📝｜prompting-help

[slow mist]

but no hands

[quiet raft]

Hey, I am using lorebar SD .. and how can I get back a side bar? I pressed the icon on the image and it went away. How do I get it back?

[slow mist]

this is the right hand

CS1o will kill me for this 🤣, and say that is not meant to be used like that
but try with dw_openose_full

it worked nicely for me with SDXL model

[near tinsel]

Oh sweet, thank you!

[quiet raft]

How do I make them look away?

I try
looking away, looking forward, looking away from camera

[rich crow]

looking to the side, looking away from viewer
both have worked great for me in the past

[woven orbit]

So basically, my boss wants me to take this project we did in 2010 ish and magically create good looking lakefront residency pictures from it. I've told him this wouldn't work but he is set on doing this so here we go.

One of my biggest problems are:

1. The AI sees what is supposed to be land as water, creating lakes where there are not supposed to be any.

2. The generated houses generally suck. My boss wants them to look all "light" and "friendly" and not as "brutal" as what the AI generates from the blocky geometry in my control net

I'm using SD 1.5 models, I've tried RealisticVision and ArchitectureExterior 4.0

I'm using controlnet, 0,5 -0,75 strength, I've included images of what I use for controlnet input.

Please, if anyone can help, I'd much appreciate it!

[slow mist]

from behind?

[analog sphinx]

https://i.imgur.com/RTlB8zZ.png i mostly use these 2 which i believe are XL

[quiet raft]

Yeah

[slow mist]

have you tried literally that as a prompt?
"from behind"

[quiet raft]

Well the character is facing the camera with the back..

But I want the character to not look at the camera

[slow mist]

so, you want front image, but you want your subject to look at some other direction?

[quiet raft]

I want view from behind and the subject to look away

[slow mist]

"from behind, looking away" should work in that case

be sure to use -1 as a seed, since some specific seeds just does not work with some prompts

[analog sphinx]

what prompt would i use if i wanna keep original image just change colors? im trying to change a hair color to match the rest but for some reason it keeps spawning a face lol\

[silver valley]

You would need to inpaint the hair

[analog sphinx]

so just try to redo it as close as possible?

[spark wyvern]

Hi!!! Does anyone know how to use ipadapter in comfyUI on AMD cards?

[silver valley]

What tool are you using to generate images?

[analog sphinx]

i use SD with FooocusUI

[quiet raft]

Hey, I've got a image.. I'm kinda happy with the character and wanna add a background to it.

I got the image in IMG2IMG

Should I switch to sketch and sketch all around and enter the prompts?

[analog sphinx]

is there a way to when im done with making adjustments to let the AI run the picture once to blend everything together? it seems like some things look slightly out of place or the focus is off

[restive socket]

come somebody help? I still couldn't get to make it work. ://

[silver valley]

can yous show an example?

did you followed the specific install guide on their github too?

[restive socket]

visual studio is installed with python and c++ and other features

i installed it perfectly and I see the extension

I see no enable face swapping option as well

[silver valley]

you dont need to, just load the image in there and then it should work

you can also try activate the gender detecttion

[restive socket]

yes I think they removed that option

I enabled it and still didn't do anything

[silver valley]

what does the cmd shows ?

[restive socket]

I'll generate an image like that now

[silver valley]

when doing the generation

[restive socket]

okay

lemme generate one and send the cmd to you

[near tinsel]

How would I create something similar to this?

[lavish cairn]

Does anyone know how I can get this effect is SD,

[restive socket]

hello, it worked when I decreased the CodeFormer weight to 0

[restive socket]

but the face I uploaded has some freckles on her face, it swapped the face but it's like she has makeup on, her skin looks smooth, its like it only swapped the face shape.

[remote badger]

This method worked well for me. Not perfect hands but pretty good results. This worked even for small hands.

This is my final product, been working on this one for a while.

[tall prism]

Is there a way to specify the gender for the character it is supposed to generate? Since, i do specify that it should be a Boy and give it strength like this ({gender}: 1.5)

[slow cedar]

I would love if someone could help me with setting up a prompt for a image that's in de dark, no lighting, maybe dim one but mostly an obscure ambience

I'm tryng to make abandoned bulding interiors and well, the AI is popping lights everywhere even if its on the negative prompt :(

[slow mist]

(1boy:1.5) - for male

(1girl:1.5) - for female

but, try without weight first

like
1boy
1girl

you can also try to put undesired in negative

for example

1boy

np: girl

1girl

np: boy

you can also try 2girls or 2boys to get 2 of them

[tall prism]

so 1boy and 1girl are the keywords for it? Thank youu! will try.

And I don't think using them in the NEgative prompt will be good for me since sometimes, there will be another female character in the generation. I am not in control of the prompts as I made a system that generates the prompts and stuff. But I will surely try replacing boy and girl with 1boy and 1girl! t

[slow mist]

you can make a list of pairs like
if positive is boy, add grill to negative and vice versa

it is very easy to implement

[tall prism]

Will do that

[slow mist]

    prompt_pairs = {'1boy': 'girl', '1girl': 'boy'}
    negative_prompt = 'ugly, blurry, low quality'
    postive_prompt = '1girl, dark hair, blue eyes'
    for pp in prompt_pairs:
        if pp in postive_prompt:
            negative_prompt = f"{prompt_pairs[pp]}, {negative_prompt}"
    print(postive_prompt)
    print(negative_prompt)

something like this

[silver valley]

Man, boy, male, woman, girl, female,
1girl and 1boy for Anime based models

[slow mist]

1girl and 1boy works fine with some realistic models

[tall prism]

I am using a Cartoon Model called "Children's Stories"

I use boy and it works 70-80% of the time. The remaining times, it generates a girl

[glacial stirrup]

When ephasising certain parts / words in a prom, can I use multiple words like so (orange striped ball:1.5) and it will read the whole thing inside the ( ) or do I need to use it with single words only?

[slow mist]

you can, exactly like that

[glacial stirrup]

perfect, thanks!

[slow mist]

just do not go too high

1.5 should be more than enough

[glacial stirrup]

I'll experiment with it, cheers!

[spark wyvern]

Hey guys one question. Last night I tried installinbg faceswaplab but it threw out a bunch of errors saying that pip could not manage dependencies for the entire project and there were conflicts between the requirements for mediapipe and ultralytics and the requirements for faceswaplab

I am unsure of what to install then since I have a couple of characters of who I want to produce more images, like in different poses to create a story with them, should I install reactor then, or will this also comflict with other installations like controlnet?

[soft zenith]

is it possible to upscale own Textures?

[silver valley]

Yes

That can be done with different Tools. You also dont need SD for that.
For example try the free and open source tools Upscayl or Chainner on Github

[soft zenith]

oh, thanks

[slow mist]

i would try "tempera paint"
but you can load that image in img2img and click on the interrogate clip button and let SD examine the picture for you and give you a approximation of a prompt

[spark wyvern]

So does anyone knolw a good way to create consistent faces, reactor? roop? inpainting?

[silver valley]

Reactor or IP-Adapter Face

[spark wyvern]

Heyy thank you, ill try reactor if it doesn't kill my install like faceswap did, also @silver valley Would you mind also helping me out with an erorr in #🤝｜tech-support

[static hinge]

How to get urban settings but not the subject standing in the middle of the street?

[restive socket]

guys I use ReActor to swap the face and its working well but it's like only swapping the face shape, the face I uploaded has a natural face with freckles, when its swapped its like she has makeup on, her skin is smooth

how can I make it more accurate

[hollow brook]

guys what checkpoint or lora model should i use to get an rpg game lookibg character and if I like that character can i pose him? like t pose front view side view back view?

[silver valley]

hey, Reactor uses codeformer for face restore that will remove all freckles. I tried it even with codeformer disabled and didnt get any freckles result.
i tried it now with Controlnet IP-Adapter Face Full, and that works

[restive socket]

alright I'll try it with it now, thank you so much

[hollow brook]

can some tell me why do I not have open pose model on my control net models panel. and if I have to dowwnload it, which one is it on civitai?

[silver valley]

here:
https://civitai.com/models/38784?modelVersionId=44811

download it with its config file

[hollow brook]

when i use controlnet, my generated images doent follow the skeleton on my controlnet image. is it because I dont use this model?

@silver valley and can you also tell me is it possible to t pose a character on side, front and back view at the same time? because i will use those images for texture projections

[silver valley]

no, if you use a pose as input image, its already preprocessed, so you have to set the Preprocessor to "None"

[silver valley]

and yes thats possible

[hollow brook]

that is already on none and also model is on none because i dont have any controlnet model, I just downloaded the one u told me

[hollow brook]

how to do that?

and if I make character image without using controlnet, can i pose the character after ?

[silver valley]

then you need to place the controlnet model + its .yaml file inside the models/controlnet folder

[silver valley]

no

[restive socket]

hello, you said IP Adapter Face Full, I picked the IP Adapter option but i dont see the face full option (im a noob im sorry)

[silver valley]

you also need the IP-Adapter models: (the .safetensor ones)
https://huggingface.co/h94/IP-Adapter/tree/main/models

[restive socket]

oh i gotta install them

alright thanks a lot

but what do these models do? and the preprocessor setting

[hollow brook]

@silver valley it gave me this error after using the model i downloaded

RuntimeError: mat1 and mat2 shapes cannot be multiplied (77x2048 and 768x320)

[silver valley]

are you using sdxl as model?

the controlnet models i send you are for 1.5 based models

[hollow brook]

uhm im not SD expert but do you mean checkpoint models?

yes i have SD1.5

but i used juggernaut xl

when generating that image

[silver valley]

yea that wont work with 1.5 controlnet models, you would need the openpose xl model:
https://huggingface.co/lllyasviel/sd_control_collection/tree/main

download thibaud_xl_openpose_256lora.safetensors, at the bottom

[hollow brook]

oh man i dont know how to install things from hugging face .d this is confusing for me

ok lets tell me this

if i want to use anylora checkpoint model and select a lora Or just select a checkpoint other than default stable diffusion 1.5, you mean i cant use the controlnet model i just downloaded?

and if i want to pose character on on the models i mentioned what model should i download from civitai

because civitai downloading is the only thing i understand

[silver valley]

no, the Controlnet models on civitai are for 1.5 based checkpoints (models)

juggernaut xl is not an 1.5 based checkpoint

its sdxl based

so you need to use an other model

[hollow brook]

so you mean we can only pose characters from default stable 1.5?

[silver valley]

you can use anylora model

its 1.5 based

[hollow brook]

oh good

im gonan try it then ill let you know

and recommand me a good lora for character creating pls

[silver valley]

on civitai every model has this section here

where you can check

[hollow brook]

so thats what i should download for controlnet ok i get it

[silver valley]

for making poses, you need the openpose model

download it + the config file below

then put them into models/controlnet folder

[hollow brook]

i downloaded this model inside stable diff
the civit ai browser+ tab

[silver valley]

the models do different things, based on their training. like openpose is for changing poses of characters

[hollow brook]

isnt that automatically putting things in right place?

[silver valley]

it should

but idk if you still need to download config file or it thats also downloaded

[hollow brook]

im gonna download a lora and try anylora with controlnet now
ill text back if there is any problem

[silver valley]

sure 🙂

[hollow brook]

man

i just downloaded a lora

but it doesnt appear on the lora panel

why would that be?

[silver valley]

did you resfrehed it?

[hollow brook]

yes

[silver valley]

was it an sdxl lora? xD

or do you have an sdxl model still selectet?

[hollow brook]

type is lora

base model says sd15

[silver valley]

change the checkpoint to anylora

[hollow brook]

already did

[silver valley]

hmm

you can check if its in the models/lora folder

[hollow brook]

its not even in the lora folder

anyway

ill try another lora

https://civitai.com/models/162141/ahri-18-skins-or-character-lora

it was this lora

man its impossible to find a model

everything is just anime girl

[silver valley]

what do you need?

any specific theme?

[hollow brook]

the theme i want is

to create dark fantasy characters

or just normal

like knight

assasin

[silver valley]

maybe that model is good:
https://civitai.com/models/1116/rpg

[hollow brook]

but this is checkpoint
can be used with controlnet?

[silver valley]

yes every checkpoint/model can be used with controlnet

loras also work

[hollow brook]

oh good

[silver valley]

just make sure that they are all on the same base like 1.5

[hollow brook]

ok i will download that

@silver valley i might need help on another thing. where can i have a picture for controlnet front view t pose side view t pose and back view t pose?

btw is it possible to make back t pose on controlnet?

[silver valley]

you need the openpose editor

extension

[hollow brook]

I already have it but it looks impossible to make it side view there

[silver valley]

you can use that for a character style

with front, side and back view

[hollow brook]

it will work but is there a version where arms up?

[silver valley]

i dont have one

[hollow brook]

oh that could be great bcs im gonna model 3d characters and use SD for textures

anyway itll work too ty

@silver valley why do i get images looking this bad?

[silver valley]

because your resolution is 512x512

[hollow brook]

why do i not get realistic, good looking images

oh

makes sense

.d

@silver valley but this is what i get from 1080x1080 image

[silver valley]

The input image isn't a sqaure

So try 512x1024

Then set preprocessor to None

And model to openpose

[hollow brook]

controlnet11Models_tileE [e47b23a8]

this is the name of model

openpose doest appear

[hollow brook]

xd bro whats wrong with my stable diff

[silver valley]

You downloaded the wrong model

[hollow brook]

really?

[silver valley]

Yes

[hollow brook]

u sent a link i downloaded that

[silver valley]

There are multiple on that one link

You have to select the right one at the top

Openpose

[hollow brook]

oh

ima check it now

İ guess u mean this by open pose and I already selected it

You can see on the top left @silver valley

[silver valley]

Check the models/controlnet folder

[silver valley]

Thats the tile one

Not openpose

So you have to manually download it

[hollow brook]

oh ok

how to do manually?

[silver valley]

Big blue download button + the config file below

[hollow brook]

i know that .d i mean do i simply put those 2 files here?

[silver valley]

Then you may need to restart the webui to see them

[hollow brook]

ok

i also download config and put there right?

[silver valley]

Yes

Best would be to do that for the other models too. Then you can use them all

Canny, sketch, depth, normal, lineart, mlsd are really good ones

[hollow brook]

i currently dont know how to use those didnt watch a guide video but will look up to that

now time to try it

this looking fine but front and back views are not straight sadly

is there a way you know to fix it? cuz i need straight views @silver valley

[silver valley]

You can give the controlnet more weight

[hollow brook]

set the weight to 2 the max i can

now waiting for result

still not fully straight

there are 2 options

my prompt is more important

and my controlnet is more important

i guess i go with controlnet

back view is great but front still not what i expected

and i dont need 2 side views bcuz im gonna mirror when modeling. is there a way to adjust controlnet pic to raise hands up? so i can also get texture of side of chest

[silver valley]

Nope because thats just an png

[hollow brook]

ill try it on posemyart

no man. when i raise arms, it always giving error anyway good night tysm.

[silver valley]

No problem, you can also install openpose 3d editor or posex extension

[hollow brook]

I already made the pose but somehow arms just kinda exploding when raising up. Weird.

[silver valley]

Strange

I can test tomorrow again

[hollow brook]

Itll be helpfull

I just need a way to generate a character which doednt have unvisible part and always turned straight to the camera so I can get the textures

If you find a way to do it pls let me know

[silver valley]

Okay I think thats possible

Gn

[restive socket]

hello I swapped the face but the nose seems to look weird

the tip of the nose didnt really satisfy my expectations, and a few big moles seem to be too visible, its not even in that spot in my source face that I swapped with

which settings should I play around with here?

[tired vigil]

Hey guys, I need a little help here, my controlnet isn't recognizing the pose. Does anybody know how to solve this ?

[tall prism]

I think you need to feed it an image, not a pose.

There is another option to feed a pose directly

I mean, you need to set Preprocessor to none and feed that Pose Image iirc.

[tall prism]

Just use inpainting to fix them no?

[restive socket]

Yeah makes sense lol

[silver valley]

You should play around with the Ending Controlnet Step

For example set it to 0.8

[restive socket]

Will do! Thanks

[hollow brook]

@silver valley hey man

I just tried posing a character on blender

with every pose i need and took normal map for conrolnet but it doesnt to whats in the controlnet picture

is it possible to get better result on reposer on comfyui?

[silver valley]

Do you have the normal map controlnet model downloaded?

[hollow brook]

yes i do

[silver valley]

And its config?

[hollow brook]

I put 5 different poses but it generated only 3 and all wrong

[hollow brook]

yes

[silver valley]

Can you show a screenshot?

[hollow brook]

ok

and let me also show cotrolnet

@silver valley

[silver valley]

Controlnet is not enabled

[hollow brook]

.d

how didnt i notice it

[hollow brook]

the generated image is so big

doesnt fi the sscreen

whats the fix @silver valley

[silver valley]

What do you mean?

[hollow brook]

The image i generated is bigger than what it shows so its cropped on sides

so its a half image

[hoary jewel]

how do I do regional prompting in swarmui?

[mossy owl]

Hi guys, I'm trying to improve the quality of my images, but no matter what I do, it never looks like the images I see on civitai or elsewhere, could anyone point out what I'm doing wrong?

Now, I'm trying to replicate this image, but even using the same model and the same inputs, the quality of the image I generate is low

[silver valley]

Yea you would need to adjust the resolution that it matches. Or use a smaller input image

[silver valley]

Hey, your not using the same settings as the image creator.
There are hidden settings that you only get when clicking the "copy prompt" button on Civitai.
Then paste it into the positive prompt box and press the white/blue arrow on the right to auto apply all settings.

Next thing is your missing the right VAE file for color correction

To get the VAE dropdown go into Settings > User Interface > Quicksettings, and there add sd_vae
Then hit apply and restart.

[mossy owl]

Ok, thank you!!!

[silver valley]

Then download the kl-f8-anime2 vae and put it into the models/VAE folder. Then select it In the dropdown

No problem 🙂

[quiet raft]

Hmm...

So I've got a image.. and I wanna try to do as close to that image as possible but with the model I am using and style ..

What controlnet type is the best for this and what settings?

[gray dawn]

is there a prompt cheat sheet anywhere? just basic stuff

[silver valley]

IP-Adapter

[silver valley]

for ngeative prompt to increase quality:
#📝｜prompting-help message

[gray dawn]

Thanks!

prompts are vastly different than midjourney and dalle. trying to get a handle on it

[quiet raft]

NVM

[spark wyvern]

Guys I need a little help on a issue sort of that I am having

I am using regional prompting as shown in the image,png file to generate a background and a character in the middle of the image BUT

The character is appearing to the side of the image, could that be an issue with the model since AW painting managed to produce the character centered?

[dull portal]

Would any of this cause me to only get ONE person even with LoRA helpers meant to generate groups? RAW photo, award winning photo, (masterpiece:1.3), best quality, ultra detailed, intricate, detailed face, professional photography, professional lighting, HDR, High Dynamic Range, photo realistic, ultra realistic, photorealistic, 8k uhd

[sterile lark]

Is it possible to input a photo into Stable Diffusion and then input a reference photo and have it color-grade the initial photo based on how the reference photo looks? If so, how is that done?

[tender flame]

May current prompt is this: "Generate a visually compelling PowerPoint background that captures the geopolitical tension between the United States and the Soviet Union during the Cold War. Split the image into two distinct parts. On the left side, include images of key American leaders - Franklin D. Roosevelt, Harry S. Truman, and John F. Kennedy, positioned against a backdrop of the USA flag. On the right side, feature an image of Joseph Stalin, the leader of the Soviet Union, set against the USSR flag.

Specifications:

Composition:
    Left side: Arrange the images of Roosevelt, Truman, and Kennedy in a way that they face towards the right side.
    Right side: Position Stalin's image facing towards the left side.
    Ensure a clear and noticeable separation between the two sides.

Flags:
    Display the USA flag as the background for the left side.
    Display the USSR flag as the background for the right side.
    Make sure the flags are prominent and recognizable.

Style:
    Maintain a visually balanced and symmetrical composition.
    Enhance the contrast between the two sides to emphasize the ideological opposition." but it doesn't work right. can i have some help ?

[quiet raft]

Is it possible to have different Interrogate Clip models so that you can change between what kind of outputs you get from an image?

[silver valley]

Yes, there is an extension that includes multiple clip models

[quiet raft]

Well, I feel like it's kinda weird. Was more thinking if I could add some something so that I can use the one in Img2Img

[silver valley]

Nope, the on in img2img can't be changed easily. Maybe with coding

[quiet raft]

Ah okay

[silver valley]

Here is the extension;
https://github.com/pharmapsychotic/clip-interrogator-ext

[quiet raft]

I got that one, but it keep messing up the prompt

adding in instagram model, names etc 😄

[marble gulch]

I want to make a guardian angel with a sword anime style

[summer moon]

Hey ! I wanted to know if that's possible to have a character sitting on the lap of the viewer/camera holder, bit like a father & son moment ?

So far i've never been able to

[wooden ore]

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

[quiet raft]

I need help in Img2Img .. I am trying to fix some hands in the image 2 image tab and the images are 1024x1536 .. I use the inpaint and mark the area and take like 10 images so that I can get a variaty.. But it takes like 4 min for 2 images or 4 ? Dunnob ut atm.. I am doing 10 and it takes 30+ min ..

How can I speed it up ?

[quiet raft]

This is the image I am trying to edit

WHy does it do this?

Or is due to denoising str 0.9 ?

[analog sphinx]

why is it so hard to let the person in the picture NOT look at the camera

no matter what prompt i use they keep looking at me

[silver valley]

set the inpaint are to "only masked" then set the resolution to 512x512, it will be much faster, and lower the denois to 0.5 or below

[sterile lark]

Is it possible to input a photo into Stable Diffusion and then input a reference photo and have it color-grade the initial photo based on how the reference photo looks? If so, how is that done? Possibly with ControlNet?

[silver valley]

yes thats possible with t2i-style-adapter for controlnet:
https://www.youtube.com/watch?v=PbDdtPTYm_4

[quiet raft]

I am trying to change the colour of some pants from Blue to pink

I have outlined the area with inpaint and use the prompt "Pink shorts"

But it will barely do any changes to it :/

I am trying to look at a video that explains this but for him everything changes.. Like in the area that he tells

[silver valley]

which video?

[quiet raft]

Been looking at 3 different videos

Currently watching a video by Sebastian Kamph

[silver valley]

ah okay

[quiet raft]

I wanna learn it more as it seems to great and I am not good with photoshop or tablet painting etc.

So if I can get good at it to be able to do commision and make sure the hands get better etc 🙂

[sterile lark]

Thanks

[quiet raft]

This is what I have painted in ...

My settings

And NOTHING is changed on the face??!

[silver valley]

whats the resolution

[quiet raft]

512x512

It does some "small" change here and there.. Like, not any good changes and small "patches" .. But nothing on the face changes.

Okay, it seems like SD-Upscaler did something

I turned it off and not it worked

[silver valley]

yea you need to disable upscalers for inpainting

[quiet raft]

Hmm okay

What prompt can I use to delete stuff?

[silver valley]

inpaint cant delete stuff, it will always change something based on what is already there

[wary sentinel]

hi guys! I am trying to use my trained lora to gerenate my dog. I used online trainer (colaboratory, 22 pictures of him, 2000 steps something). I think the lora works and it captures the details of my pug. HOWEVER, the prompt I made could not make him to be different texture or style.

I used revAnimate to train my lora and this is the same checkpoint for these two. The prompt is the same.

First:
((Elegant)), ((Exquisite)), ((4K)), ((Vibrant)), ((Stunning)), ((Astounding)), animal, Teeth, Apex Predator, Shark, Watercolor creation, Aquarelle painting, Wash painting, Dynamic blur, Blurry motion, Speed blur, Extreme reach, Super telephoto lens, Distant subjects
Negative prompt: (worst quality:1.2), (low quality:1.2), (lowres:1.1), (monochrome:1.1), (greyscale), multiple views, comic, sketch, (((bad anatomy))), (((deformed))), (((disfigured))), watermark, multiple_views, mutation hands, mutation fingers, extra fingers, missing fingers, watermark
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3797611897, Size: 512x512, Model hash: 4199bcdd14, Model: revAnimated_v122EOL, Version: v1.7.0

Second:
((Elegant)), ((Exquisite)), ((4K)), ((Vibrant)), ((Stunning)), ((Astounding)), animal, pug lora:Laklak_Lora:0.2, Watercolor creation, Aquarelle painting, Wash painting, Dynamic blur, Blurry motion, Speed blur, Extreme reach, Super telephoto lens, Distant subjects

i did realize that I can lower the weight to make my pug less photo-ish, but I already lower it to 0.2 and it still looks so real (it is good in some sense but not good for this prompt)

[quiet raft]

Anyone here who knows how to take 1 image and make it into pixels? 😄

[gray dawn]

anyone have luck using prompt generators? So far it seems like they make things worse for me. not sure sure if wrong ones or if im using improprely...i would guess the latter is true

[proven jasper]

what is a prompt generator

[quiet raft]

What is the issue?

[quiet raft]

You want different text?

[proven jasper]

better to just type your prompt,that word salad of a prompt looks like a mess, less words = better

[gray dawn]

i just want a prompt that will work that first prompt yields this result:

it endes up with a dude sitting with a laptop.

[quiet raft]

Use the prompter instead

Go to Img2Img and put in a image and press the Interrogate Clip

The Prompt Generator is useless

[gray dawn]

thats what im asking, it doesnt seem to be worthwhile so before i abandon i was getting confirmation on whether it was garbage or if i was using it improperly

[proven jasper]

yea its not good

[gray dawn]

so i can take a similar image from google images for example, go to txt 2 img and interrogate and it will build a prompt?

[quiet raft]

Nope

Img2Img

[gray dawn]

yep sorry just noticed that typo.. ok ill try that thnak you

wow it actually handles webp files

im not seeing interrogate clip is that an extension i need to add?

found it

this process is not fast right? maybe 1-2 minutes in so far

[quiet raft]

If you have not used it before, it will probably download it first time

[gray dawn]

oh ok cool.

this has been one of the most challenging things ive tackled. I self taught myself HTML/css coding back in like 2003, solidworks, all graphics editors, after effects video editing, python scirpting, but all that is like this is the correct way to do it, there is a definiteive language, a definitiive process...this is a totally different can of worms there are endless ways to do things, and so many different processes it has me totally flummexed...i mean heal i binged a 16 hour course on solidworks and was quite proficient..i bet im 30+ into this and i feel like i havent even got the basics down

ok so i have went to a prompt website, prompt hero and reviewed some images that were lit and had the detail i want to achieve and i kept the prompt simple utilizing their positive and negatives and it yields nothing remotely close. is this a settings issue or am i missing something

[silver valley]

Hey, sdxl models are trained on 1024x1024 resolution. So dont use them with 512x512

[gray dawn]

oh really how has none of the 74 damn tutorials ive seen mention this LOL

i sent you a dm also, just a thought, didnt want to share it public though

[copper ginkgo]

can someone tell me how i can use openpose controllnet with xyz plot to make pultipled poses?

[eager jolt]

hi, help pelase with wildcards, when i put this for example, it only takes 1 entry of the list and repeats over and over. why?
https://media.discordapp.net/attachments/1031106064776712226/1196111677142347846/image.png?ex=65b670ce&is=65a3fbce&hm=1405ae3ed58d7e3bff177c2059f6d833a443d02268619b921992e1a98898d140&=&format=webp&quality=lossless
it only takes this one
https://media.discordapp.net/attachments/1031106064776712226/1196111857883304017/image.png?ex=65b670f9&is=65a3fbf9&hm=8b5af74d2a5773ce9ad95740b57c8ddab62126aba41a332959e5e852fe4047ea&=&format=webp&quality=lossless

[hollow brook]

@slow mist my lifesaver I need help on something

[mild gorge]

hey guys any idea why she's not "shocked"? any idea how to make her have a shocked expression on her face?

[slow mist]

https://civitai.com/search/models?sortBy=models_v5&query=facial expression

take your pick 🙂

[slow mist]

go ahead

[mild gorge]

Thank you!

[rocky ingot]

When wording a prompt by separating desired elements with commas, do the order of the elements matter, and if so in what way? Say I have a prompt for spaceship, 3d render, star would switching the order around affect the results in a meaningful way?

[hollow brook]

I need to pose a character on every angles. Like side, back, front view. And also I need the unseen parts on those angles like bottom of the arms, bottom of foot. So I can project that texture to my 3d characters and make easy retro style character. But controlnet is not doing all those poses at once. And if I do more the one generaion, then charracter is not being consistent. I need help on this what should I do?

[slow mist]

SDXL or 1.5?

https://civitai.com/models/140117/latent-layer-cameras <- this one is good, but unfortunately, it is for 1.5 only 😦

[serene fractal]

Yes, spaceship, 3d render, star is creating a different vector the model uses, even if you used the same seed with 3d render, spaceship, star. As it is a statistical model though, the term "meaningful" is always a little open to discussion. It can cause large changes, or only very minuscule ones, depending on the seed and checkpoint you use.

[hollow brook]

the setup said 1.5 + sdxl

[hollow brook]

im checking it out now

[slow mist]

what model do you use?

if you use 1.5 based model it would work in most cases

even if model mostly work with it, some specific seeds might not work

[hollow brook]

has to be 1.5

im launching sd now

@slow mist oh thats a checkpoint download and u know bout my inernet .d I will do it later. Saved the model. thx again.

[slow mist]

that is not a model, that is hypernetwork

it goes to this folder

[hollow brook]

yeah just noticed but Itll take some time and I gotta do something now

[hollow brook]

ill download when Im free

but first time

[hollow brook]

seeing hypernetwork

dont know what itis

[slow mist]

it looks like that on my disk

[hollow brook]

ok thx ill put it here

[slow mist]

just put it as i did

or ask if it is ok when you do it

[hollow brook]

ok

[slow mist]

and see the video from start to the end

it will explain how to use it

__np__ and similar things in negative are negative prompts, if you wanter what those are

[hollow brook]

Ill look up to that thx again

[stark shard]

What is the equivalent of styles.csv from a1111 in comfyui?

[slow mist]

that thing can be useful if you extract the values and do not use them like a "click-click-click" engine. If you add a bunch of them with clicks, it can be really messy, so I stopped using it. imagine you want something "realistic anime using a non-anime model, with at least 3 different negatives," and you end up with 7-10 styles selected and your prompt will be full of junk

yes you can specify where the prompt goes, like this:

Style: Anime,"anime artwork {prompt} . anime style, key visual, vibrant, studio anime, highly detailed"

but that is not enough for me

for example, if i want to add
Style: Realistic Anime i need to be sure which one i clicked first and which one i clicked second to get what i want, since there is no way to reference another style like prompt

[slow mist]

but i guess you want this https://github.com/theUpsider/ComfyUI-Styles_CSV_Loader

[stark shard]

Can you give me example workflow/guide with styles? I've seen nodes with it but didn't know how to use them

I get your point and i agree but im looking for solution that takes a burden of writing extensive style keywords each time

[slow mist]

i guess you need to connect this one with prompt

[slow mist]

i am using a little "dirty" solution, with empowers the dynamic prompt extension

[stark shard]

Which one? I tried gpt prompt expansion but it is heavily biased towards digital art

[slow mist]

this is just something i am testing right now

so, when i need my default netative, i just add

__vkg/np/default__

__vkg/np/default__, __vkg/np/nsfw, __vkg/np/n75__

i guess you understand the point

it is similar to what you are using, but with much more control

[stark shard]

Where do i put this ;d

[slow mist]

https://ericri.medium.com/stable-diffusion-sdxl-wildcards-and-comfyui-e8483e00e1c

i guess you can add as many wildcard files as you want as in a1111

but i haven't try

and this is the dirty part i was talking about

__vkg/xl/w__0.8>

will became

<lora:SDXLrender_v2.0:0.8>

[stark shard]

Thank you so much, it's what i was looking for tbh

[quiet raft]

What prompts can I use to avoid sleeves?

I have added without sleeves

In negative I added, Sleeves, gloves and long sleeves.

[gray dawn]

i am trying to achieve some images like these

70's style vintage photo of young skater mid-action, performing a Frontside Air trick within a large skate bowl that is actually an empty swimming pool in the backyard of a southern california home. The skater is dressed in a short-sleeved black t shirt, shorts, and tall athletic socks. A few other skaters should be seen observing from the pool's edge, all under a bright and sunny sky, old photo of object, 70s, scratches, vintage, colorized photo, grain photo, faded colors, muted colors, Kodachrome

worst quality, normal quality, low quality, low res, blurry, text, watermark, logo, banner, extra digits, cropped, jpeg artifacts, signature, username, error, sketch ,duplicate, ugly, monochrome, horror, geometry, mutation, disgusting, bad anatomy, bad hands, three hands, three legs, bad arms, missing legs, missing arms, poorly drawn face, bad face, fused face, cloned face, worst face, three crus, extra crus, fused crus, worst feet, three feet, fused feet, fused thigh, three thigh, fused thigh, extra thigh, worst thigh, missing fingers, extra fingers, ugly fingers, long fingers, horn, extra eyes, huge eyes, 2girl, amputation, disconnected limbs, cartoon, cg, 3d, unreal, animate

i am getting nothing remotely close

this is what this is producing

[frozen sonnet]

Dreaming for @nova holly... /dream prompt:\\\" a yong monkey angry face talking with a white rabbit , both are talking with each other in the jungle\"/ style:Photographic width:1344 height:768 aspect:16:9 seed:3489038418

[tired vigil]

is (worst quality, low quality, extra digits:1.4) basically each tag within the parenthesis with a 1.4x multiplier? or is the multiplier only affecting the extra digits tag?

[quiet raft]

Hey, anyone seen this before?

You see at the eyes etc?

[silver valley]

seems like you used facerestore

[quiet raft]

I don't even got that as an option?

Forget about the full settings

The images above

Was with Hires OFF and Sampling 45 steps

Other than that, it's the same stuff..

[silver valley]

facerestore can be hiddenly activatet when you loaded in another image with png info

[quiet raft]

How can I find it?

Or How can I turn it off?

[silver valley]

its in the settings under facerestore

[quiet raft]

I need to install it then..

Cause I don't have that option

[silver valley]

your sure?

[quiet raft]

I don't see the face restore option?

Under settings there is this

[silver valley]

in auto1111 its called Face Restoration inside the Settingsd menu

[quiet raft]

Yeah I just looked, but I can't turn it off?

[silver valley]

thats how it should look like when its off

[quiet raft]

So GFPGAN = off?

[silver valley]

no

just the checkbox not checked is off

do you use sd.next or Lobe theme?

[quiet raft]

I can't check it off? I can only choose Codeformer or GFPGAN

ASGasgfA
OHH

Dafuq

I've had it on all the time :/

But I don't know if this will be better or not :/

What is your opinion?

[silver valley]

face restore makes every non realistic faces bad and deformed

so keep it off

when you want good realistic faces,upscaling will do it better

or use the Extras tab, there you can use Facerestore

[quiet raft]

In extra tab it's the GFPGAN or CodeFormer then? 🙂

[silver valley]

for realistic images codeformer

[quiet raft]

Cool 🙂 Thanks, I've learned something new 😄

I rarely do any realistic models 🙂

[silver valley]

np, then your in no need to use any face restoration

for animated stuff use hires fix or sd upscale script

or inpainting

[quiet raft]

I found inpainting this saturday 🙂

Really cool

[silver valley]

yea its really nice, face inpainting is done in seconds

[simple finch]

How would to prompt this?

any ideas?

[stark shard]

Some kinda of modern minimalistic hotel hallway somewhere (dont have any idea where it could be)

[simple finch]

Ok cool, I'll try that

[mild gorge]

hey any idea why her face is looking like that? 😭 im using juggernaut with this prompt

and 35 steps and 1024x1024

[restive socket]

guys how do these AI influencers have consistent bodies? like the body shape and details are exactly the same in every picture.

[elfin star]

AI influencers?

[restive socket]

ai instagram models

[restive socket]

Do I need to train a lora for it

[stark shard]

Quick question - started using comfyui and i wanted to latent upscale image before running it through ultimateSD, but it comes out really... messed, what can i tune to get better results? Same model and prompts as base gen //edit denoising is the way xD

[next linden]

Questions for those familiar with training models/ photorealism....

I’m working on a project of taking a single reference image, turning it to a 3d model and making a LORA out of it. The reason I’m going this long route is to have a fully photorealistic consistent character able to express convincing emotion and look consistent at all angles.

I need help coming up with specifics to convey to a 3d modeler on what images I need so that they can be used for LORA training.
Aside from needing an idea of basics such as specifications on facial expressions, angles of face, lighting scenarios, resolution/aspect ratio...

I’m unsure how “realistic” the 3d model for training needs to look. My current workflow idea is:

1. Get the output looking like the image below (no skin textures, only the head, should I include hair? ) with a focus on only facial expressions and capturing different angles
2. Import that LORA into stable diffusion to apply skin/photorealism to the model
   3)Train a new LORA on the photorealistic outputs of stable diffusion.

[slow mist]

Can someone give me a link to any FULL 1.5 non-pruned model (more than 6GB) on Civitai or Hugging Face? i need it to test something

[stark shard]

What CFG scale should I set in ultimateSDUpscaler in comfyui so my images won't come out like that?

[slow mist]

i would lower denoise

[stark shard]

was 0.4, too much?

[slow mist]

it should be fine

[runic jackal]

does anyone know how to get good looking text?

[prime ingot]

How do we write down a correct prompt? haven't used this for a while

on discord

[silver valley]

Lower the denois to 0.3 or lower

[stark shard]

I'm trying something like this: dreamshaper turbo -> generates base -> latent upscale x2 -> dreamshaperXL with LCM lora and LCM sampler and it tbh breaks image, am i trying something stupid/something i shouldn't do or is it a problem of parameters of samplers?

workflow

[silver valley]

I think you dont need the LCM sampler if you already use the lora

[stark shard]

I though you need both so it works as intended o.O

[silver valley]

Nope

You could also use a LCM merged model. With that you won't need a lora or the sampler

[stark shard]

Doesn't look like sampler

[silver valley]

Is that with the LCM Lora?

[stark shard]

Yep

[silver valley]

Try set LCM lora strength to 0.5

[stark shard]

Will do, now trying base model without lcm, maybe it doesn't like images from turbo for some reason

Okay, base model works, ill try now lcm strengh

Not bad! It was lora strenght, kinda wondering what to do about smoothing everything but thats for tommorow (today later)

[silver valley]

Upscaling would be the last step then maybe

[stark shard]

what model would you recommend?

[silver valley]

4x-UltraSharp or NMKD-Siax-200
Or Resrgan4x+

With a denois of 0.35 up to 0.5

[stark shard]

Okay, will see how long it takes. Btw can I use lcm with USDupscaler or shouldn't? I have 1060 6gb so upscaling 4 times from 512 takes a long time for me ;d

[silver valley]

You could try upscale by 2 first if 4 takes to long

[stark shard]

512 -> latent x2 -> ultimate x2 - so 4x ;d

[silver valley]

I didn't tested LCM lora with upscaling

[stark shard]

hmmm, denoising 3.5, steps 20 euler cfg6, ultrasharp4x

[dull portal]

Guys, am I wasting my time with some of my prompting/settings? I like my output on the end, but a lot of the details in the background changed, not that the first Kermit looked great though. LOL, 2x 'Upscale Latent By' on the second KSampler. I was trying Highres Fix script alone with the first KSampler before...but I don't know, it seemed to have stopped working and lagging my system to hell, but I'm just on CPU mode. Highres gens with just 1x KSampler was taking maybe 20 minutes total, now I'm like 1 or 1-1/2 hours without it, but I was getting the same boxes Holowed was getti...wait, was it the LCM that was causing it?

[steady spire]

How my images look better?

I want then line this

[upper flame]

https://youtu.be/xvFZjo5PgG0?si=cS2waMuUnu2s1lgo

https://youtu.be/xvFZjo5PgG0?si=cS2waMuUnu2s1lgo

[stark shard]

Getting back to it. I managed to get nice results with latent upscale from dreamshaperXL model with dpmpp_sde, but ultimate still breaks the image

[silver valley]

You need to lower the denois

To like 0.2

[restive socket]

guys which lora should I use to have more specific backgrounds? like if I want the background to be a snowy neighborhood in canada or something

[stark shard]

better, but not there yet , 0.1? sounds like nothing will change

[silver valley]

you can try 0.15

[stark shard]

thought that maybe prompt is a problem but nope

Hmm, i've found node ClipTextEncodeSDXLRefiner maybe the problem is that I use normal ClipTextEncode?

[stark shard]

okay! I got the problem, had tile size 512, increased to 1024 and its better, i dont like hair but will tune that

[silver valley]

ah true, your using sdxl, therefore tile size needs to be higher than 512

[stark shard]

is there any formula how to calc it, base img has 512x768 and up to x4 or just 1024x1024 tiles are ok

[silver valley]

using ultimate upscale script is for tiling, so its based on the model you use. sdxl models are trained on 1024x1024 and 1.5 models on 512x512
so use these resolutions as tile size, or use 768, that should work for both

[stark shard]

Okay, but i shouldn't take notice of aspect of resolutions of images to upscale?

[silver valley]

not, the tile size should be a square

the input image resolution doesnt matter

[prisma grail]

Is there a way to batch prompts? I want to make multiple images of different prompts without needing to wait for the first one to get finished?

[silver valley]

only with that extension:
https://github.com/ArtVentureX/sd-webui-agent-scheduler

[stark shard]

@silver valley do you know if i can somehow in comfyui make variable in file path for save image? Ex. comfyui/{gen_seed}/...

[vital charm]

Hi People,
How would i go about making this creature look to the front and maintain the same style and consistency? I tried using control net but as soon as i enable that everything gets slow and tends to crash Open pose doesn't really work for me. How would you go about this task?

I wanted a front, side and back view.

One i dea i had is to only img to img and get some base pose ragdoll and let it dream on that.
https://gyazo.com/9c83d101a2929b3ebc4889e408493b16

[silver valley]

Oh I dont know

[mild gorge]

hey guys any idea why the girl's body is not fully visible? i tried lots of things and it just doesnt want to show her full body

[hollow brook]

@slow mist downloaded the hypernetwork you said on civitai tab on webui. What is it and how to use?

[silver valley]

try prompt for her shoes

[mild gorge]

wdym? i wrote "red shoes" and it didnt work

[silver valley]

hm do you use txt2img or img2img?

try add:
full body

or cowboyshot

[kindred blaze]

Hi all, I've been making some simplistic cute style robo-characters on a phone app and trying to replicate a similar look with Stable Diffusion, any suggestions of LoRA's or Models please?

[mild gorge]

faceswap so img2img i guess?
this is my current workflow

[slow mist]

if you meant https://civitai.com/models/140117/latent-layer-cameras
then watch the video and see how to use it https://youtu.be/JB-BmFUdT7o
you can try to use it with the model made by the same author https://civitai.com/models/113034?modelVersionId=122075
it probably works the best with this model, but it works with other models too

[hollow brook]

Ill watch it thank you

[stark shard]

It also helps setting aspect ratio so picture has more height than width, symetrical don't usually work for me, ex sd1.5 512x768 works great for full bodies

[mild gorge]

oh okay nice idea!

[stark shard]

and put full body at the beginnig, so it starts from body and later gets to detailes, otherwise it can break depending on the model i would try yours prompt in a form like
full body picture of a standing beautiful girl, cyan hair posing, green eyes, holding an umbrella in a rain, wearing black pants, red shoes

worked for me

[mild gorge]

ohh smart i will try that!

do you have any idea to have more realistic faces? im using realism engine and its rally not realistic

got that with foooocus

[stark shard]

Oh boi, not sure, not an expert, fooocus has XL model, that's one thing, other is that propably has good negative prompt and styles settings

[mild gorge]

oh okay xD thanks for the advice

[stark shard]

you can check in fooocus logs how the final prompt looks like, so u can take keywords that styles implement and use them

another thing, check if fooocus didn't use a refiner on that image

[mild gorge]

oh didnt know that!

[stark shard]

Are you using sd1.5 or xl version of RealismEngine?

[mild gorge]

realism engine sdxl

tbh i just want to generate consistent images of a face

[stark shard]

Key, will download and try to get something like you want

[mild gorge]

well if you have a good model that generates realistic faces i would use that

i just tried realism engine because i heard its good for realism

[stark shard]

Not rly, im playing around more with digit art

[mild gorge]

but my goal here is just to have realistic faces

ohhh

right

[stark shard]

First thing - change sampler to dpmpp_sde karras

Thats what author of the model used and propably works best

You can try lower cfg

[mild gorge]

okay

[stark shard]

Not too sure about ipadapter in comfy as i didnt use it yet, but maybe a little noise would help?

[mild gorge]

i wil ltry with that and show you the results

what would you use if you had to face swap? and have consistent results

[stark shard]

Didn't play with that yet but I heard many mentions of reactor or sth like that

[mild gorge]

yeah i also tried that

the results were bad

i mean its a lot faster or something

i think ip adapter is more advanced

[stark shard]

https://www.youtube.com/@OlivioSarikas this guy makes really good tutorials and there is IPAdapter one, maybe check it out and this one https://openart.ai/workflows/academy

[mild gorge]

okay i ll look into it

[restive socket]

which promts should I use to change the camera angle?

[tall prism]

Hey, I have doubt. I want to train a LoRA model using Kohya SS. I am using Spongebob Comics as the training images. So, I need to caption the images. Any idea on how to approach this?

[stark shard]

https://www.pixcores.com/2023/03/camera-angles-in-stable-diffusion

[restive socket]

Thank you 🤝

also I have another question if you dont mind, do I need a lora to generate this suit? if yes, which one can I use?

[silver valley]

i would prompt for
woman, in office, wearing black suit, open suit, white blouse, cleavage, standing infront of window, skyline background,

[restive socket]

alright thanks a lot

I already inspected it in automatic111 but the results were not accurate at all thank you ill try them now

is there another extension that I can use which does inspecting more accurately?

[silver valley]

there is this extension, that features multiple clip models:
https://github.com/pharmapsychotic/clip-interrogator-ext

[restive socket]

thank you so much!!1

[quiet raft]

Hey, I got a question about the X/Y/ plot thing..

If I wanna test a image with different Sampling Steps

Is this right?

I can't find Sampling in the list

[stark shard]

I think in a1111 it was simply steps, will check that in a sec

Yeah, it's just "steps"

[quiet raft]

Alright, thanks 😄

[slow mist]

what is the difference between this and the default one?

[silver valley]

the default one has only 2 models while the extension features numerous

[agile ferry]

I'm wanting to use historical clothing prompts, is there a particular checkpoint for getting good historical clothing?

[agile ferry]

I'm trying the suggestion given here to interrogate based on pics I want, hopefully I can figure out something that way about finding good historical clothing ideas.

[agile ferry]

There are so many very large models to choose from, advice on how and which to pick?

[silver valley]

not the biggest ones, as they use gpu vram

could freeze or oom

idk if --use-cpu interrogate would help with that extension, but that command can be used for img2img clip interrogate or extras

[agile ferry]

ViT-L-14/openai was one I randomly picked and it worked once, failed 2nd time

I'll try that, I'm not in a rush when doing that sort of thing

[stark shard]

Any good guide on outpainting in comfyui?

[mystic anvil]

I had this idea that i wanna see an ai create but so far none have been able to actually understand what im asking. Can i have some help creating a good prompt for this? Heres my idea. I think something that would be so creepy would be to have this realistic gorgeous looking game but if you strayed too far or lingered in one area for far too long or some other random arbitrary condition youd see this one npc with that ps1 era style blocky body and flat texture face. And they dont do anything creepy. They just exist in a totally different style than the rest and that alone is unsettling.

[agile ferry]

Since you're wanting time to pass, consider comic book style frames showing progression of time? I don't think you can get across what you want in a single image, but who knows.

[mystic anvil]

Did you mean to reply to me?

Im moreso wanting a photo of like a scene from a realistic game. Think spiderman ps5, with npcs in the background but just one of them is low poly. Just standing there. Out of place

[agile ferry]

Yeah, you mentioned straying or staying too long. These require passage of time. But if you just want a photo of that, I would try in-painting rather than a single prompt. Get yourself a good base image, mask off the character you want to make low poly, then replace it as you like with a new prompt including low res, low poly, pixel art details.

[mystic anvil]

Ohhhh good idea!

I forgot inpainting was a thing tbh. Lol thanks

[agile ferry]

hehe, good luck on your vision

[keen panther]

I have a canny image that I want to use in automatic1111. However, the only option avalible is for automatic1111 to generate a canny for me. How can I use my own canny image instead of generating a new one?

[silver valley]

For that you need the controlnet extension and the controlnet models like canny.
Here is a quick guide:
#🏞｜general-with-images message

[restive socket]

which prompt should I use to add a little bit of snow on top of the head?

[hollow brook]

guys all of a sudden my stable diffusion is not launching

why could it be

I have the installer. should I click it to download missing files or anything like that

ok fixed it

[restive socket]

I am generating a woman in a suit, I used the "wearing black suit" prompt but its INSISTING on generating a white suit... can somebody help?

[stark shard]

Do you have white color in your prompt somewhere else? Maybe it's bleeding

[hollow brook]

Downloaded a lora but doest appear on lora section

[restive socket]

yeah i put black suit while blouse

[hollow brook]

how to fix?

[restive socket]

did u refresh it

[hollow brook]

yes

[restive socket]

oh idk then

[hollow brook]

https://civitai.com/models/122181/head-textures-xl

this one

maybe it says base model sdxl 1.0

is that the problem?

[restive socket]

what version are u using

1.5?

i think it could be the problem but im not sure since im a beginner

[hollow brook]

same im begnner too

and yeah

1.5

[restive socket]

it could be the problem since its not compatible but wait until someone competents answer

[hollow brook]

Yeah

[stark shard]

so yeah, its bleeding, try reversing the order, but it propably will make blouse black then, inpainting propably would be the fastest way

[restive socket]

Alright

[stark shard]

How do I outpaint correctly to get lower part of body of this character?

[slow mist]

would you believe me if i said:
"a little bit of snow on top of her head"

[restive socket]

Lmao

[quiet raft]

What would you say is the best Lora's to use to alter the characters? 🙂

[dull portal]

How do you stop too many fingers and toes? I've got it in my negative. 😛 Well, I just switched it from 'too many fingers' to '(too many fingers:1.2)'

[stark shard]

control net or inpaint i think

[keen panther]

I wanted to change the artstyle of a character I made. The character is drawn in lineart with with colours filled in an a separate layer so the colours can be removed and adjusted. The artstyle I want to change the lineart to is to a Chinese inkwash painting. However, the Automatic1111 kept distorting the design of the character as their appearance is unique despite modifying the prompt several times. I tried Stable Diffusion XL Base 1.0, Stable Diffusion v1-5, with no avail. Do you know how I help in this?

To give the details, the face of the character is a blend between human and animal like the Na'vi from Avatar. Models for anthropermorphic animals tries distorts the character to look more animal like while Stable Diffusion models distorts the character to look like a human hobo.

I had also tried using canny controlnet to maintain the details but it does not help with the distortion.

[orchid scaffold]

Whats the best way to sub out the background of an image but keep the subject? some sort of controlnet?

[restive socket]

guys I have a problem, when I use the "suitOL-AIX" Lora for female suits, its working well, when I use IP-Adapter for face swap its also working well. but when I use them together, its giving me a mixed blurry glitched picture. why is this happening?

[silver valley]

I think Controlnet IP-Adapter can help maybe

[prisma grail]

What's a good tag for "zoomed out", zoomed out is not a keyword

[abstract terrace]

I'd like to see that as well

What is a good way to prompt to keep the subject in the center and not off the side?

[wind valve]

describe like "a big city with a girl in the middle" then describe de girl

[wind valve]

a distant view of a girl in a city

i guess

[prisma grail]

distant view?

[wind valve]

a distant vision of

[restive socket]

why am I getting a result like this when I use IP-Adapter for face swap??

without ip adapter everything is fine

can somebody help

[wind valve]

its like this that you want?

or more distant?

[prisma grail]

Further away

like practically in the background

[silver valley]

Can you show your settings?

[restive socket]

hey sorry for the trouble, I set denoising strength in my img2img tab to 0. and it fixed. 😆

but I have another question tho, how can I switch both faces and the hairstyle with IP-Adapter? its changing the face only

result:

how can I swap the hair along with the face?

[silver valley]

Therefore you need to use the IP-Adapter that hasn't the Face included as second Controlnet unit

[restive socket]

okay so I just added another controlnet unit

[hollow plank]

i want it to do a blank, more modern mannequin, but it keeps doing that "old-ish" mannequin style

modern feminine mannequin, light blue dress, white fur trims, body shot, photograph, 4K

changing it to "blank feminine..." or "white feminine..." doesn't seem to change much

i want mannequins like this, not that lol

(that's from wikipedia, not a prompt)

gonna try "mall mannequin"

k, still did a face. gonna try adding "faceless" to the prompt as well

the fact that it's actually white and not Caucasian white is good tho, closer

not sure why it put headless mannequins in the same shot either 🤪

now it's just removed the damn head, i said "faceless" not "headless" lmfao

[silver valley]

xD you can also try different comunity models. these are better trained as the 1.5 base model

[hollow plank]

are there ones that are better for photo shots like this?

[silver valley]

i used Dreamshaper v8 for this:
https://civitai.com/models/4384/dreamshaper

[hollow plank]

nice

will try it

[silver valley]

on civitai.com you find models for many stuff, art, photorealistic, etc

its the biggest model database

[hollow plank]

i'll also probably shop around there as well, see what else ppl have made. it's crazy to me the fact that we can achieve local ai image generation (mine takes 1-2 min per image, but I also don't have a hyper-expensive GPU)

[silver valley]

yea its awesome ^^, whats your gpu btw?

[hollow plank]

Radeon RX 6600 XT, got it a little over a year ago during black friday sales (bought it new, not used)

[silver valley]

ah okay, then i would recommend staying with 1.5 based models

[hollow plank]

ah, sdxl was working fine for me LOL

will remain with 1.5 community models tho, don't wanna fry my gpu 😛

[silver valley]

haha yea 😄 1.5 is also a lot faster

[hollow plank]

using "full shot" with dreamshaper helped a ton, still no head but my main focus was the dress anyway (testing its ability to synthesize clothing)

[silver valley]

ah okay, looks good 🙂

[hollow plank]

how does one go about "fixing" faces?

in-painting only seems to make the face look worse...

i like almost all of this image, save for her face.

from a glance the face looks fine

but looking closer...

found that face fixing option in settings, is there a way to run it on an existing image?

[stark shard]

Inpaint but select to inpaint only masked area (Inpaint area -> only masked) then it inpaints it in higher resolution

[hollow plank]

should I change the prompt to just describe the face or no?

[stark shard]

Yeah https://stable-diffusion-art.com/inpainting_basics/ i recommend this if it's new tool for you

[hollow plank]

will look at that, thanks

[stark shard]

There is also adetailer to help with faces and hands/bodies during generation but i heard its not recommended for SDXL not sure why tho

[hollow plank]

ah, that wasn't an sdxl image

that was another model I found on civitai

using 1.5 as a filter

is inpainting more computationally expensive than generating?

keep getting "out of vram"

not sure why you deleted your message, but I'll try the SD inpainting model

sd inpainting model also immediately runs out of vram

[coarse briar]

hello, beginner here, how i can tell StableDiffusion (for now i use bot on this discord) to change only one part of generated image in last prompt, for example: change hair color, or add item in hands
Or meybe SD is not tool about that?

[hollow plank]

is there a good way to get more "average" looking people in SD's generations? i really like this one but he's a bit too handsome, would rather him be more average lol

(i didn't have any person in mind with this, just had an idea for a cool-looking guy lol)

funny how some of the "in-betweens" that I see as it outpaints slowly are more average-looking than the original picture lmfao

[wind valve]

more ugly, put (ugly guy:0.6), mean 60% force

[hollow plank]

putting things like "perfect" and "model" in the negative prompt also helped

[restive socket]

Hey sorry can you explain it again like ur explaining it to an idiot please

First controlnet unit: I uploaded the face, IP-Adapter, "ip-adapter_clip_sd15" preprocessor and "ip-adapter-full-face_sd15" model

What should the second IP-Adapter controlnet unit be?

[scarlet latch]

anyone know what model that this image used? I tried a few but still looks not realistic

[silver valley]

IP-Adapter-plus_sd15

[restive socket]

Alright 🙏

[stark shard]

Im trying to inpaint an eye color in sdxl model but im getting some mutants, is there different model for inpainting for sdxl same as in 1.5 some models had standalone inpaint models?

[silver valley]

Set the inpaint resolution to 768x768 or 1024x1024 for sdxl

And no there is no base sdxl inpainting model

[stark shard]

I read that it's fairly easy to merge inpaint model - would you recommend it? (Trying with different resolutions as you mentioned atm)

[silver valley]

Yes for 1.5 its easy. It could work the same for sdxl ones but I didn't tried it yet

[gray dawn]

what term is used for a straight on perspective. trying to generate some graphics and it keeps putting them at an angled perspective.

[hollow plank]

Usually I add in "head on photograph" for generating headshots from that kind of perspective and it works well

[gray dawn]

im creating icons, like button icons, wonder if that command would work...ill give it a whirl

[stark shard]

Why does inpainting in fooocus dont work

[silver valley]

whats your GPU?

[stark shard]

1060 6gb, but i changed inpaint to modify content here and it worked

[silver valley]

ah okay

[stark shard]

Still not sure why inpaint didn't work

[keen panther]

Hey guys,
I have made the front view of a character that I want to generate a model sheet of. What is the best and easiest way to generate a character turnaround
of said character?

I did try CharTurner and inpainting, though Automatic1111 seems to ignore the original character and make its own design.

[static hinge]

Haven't tried that specifically, but I would go with ip-adapterfaceid. Insightface is pretty powerful and will stick your character face on some way. Plus openpose or some other control method for the poses.

Also you can try with ip-adapter for the general style, and ip-adapter face and face plus vith

[keen panther]

My concern for using the IP you suggested is that the character is non human in a sence having elements that I am not sure a face transfer can achieve (i.e. horns, rabbit ears). In addition, the art generators really like to change the artstyle I envisioned.

I will give it a go anyways to see if it works. Thankyou for the suggestion.

[static hinge]

oh yes in that case ip-adapterfaceid probably won't work as it won't detect the face. Maybe ip-adapterface will try to work something (ip-adapterfaceid just don't work if it doesn't detect the face). Or just general ip-adapter may help with adding the details of the character.

[keen panther]

Would that work for rotating a character?

Upon searching using IP-adaptors for rotation, I stumbled upon this:
https://www.reddit.com/r/StableDiffusion/comments/18wy1ww/my_stablezero123animatediffipadapter_rotation/

Which lead to discovering about Stablezero123
https://www.youtube.com/watch?v=kIGTqyND21g

It looks very promising

[static hinge]

It will try to replicate the reference, so it will kind of help but might need tweaking and several tries, and when you have some kind of base you can translate it to inpainting

[static hinge]

well great, hope it works

[quiet raft]

Any way to get better tongues? 😄

[spark wyvern]

Hello! Does anyone know a good prompting guide?

[dull portal]

I wasn't expecting magic but this, from 'male, Finn the human with (long blonde hair:1.2), (standing upon a mountain peak:1.2), above the clouds, holding onto skull flag on a post' LOL, that is an ocean, and maybe a USA flag, it's mostly a test though, I want to see how bad the upscale is gonna add scars and stuff. :/

[dull portal]

Not as much as I thought, but I don't know what setting is causing the scars, Maybe it's obvious to one of you guys?

[static hinge]

is there some magic setting for making Outpainting mk2 Script to actually generate somethings like goes seamless with the rest of the picture?

oh nvm, so like I'm experiencing, just need to try until something make sense

And just barely, so I can then inpaint

[quiet raft]

Hey, what prompts that I use to get green toppings in the hair.. This image that is where I have added like

(green topps:1.2) and (black hair with green toppings:1.3) .. How could I enter it differently?

[agile ferry]

it's "tops" rather than "topps" but why not "green highlights in her hair" or such? I think "toppings" would be more like something spread over the top so I'm not sure about that either. Consider "green streaks in her hair" also

So here's a tip for whomever needs to hear this. I kept fighting my prompt with "upper body shot", full body shot, and I kept getting portraits until I changed ratios or such like that. Finally I found adding "nice hips" to the prompt consistently includes hip level in pictures generated. I may have found this later than everyone else, still... Hope this is useful!

[quiet raft]

Well, it seems like the ai don't want to do it at the bottom :/

Something like this

[agile ferry]

even with green streaks and highlights both?

[agile ferry]

Oh, you meant "tips" not "tops"

[quiet raft]

If I google black hair with black topps I see those images so I tought I was correct 😄

[agile ferry]

Yah, try tips

[agile ferry]

Also try "green hair with black roots"

[quiet raft]

Should it not be black hair with green roots? @agile ferry

[agile ferry]

Roots are the base of the hair, I thought you wanted tips. Anyway, experiment! I gave you prompt ideas. One should work. lol

[quiet raft]

Ohh

I will try some and see how the turn out 🙂

[agile ferry]

General advice on prompting, too. If you don't know the correct terms, such as hairdressing in this case, google hairdresser terms. Like I wanted braided hair once, so on google I found dutch braids, fish braids, ring braids, french braids, cornrows and other names from a web page to get exactly what I was wanting in my image generation.

[orchid scaffold]

any tips on how to inpaint and close a characters open mouth?

[orchid scaffold]

it might not be sd

[zealous field]

Hi, I'm new to stable diffusion. this is a prompt I wrote, what would've you done different to make the prompt more effective? Any keywords, format etc.

a grey haired, beautiful, smiling, innocent sorceress doing magic, wearing silver jewlerry, princess crown+, black dress, digital art, fantasy art, light academia, highly detailed, movie style lightning, shadows

[static hinge]

Hey hi, I'm looking for help with regional prompter (or multidiffusion?) to get two different characters with different set of Loras, but usually just get a mess or one single character that takes the two regions.

Is it OpenPose a must? I feel kind of lazy of posing the characters but...

Latent takes too much and doesn't necessarily delivers a better result

[static hinge]

It seems I get two characters more easily without it, but they share prompt and lora, and when not even asking for two characters in landscape 16:9

[static hinge]

yes with something in the prompt that may indicate other person, it does double generation, and it seems more difficult to have a double generation with the extension 😵‍💫 as it tends to create one single character that has all the characteristic in each region

I actually had to check if the extension was on as I keep getting double generations, but it seems it doesnt (I though it got stuck or something), just double generating from 16:9 (it seems it is more biased to double generation)

[silver valley]

You should use Multidiffusion with prompt control setting

[restive socket]

guys I have a noobie question about Loras, how do I use loras with no trigger words? I downloaded this one lora for female suits, it doesnt have a trigger word. Do I just add the lora and not specify the clothing of my character? or do I add the lora then I add things like, full black suit, white blouse, etc. ?

[silver valley]

both can work

some loras dont need a trigger word

but promting a bit to help the lora is good too

[restive socket]

alright

what about changing the scale of it like 1, 0.5

does it matter?

[silver valley]

yes

[restive socket]

its just clothing after all right

oh how does it affect it

[silver valley]

with that you define the strenght of it

[restive socket]

yeah but what confused me was its just clothing

[silver valley]

some loras are "overtrained" means that the effect of the loras on :1 is too strong and resulst in bad quality images

then you can reduce the strenght

[restive socket]

okay ill just play around with it then

its an sd 1.5 lora, I'm using juggernaut XL, it won't work right

[silver valley]

that wont work yes

[restive socket]

thanks a lot

[silver valley]

no problem 🙂

[static hinge]

I'll try that in next session I'm into it, thanks! Regional prompter seems pretty frustrating tbh

If you want specific character and poses, I have experimented with it for composition and loose ideas, and it kinds of works sometimes, but not for specific things it seems

[silver valley]

I dont mean regional prompter extension.
I mean multidiffusion as it has a in build in regional prompter tool

Much easier to understand than the regional prompter extension

[static hinge]

yes I was talking about what I was trying, haven't tried multidiffusion

[static hinge]

I tried to understanding it but it seems not matter what it just doesn't work in many generations or prompts/Lora bleed into other sections like nothing, or it doesn't understand you asked for two characters

like I said it is pretty funny that I get more double generations without it, it is like it is more useful to get a single subject or thing with certain characteristic in certain regions, than different characters

[silver valley]

yea for me it was pretty confusing ant not intuitive to wor kwith

[silver valley]

give it a shot, its exactly what you want to do

seperated characters with different loras for each for example

[static hinge]

I'll try it, ty. And relieved that at least "it wasn't me", the reason it wasn't damn working

[rich crescent]

i want genrate texture image for 3d image
any prompt idea ?

[restive socket]

guys I put "clear background" in my positive prompts and I put "(((blur))), (((blurry background))), bokeh, depth of field" in my negative prompts and I STILL get a blurry background, what can I do? (cfg scale 8)

[agile ferry]

Experiment with the commas, try putting fewer with each idea. Make sure of your spelling, it might not understand jewlerry vs jewelry, try 'cinematic' if you are going for movie lighting, and if you want more weight on the crown, use (crown:1.3) instead of crown+. Just my thoughts. So I'd try "a grey haired beautiful smiling innocent sorceress doing magic, wearing black dress with (crown:1.3) and silvery jewelry, fantasy art, digital art, light academia, highly detailed, cinematic, shadows" but experiment.

[agile ferry]

Try describing the background? I'm not sure what 'clear' would do, but consider white or black if you want nothing in the background, but I put "sharp focus" in the prompt as well. Seems to work for me.

I use baddream, unrealisticdream from civitai as negative prompts as well.

[restive socket]

I want to use those but I'm using SDXL, would it still work?

[agile ferry]

Yeah, I always use SDXL

[restive socket]

okay I'll use those as well and will try the things you mentioned, thank you

[agile ferry]

Good luck

[zealous field]

this was a very helpful feedback thanks! Though nearly every prompt I saw on civitai was in a "booru" tag format. (1girl, long hair, flowers) etc.

[agile ferry]

It's all experimenting. Honestly, with the right model and right negative prompts, you can sometimes trust the most simple prompts. I'm doing this one now "art by H.P. Lovecraft,priest,sfw," and I'm getting good results.

[agile ferry]

Oh, and try going through https://stablediffusion.fr/gallery/artists-page05 for the right artist. That will fix so many issues if you find the style you want.

I've been building a huge number of images going through styles and saving what I like

[zealous field]

thanks again, you're a life saver

[agile ferry]

Let me show you precisely what I'm doing actually... art by H.P. Lovecraft
,priest,sfw,
Negative prompt: nsfw,baddream, unrealisticdream, closeup,
Steps: 8, Sampler: DPM++ SDE Karras, CFG scale: 1.5, Seed: 3876852312, Size: 832x1024, Model hash: b62ddba1a6, Model: turbovisionxlSuperFastXLBasedOnNew_tvxlV431Bakedvae, Version: 1.7.0

For a friend's Call of Cthulhu game

[restive socket]

it says that baddream and unrealistic dream is for sd 1.5, and they do not appear in my Textual Inversion tab, could it be because I'm using SDXL? if so, how are you using it? I'm confused lol

[agile ferry]

Well, it's just a negative prompt so I add it to the negative prompt and I think it helps, that picture I sent is good and I get consistent good results regardless of model.

It isn't a lora to add to the positive prompt or anything like that, so I dunno.

[restive socket]

but it doesnt appear for some reason tho ill see thank uu

[zealous field]

I'm always impressed by good hands

[agile ferry]

Yeah, and you see the prompt I used. That wasn't cherry picked, I ran a random prompt for you with that setup.

But you dl the textual inversion, put it into your embeddings directory, use the keywords in your negative prompt, step 3, profit.

[zealous field]

step 3 profit?

[zealous field]

Oh you mean like earning money from it?

[agile ferry]

That was a South Park reference

Look for underpants gnomes if you're like very curious. Should be easy to find

[zealous field]

oh ok lol I found it

[restive socket]

I would like to generate a gaming room with a gaming chair and relevant equipments like headset etc. for my AI girl, and I'm on SDXL. I couldn't find a lora for that at all, there is only one and thats 1.5 only. is there any way I can do that?

[zealous field]

you don't necessarily need a LorA for that, a detailed prompt and a good base model would do the job for you

[restive socket]

I use juggernaut XL

[zealous field]

LorA's are of course handy, but in my experience they're better to use for specific poses, characters, body parts, textures etc.

[restive socket]

bet

I'll try it that way 🤝

[zealous field]

You are using a highly realistic base model, maybe try Anime Art Diffusion XL

This is surely just my guess, I'm no expert as you know

[restive socket]

yeah but im trying to generate hyperrealistic images lol

thank u for the help tho lol ur appreciated

[zealous field]

my bad, I read "AI Girl" as "Anime Girl" therefore I thought you were going anime style

[restive socket]

lol no worries XD

[sullen gazelle]

Is there a way to prompt using descriptive prose? I'm a writer so it'd be great if I could just copy and paste my description of a landscape or character rather than having to "reverse engineer" it into a or tailor it into a prompt for SD to better understand

[undone bronze]

Hello, I'm trying to make the background blurry, but I can't do it more than a little, my intention is to make it moderately blurry, I already eliminated words similar to "blurred" from the negative prompt, but I still don't get the result I want, what? could do? thank you

[timber rivet]

not a wiz but ive gotten things like "focal length 20" to work, sorta

[crystal echo]

Using control.net the way to go for that

[crystal echo]

Add "Bokeh " or even better, a " a portrait photograph shot by a professional photographer with a DSLR camera with a 55 mm lens, Bokeh effect"

[sullen gazelle]

It's been almost a whole year since I've used SD, are you talking about controlnet? I thought that's used for posing / reference images

[crystal echo]

are you taliking about png info or the interrogate clip?

[grand moon]

Descriptive prose works as a prompt, more or less. I'd suggest trying it and seeing if you like the results.

[sullen gazelle]

Uhh I have no idea what you're talking about - I'm trying to gen an image from my description that I wrote in my short story

[quaint sun]

How do I no messed up limbs

[dull portal]

I hear ControlNet and Inpainting, I've figured out neither of them. 😄

[scenic flume]

hello, i have been trying to train a lora (dragonborn from baldur's gate 3) but the face is always distorted, i heard it has something to do with overtraining? could someone help me

[iron blaze]

what prompts should i use for reflection images like this

https://w0.peakpx.com/wallpaper/1020/167/HD-wallpaper-reflection-in-the-water-pretty-moon-water-girl-anime-elf-reflection.jpg

i tried many different prompts, but all i get is character half submerged in water.

[restive socket]

how can I remove that glitched cloth on the wrist?

[restive socket]

Can somebody help?

[agile ferry]

inpainting

[restive socket]

yeah but I couldnt figure out how exactly

i painted the area and used many different prompts

[agile ferry]

You'll have to watch youtube videos. Inpainting is a whole skill set you need to develop

[restive socket]

okay will do

[agile ferry]

You have to consider masking, blur, latent noise, fill, original, weights and denoise if not more

[restive socket]

alright

I thought painting that area and setting prompts as "empty wrist" or something would be enough

[agile ferry]

I'd mask only, set a low denoise maybe .5 or so, inpaint mask, 20 blur and see

with a prompt of "a wrist"

[restive socket]

alright will try now, thanks a lot

[agile ferry]

Now I have a question. How do I get a pic of a woman about 40 pounds overweight? All I can get is either athletic or very much overweight.

Everything is either around 100 pounds or 250

[restive socket]

idk much but I would use the "slightly feminist" prompt

mb mb lol

[agile ferry]

Curvy, yeah. Might help. Rubenesque would be too much. I tried thick and got the 250 result

slightly overweight is 250. 😛

[restive socket]

oh fr curvy is like right in the middle

maybe slim-thick

[agile ferry]

That hit at around 190 pounds, so getting there. Ah maybe weights. (curvy:.4)

[restive socket]

yeah like

maybe (chubby:0.4) ?

[agile ferry]

curvy gave me the 250. I looked at the preview. 😛

Checking that

nope, gave the 250

(chubby:0.1) asian girl, full body,masterpiece,sfw,(Bauhaus Art:1.3)
Negative prompt: nsfw, baddream, unrealisticdream, closeup,watermark,signature,watermark,signature

That does not seem like 0.1 chubby to me

'average' looks like under a buck soaking wet

[silver valley]

maybe with:
obeese

[agile ferry]

I'll try.

Nope, that's way bigger than chubby

[silver valley]

oh ok

[restive socket]

lmao

you think would using a different base model work?

or maybe there could be a lora out there that lets you adjust the weight

[agile ferry]

I think this is going to be my best bet

Or a lora

[restive socket]

yeah lora too

good luck

[agile ferry]

How did the inpainting go?

Or... I could find an artist with slightly heavy women

Like Reubens

Even michelangelo

[restive socket]

sounds reasonable

inpainting didnt make a difference at all

i think ill watch a few videos and get detailed knowledge about it

[undone bronze]

thank you very much, I will try with the prompt you say

[scenic flume]

can anyone help?

[true star]

Someone has any tips for better results? these kinda suck

[restive socket]

hey does anyone know how I can change the face expression while keeping the same person? I put my character on img2img and I set "shy face" as a prompt, its doing well but the result is a different person.. I couldn't figure it out

[zealous field]

I believe the best way is to use "unified canvas" to mask the face and generate it like that.

[restive socket]

is it usable in A1111?

[silver valley]

Use controlnet reference_only or IP-Adapter for that

[restive socket]

I already generated my image on txt2img, then I sent it to img2img to change the face expression

so should I upload the same image on controlnet on my img2img as well and then set reference_only?

or should I use txt2img again and upload the generated picture on controlnet?