolling thorugh civitai, thought maybe it could help

[agile ferry]

Thanks! I didn't find that yet.

[sharp spade]

Hey everyone, was wondering, how do i prompt for a second character? Like separate in the prompt this is that character and this is the other one

[hollow plank]

is there a good way to "realistic-ize" images using an SD 1.5 model? i'm using automatic1111

[agile ferry]

I have a lot of difficulty with that, but you can try controlnet with two figures. That seems reliable if maybe too structured.

[silver valley]

Controlnet IP-Adapter can do this

[hollow plank]

what preprocessor should I use? i'm still pretty new to controlnet

[silver valley]

You first need the controlnet models

[hollow plank]

i've got 1 or 2 of those, one was the reference model from earlier

[silver valley]

Reference doesn't need a model

[hollow plank]

ah right, i meant segmentation

[silver valley]

Make sure to get the controlnet models from here if you need one:
These are the smallest for 1.5 (700mb each)
https://civitai.com/models/38784/controlnet-11-models

for IP-Adapter you need these:
https://huggingface.co/h94/IP-Adapter/tree/main/models

[dapper bluff]

hello, how could I explain to an AI the position of the kid in this image?

[runic jackal]

anyone got good models for berserk style?

[stark lark]

this one i did put a 4x upscaling as i thought maybe thatd help. @silver valley - going back to 2x now again of course. was just a test

[silver valley]

Yea better to use upscale by 2.
Also maybe try clip skip 1 for realistic models

As clip skip 2 is used mostly with anime models

[stark lark]

oh okay. yea i mostly just copied stuff people on civit were saying that it worked for them. i also copied the negative promt from there

[stark lark]

i cant get this odd look off of the image and make it have a nicer to look at resolution. already put low contrast in the positive now

[ocean gate]

Yo

I dont know where to ask thats why I am asking here

Can I add this bot to my server?

İf so how can I add it?

[silver valley]

Hey, no you can't. Its only available here. But there are some community made bots in other Discords

[ocean gate]

Oh okay

Then can I make images here?

[silver valley]

Yes in the bot channels with the / dream command

[ocean gate]

Thanks for the help mate

[silver valley]

You can try to use an other VAE or other negative prompt

[silver valley]

No problem 🙂

[ocean gate]

I think I have a little problem

Bots are down

[stark lark]

didnt understand VAEs yet so i left that at automatic for now. RealisticVisionV6.0 is supposed to work best with DPM++ SDE Karras but using Euler A almost completely fixed the image, only sampler those promts seem to work with how ever. isnt tht strange?

[silver valley]

Oh right xD I forgot. They are under maintenance rn

[ocean gate]

Oh okay

Sorry if I am being a pain in the a**

[stark lark]

i bet youre not. I am right now ..😅

[stark lark]

this is with euler a. everything even euler, heun and stuff looks basically just like those.

not happy with it just yet, but thats proably the best way to go with as a foundation and start adding and taking new prompts.. 🙂

[silver valley]

No not at all 🙂
If you want to generate images for free you have some other Options:
Install Stable Diffusion localy on your PC.
Use Stable Horde client or website.
Use websites with daily free tokens for generating like Leonardo ai

[ocean gate]

Damn my guys taking all the blame you're a huge w

[ocean gate]

I appreciate your help my guy

[silver heath]

Use ChatGPT (or Jai.ai) to make some SD prompts.https://youtu.be/VpTEdsspxbw?feature=shared

You can change Automobiles to anything you like - haute couture, vintage cars, real estate, landscapes - etc. Just filter out or add terms as needed

[tender kayak]

Engine: stable-diffusion-xl-1024-v1-0 prompt: A group of people engaged in a lively discussion about cryptocurrencies.

Negative prompt: "text": "blur, blurry, ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, extra limbs, body out of frame, blurry, bad anatomy, blurred, watermark, grainy, signature, cut off, draft, closed eyes, text, logo",

cfg_scale": 7,
"height": 1024,
"width": 1024,
"samples": 1,
"steps": 40,
"seed" : 0,

So anyone knows how can I fix this ?

[static hinge]

hey is that work with SD XL? I get TypeError: must be real number, not NoneType

[silver valley]

Idk

Would need to test it

[unique solstice]

What are the best textual inversions for anime images? I am using simplenegative_animev1, easynegativev2 and bad-hands-5

[left olive]

anyone got interesting negative prompts?

[static hinge]

I think it doesn't work.

[modest jungle]

d

[shut lotus]

ive been reading a bunch of the articles people publish on civitai and im really struggling to understand effective ways to prompt
https://civitai.com/articles/3770/how-to-prompting-with-style-and-quality this one was most comprehensible for me
it seems like things have pivoted heavily towards LORAs, but my main interest is in creating images of original characters etc, not existing ones with populated booru tags
i have plenty of artwork but i feel most of my art probably isn't very suitable for use in a training dataset? ive seen people use generation to create images for datasets, but i would need to get consistent and precise results in the first place for that to be a viable option
sorry if this is really broad this seemed like the best channel. i guess im mainly asking for advice in getting specific details to generate consistently and accurately without an existing lora? but i wouldnt mind advice for how to approach creating a lora/dataset in my situation

[static hinge]

I think I'm done again with trying multiple characters with latent couple, regional prompter, it is just a damn mess. I even mask the regions of a referecen image that I put also in ControlNet openpose, depth canny. It almost never did the poses, and did some mess when it did, also switched the character regions, pose the characters any were in the img, duplicate and merge characteristics, god damn

Adding other character in a specific pose that make sense with inpainting is also an herculean work, I did some but it never quite make sense in the style. It can be done but with so much effort. But I'll keep trying some other time.

In those times I think OpenPose is a joke, just doesn't pose it in the way you ask

not even close, wtf

Now I think one character was lying, so maybe I should had add "laying down", that may caused a mess (that character was usually a mess when it was drawn)

But I though Openpose would handle just with the pose

[crude storm]

How can I generate the opposite of a picture? fow example switching the grass and the soil on this picture:

[undone bronze]

Hello, how could I make the photos look unprofessional? but rather that they have the quality of a poor quality cell phone, or perhaps with bad lighting like that of a room, I have tried setting "low resolution" and "bad illumination" but it does not give me the results I want. I leave an example of what I require, thank you

[dull portal]

'a webcam photo with dim incandescent lighting, lighting from above' maybe something like that, drastic so you can hopefully see a difference

[robust lion]

[undone bronze]

Thank you very much, I will try what you say 😄

[lavish magnet]

I can't generate images

[dull portal]

What you working with, SD-WebUI, ComfyUI?

[floral rover]

trying to get a girl with curtain hair but then curtains pop in the background, what do you do to stop that? negative prompt?

[steady cape]

Hello, can you help me? I want to create a cyborg with detailed body. The "typical" plating with cables and mechanics visible at the joints. But I couldn't get it to the AI. Can you please watch over my prompts (I am a beginner) and give me tipps what to improve, write better?

(full body:1.5), a cyborg with white plating and female full mecanical body form standing tall. Gaps between the plating, Between the plating should be some glowing blue light., completly mechanical body, mechanical interior, porcelaine matte material, high resolution, neon, cold colors, studio lighting, glowing blue eyes, digital painting, detailed face, lora:add_detail:2, face uncovered, neon pink hair, toned physique, neutral grey background, expressive blue eyes, small shoulders, cyberpunk, lora:extremely_detailed:1, glowing blue iris, highly detailed

[copper ginkgo]

does anyone know

how i get this quality

it looks made with a phone

[clever sail]

I dont know the range numbers of what is low or high in advance sttings on all settins, what numbers are low or high prompt strength, generation steps or seeds what numbers sould i put in order to what results? Alsi, what is the difference in models SDXL V1.0 ou stable diffusion v1. 6? What all of this means?

[oblique hound]

How actually to prompt those decorations (accessories) on legs and on arms? Any ideas?

[round lake]

I am trying to make an image of the standard "briefcase full of money" often seen in movies/games/etc but every time I try there is also bunch of money scattered around the briefcase on the table/floor/whatever? How can I avoid that and only have it be in the briefcase.

[fiery pecan]

Still cant generate?

[fiery pecan]

What model are you using?

I think lora can help you with that.

[silver heath]

Anyone using Dynamic Prompting in A1111 here is a Dynamic Prompt about cars I made with the help of CGPT3

https://drive.google.com/file/d/1bKYGy0XAEL91BfR0LHkRikSJovhRL_Sn/view?usp=sharing

[silver heath]

AUTOMATIC1111 using NightvisionXL model

[steady cape]

I used Dreamshape (I like the style alot and it seems better for me evading the uncanny valley). I have searched but didn't found a good lora ...

Unfortunatly I am now only on a laptop with integrated GPU so no testing the next days 😦

[fiery pecan]

No prob. i can test a few thing. see if it work. May i know what gpu are you using? Can you use SXL base model?

[steady cape]

Its a radeon on board gpu. I have just tried Web gui which does't start and throwing the error with the onboard gpu..

[fiery pecan]

when you said you cant do testing the next day. Does that mean you have access to a better gpu later on?

[steady cape]

Yes, my private PC has a 2060 in it.

[gilded tendon]

cat

[silver valley]

That should work with SD, but you would need to use my AMD install guide. Its in the Pinned Messages of #🤝｜tech-support

[steady cape]

Thank you

[lime wadi]

need some help with Highres fix... how can I make it so it doesn't go too far from the original picture? 😦 If i turn the denoising down, the image getr rllyy bad

[silver valley]

Lower the Denois and dont use to much hires steps

For example
10 hires steps + 0.35 denois

[lime wadi]

what do the hires steps do? I was keeping them at 0...
tried 0.35 and 4x (was using latent) and it seems better now

[silver valley]

Hire steps are the upscale steps.
0 will use the same amount as the normal steps. Which is to much and can distort the image.
So 10-15 is enough

Then play around with the Denois

For latent 0.5 is good

For esrgan based upscalers 0.35-0.5 works also good

[steady cape]

I have generated that image. I want substitude the skin part with mechanical joints. So I loaded it in inpaint. marked the skin (and the parts around and prompted:

"mechanical joints under armor plating, high resolution, detailed, steampunk, lora:add_detail:1 lora:extremely_detailed:1"

It only changed the color. Can you help me understanding how to refine this image? I am new though so I have to learn alot and have no clue where to start ...

[silver valley]

are you on nvidia or amd right now?

[cobalt narwhal]

#1080946152318443610

#🏞｜general-with-images Dream houses

[steady cape]

Now on my personal NVidia-PC

[silver valley]

okay, then the first thing you need to know is that inpainting is like generating an image into an image.
so you have to set the Inpaint area to "only masked", if not it will take much longer as it regenerates also the not inpainted stuff-
Then you need to set the resolution to a sqaure, like 512x512 or 768x768, for sdxl 1024x1024
because thats the resolution the models are trained on. (the resolution of the input image doesnt matter for that)
The denois slider will define how much the image should change to whats already there. start with 0.5 and then go up or down to your liking

[steady cape]

OK, thank you. I thought that there is a tool where the AI repaints a part of the model regarding its older state (as I meant it should be another armpit, but with mechanical parts). Is that possible?

And thank you for the posts

[silver valley]

follow my steps, then inpaint that area, and prompt for :mechanical joints, armor, cyborg, android,

that works

denois was 0.6

[steady cape]

Thank you. Got this.

A little missaligned but now I know hot to do it

[sharp ember]

can i ask for lora creation help as well?

or is this not the right channel for that

[mellow wave]

Is there a positive or negative prompt that helps me with the teeth? Cause the teethes of every model looks horrible, or if it looks more or less good, they have bunny teethes

[static hinge]

I'm getting crazy trying to inpaint in SD XL, I though I knew how it worked, at least for changing basic stuff, but know I want to inpaint other characters and it is a nightmare

I'm using basic models and juggernaut inpaint model. They kind of work but it seems only way to introduce another character is to make some photobash (that it is a lot of work), or praying that OpenPose would do something

Also I don't know anymore what fill /original / latent noise / latent nothing do as I tried each and sometimes they work for what I'm doing, and sometimes they don't. Latent noise sometime just do noise, sometimes it produces something... sometimes it seems original with full strenght works better

the problem is that I give a lot of context, SD thinks that I'm talking about the main character, if I give little context it generate something unrelated, and so on

[static hinge]

that is the bigger issue, as it only generate the other character when I give it little context, but it is a total mess

[static hinge]

well I checked some tutorial and relaxed, I think the key is inpaint with fill, only masked, to create the new stuff (so SD doesn't see the main character), and then original at giving variable context so it merges

[marble imp]

I have tried to face matching in stable diffusion with Reactor Extension.. Some face matches are really good but not all of them.. Is there a better extension for face match or a better method for face matching?

[silver valley]

There is FaceFusion or ControlNet with IP-Adapter

[marble imp]

Thank you, will try that out

As we have many different sampling methods, do we have some documentation about comparison of those sampling methods?

[tired vigil]

Hello can anyone help me with prompting every time i generate picture the face always being like this how can i get a normal face ?

[analog sundial]

hello guys, I have a problem and I'm hoping I can find some help here.

basically I made 2 loras, each of a different character. I'm trying to do images with these two characters together in the same image, but they keep mutating into each other (swapping genders, clothes, hair colors, eye colors) and a lot of extra heads/arms/legs/hands are generated as well.

I've tried using control net and I've tried regional prompting, but nothing seems to be working for me and I can't find much info online

Could anyone please provide me with a prompt that might work? I'm using SDXL and this checkpoint https://civitai.com/models/261336/animapencil-xl

any help would be very appreciated as I've been unable to figure this out and it's been driving me crazy for a few days!

Thank you in advance

[fiery pecan]

What render resolution do you use? What model and vae did you use?

[tired vigil]

512x768, i'm using fantasyWorld_v1 model and the vae is vae-ft-mse-840000-ema-pruned
I'm following a guide to learn stable and all his generated was perfect even tho i tried the same prompt and negative prompt

i'm using the last model of stable

[fiery pecan]

What sampler did you use?

[fiery pecan]

Did you use the same as the guide also?

[tired vigil]

yes i did

[fiery pecan]

Which sampler is it?

[tired vigil]

DPM++ 2M Karras, i did try Euler still the same weird looking

[fiery pecan]

I am outside atm. Will be back home in 10-15 min. Then i can take a look in further detail.

[fiery pecan]

Have you tried with other model? Does it still have mess up face?

[silver valley]

Yes here:
https://stable-diffusion-art.com/samplers/#Samplers_overview

[tired vigil]

no only this, me and my instructor were using the same model his face that he generated all of them were good looking only mine that didn't look good

[silver valley]

Do you have face restore enabled by accident?

Check the settings under face restoration, the first checkbox should be unticked

[steep inlet]

Blue, transparent waves

Blue, transparent waves

[tired vigil]

yes it's enabled should i uncheck it ?

[silver valley]

Yes

Its not made for anime faces

And will make them distorted instead

[tired vigil]

oh okay thanks, will try and see <3

[silver valley]

No problem 🙂

[fiery pecan]

check in the guide if he enable it or not. if its not. then thats your culprit.

i am on my pc now. i can check in further detail if you still having issue

@tired vigil Speaking of face restore. You should checkout Adetailer. it can use to repaint the face automatically into higher res.

but check first if the face restore option is the one messing up the faces.

[tired vigil]

Ok so when i uncheck the restore face it got e little better but still i'm really unhappy for how the eyes look

[tired vigil]

i will try it thanks <3

[silver valley]

Yea its better, the rest is due the low resolution.
You can use hires fix, or inpaint or the adetailer extension to fix the face

Also the nearer the face the better the quality

Portrait images always look better than full body for example as you only have 512x768 pixels for details

Thats why most people upscale the images for better quality

[minor sphinx]

👍

[tidal fiber]

how would i get the hoodie to have blue trims?

[cunning hawk]

Hi, anyone suggest which AI tool can create such images where I put the amalfi pillow with sofa against a background of the amalfi coast? similar attached image

[frosty summit]

I would put the img of the pillow into ControlNet and then prompt "on a couch in front of the amalfi coast" or whatever and hope it works

Or loosely photoshop the pillow onto a couch first

[static hinge]

inpainting with ControlNet with the reference, the best controlnet model I don't know, ip-adapter, canny, depth, maybe the color adapter one

[oblique rampart]

Right so I'm trying to use the animage SDXL on Automatic111 and it keeps churning shit like this out even with just these inputs

Could it be the VAE

[static hinge]

VAE

[oblique rampart]

It was indeed, the VAE

[prisma grail]

how to stop model from making hyper saturated/flash banged images

?

[fiery pecan]

what cfg did you use?

[prisma grail]

6

using krita extension

[fiery pecan]

Whats Krita extension?

[prisma grail]

https://github.com/Acly/krita-ai-diffusion?tab=readme-ov-file#installation

what's with these random bot accounts

[copper ginkgo]

Okay question

who knows

how you get this extreme realism look to a picture

like this

the lighting is just so realisti

why style to prompt this with

[fiery pecan]

Like what? there is no picture.

[echo nacelle]

Does anyone know why this happens?

[twin yoke]

#1047610792226340935 message

[lunar dirge]

#1047610792226340935

[rough forge]

hello

i need help in promting

anybody there

[torn pilot]

When creating an image with t2i, it sometimes stops at 95%. What is the cause?

[silver valley]

make sure the webui is whitelisted in any browswr adblocker

also whats your gpu and whats inside your webui-user.bat?

[fiery pecan]

Need help to prompt what?

[rough forge]

can you tell me documetation for promt wirting

i am having hard time to writing promt

idont know what does what

[fiery pecan]

i dont have a documentation for prompt writing. i mostly get better at it through self test and see what other people are prompting. May i know what kind of images are you trying to make?

cuz realistic images or anime style of images has different way of prompting.

[rough forge]

oh oh how can i see what other people writing

[fiery pecan]

ohh you want nsfw images? i can help. but i mostly deal with 2d or digital painting kind of images. for realistic. i can tell you want you want to look out for. pretty much same way for digital or realistic images.

[fiery pecan]

best way is on civitai website. go to the model you want to use and scroll down to the bottom and you will see user created render

some of them has metadata included and you can see what they prompt to make it.

or the images included by the model poster. they usually include render using said model without any lora

so much easier to follow.

[rough forge]

https://tenor.com/view/anime-acchi-kocchi-cute-shookt-sighs-gif-17056450

thanksyou

[fiery pecan]

no problem, let me know if you need any other help. Have fun!

[rough forge]

does increasing steps give me good result

[rough forge]

i will send you my nsfw as token of grattitude

[fiery pecan]

it does up to a point. then it is pointless. refer to each model description as to what the appropiate step is.

[fiery pecan]

haha, send away. curious to see what you cook up.

[rough forge]

https://tenor.com/view/frieren-beyond-journey's-journey-end-gif-16585420796092983366

[torn pilot]

RTX2080 super
@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers

call webui.bat

[silver valley]

then add --medvram-sdxl --no-half-vae
and then try again

[torn pilot]

Where should I add it?

[silver valley]

behind --xformers

sperated with space

[torn pilot]

set COMMANDLINE_ARGS=--xformers --medvram-sdxl --no-half-vae

ok?

Still, the image is generated after some time passes after it freezes at 95% ETA:1s.

[severe salmon]

Can we mod chat gpt for un cesored words for our templates for SD?

[silver valley]

yes looks good

[silver valley]

is your webui whitelisted in any browser adblocker?

[torn pilot]

I tried it but it didn't change

[silver valley]

hmm whats your vae?

[torn pilot]

It will go smoothly until it reaches 95%, but it will not reach 100% until some time passes. In the command prompt, all items are to 100%

[torn pilot]

anything-v4.0

[silver valley]

how long does it take. from 95 to 100?

[torn pilot]

It's about 2 minutes to 10 minutes.

[silver valley]

okay, whats your browser?

[torn pilot]

It's firefox, I also tried it on edge earlier.

the result was the same

[silver valley]

hmm using firefox too

[torn pilot]

Even with anything v4, will the contents change if it is published by another author?

[silver valley]

nope

[torn pilot]

🤔

[silver valley]

but try out clearvae

to test

[torn pilot]

Is there no problem with the VAE issued by anyone?

[torn pilot]

Is it possible that high-resolution supplementation is performed after reaching 95%?

[silver valley]

last step is vae apply

[torn pilot]

It may have been due to high resolution assist

I don't know what to say in English though

[silver valley]

hires fix ?

make sure to set hires steps to 10-15

[torn pilot]

what the hires step

[unique solstice]

Hey guys, all the faces SD generates look eerily the same in my case. I tried Anything V5 and various other models and the faces all have the exact same proportions no matter what. How does one go about making slightly different faces? I use booru-style tags.

[fiery pecan]

Can you share your image here so we can see the metadata.

i usually tend to make 2.5d render so if i want different faces. i input like nationality. that help to make new faces.

[unique solstice]

Of course, it's a raw image, no modifications or anything, so the hands and limbs in general might look weird.

[fiery pecan]

hold on. in a match. ill take a look at it in 10 - 15 min

[fiery pecan]

i want to clarify. the "merlin" tag you use. What is it? is it a character from an anime or something?

[woeful epoch]

ok so i reinstalled stable difssuion. i put everything i had before on it. but now my images just.. look like shit and idk why

both are using the same models and loras

the one on the right is the older one and im not sure how to fix.

[silver valley]

try adding portrait, cowboyshot

and use some quality tags like, masterpiece, highres, best quality

and some negative tags like, blurry, deformed,

[woeful epoch]

i.. idk why that magically fixed it so easily

however the faces are still a big blurred

current promts are 1girl, lora:rwby_cinder:1.2, long hair, black hair, yellow eyes, full body, detailed face, portrait, cowboy shot, masterpiece, high quality, and negative promts are: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, (worst quality, low quality:1.3), (depth of field, blurry:1.2), (greyscale, monochrome:1.1), 3D face, nose, cropped, lowres, text, jpeg artifacts, signature, watermark, username, blurry, artist name, trademark, watermark, title, multiple view, Reference sheet, By bad artist -neg, dark skinned male,

[hollow tapir]

There's so many settings that could be different between the two images beside the prompt.

obvious one being clip_skip

[silver valley]

the nearer the face to the viewer the better the face quality

[woeful epoch]

what about clip skip? i was always told to isntal it but idk what it did and never use dit o.o

[hollow tapir]

it's not something you install.

[woeful epoch]

oh then i have no idea what it is or how its used

[hollow tapir]

Regardless we need at least two .png to compare the different settings as .jpg do not embed generation parameters in its metadata

[proven jasper]

lora:rwby_cinder:1.2, try to reduce the weight of this lora,if u go over 1 the img becomes overcooked,try with 0.7 to 0.85

[woeful epoch]

how do i set it to jpg not png?

[hollow tapir]

it's .png by default. Regardless it would be too late to switch it now as the .jpg is the old one.

[woeful epoch]

[hollow tapir]

So yeah either you find a .png of one of your older gen or you'll be stuck wondering and tweaking settings blindly trying to recreate it.

[woeful epoch]

i.. have thosuands of images. it jsut made them all jpg..

like i never touched any settings like that so idk

[silver valley]

Did you downloaded the images trough the browser?

[hollow tapir]

should be png by default

[woeful epoch]

but any other ways o can make it not blurry and bad looking?

like promt wise since that seems to be part of the issue

with promt recomendations this is what it looks like now

[woeful epoch]

yeah its set to that now. but i guess before it was not 😛

[fiery pecan]

Can you share an image? what resolution do you render at? the only issue is blurry face only right? @woeful epoch

[woeful epoch]

yes. thats my only issue atm, currently trying an image with sampling steps maxed to see if that fixes.

[fiery pecan]

btw, what gpu are you using?

[woeful epoch]

normal gen res is 512 by 712. but i also have hires fix upscaling with r-esrgan 4x+ anime6b

[woeful epoch]

nvidea 4080

[fiery pecan]

i see, are you using SD1.5 image or SDXL model?

[woeful epoch]

1.5 currently becuase the lora i use dosnt work with xl 😞

heres an image with it set to 25 steps

[fiery pecan]

I see, do you use hires fix when rendering images?

[woeful epoch]

yes

[woeful epoch]

said so here

[fiery pecan]

lol, sorry. kinda late where i am at. feeling sleepy.

[woeful epoch]

all good

should i put the refiner on as well?

[hollow tapir]

So then you did change some settings, check the clip_skip value too as the model you're using recommends a clip_skip value set to 2.

Also 80 steps is ridiculously high.

[woeful epoch]

i have not touched clipskip. idkw hat that is

[fiery pecan]

refiner is only for SDXL.

you dont need for SD 1.5 images

[woeful epoch]

ok

but so how do i check the clipskip thingy

[hollow tapir]

check the settings tab

[woeful epoch]

what would it be under?

[hollow tapir]

iirc, Stable Diffusion

You can also add a shortcut for it at the top of your page by adding , CLIP_stop_at_last_layers to the Settings -> User Interface -> Quicksettings list

[fiery pecan]

@woeful epoch Can you share the link to the lora you use. the rwby cinder. i would like to test personally.

[woeful epoch]

https://civitai.com/models/47955/cinder-fall-or-rwby

[woeful epoch]

i.. dont see anything in any of the settings saying clipskip?

ohh think i found

its at.. clipskip 1?so i keep it at 1 right?

[fiery pecan]

did you found it? did you manage to add clip skip at the top of your ui?

it should look something like this

[hollow tapir]

(only for some models)

[woeful epoch]

no mine dosnt have it there

[hollow tapir]

because you didn't add the shortcut like I told you there #📝｜prompting-help message

(you ll need to restart sdwebui afterwards)

[woeful epoch]

restrating it now

ok its there now

[fiery pecan]

@woeful epoch i check your image metadata. you use hires fix with r-esrgan 4x+ anime6b set at 0 step with 0.7 denoising right?

[woeful epoch]

yes

i dunno what any of that is so i didnt wanna touch it and break something

[fiery pecan]

typically when using upscaler model. you use like 0.2 - 0.25 with 0.3 denoising. change the step to like 15 or 20 step.

0.7 denoising typically use for latent upscaling. for using with upscaling with model such as r-esrgan 4x+ anime6b. it is quite high.

[woeful epoch]

changing it now

oh god what did you doooo

hm.. now the background seems more detailed but the character seems to have had its quality lowered?

another image as example of what i mean

[hollow tapir]

There's no hint of a hires-fix in the initial picture's metadata

However I'd say that 1.2 strength is too high for the rwby_cinder lora. And I get better result when using Euler A instead

[fiery pecan]

ngl, i think it just the model. i am using galenaredux. and i am having better result.

though it is understandable if this style if not what you looking for.

maybe need to try other model.

[woeful epoch]

the model im using is aing diffsuion.

[hollow tapir]

I'm still tweaking the adetailer params to try to get a proper output of their initial image using DPM++ 2M Karras sampler

[woeful epoch]

trying now putting it down, from 1.2 to 1.0 and using eular a

[woeful epoch]

would me screenshoting my ad detailer stuff help?

[hollow tapir]

it's already included

Usually I'm not using adetailer so it's taking me a while to get everything right

[woeful epoch]

ah. i just use it to fixface and hand stuffs

using your recomendation ary this is how it looks

[fiery pecan]

@woeful epoch would you agree that this is a the cinder fall character? i dont watch rwby. so it might not look like the character.

i am rendering the above using sdxl.

[woeful epoch]

kinda?

[fiery pecan]

aight, i dont use lora for above. just hoping the model knew that character well enough.

[woeful epoch]

like yellow eyes and black hair are really the only thigs sort of like it. i use lora for betetr result

[fiery pecan]

i use another model using the same prompt and lora you provided. is this has the clarity you need? i use a model call blazingdrive.

the look can be fix later

just the visual clarity.

[woeful epoch]

i think i may have figured it out.. maybe

messed with the promt a little and that seemed to helpa bit

[fiery pecan]

is this good enough?

[woeful epoch]

not really what im looking for? i like the model im using as it seems to wrk the way i want now 😛

[fiery pecan]

i use the same model. aing diffusion

[fiery pecan]

i only add a few visual prompt and change the sampler use.

[woeful epoch]

seems to be working much better now. any other suggestions just for improving hte quality? (images atached to show how mine are looking now

[hollow tapir]

There you go using your initial left prompt as a basis but with some modifications :

• clip_skip : 2 (this alters the image significanty !)
• lowering the lora strength to 1.0
• lowering sampling steps from 80 to 40 (still a lot but much more reasonable)
• using the actual trigger world (cinder fall) as described per the lora description / readme
• using a positive and negative prompt for adetailer
• setting Euler A as the sampler for ADetailer
• using 50 steps for Adetailer (the real number of steps done by ADetailer follow the same rules as the one for img2img : denoiser strength*steps+1 => so in my case 0.4*50+1 = 21)
• using cutoff to avoid any color contamination,

And I think that's all the modifications I've done. Every additional tabs that I have are inactive if not deployed, you can safely ignore them.

ps : I don't watch rwby and civitai refuses to show me the lora's preview for some reasons so I have no idea if it looks even close to the actual character XD.

[fiery pecan]

if you like an image. bring it to img2img and use ultimateSDUscale and inrease the res thus add more detail.

[hollow tapir]

If you want something closer to the actual composition of your initial image (this one). Then I'd suggest using img2img to fix it... Because it being generated with clip_skip set to 1 means you won't be able to recreate it with clip_skip 2 in txt2img.

And because I see that someone mentioned ultimateSDUpscale. Let met just drop this here.

How does Ultimate Upscale Script works for mimics and alike.
Basically it's gonna split your (too) big image into multiple smaller ones called tiles. Upscale each one individually and then stitch them back together to make the higher resolution output.
What do those settings do ? cf screenshot
why blur the edges ? Because it helps them blend better. SD will figure out something correct from the blurry mess, don't worry.
https://cdn.discordapp.com/attachments/1091193032273043516/1194628488200724550/image.png?ex=65ccbafa&is=65ba45fa&hm=135890c7d54e5672337696c1aa7dc4b6c6efae312a9332da5fba38ed729f8b9c&

This guide is assuming you already know how to do basic upscaling tho. Meaning you should already know how to tweak the denoiser strength and set the desired output resolution.

questions ? @woeful epoch

[woeful epoch]

what do you suggets for psotive and negative for add detailer? i just have them left blank but am using it for yolohands and yolofaces

[hollow tapir]

Considering that with your settings adetailer is only gonna add details to the face. It means it will basically """"zoom in"""" the face and only repaint this part of the image.

Thus I'd say it's better to make positive and negative prompt that only describe the desired face.

no need to describe the dress, background or whatever.... It's only gonna confuse the AI.

[verbal pawn]

hi why when I use open pose it gives me a black screen

[woeful epoch]

what about for hands?

[hollow tapir]

you're only gonna add details to the face using a "face" model for ADetailer

[woeful epoch]

but my add detaler has a 1st and second?

so i have 1st as the face and 2nd as hands

[hollow tapir]

that wasn't saved in your metadata.

Then yes you'd want to use a face model for the 1st and hand one for the 2nd (or vice versa)

[woeful epoch]

but aany sugegstion for what to put as psotive and negative promts for both?

[hollow tapir]

same principle as for the face

ADetailer is gonna zoom in each hand and """"redraw"""" over them

So use prompt that describe the desired hands

[still pike]

I used for each image a different model but they all look not sharp

I used

[hollow tapir]

eg

[woeful epoch]

its.. oddly peacul now. last time when i used stable diffsuion it was only using my cpu. still was fast but got so warm and loud. but now that its actually using my gpu its so much faster and nicer.

and ty all for th ehelp

[hollow tapir]

what tool are you using ? I see that in your prompt you're :

• talking to the AI like it's a human. You don't have to make actual sentence, use keywords instead. If anything it makes it more difficult for the AI to understand what word is actually important in your prompt. Like In the foreground, there's a range of majestic mountains covered with a thick blanket of snow ..... What word is important there ? Me dumb AI.... In is the first word so it must be important... there's a Oh I see that a lot it must means mean something then....
  No instead you should write majestic snow blanket foreground
• , are used to cut your prompt into multiple "parts". So again when the AI read In the foreground, there's a range of majestic mountains covered with a thick blanket of snow it's gonna try to put two thing in the output image : In the foreground which means absolutely nothing, and there's a range of majestic mountains covered with a thick blanket of snow
• I don't see any keywords such as photography, realistic, detailed in your prompt
• Nor do I see any keywords such as blurry in your negative prompt
• Also your output are 720p so yes it's bound to be kind of blurry

[hollow tapir]

share your input image please

and check the console logs for any errors

[still pike]

thx you for this text

is there any prompt generator or something which helps?

Also why does it now understand text like dalle or something?

not

[hollow tapir]

There's stuff such as https://promptomania.com/ or if you really have to you can use chatgpt I guess .... but then what's the point ? having an AI write prompt for another AI to draw them..... Instead take a look at examples from website such as prompthero or civitai and learn from them.

[still pike]

okay thx I tried chat gpt but it does not know how to make prompts

[hollow tapir]

Each SD model (SD1.5, SD2, SDXL) comes with a different """"interpretor"""" that will understand your prompt differently. Sure you can use "natural language" to some extent. But most of the time you'll get better results if you break down what you want into small chunk / keywords. It makes thing easily understandable for anyone, including :

• marketing guy who loves to create powerpoint presentation with bullet points
• tiktokers with a 2 seconds attention span
• your average AI.

[still pike]

thanks a lot

[still pike]

Would this tools work to improve Stable Diffusion Prompt or Generate them https://github.com/TonyLianLong/LLM-groundedDiffusion https://github.com/aws-samples/stable-diffusion-prompt-rag

[hollow plank]

what controlnet preprocessor and settings would I use to change the style of an image

i'd like to upload an image of myself and have the AI do its best to create vector art of it (I have a LoRA for vector art)

my current pfp is a vague description of me I gave to stable diffusion, but I'd prefer it to look more like me 😄

i've got the models and such all downloaded

just not sure how to use them

[restive socket]

hey, what causes these ugly pixels on the screen? (sdxl, dpm++ 2m sde)

[silver valley]

you can try IP-Adapter + canny or + openpose

[hollow tapir]

could be an incorrect cfg value or resolution

[crude zodiac]

can you assign certain lora to certail models in a prompt?

if i want to have 2 characters that i have seperate custom lora for?

[silver valley]

That can be done with the Multidiffusion extension it has a Regional Prompt control function

[fiery pecan]

seems like a Vae issue. look at the model description and see what the recommended vae is. and also try to use other sampler. like euler a or dpm++ 2m karass. see how that work.

[restive socket]

it says vae is included

and i will try other samplers, thank u sm!

[fiery pecan]

Np, did it work out?

[slate radish]

Hey, I don't know if this is the right channel but here goes, I've been trying to make seamless tile patterns but it turns out it's very visible where they connect. Is there a way to correct that? Here's a screenshot of what I mean

[earnest ibex]

hey there ... i have a question about comfyUI prompt control ... not sure about the syntax for LoRA loading
does this lora:impatientGF:0.9:0.1 interpolate from .9 to .1 till it leaves? ... could i do 0.9:0.1:0.3 ? ....

[rough forge]

what couse this problem ony happen when i use upscaler

looks like problem coming from reactor face swap

[simple sail]

Hi guys. is it possible to mask an output image directly with any given shape (i.e. a circle, triangle, any other shape? I don't want to modify that masked area, the output image should be inside the mask and the rest transparent. Like masks in Photoshop. Thanks in adavance

[fiery arch]

Hi

[still pike]

what is wrong here

like there is no fire and the can look strange

[frosty summit]

Try adding weight to (surrounded by fire:1.2) for example. It might be confused that it's both surrounded by water and fire lol

[tired vigil]

That is ONE result... how many dozens did you generate? You did generate dozens right, if not a few hundred? This isn't a magic genie that grants your wishes... it won't give you exactly what you want every single time... In fact it will hardly EVER give you exactly what you want. Going into it like that is setting yourself up for failure. The right way to use these tools is to expect to be surprised and see what you can get.

[leaden socket]

Any suggestions on getting the same portriats?

[idle glade]

how do i make images like this? (notice how there is a message)

[tired vigil]

I believe there are loras for that same 3 views, and multiple views is the prompt you're looking for

[leaden socket]

You just reminded me, thank you.

https://civitai.com/models/290836/multiple-views-sdxl?modelVersionId=326946

question for you guys, when a model says "Trigger words" should I be dropping it in my prompt or is just using the lora good enough?

[dapper tree]

you need toi add it to your prompt or the lora wont get "triggered".

[leaden socket]

Last question for the day, when prompting img2img anyone have any suggestions on how to remove artifacting like this? left output right prompt img.

[oblique rampart]

Does anyone have any ideas what prompt I could use for animage SDXL to match this style from one of my older SD models?

[wise vigil]

Did anybody else's prompts basically improve a huge amount once they started using keyword weights instead of spamming parentheses? I feel like it is so much easier to get the AI to do something that it's a little bit reluctant about and when I've overweighted a word I know exactly what I did

It's a lot easier to subtly influence and mix pieces as well

[fiery pecan]

yeah, thats basically what season user use. It is alot easier on the eye also when reading.

(keyword:1.4) is alot better looking then ((((keyword)))). i take it your older prompt is just a jumble of parenthesis? lol

[wise vigil]

It wasn't until either I started using NegPIP or needing to tweak LORA weights more that I realized that it could apply to everything

Also the fewer similar types of brackets, the better ( (), {}. [] )

[] is totally redundant with this iirc, {} I use for dynamic prompts

probably going to change the wildcard token to @ or something instead of two underscores because that looks nasty imo and there's no clear separation between them

[fiery pecan]

yeah very important to distinguish each bracket uses or later you gonna be confuse why stuff is not working lol.

[wise vigil]

I don't even remember how strong a given pair of brackets/parentheses even is and who wants to do that math on the fly

[fiery pecan]

i hope you been getting into ai image relatively recently not like 2 years in and you just discover this. 💀

[wise vigil]

In fairness I spent a good part of that on NAI which I don't think lets you actually do that

And tbh
I did actually use prompt weight adjustment to an extent, with All-In-One Prompts, but it's WAY easier to do it manually

It's more that only recently I discovered it was way more powerful than I imagined

Some of this is cope
how much? that's up to you to decide...

[fiery pecan]

lol aight. i take it you switch to A1111? is it a better experience overall than NAI?

[wise vigil]

specifically I use Anapnoe's UI-UX because it's pretty much compatible with many of the same extensions but it's basically the same guts as A1111

and I am a bit of a snob for UIUX design

plus I accidentally made kind of a nice solarized theme

I couldn't really get into SD.Next and Invoke doesn't automatically implement the token limit workaround that everything else seems to do so that was a deal breaker

[fiery pecan]

ohh that nice. never heard of anapnoe ui before. will keep it in mind.

[wise vigil]

yeah one sec let me boot it up and show you what i made

i find the vast majority of gradio themes either hideous or to not work outright on a1111 so

[fiery pecan]

be the vegan of AI renderer and get into ComfyUI. haha

[wise vigil]

tbh ComfyUI just feels like reinventing the wheel far too much for me

[fiery pecan]

i didnt know that there are variety of ui similiar to A1111

[wise vigil]

like, I'm willing to branch out on settings, samplers, extensions etc, but
my ADHD brain DOES NOT like workflow spaghetti

[fiery pecan]

is it basically just a reskin?

[wise vigil]

It's basically its own new front-end, so, like
More than a recolor/slight reskin

[fiery pecan]

i see

[wise vigil]

1 sec and i'll post a screencap

[fiery pecan]

sure

[wise vigil]

the extra networks browser is one of the easiest ones to use, the whole sidebar can collapse to just a small column when not needed

and for even more space you can make the left options just icons as well

i guess it's not blue enough and too much orange to be 'solarized' the way most terminals define it but i like it and it was a total accident

you can change the ui colors on the fly without a restart

[fiery pecan]

Neat! And it can do detailer, face swap and all the fancy stuff? Basically if it can be done on A1111 it can be done here?

[wise vigil]

yeah essentially
the panels are modular so while you might need to resize the input area a little to get everything to fit comfortably, the underlying processes are the same

[fiery pecan]

I see

[wise vigil]

i havent personally used adetailer/faceswap/etc. but inpainting, photopea embed, worked etc

[fiery pecan]

Alright. And extension from github like the one in A1111 can be install here?

[wise vigil]

yeah
it has the same extension search/install page as a1111 with the same sorting options

the only things i would explicitly expect not to work would be ui overhauls like lobe theme or smth

[fiery pecan]

Alright. Do you know if i need to transfer my model into its installation or those path can be redirect to my A1111 model folder.

Comfyui can work by using A1111 model path so i dont need multiple file on each installation.

[wise vigil]

I use Stability Matrix so my models folder is symlinked to it
You may have to set the model directory manually in config files, I don't see an option to define the model directory in the UI itself

but since it has similar underlying code to A1111 I would be really surprised if you couldn't set the model folder

[fiery pecan]

Alright noted thanks!

[silver valley]

Symlinks are your help

Can be done with cmd commands or with a tool called Shell Link Extension

[still pike]

thx

[still pike]

thx

[dull portal]

Am I wrong? Wasn't there a way to do mixed race with like (asian|spanish:-1 to 1)?

[trail finch]

Longnian wallpaper

[hollow brook]

guys is there any extension to use a generated face on any image we create

doestn matte the checkpoint we use?

[terse mulch]

Need some help with LORAs, some of them simply don't work. Both SD and LORAs 1.5

[silver valley]

Controlnet IP-Adapter

Or Reactor

[gaunt ledge]

anyone have any idea how i can put the hood up on this model I'm trying to make it a spade shape but the ai wont fricking budge i also dont know how to describe a screen that has leds

[ember garden]

"Let the warm, amber glow of a Himalayan salt lamp wash over your space and transform it into a haven of tranquility. Share your before-and-after images and let the benefits of Himalayan salt shine through." please shape is natural and realistic

[restive socket]

Guys is there anything I should know before using multiple Loras in the same prompt?

[static hinge]

test the Loras with various fixed seeds and prompts, and try all possible combination of strengh of the Loras, until you start getting some good result and some magic strengh number

maybe start with both at :1.0, then lower one, then lower the other, lower both to 0.5, etc

0.4 in both usually make then not to break everything and work, some might work at full strenght both

but it really depends on the Lora combination and the prompt

[grave current]

for multiple loras i've found using lower values for strength helps, you can also try different schedulers since they all work a bit different on when are they loaded, linear/normal load different than exp/karras etc

[knotty night]

how do i do prompting here?

[restive socket]

Alright thank you so much guys

[leaden socket]

What would you consider a lower value?

[worn dome]

I've got this image. It's Leon and Jessie, from "Brawl Stars" videogame. I'm trying to create a prompt to generate a remix of that image. I want the characters to be watching a sunset, they are looking at the sun so I want an accurate ligtning. In the images I've generated, SD hasn't understood that Leon (the boy character) has a sligthly brown skin tone. He's got his skin yellow. I want to fix that. On the other hand, characters are supposed to be sat down (just floating on the air

This is the first image I've generated ( Denoising strength: 0.55)

Some important facts:
-I'm doing an img2img of this image.
-I'm using openpose and lineart (multi control net)

Prompt:
leon, jessie, leonxjessie, (cute:1.2), young, brawl stars characters, holding hands, sunset lightning, sky background

That's the prompt I want to make better. I'm not so good at prompting, so any advise is welcome

This is another image I've generated using Denoising strength: 0.7
It looks horrible, and Leon is supposed to wear bue shorts

[leaden socket]

Your prompt doesn't contain "Blue shorts" so it's not going to generate blue shorts.

[Question] I know it's hard to give an answer, but any suggestions on issues with upscaling I should try?

[worn dome]

I don't understand the question

What do exactly you want to do?

[leaden socket]

Not have these types of issues:

[worn dome]

https://youtu.be/_yALE8KziEU?si=gRTr51rfF6ID0dJM
Check out at this tutorial

[leaden socket]

Thanks but I don't really inpaint, looking for an solution through generation.

[worn dome]

Also, I think this channel is only for prompting. You should go #🤝｜tech-support

[leaden socket]

It's a prompting question...

[worn dome]

It obviously is a very difficult question for me. Hope you get it

[frosty summit]

Upscale with a lower Denoise number (0.2-0.3) to not introduce new things. That and just keep rendering more images until it looks good.

[leaden socket]

Thank you, this is my upscale and sampler:

[frosty summit]

0.05 denoise... I feel like that shouldn't be doing anything at all lol but I'm not familiar with that Comfy setup unfortunately

[leaden socket]

I'll try upping it to .5 to see if that does anything.

[restive gust]

@oak verge

[oak verge]

yo

[restive gust]

Send photo and write prompt here

And explain a bit.

[oak verge]

this is the style im trying to recreate

these are the irl toys

[restive gust]

I m not much of a prompt expert.

What about insects made of umm...jelly?

Slime insects..

[oak verge]

jelly doesn't work too well

[restive gust]

Ok lets for someone to help then.

[oak verge]

i can get as close as this

but it's not quite there

[restive gust]

What prompt is it

[oak verge]

90s3d image of a bug made out of (((gummy))) candy, simple, very sharp, high quality, lora:hj90s3d_v10:0.8

i tried using this 90's aesthetic lora i found

[restive gust]

And whats not good about this image yet

Is it too sharp?

[oak verge]

it just doesnt quite capture the real thing

i want it to look comparable

to an irl toy

i can do ceramic figurines just fine

for example

but this is out of my wheelhouse

[restive gust]

It looks like a jelly

And toy

[oak verge]

it's close

[restive gust]

Whats missing i cant seem to know

[oak verge]

but like i can make a figurine look ike i just took a photo of a toy

this is just missing the realness

[restive gust]

Hmm

So I don't know promptwise.

But as photoshop artist if i look at this image there are few things which are wrong to me.

First its more sharp than it should he.
Its so sharp that it made kt noise.

Second its overly vibrant to be real.
It is too much saturated.

Its supposed to be saturated yeah but not this mhch.

Third is sometimes background make things more real.
For example a table or something

Apart from that you can try changing model ig.

[oak verge]

thanks man

ill keep playing around

[tall prism]

Every generation has the characters present in the screen looking at the screen if ykwim. Is there someway I can avoid that?

[tall prism]

My prompt is

((Comic Page)), ((Comic Panels)), [(((1boy))), 12 years old, Young Child, Fair coloured skin, Black eyes, Flow Haircut hairstyle, Black coloured hair, Red coloured clothes, Cat Eye styled glasses, John, practicing calligraphy with red ink, making birthday cards for friends, winning first prize in a writing competition, dreaming of becoming a master calligrapher]

And I get generations like these:

The main issue is that the characters stare at us all the time. I have tried so many prompts in negative and positive prompts to try and fix this but unable to do so. It should seem more natural like interactions with objects and people in the scene.

[lofty agate]

Is there a good way to ensure that no extra characters are generated? I try keywords such as "alone", "solo", "1girl" or "1man" and as negative keywords I use "extra characters" and "multiple characters" but it still generates multiple or merge more heads into the body of the character if I make the image "too big".

[fiery pecan]

what model do you use?

[fleet walrus]

教程

[next halo]

hi guys, I am struggling to get what I want. I need a flat seamless silk material, but no matter what I write, it have folds in it

[fleet walrus]

Tutorials

[next halo]

more specific?

[worn dome]

I want the girl to wear boots but the results I'm getting are so bad. I'm using controlnet inpaint. I'm looking for a prompt that may work.

8k, UHD, high resolution, delicate detail, realistic, highly detailed, perfect composition, beautiful, detailed, insanely detailed, octane render, trending on artstation, masterpiece, intrincate details, (a cool brown boot:1.1), cowboy boot, solid white background

That's the prompt I've got now

Or maybe the problem could not be prompting?

[crisp spruce]

hi guys -how to create the image -what to type in the text box to getthe prompt input box--before it was # dream something --- i wm not wble to input my prompt --can you help pleaase

[dapper tree]

#1047610792226340935

[crisp spruce]

#1047610792226340935

hi,i want to create a image --how to do it -not able to do it

[dapper tree]

Click on that link i sent you and read the last post.

[crisp spruce]

we cannot create images now?

but what to first type in text box to get the option to input text

[dapper tree]

🥄 The bot is currently down while they update it. There is not ETA on when it will return. You cannot make images using the bot until its back up.

[lofty agate]

At the moment Mistoon Anime, but it happens to any model but the resolution it "breaks" at can vary slightly. I assume it is SD that try to utilize all available space vertically rather than only use some of the space. That I use weird widths and heights probably also contribute 😛 768x2048 give me this, only difference is "full body" on the second image. Prompt: "solo, alone, 1girl, looking at viewer, smiling, red hair, blue eyes" and negative prompt: "multiple characters, more than one character, several characters, extra characters, disfigured, deformed"

What I would want is to have the character at the bottom/top of the image and then utilize the space below it for something like having a sky above the character.
Maybe there is some extension I can use to generate these weird resolutions at and keep the character in a certain area of the image rather than it tries to take up all space?

[silver valley]

Thats normal SD behaviour.
1.5 based models are trained on a resolution of 512x512
Going much larger than that will result in duplicates and deformations

Thats why most people use lower resolution and then Upscale the image to get better quality

Or using sdxl models because they are trained on 1024x1024

[lofty agate]

Is there a way to tell SD that it should only put the character in a certain region and not go over into other regions? Or should I render a empty background without characters first and use inpaint to add a character?

[silver valley]

Thats possible with extensions like Regional prompter or Tiled Diffusion

Or you generate at normal resolution and then use Outpaint

To get a larger image

[lofty agate]

I tried regional but it put my character all over the place anyway. But I inly tried it very briefly so I might have done something wrong.
I will put some more effort into regional, and try tiled and outpaint.

[left prism]

I'm struggling ATM to prompt a man with a camera for a head without using a LoRA. Can anyone accomplish this?

[leaden socket]

Could anyone recomend something for my upscaler, it seems to just blur everything and remove detail:

[leaden socket]

Try an img2img?

[left prism]

Tilediffusion, which I'm fighting to get working since the last update.

[left prism]

Gotta greenscreen it. IMG2IMG leaves too many artifacts.

[leaden socket]

I don't understand what greenscreening would do with prompting, as img2img would be perfect / inpainting.

If you're getting artifacts you need to play with your settings, this is an example of a img2 img I'm doing. No artifacts.

[left prism]

I'm using a greenscreen LORA. Unless I Set IMG2IMG to 100 denoise, there are bg artifacts.

[leaden socket]

Okay, but you said you can't use a lora.

You can use multiple loras.

[left prism]

There isn't one. I'm making one and can't find enough source images. Dall-E 3 is good for it, but slow.

[leaden socket]

Lower your weight of your lora to do img2img

[leaden socket]

Sorry but this doesn't make sense.

There's hundreds of loras out there.

[left prism]

I'm genning source images to make my own.

[leaden socket]

What does that have to do with anything?

You can generate a source image with a img2 img

[leaden socket]

Left is a template I got online, right is my prompt

[left prism]

This is what happens when I'm up for three days. I'm not thinking straight. TY.

[leaden socket]

Grab a picture of a man in a pose you like, photoshop a camera on his head, and use img2img

[next halo]

I am lost, I cant get the SD to make a plain flat texture, what prompt to use ?

my prompt: silk fabric, texture, material for blender , seamless, flat, plain

negative prompt: text, logo, folds

[misty cypress]

i need a prompt for an astronaut going in space

[leaden socket]

an astronaut going in space

[worn dome]

Epic

I'm looking for a prompt to describe SD that thing, but I neither know what is it

also I'm looking for a prompt to make arms thinner. That arm does not like good

[leaden socket]

Golden bead hair accessory?

[worn dome]

Those are results I found about it searching "Golden bead hair accessory", but I think it's not exactly that. However, it's a nice start

[leaden socket]

If you're just trying to remove it, use inpaint model, or img2img with the template you like.

Generating from a raw prompt is difficult if you want to get specifics.

[worn dome]

Actually, I'm doing img2img from a render a did to a 3d model. Also using controlNet. Lineart realistic I think. However computer is already shut down so I don't remember

[leaden socket]

https://civitai.com/models/116230/bad-hands-5

I always use this for characters, may help your arms

[worn dome]

But, I'm using another model, how should I apply?

Maybe an inpaint changing model?

[leaden socket]

That's an embedding.

[worn dome]

I don't remember what exactly is that

Can I use it without quitting my model?

[static hinge]

How to set ip-adapter-face-id-plus so it doesn't break and does't do a kind of cartoony face?

Sometimes I lower the Lora, sometimes CN strenght, it changes the style and closeness of the face, sometimes it breaks, more or less, then it doesn't break but makes some kind of cartoony faces.

0.6 in the Lora en 0.5 kind of perform, quality is aceptable, but it feels it can do better as it does sometimes

Sometimes I don't understand if it breaks because CN or the LORA is too high or too low

[crisp lily]

Try a low CFG, in my example code I've used 1.0, I find it works better then, you might want to couple it with a model that works with a low CFG like a LCM or Turbo model.

[static hinge]

Great I was thinking about CFG, I notice it affect it too but didn't lowered it much, ty

[static hinge]

Wow what it was actually broken was insight face. After something about ssome Onnx model broke Automatic1111 I tried to install ReActor, I reinstalled Automatic1111 and now it works, it was broking and making funny faces becaause of that it seems

[restive socket]

when im using inpainter and I wanna change the color of something, can I use color codes? for example "#4682B4"

[fathom spruce]

Does anyone have any Lora suggestions or tag/prompt suggestions for high heels or feet on v1.7.0? I primarily use the galenaREDUXv2.0 checkpoint and when i do something like, "standing/sitting, full body, red footwear, high heels" etc etc, it always comes out pretty bad

[full lily]

Hi, I have a question. How can I have two or more characters in my image using Loras? I am new to Stable Diffusion and when I try to put two characters their clothes or faces merge/overlap. I have seen that it can be done but I can't do it 😥

[bright knot]

help i typed

dog

in the prompt

and this is what came out

[silver valley]

What's your model?

[kindred juniper]

Does keyword influence consume tokens? For example - (solo:1.3), would the brackets and number contribute to token limit, or will SDUI count them as a system command and only "solo" word will consume tokens?

[bright knot]

Sdxl but I fixed it no worries 👍

[restive socket]

guys how can I add LED lights or that kinda lighting in a room? i tried to use "LED lights" prompt but it doesnt work. im on SDXL

[timber rivet]

like accent lights, maybe light strip

[restive socket]

alright will try

one more question, when im using inpainter and I wanna change the color of something, can I use color codes? like #4682B4

[hollow brook]

I have webui 1111 installed. How do I also install comfyui

[dapper tree]

Maybe try StableSwarmUI instead of 1111? It uses comfy as a backend so it's auto installed.

[hollow brook]

Ill take a look at that ty

[dapper tree]

#🐝｜swarm-ui message

[stark lark]

What's up with the LMS sampler? For me at least it works pretty well and fast. Images aren't much worse than the other ones that work 'good'

But I hear no one ever mentioning it, why is that?

Thinking about it, it probably makes sense tho.. might be the reason for why there there are so many samplers and upscaler because different people prefer different results?

But for pure txt2img are there top 3 samplers/upscalers? Most stuff on the internet nowadays is quite outdated:)

[unique solstice]

I'd like to request Negative prompt examples for 1.5 checkpoints. I'm using one I found on a forum.but I'd like to see if there are any better options.

[tired vigil]

How do you guys fix the teeth problem in dreamshaper XL?

[prime ingot]

Which channels do we make a prompt? also, how do you write a correct prompt?

[restive narwhal]

Super newbie inquiry.

Anyone can offer tips and tricks keeping the imported image intact but have the QR code diffuses?
Meaning I want to keep the imported image as is.

[delicate arch]

Why do my fingers keep coming out like spaghetti

[frosty summit]

I would consult a medical professional

hands are notorious for being bad in AI but I haven't seen much spaghetti

[delicate arch]

Ah

Wait till you see my generations

Also how do use clip skip?

[frosty summit]

I don't!

[cold rock]

hey guys ! how and where can i prompt an AI modification of an image I want to import ? big thanks for any tips :))

[delicate arch]

If you have automatic1111 already set up, you can upload in the png info tab and send to img2img

[frosty tiger]

guys i need prompt that shows of a rocket flying in space. Inside the rocket, show a big window where two people are floating, view from far but zoom alittle on rocket window, chatgpt not help somehing like that but in realistic art style, and not that smal circle windows but one panoramic photo of rocket and in this window we see them from first person

[mellow wave]

Is there an extension or something that describes an image? For A1111. So image to text instead of text to image? So if i upload a sandal for example it writes me a text like: a close up of a pair of sandals with a bow on the side, gradient light blue, leather belt, etc..?

[static hinge]

the last two, the clip and the box, are built in, clip interrogator, deep boru interrogator, in img2img tab

Or this extension, I haven't tried it https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/7968

[static hinge]

you will need to use some controlnet model (canny, depth) with a reference image that look like the want you want, and then inpaint the window and the characters

very difficult to get it exactly as you want with just prompts

[bleak goblet]

Hello all please can anyone give me this type of prompt please just one i will practice
i tried all but getting bad errors so help me out guyz

[ruby bough]

Looking for a prompt that would allow me to generate realistic humans/animals, but with tumors all over or additional limbs/eyes/mouthes/ teeth/ hair/etc.

Its for a grim dark setting of a dnd game I'm running.
So far only keep getting normal images, with no grotesqe modifications.

[marble imp]

dm, I can share

[frosty tiger]

where i can find some templates of prompts for every style? like e.g cyberpunka and matching to this style lights, etc

or prompts with the words quality universal, because adding them every time is an ordeal

[marble imp]

You can try referring this,

https://supagruen.github.io/StableDiffusion-CheatSheet/
https://civitai.com/images

For style templates, you can use this or customize it as per your need.

Store the styles.csv file in the root of your stable diffusion installation directory

[frosty tiger]

yeh i see that tempaltes what use sebastian on yt, this is nice because u can put this to SD but i use most of the time comfyUI and there no options to create tempaltes in comfy?

[neat viper]

any idea how i can get sd to stop generating the little middle tuft of hair? I want it more like what is shown in green, but i have no idea what to prompt it to do so. I have tried things like a negative prompt of middle part, and a positive prompt of side part, but nothing seems to work

[marble imp]

No for comfyui so far I guess, I am still exploring some options similar to SD for comfyui.. Will let you know if I come across any

[marble imp]

I guess you can try inpaint option and try to select the area that you have highlighted in pink and choose masked to regenerate the area..

[neat viper]

i haven't messed around with inpainting yet...

i think i am starting to get good reslts though with another model

[neat viper]

ill try this.

[rich crow]

bangs_between_eyes is a popular booru tag, you can include that in your negative prompts and increase the weight accordingly to see if it makes any changes

[neat viper]

awesome

[rich crow]

i coulda sworn ive seen v-shaped bang or some variant of that before too in booru sites but i cant remember exactly what it is

nvm scratch that, thinking of something else. bangs_between_eyes in negatives would probly be your best bet

oh yeah this variant is also super popular

[sly crest]

can anyone help me

[marble imp]

What's the issue?

[viscid trail]

hey ppl, how do I render boxes without those things? I used "simple" "--no handles" "no lockets" but it just ignores this
prompt: "isometric set of 3d rendered worn very realistic repair tool kit instruments kit stored in simple metal boxes, box without details, less details, no handles"

[dry goblet]

How to force SDXL model to generate picture completely full of flowers? I tried "picture full of flowers, colorful, high quality..." but most pictures I generated have quite big gaps between flowers. Help me!

[stoic belfry]

hello, I am super new to the community and AI photo development and looking for someone that can assist me.

I am trying to get Enfugue to work, does anyone have experience with it? Again, so new this the the first WebUI I'm trying.

[frosty summit]

I find variations on "fill the frame" to help, completely fill frame

[unborn relic]

does anyone know a prompt to get an image like this or how can we do it ?

[keen panther]

Hey guys,
Is there a way to use ComfyUI and Blender in tandem with each other? Export depth, diffuse map, and freestyle map from Blender which will automatically be transferred into the right nodes in ComfyUI to generate an image. When an image is generated, it will go into Blender texture projection to paint onto the model?

I know dream textures exist but it is very slow compared to ComfyUI and it requires me to make the textures from one uv to another as well as edit it if there are any artefacts.

[frosty tiger]

look https://cdn.leonardo.ai/users/768524ad-0f5c-4963-b130-4b943f1643b0/generations/c6ced90a-642e-4095-b179-11baf8077111/variations/alchemyrefiner_alchemymagic_2_c6ced90a-642e-4095-b179-11baf8077111_0.jpg

prompt high quality, 8K Ultra HD, vivid colorful, Seamless patterns, fabric art, Art station, stary night many colorful detailed Design combining flower and leaf, splash arts, aesthetic for Tshirt design, white tone, photorealistic, ultra realistic, by yukisakura, awesome full color,

[soft kernel]

hey guys, if i have an image of a shirt, and i want to create a similar art style shirt just with a different material, what kind of prompt i should use?

[orchid zephyr]

Hello! How do I generate 3D models?

[keen panther]

There are three places I found for generating 3D models that are free and does not require credits.

If you have an image of a human character you can try PIFuHD.

If you want to generate a non human mesh you can try Lumalabs and Shap-E. Keep in mind that Lumalabs generates a texture that is wrapped on the mesh while Shap-E uses vertex painting to colour the mesh. Lumalabs and Shape-E generates images based on prompt and not from images.

[wise hollow]

Crowded office space with everyone wearing VR headset

[restive socket]

guys whats the best way to get rid of mutated fingers/toes? I just generated a perfect image but the fingers and toes are mutated

[fiery pecan]

What model did you use?

[restive socket]

EpicRealism

[fiery pecan]

in my experience. some model can handle hand better than others. You could use Adetailer to fix the hand. some time it work some time it doesnt. depend on each image

[fiery pecan]

Are you using A1111?

[restive socket]

Yes

I tried ADetailer but it didnt make much of a difference

[fiery pecan]

Can you share an image? Would like to see the metada. See if the prompt can be adjusted,

[restive socket]

Sure 1sec

[fiery pecan]

if you can run SDXL model. you have better chance of getting good hand.

like alot better in my experience

[restive socket]

i can use SDXL

[cunning bolt]

help me

[fiery pecan]

let me take a look. typically SD 1.5 have a hard time making hand look good when holding object.

[fiery pecan]

Ya got to be more specific. "help me" alone not gonna be helpful to assist your issue. . list down your issue and hopefully someone can help you.

[fiery pecan]

The image has no metadata

[restive socket]

Oh

I screenshotted it just for u to see it my bad lol

1sec

tbh I didn't add anything about holding a bag but fingers are mutated either way

[cunning bolt]

how to create a image

[fiery pecan]

You may not add it. but it it randomly put in a bag thus making the hand harder to look good

[restive socket]

yep

[fiery pecan]

you may need to adjust negative prompt. to exclude bag or hand bag

[restive socket]

yeah ill try that

other than that, is there a lora or an embedding for hands and toes that u know of? sd1.5 or sdxl doesnt matter

[fiery pecan]

the easiest way i suppose is to go for SDXL model. they have a much better hand rendition. i pretty much make 2D or illustration images. and the hand i got are amazing.

one of the few realistic render i had.

you pretty much dont really need to use lora for it

but the con is that SDXL dont have as many lora option as SD 1.5

the first image is ponydiffusion model and the second image is Reproduction SDXL i think.

[restive socket]

ohh alright

will keep that in mind

thanks a lot

[fiery pecan]

welcome. Sorry if i didnt offer much. but the simplest way is for the hand to not hold anything. if you want it to hold something be prepare to render alot

or use controlnet to guide it.

you probably need to use depth map or a openpose to make a skeleton hand of it so the ai can follow.

[restive socket]

yep

i was also thinking of using dw open pose it makes a good difference

[fiery pecan]

you should have better chance for the hand to be somewhat coherance than not using it. also. you probably need to increase the resolution. higher resolution more stuff the ai has to work with so it should improve your chances.

[restive socket]

oooohh makes sense

i hear the resolution part for the first time

then I have no choice but to use SDXL lol

[fiery pecan]

yeah pretty much. Do you use a bunch of lora in your render?

[restive socket]

i use 2

usually

[fiery pecan]

I see, hopefully you can find a replacement

though if it is something specific. chance are, you wont

[restive socket]

one for dof slider and one for zoom slider lol

[fiery pecan]

Maybe you can prompt SDXL for that.

hoepfully no need lora.

[restive socket]

no prompts managed to help me to remove dof and blur lol

[fiery pecan]

nice also. here one render i just make, look at the SDXL model manage to make good hands. Sure the knife holding is kinda wierd. but the hand look awesome. But i never seen SD 1.5 manage to make hand look good holding a knife without some controlnet shenanigan.

[restive socket]

true

sd1.5 sux when it comes to hands indeed

i was thinking of generating my image first without dof and zoomed out, then send it to img2img and switch to sdxl and make adjustments

[fiery pecan]

that can work.

but still, that is quite abit of effort need to put compared to just getting it down from rendering straight away.

[restive socket]

true

[tawdry pier]

Hello, 1 year ago, I generated images with Dreambot, since that period the dreambot channels were deleted, how can I find my pictures ? And where can I prompt again ?

[keen panther]

Hey guys, I have a 3D mesh of a character that I want to texture. I wanted to use texture projections However, I am concerned about generating the textures with lighting, highlights and shadows which would require me to edit them. Is there a workflow for generating the texture for a character without lighting information. I also want to include ambient occlusion for detailing. I use Automatic1111 and ComfyUI.

[tawny quail]

You could texture then compute albedo using another AI model...

Albedo will strip out the lighting info

[keen panther]

What AI models do you suggest?

[finite prairie]

Any ideas on how we can improve masking (https://platform.stability.ai/docs/api-reference#tag/v1generation/operation/masking) It seems to generate a completely different image every time and ignores the masking.

[surreal condor]

hey guys, i've been working on a render but ı don't wanted to make it look different . don't wanna change anything, same building but more realistic. i'm using control net but still can't get the similar and more realistic image of my input.

[hollow tapir]

and what settings are you using to get to that results :p ?

screenshot your webpage

[surreal condor]

i did not change anything on my final work but ı randomly play with the threshold settings on my previous project

is it happening because of my model ?

[surreal condor]

and this is the prompt that i used: don't change this architectural building. base color of the building is white, dark grey corners, fancy car in the bottom floor of the house. Photorealism, realistic modern house, modern building, blue sky and clouds, grassy field left bottom corner, two floor and one roof, (((realistic render)))

[hollow tapir]

Can you rather screenshot the whole webpage please ? I'll afk for a while, got stuff to do. So don't expect answers from me for the next two hours or so.
In the meanwhile someone else might help you.
(But from what I see you might want to tweak your canny thresolds and the weight of controlnet over prompt)

[surreal condor]

i find the solution. thanks for responding

[drifting bay]

What are some good prompts with 1.5 for first person interactions? I can make this work in other models, but wondering if there are good prompts to make this happen consistently in 1.5 and if certain fine tunes are better at it.

[tepid osprey]

if you are using a 3d rendering tool, export a depth map instead of a beauty render

[kind idol]

Hello, I tried to make a deepfake with reactor extension, but there was tremors in the hair and there was not much change in the face. It does not look exactly like the face in the photo. Can you help me?

@stable prairie I'm sorry I'm tagging you but I need urgent help.

[terse trellis]

The evil one

[maiden cape]

oh shit

you trying to convince me the first one was not?

:3

[terse trellis]

she is a cutie

[maiden cape]

haha

[vital remnant]

What are some prompts that you always start with when making a new piece?

[kindred blaze]

Wondering if anyone could help please? I got this new checkpoint, but so far all my results are coming out with fuzzy blurred edging...how can I get a more crisp result?

[silver valley]

Hey, first you need a VAE file for better colors.
To activate the VAE menu go into
Settings, User Interface, Quicksettings, there add sd_vae
Then hit apply and reload ui. Then you can download a VAE (like kl-f8-anime2, 840000-mse, or clearvae) and put it into models/VAE folder And select it in the new dropdown.

The actual blurry edging comes from the wrong settings of your Hires Fix.
You have to set the hires steps to 10.
Then choose Resrgan4x Anime6b for example as upscaler.
The Denois should be 0.5 or lower.

For latent upscalers try bicubic and stay at 0.5 denois

[kindred blaze]

Thank you for the advice, I'll try that first thing in the morning I need to get a little sleep here first it's been a long night 😊

[silver valley]

Sure, np 🙂

[fiery pecan]

What do you mean by making a new piece? Like if i want to render a character. What prompt i usually use but change the description for each character?

[vital remnant]

Well I assume people start with things like "masterpiece, highly detailed" etc and then go from there

That set of prompts that people start with is what I was asking about

[fiery pecan]

i see, that is really depended on the model and what kind of render you looking for (realistic/anime/illustration/etc)

for me i been using PonyDiffusion quite alot lately.

what model and what kind of images do you render?

The prompt i use. the green prompt box contain the "style" that i use and i only change the black prompt box to render the stuff that i want.

@vital remnant i pretty much render illustration or 2.5D images only. if you looking for realistic render prompt. i am afraid you gotta ask another guy.

[vital remnant]

Nah that's helpful

I mostly render dark stuff, and right now I'm using NightCafe because my local Automatic1111 keeps shitting itself

I do a fair amount of 2.5d too

It's among my favorite styles

Before I stopped using my own art maker, I was using Dark Sushi Mix from CivitAI for basically everything

Mostly I'm just fishing to see if there are consistencies among the 'basic' prompts that people often start with before tailoring it to what they want

[fiery pecan]

What do you mean by "dark stuff" like kinda horror stuff? or like actually image that has low light lol

[fiery pecan]

ohh, i'll keep that in mind. always on the look out for more model'

[vital remnant]

Well, this is my most recent thing

[fiery pecan]

i see, neat!

[vital remnant]

I guess both are true

[fiery pecan]

Can you post the prompt for the image here? i am kinda lazy to open up A1111 to see if the image has metadata or not

[vital remnant]

a front facing ancient arch leading to a dark tomb, black and white illustration, Gustave Doré, symmetrical, cosmic horror, dread, hyperdetailed, unsettling, darkness, fantasy

[fiery pecan]

Have you try out comfyUI? it is typically more smoother and faster than A1111

[vital remnant]

I did try it but it wasn't particularly intuitive

I guess A1111 wasn't either to begin with

[fiery pecan]

true, I suppose it is kinda hard to get into it. If you want. i can show you the rope. to use it for regular txt2img and img2img stuff is pretty straight forward.

its only get challenging once you started to use Ipadapter, controlnet, Adetailer or some other shit in your workflow.

[vital remnant]

Nah, I appreciate it but I'm fine using NightCafe for now

I'm not mainlining AI art right now, just a few when I have an idea

And hopefully the next time I do do it a lot, A1111 isn't bricked

[fiery pecan]

try using SD Next. Apparently it is a more eficient fork of A1111 or something

it is also has other installer for various AI tool if you need it like txt2vid or something like that

[vital remnant]

Thanks, I'll have a look

And hopefully my issue isn't shared across multiple SD webuis

[silver valley]

For any issues with auto1111, you can ask in #🤝｜tech-support

Share your GPU name, and and screenshot of the error you get

[terse trellis]

don't use latent upscale. If you insist, don't use with such small denoise

[kindred blaze]

Thanks! The prompts are coming out alot better now ^^

[silver valley]

perfect, np!

[tepid osprey]

are there pre-existing comfyui nodes for "find bounding box of image A inside image B"? kind of like grounding dino except with images

is there a clip vision for clip seg approach?

[static hinge]

I usually start with the most basic stuff I want, I use SDXL so I prompt the basic and no negatives.

So the most basic idea I have, sometimes with no style at all to see what styles it generates.

Then I start tweaking, the style, the subject, the photography stuff like angle. Use negatives for things I see it is generating and I don't want.(1)

Then maybe I add "masterpiece, best quality" to see if it has some effect.

But (1) negatives should handle with care, as they may mess with the things I want and chance the style, subject. But they are powerful.

Also it depends if I have a really crafted visual idea or I just throw an idea to see what SD makes. i.e: librarian girl in a park eating cookies vs Painting by (painter), scene from (whatever) movie, like the second it is really more random, to see how the combination works, while the first it is more controlled.

[fiery pecan]

the original SDXL model?

[static hinge]

no, I use the highest rate models on civitai, juggernaut, albedo, zavyschromax

[brisk peak]

.

[zenith stirrup]

hey guys, I'm looking for styles for SDXL, is there any repository where these are collected?

[quiet raft]

Hello, I've got a question.. If I wanna use same steps, same hires and etc.. Can I with several prompt use the X/Y/Z plot and generate like 5 different types?

OR casn I download something and uses files etc?

Anyway around it?

[soft kernel]

I wrote "add cool hair", what am I doing wrong? using controlnet

[runic bear]

im trying to get this crime noir comic style working, but the results are totally disappointing. has anybody done this style yet and can give me some advice ?

[runic bear]

did you already try https://civitai.com/ ? thats where i get my models and stuff

[runic jackal]

so i have a question, in the last few days i was using loras, but they dont turn out how i see it on civit.ai so my question is: do you need a special model in order for the loras to work?

[runic bear]

most loras have example images where you can see the whole settings, model,lora, denoiser, cfg scale, steps etc., just check the parameters on the examples. but yeah, i noticed that too, my results differ strongly from the images posted for the specific lora

[runic jackal]

i see, thank you, is there like any model which can be used for anything?

[willow ledge]

Hi guys, i need advice to create prompt for style image like that , hope i'm in good section

[runic bear]

looks like a woodcut but in 3d, maybe that keywords are a good starter

i want that cute dwarfs, trying to get that style

3d woodcut of some dwarves, its the right direction

[runic bear]

ok i looked a little bit more into it, its a "wood carving" like this real pic

i guess using "wood cut" instead of "wood carving" gives a more 2d style like a picture

[runic bear]

dude thats the best i got, some dwarves my grampa would have in his collection, but wood carving is the best starter, didnt figure out the keywords to get it unpolished and fresh out of the trunk, but thats due to my toaster computer, that takes 1 minute for a picture

[dim mica]

Any suggestions for getting something to breath fire? I was listening to some Weird Al but I can't seem to generate a fire-breathing hamster.

[kindred blaze]

Hey all, so I've been playing around with Bing's AI image generator a little, got some cute sheep-girl results and I was wondering if anyone would have any advice of how I could get similar in stable diffusion?

[twin yoke]

All of the above are variations of:

Positive:
anime lamb girl with long soft platinum wavy blonde real human hair figurine closeup, white fluffy furry sleeve sweater, (long porcelain fingers:1.2), bright blue eyes, sheep ears, (bedroom:1.1), (real human hair:1.1), light blue bow hair ornament, light blue pleated skirt, eyeliner

Negative:
lamb, horns

[tough hound]

Can someone help me? I don't know if this is the right place to ask questions

I need to produce some concept art of a creature, for a short horror film that I'm going to start recording soon and I would like to generate some ideas like this artist that I found on Instagram, he can generate images outside the standard ones that AI usually generates for everybody

I need something along these lines, what would I have to look for in prompting?

[unreal delta]

Hi everyone.. i can't get any bot

[upper merlin]

Hi friends, a quick question regarding prompts and real estate. Let's consider a scenario/prompt (img2img):

• A poorly furnished living room photo that I want to enhance with AI using a given prompt. Most of the time, the AI not only improves the photo but also doesn't adhere to the room dimensions and adds elements that were not requested.

My question is how to instruct the AI to stay true to the original photo. That is, to respect the room dimensions and not add imaginary windows. I've read some articles about negative prompts, but I haven't been able to achieve convincing results so far.

[kindred blaze]

Thanks I'll give that a try!

[timid trench]

Hi everyone! What should I write as prompt, if I want this character's side/back profile? This is what got me to this character so far.

[rain cradle]

[I used Google Translate]

hello teacher

I want to create animation => real life in img2img tab.

But it's not going well.

They don't resemble each other at all.

help

[option]
Denoising strength 0.51
control weight 1
majicMIX_v6 checkP
Lora X

[willow ledge]

Thank i , i search too but i cant get the same result . Square and front of view

[willow ledge]

nice one i search in this way

[willow ledge]

I'm working on it but not the same result, if soem expert can help 🙂

[west flower]

how to create a picture

[peak igloo]

Hey does anyone knows when bot comes back to normal. Is there any prompt writing method that I can add an reference image so that the bot can build up a similar image to that, anybody knows ??

[silver valley]

you would need Controlnet IP-Adapter for that

[kindred blaze]

It's subtle in this image I made in Bing, but I would like to duplicate the faint soft "fuzz" she has on her face and hands in stable difussion, almost like a felt or velvet material.
Any helpful advice is greatly appreciated!

[upper merlin]

Could someone please provide me with keywords to include in my prompts to instruct the AI to stay faithful to the original image?
For example: A photo of an ugly living room that I would like to enhance with Stable Diffusion but it must remain faithful to the original image.
Thank you in advance.

[grand moon]

What you are looking for isn't a prompt, it's a setting called 'denoising strength'.

[willow ledge]

Hi, any advice to setup the light on the center of the caractere , same light each side :

[plucky vapor]

what happened to the prompt bot ?

the ones where you said "/dream etc etc" and it genrated you an image

[delicate ledge]

What is it about your example that you are looking for? If it's composition try Regional Prompting. If it's content try different models and loras.

Hello All,

I'm trying to show three political figures standing on the planet with money swirling around them. I was using standard prompting and was able to get three men standing on the planet but it tended to be the same man three times. I've switched to using Regional Prompting and am only able to get the center figure to show up, but it is always the figure I've specified.

Here is my Prompt:
Photorealistic, intricate and detailed, extreme perspective, POV looking down ADDCOMM
blue sky with money flying through it. ADDROW
blue sky with money flying through it. ADDCOL
(Alan Greenspan:1.5) standing on the planet with his hands outstretched, large face ADDCOL
Jerome Powell standing on the planet with his hands outstretched, large face ADDCOL
(Ben Bernanke:1.5) standing on the planet with his hands outstretched, large face ADDCOL
blue sky with money flying through it ADDROW
Planet Earth as seen from space with money flying around it up into the sky

.. and settings with yet another single person standing in the scene are attached.

[delicate ledge]

If you want an animation to look similar from one frame to the next try Deforum. It's availale in Automatic1111.

[kindred blaze]

So after many attempts, I am still not quite able to match some bing creator results to my stable diffusion results..
Does anyone know of a lora or checkpoint that can "imitate" bing results?

Trying for this style and look result:

[cinder olive]

Best Quality, Masterpiece, 8k, Perfect, Landscape, Festival on Mek-Sha, Cyberpunk, neon signs, ((sci-fi)), a market festival in a busy sci-fi city, busy alien market, Mek-Sha, Scum and Villainy, Star Wars, Sci-Fi, starry night sky,

I'm trying to get a Star Wars style festival scene. What are some more prompts I could add here?

[willow trout]

I need some help. I'm trying to generate some designs but a lot of them are insitu (e.g photos of magazines, billboards, etc). How can I format my promp to emphasis a flat image (without borders etc)? I think it's challenging because if for example I mention "2D" of "flat" then the design itself will be like that.

[runic jackal]

does anyone know what kinda specific model bing ai uses?

or which model it could be

[astral oar]

on the first day of spring

[whole estuary]

how do you get horns to come out of forehead more consistently? Ponydiff

[bleak goblet]

guys please someone help me out to get this type of results
https://www.instagram.com/p/C3cuSRpvGOX/

aiart.mania

Reyansh as spider man.

Transform your sweet memories into visual spectacle.
Did you miss the photoshoot for the past memories for your kids or for you or for your family.
Dm if your interested to get for your loved ones.

#spiderman#marvel#spider
#aiartmania #aiartportrait #aiart #aiphotos #AiPhotography
#aiimages #aiartwork #aistillsCustom...

Instagram

[rough axle]

hey, anyone got some input on how i can generate background images for car advertising, so i basically want a flat surface where i then can inpaint the correct car. i want the images to look as real as possible 😄 im hitting kind of a wall and ive tried tons of different prompts and nothing really works good, im using automatic1111 and locked into a instance where im forced to use epicrealism as a model due to some constrait with my friends setup

[slender pasture]

I'd love some suggestions on what I can do to make the background more full and complete for my gens~
indoor scenery, scenic background, Castle Chamber, (stained glass window), (plants), ((roses)), furniture, checkered marble floor, candles, marble, BREAK

I also have the Detail Slider LoRa active.
This is the part of my code with the "physical" background items~

[hot dirge]

You're kinda right, just by even slightly concentrating to the background, i begin to see how it really makes no sense

[silver valley]

maybe if the person is smaller or the image is wider, there would be more space to generate more things properly instead, it also can be model dependend:
here is what i got:

[burnt nova]

It's not getting the results I wanted (both girls come out either tan or white and not one of each). Is there any settings I need to adjust? Or is it purely a prompt thing?

[finite gale]

prompt help? maybe its the LoRa trying to get more closer up pics from the waist up, but it keeps spitting full body shots

nvm cowboy shot kekw

[muted haven]

any suggestion for negative prompt to avoid artwork to be rendered as artwork in frame in a room?

[lavish mantle]

no mockup

[restive socket]

I have a question, i inpainted a skirt, i just wanna change the color of it, what prompt should I use? its a detailed skirt, kinda like tech wear style and I wanna keep it, just wanna change the color of it

[silver valley]

You can do that with extensions like regional Prompter or Multidiffusions prompt control.
There you can specify parts if the image where which person should be

[slender pasture]

may I ask how you achieved this?

[slender pasture]

Rules of Thirds and Depth of field have been added

[slender pasture]

adding BREAK to the end of my sections has helped too

[real echo]

anyone knows of any lora or way to make an anime smile in V like this?

[silver valley]

I used your prompt and an other model, .I described the character, added break and then your background stuff.
Used xpero endless model I think

But got good images on other models too

[random verge]

What's wrong with my problem "Masterpiece pixar nickelodeon cartoon of ((a young brown haired girl, 20 years old)) on the bed in her bedroom, on the floor of the bedroom is a small mouse, digital painting, artstation, concept art, sharp focus, illustration, volumetric lighting, epic Composition, 8k, oil painting, cgsociety ". I either get images of the girl with a mouse on the bed with her, or both are on the floor. Sometimes the girl is even a mouse.

[random verge]

Another one that's giving me trouble : "a teenage girl and boy, standing at the entrance gate of a spooky mansion ((tim burton style))" . Somehow, SD always draws three people (?).

[thorny ledge]

#📝｜prompting-help a book

[quartz coral]

"(full-body dress:2.5)" and "flat chest,( AAA-size breast:1.7)" gives me this... what do I do?

[hollow tapir]

2.5 is crazy high strength.

Try to rephrase the whole thing, remove some "useless" words/tokens so the strength is more evenly distributed across each token.

Put cleavage into negative prompt

etc

[keen panther]

Hey guys,
I have a 3D model of a character who I want to texture using Stable Diffusion or ComfyUI via texture projection. However, I am encountering some difficulties in achieving this as the generated image does not conform to the shape of the mesh. I have tried using using depthmaps, canny, normal, and sketch control nets to no avail. The images for the normal, depth, and sketch are created through the Blender compositor. I just wanted to have fur maped onto the face of the character without adding any additional details. I have a reference image from a previous generation that illustrates what I want the fur to look like. I am happy to discuss privately via direct message on Discord to resolve this issue.

[sturdy mural]

Hii i need help with my prompts

portrait of one woman holding her baby in her arms (Baby: Black hair, red eyes), posing for the camera
(Karin Uzumaki) red hair, glasses, long hair, red eyes,glasses karin, red lipstick, earrings, queen, big moon, very romantic, , uchiha symbol on her dress , very long hair, , glitter makeup, eyeshadow, glitter,blush shy, round cute face, cute, blush, (violet sleveless jacket, arm gloves, overknee socks, heels, forest queen)
cherry blossom smile, happy smile, waiting for kiss, seductive eyes, accessciors, soft and warm color, movie still,bright, vivid,dreamy, evocative use of light and shadow, incredible beauty, anime style, vibrant, studio anime, beautiful, textural, ray tracing, absurdres, offical art, illustration, (((masterpiece))),(((best quality))),((ultra-detailed)), hayao miyazaki style, perspective, veil, moonlight playful, holding her baby boy with black hair and black eyes in her arms,
lora:Karin:1
this is what i prompt

and every generation looks weird like this one

all 42 look bad, i tried with various diffrent prompts, adjusted them etc. But i didnt get the outcome I wished to get

[shut nexus]

Try removing these prompts
(((masterpiece))),(((best quality))),((ultra-detailed))
and be more specific in what you want to achieve. (I didnt test this, but usually these prompts dont give desired results from my experience)

@sturdy mural

basically those prompts are memes at this point

[sturdy mural]

portrait of one woman holding her baby in her arms posing for the camera (Baby: Black hair, red eyes),
(woman: Karin Uzumaki red hair, glasses, long hair, red eyes,glasses karin, red lipstick, earrings, queen, , uchiha symbol on her dress , very long hair, glitter makeup, eyeshadow, glitter,blush shy, round cute face, cute, blush, (violet sleveless jacket, arm gloves, overknee socks, heels, forest queen) big moon, very romantic, cherry blossom smile, happy smile, waiting for kiss, seductive eyes, accessciors, soft and warm color, movie still,bright, vivid,dreamy, evocative use of light and shadow, incredible beauty, anime style, vibrant, studio anime, beautiful, textural, ray tracing, absurdres, offical art, illustration, hayao miyazaki style, perspective, veil, moonlight playful, holding her baby boy with black hair and black eyes in her arms,
lora:Karin:1

i adjusted them a bit , more like this?

the problem is the baby doesn't looks like a baby at all 😅

[shut nexus]

I'm no pro, im actually new here but that is what I've been told. Just keep experimenting, try different prompt orders, (re-arrange the prompts, you'll get different results) I HIGHLY suggest trying the XYZ plot script and learning how to use it.

@sturdy mural

[sturdy mural]

thank you so much for your advice. I'll try

[shut nexus]

Hey something else to try is changing the resolution, it seems to make a big impact in my testing. you can always upscale later.

[sturdy mural]

Thanks for that advice, I have upscale 2 and R-ESRGAN 4x+ , what resolution did you used?

[shut nexus]

I haven't tried your prompts and no idea what your settings are so Idk, it depends on what your model is, lora, etc.

this is def not recommend but I like to make wallpapers for my desktop.
I use
Width: 1024
Height: 576
Hires. Fix: 2.5x (this is 1440p, 2560x1440)

[sturdy mural]

i use the same width and height for hires, thank you for the info i'll try it out

[hollow tapir]

try using regional prompting to seperate the two subjects (woman and baby)

[tired vigil]

Hello! I am new to this, I just set it up on my PC and downloaded some basic models, is there any guide where I can read about prompt usage?

[fiery pecan]

what kind of image are you looking to render?