📝｜prompting-help

[silver valley]

did you tried blue wall background for example?

[tired vigil]

just tried that and several variations of it, and so far no luck. I may just fall back on my photoshop abilities to mask things and quit beating my head against a wall.. oh well. thanks though 🙂

[crisp linden]

Have you tried (selective color grey background)?

[gaunt sail]

Hi everyone, One of the users of one rp group asked me to make him PC on AI , well, I accepted. And now I want to ask... How to promt to make simmilar jacket , fantasy - like, with white shirt under it

( Not exact this, simmilarly-badass )
I use protogen 2.2

BTW , this is my first atempt

[south tinsel]

Hi

So I saw a challenge in my country to do a cool prompt of a blue Camel

Any good prompt ideas?

[tired vigil]

I had not. However, I tried that and ... at first it didn't work with just one set of ( ), but I kept adding more, and at 3 sets, that got the result I wanted. Thanks so much for your help!

it occurs to me that the number of sets needed might possibly depend on the remainder of the prompt, but I will just have to try it out and see. 🙂

[steel root]

Does anyone know how to leave the image as if it were midjourney's v4?

[crisp linden]

Glad to be of any help, yw

[short obsidian]

hey does anyone know if there are certain prompts for making a side profile of a person? I cant seem to let stable diffusion generate the person facing to the left

[silver valley]

side view worked for me

[short obsidian]

tried it but I think it just doesn't like the input image I used guess ill have to get creative

[silver valley]

you tried it with img2img ? i think that wont work if the character isnt facing sideways in the first place

[frank moat]

https://www.reddit.com/r/StableDiffusion/comments/10cxfpn/stable_diffusion_doc_guide/?utm_source=share&utm_medium=web2x&context=3 To help others I have posted this doc on the reddit for stable diffusion. If I'm not allowed to post this please either delete it or dm me for me to delete it. anyways i jsut wanted to ask you guys to upvote to give people the opportunity to use something I've spent months researching for. (I DO PLAN ON UPDATING IT MORE DONT WORRY ILL JUST BE SLOW)

[olive sable]

Hello. Are there any manuals about prompts for multiple characters scenes?

[frank moat]

that's quite complicated to do but i do know how so if people want enough ill put a how to in my doc

[olive sable]

It seems the two characters and character and a ‘mass of people’ scenes are possible. Three and more scenes are give a low quality of picture.

But the most compact description is very important.

The real problem is the characters with non-standard body construction (e. g. cyborgs).

[frank moat]

not entirely true there is a way to have more

[olive sable]

Which one?

[frank moat]

when it comes to making images with more then 2 it is possible for them to be good quality but the start takes time to do

[zenith smelt]

Hi, I am getting a lot of images that have text on them (not readable text) any idea why that might be?

[olive sable]

Probably, they have been generated as ‘covers’ or ‘comics’.

It seems the text is the more or less deformed parts of prompt.

[olive sable]

Can you give some prompts examples?

[frank moat]

just add in the negatives "text, logo, title"

[wispy creek]

Hello! I am very new to this. Is it normal to have about 1 good image generated per batch? Are there any tutorials that covers what makes a prompt good or bad?

[tired vigil]

in short; yes, it's very normal. And there are so much info on the subject that I wouldn't even know where to start. Other than saying this: don't give up, just have fun creating and learning :)

[frank moat]

i can teach you if you can vc rn

[wispy creek]

Thank you, so it is a bit like Goldilocks? Is it better to have run on sentences or ideas separated by commas?

[frank moat]

when you first start yeah

the longer you refine the prompt the better

[wispy creek]

Thank you for the offer Musical Sleep, but I on the clock at the moment, just on break right now

[frank moat]

i actually have made like 3 perfect prompts in my entire time making them

[wispy creek]

I have been flipping through your document

[frank moat]

awesome

[tired vigil]

when writing prompts, there's a lot of different ways to get the same results, just learn and understand how it work the way to you is the important thing. :)

[frank moat]

idk if you saw but i posted a reddit post of it and i really want it upvoted to get t popular

[wispy creek]

Does the AI understand things like Ratfolk? I have been trying, but it seems to only like to make Rats of Unusual Size

[frank moat]

lemme experiment

[frank moat]

yeah it seems to know what it is and it definitely work better then humanoid rats

[wispy creek]

Thank you for your time 🙂

[frank moat]

btw if anyone needs a how to on making prompts i can vc rn

[wispy creek]

Will you be available in 2 hours?

[frank moat]

potentially

[wispy creek]

Are you still available Musical Sleep?

[copper heart]

Hi all

SD seems not to understand the concept of a "cantaur", since every time so far i try to get half man half XXX, i get nothing i want

tried to paraphrase the concept, for example for a man-car creature, but i always get the car, sometimes the man, but not "emerging" from the car

anyone had a similar issue?

maybe inpainting would be better?

[silver valley]

Yea or an other model, or try human Bull hybrid

[copper heart]

i just remembered a thing i wanted to try, image-to-image with a collage i did for the result i'd like

let's see...

[copper heart]

ok, with an image source it works better, ofc, but I'll try other solutions anyway. thanks for the input, in the meanwhile 👍🏻

[gusty marlin]

regarding the order of the tokens, I've noticed something, probably something already noticed a lot... But it seems that the first token usually have an effect on the whole picture, while the latest token have an effect on details

[tired vigil]

I figured out something kind of by accident that seems to consistently work better: ((blue sky in the background)) Just in case anyone else has the same issue.

[astral rock]

hey guys, on a negative prompt, what does it mean if an argument has (one) or ((two)) brackets?

and do you guys know how i prevent duplication of a model which is trained on 512x512 when i use a larger image size while creating?

i always end up with 2 bodyparts

i guess i figured it out by flooding with negative promts what i don´t want 😉

but still, anyone can explain the (one) ((two)) thing for promts to me?

[quaint coral]

the "( )" will put more attention to the word, for example: if you put (red) in your prompt it will give you a redder image, if you put ((red)) it will give you a even redder image

you can also decrease the attention with "[ ]"

[gusty marlin]

I haven't read it yet (shame on me), but I was wondering if that attention mechanism was similar/the same as in the paper attention is all you need

[astral rock]

ok thanks, so is there any percentage what one ( ) or two (( )) mean for the prompt?

or "weight"

[gusty marlin]

every set of parenthesis is 1.1

so (()) is 1.21

[astral rock]

ok

so it has the same function as (word:1.2)

when i do ((word))

[gusty marlin]

(word:1.21) yes

[astral rock]

true 1.21 😛

i like the doubleclick ctrl+arrow keys function

[gusty marlin]

something funny though, I don't remember if it's still true, but (word:1.0) is not the same as word, or at least it wasn't

[astral rock]

i guess ill use that one instead of the (( ))

[gusty marlin]

yea it's best for (((((((clarity)))))))

[astral rock]

i will look up some more info about weight of those promts, if i find something usefull i will let you know

i found on reddit, but i don´t get it

[round bronze]

Is there anyway to have characters be generated with the "greater than less than" eyes or 😆? I've tried the danbooru tag (> <), "greater than less than symbols for eyes", and even the emoji above, but I haven't had any luck.

[gusty marlin]

have you tried "emoji face" or something like that ?

but tbh, I think you need a fine tuned model for that

or an embedding

[silver valley]

i had succes with something like (:P), but it depends on the anime model

[round bronze]

Which model did you use?

[silver valley]

AnythingV3

[round bronze]

I'll try it out soon

If all else fails, I could potentially try and make an embedding/textural inversion of the >_< eyes

[silver valley]

yea then send it to me haha

gonna try it now with >_< and anything

dont work :/

[round bronze]

Yeah

I tried it as well

Watching a video rn of how to do textural inversions to see the process and see if I could do it with the >_< eyes

[silver valley]

i tried (((cute annoyed))), and it kind of worked xD but not with the eyes

[round bronze]

Ok, I'm only halfway done the video, and it looks really complicated. Especially considering I want to make an embedding of a really small detail (At least compared to what embedding usually is used for AFAIK)

[silver valley]

yea ive never seen a face embedding xD

[rain flint]

can I modify my own imagen and make variations of it?

[tranquil folio]

where's a good place to find prompt keywords? lexica and prompthero aren't really doing it for me 😕

[ebon jackal]

what style or medium or even artists do you like? instead of just typing in the magic word, look up articles on best art/artists or whatever and see what words they use to describe the cool things like colours, composition, emotions, framing. all the things that jump out at you in a piece you like

like heres an example. say you like studio ghibli stuff. https://animehunch.com/why-we-think-studio-ghibli-is-better-than-disney/ look at the kind of evocative and descriptive words used

[shy vortex]

idk if this is the right channel for it but what do yall think the best model would be for generating Personas from the hit JRPG Persona. I tried making an embedding but it didnt go so well

[ebon jackal]

well its art is very postorized/t-shirty, so you could try using terms suggesting thick outlines, silkscreen, CYMK pallet etc.

[half current]

What negative prompts should i use to turn this image into

This, (i mean the colors and stuff)

map of kingdom, made by leonardo davinci
Negative prompt: cartoon,ugly, disformed, multiple,duplicate, animals,saturated,noise, grains, mutilated,disfigured
Steps: 20, Sampler: Euler a, CFG scale: 7.5, Seed: 4046394969, Size: 768x512, Model hash: c86a1a99b8, Model: dreamlike-diffusion-1.0

[silver valley]

Try more steps, also tags like intricate, highly detailed, masterpiece, digital art,

Oh you mean to get the used look, then maybe washed out, pale, old paper, Renaissance, old map

[ebon jackal]

literally just "weathered map" probably go light with that weighting. vignette is another thing to use very lightly, to give that kind of dimness around the edges

that kind of edge detail. 🙂

[orchid scaffold]

what do you guys think of these thoughts by ChatGPT on prompt crafting?

"In terms of creating the best prompts for the Stable Diffusion AI, it's not necessary to include the language "and is a" from the part "and is a 4K UHD image" in the script. The purpose of the script is to provide descriptive information about the visual elements of the animation and the overall tone and mood. The information about the resolution of the image is not as important as the other visual elements, and it's not necessary to include it in the script.

The more important information to include in the script would be things like the setting, the characters and their actions, lighting, camera angles, and overall mood. This information will help the AI generate images that match the desired look and feel of the animation. The goal is to give the AI as much information as possible to generate the best images possible.

Additionally, using different languages other than "and is a" to describe the resolution of the image could be more informative and less repetitive, for example: "The image resolution is 4K ultra-high definition"."

would "The image resolution is 4K ultra-high definition" work or be illogical for SD?

[silver valley]

Depends on the model, i would suggest to just tag 4k, HD, highly Detailed, cg wallpaper, and so on

[orchid scaffold]

thanks!

[gusty marlin]

@round bronze @silver valley face embedding ? like that ?

[gusty marlin]

What do you think about requesting a 4k UHD image in the prompt for a lower resolution image ? (4k is 3840x2160, which goes beyond what you usually do with SD)

[silver valley]

Ah nice, but what we ment was a Face expression model. Like a specific facial expression

[tranquil folio]

if I wanted to make a hybrid cat-person is there a way to accomplish that in 2.1?

[orchid scaffold]

using 4k helps sometimes imo, it has nothing to do with the actual image resolution

[tranquil folio]

i see something about blending in the prompting features in the manual, but it doesn't seem to have any effect

[white meadow]

Hey guys, pretty new to SD, could someone help me with generating a prompt to get this sort of image?

[silver valley]

Definitly try (low poly), man holding katana,

[mint rover]

how could I upscale this image to be closer to the original? I'd like the details to be more realistic - something like the second image

[tranquil folio]

anyone had any success with tattoos in v2.1?

[upper owl]

How do I get Stable diffusion to create a word in a calligraphy art style kind of a way?

i know its not that good with letters

but has anyone created a prompt that work?

[silver valley]

You could use inpaint

[elder dew]

Any tips for prompts for getting good logos? “Logo” rarely seems to work

[zealous plank]

Hey I'm generating a robot, I included "glowing green eyes" in the prompt, but rather than just the eyes it does many other areas green too. Any way to specify just the eyes?

Using the robo diffusion model if that makes a difference

[cinder imp]

I think there’s a way to connect keywords into like one big keyword but I don’t remember how

Dashes maybe?

Maybe parenthesis like the guy above is doing

[zealous plank]

i think parenthesis just add weight to a keyword, at least on auto1111 which im using

[cinder imp]

Underscores maybe?

[zealous plank]

oh maybe

ill try that once im done failing at inpainting

[velvet mural]

hello guys, is useful use hight sampling steps( > 50)? or usually will do nothing?

[silver valley]

depends on the model

[velvet mural]

anything v3

[silver valley]

hmm then not very much

30-50 is enough with euler a

[velvet mural]

and from 30 to 50 usually there is a difference?

[silver valley]

yea

[velvet mural]

depends on the type of image? like if it is a image with a lot of detail or more resolution

[silver valley]

yea for higher resolution more steps can be usefull

[velvet mural]

okay thank you very much

[silver valley]

np, i would suggest you try to generate a few picturs with the same seed and just higher the steps

then you can compare them

[cinder imp]

I feel like more steps makes it more generic tbh

[zealous plank]

Pretty new, I have an image here and I want to upscale it while making it look more detailed, how can I do that in automatic1111?

[zealous plank]

ooh nevermind found some good settings

[valid bridge]

Ok does anyone know what denoising in automatic1111 translates to when using the official model (since I found a good image to image prompt/settings but don't know how to set it for the API)?

[median bridge]

Hi, does anyone knows a good place to browse images and see what prompt they used? I used Lexica before, but since they have trained their own model, now everything looks the same, it became super boring.

[ebon jackal]

you can filter for SD 1.5 on that site.

but yeah, most images are glossy magazine type things, like if anime and dnd fans made penthouse or playboy. look up actual artists and reviews of them, start collecting terminology and striking phrases that pop out at you

other people's prompts are good for negative prompts i will say though

[halcyon ermine]

any good prompts for more interesting camera angles?

[ebon jackal]

try "informal composition"

or describe an object other than the subject, like an arch in a garden, and say "framed in arch". sometimes it goes a bit wonky, but othertimes it composes around that object really well

[median bridge]

oh! awesome thank you!

[halcyon ermine]

thanks. will do

[lavish hedge]

@stoic star 'Girl' is a forbidden world, try 'young woman' instead

[stoic star]

ok

[vestal rose]

Does anyone have experience in prompt engineering for depth2img diffusions?

[potent badge]

My latest Prompt Tutorial https://www.youtube.com/watch?v=1dJzaX6Hv0Y

[viral steppe]

does anyone what prompt i need to write to always get a white background?

[pine scarab]

Does anyone know if Stable Diffusion can put a specific phrase or statements in an image? Something like this. Any format to achieve that statement always be in the image. Thanks

[silver valley]

you can try it with the img2img tab and then Inpaint

[quaint coral]

gotta wait till the deepfloyd model gets released to the public

[burnt tiger]

black super hero

[whole magnet]

prompt for Full body, like from head to toes

[silver valley]

@whole magnet try focus on face and shoes, like blue eyes, highheels,

[velvet mural]

Hi I was trying to inpaint a new character in addition to my character, but I can't get it to add a new character, it just redraws the area. in the prompt I am describing the new character I want it to draw, is there something I am doing wrong?

[half trellis]

Hello everyone, I apologize in advance for my English, I'm a Frenchman working with a translator! 😅
I hope my question will be clear, but I can't find the input to put to generate text ON the image. What should I write in the prompt to have for example "PandaLove" written in the bottom left corner of the image. Thanks in advance

[silver valley]

hey, you should use inpaint in the img2img tab for adding words.
The next version of SD called DeepFloyd will be able to generate every text on an image.

[half trellis]

Thank you very much ! I'm gonna try it

[north lion]

how to get chatgpt to put out really good and detailed prompts like the examples from the kenshi 10 model?

[silver valley]

@north lion Someone send me this and it works very good: First say this to ChatGPT:
I have an AI that outputs an image when I enter a prompt. For example, an example of a prompt that displays a beautiful image of a girl standing in the breeze is "masterpiece, best quality, ultra-detailed, illustration, 1girl, solo, sitting, an extremely delicate and beautiful, beautiful detailed sky, beautiful detailed eye , side blunt bangs, hairs between eyes, close to viewer, breeze, Flying splashes, Flying petals, wind” With this in mind, please give me a prompt in 225 elements to produce an image of a high school girl. (1) Words to describe the drawing method, such as watercolors and oil paintings and digital art (2) Words to describe the girl's appearance characteristics, decorations and belongings (color and pattern are also specified) (3) Girl's age, hair color, hairstyle, and hair style Words to describe length, eye color, eye shape, facial expression, chest size, clothing (4) Words to describe the background in detail (5) Words to direct the girl's pose from head to toe can also be added to the prompt. Be sure to include "masterpiece, best quality, ultra-detailed, illustration, 1girl, solo," in your 225 element prompt. You don't need """ and line breaks in your answer.

then you can say this everytime after that and adjust what you want to see:
now make the same for a attractive female liberian in the library with office clothes and it looks elegant.

[north lion]

yeah basicalls changing the subject

that might actually help me later, pretty confident it would actually need even less explanation but better be safe

[silver valley]

yea, for met it worked good, and even skipped the nfsw censor if you know how xD

[potent cloak]

working for me

sometimes it spams though 😄

[true gulch]

How can I stop the AI from creating extra digits on hands or making extra limbs?

[dire warren]

try negative prompts, or the bad prompt embedding?

@true gulch https://huggingface.co/datasets/Nerfgun3/bad_prompt

[shy pine]

hi! how to make a true seamless pattern?

[dire warren]

@shy pine use the tiling option if you're in automatic 1111

[north lion]

thank you my man/girl/it/insert gender

[coral sinew]

Is there a "native" way to add a random weight to a prompt? Right now I have a wildcard that basically assigns a random 0.1-0.99. Like (forest:RNG). Works super well.

[twilit gale]

How can i change a draw that already existis?

Where do i go to make an image based on another one?

[coral sinew]

If you're using Automatic1111, img2img is best bet.

[charred wigeon]

Hi! I'm currently working on an anime retexture project for elder scrolls 2 daggerfall, and I want to try to convert the player body to an anime body, however I need the shape to be the exact same as the original, so I have to set denoising strength at below 50, however with the body texture I am currently using, I get a lot of unwanted highlights and shadows that I can't seem to get rid of. I more or less just want the body to have a single color like most anime figures, does anyone have some sweet prompts I could use? This is the body in question for reference:

seems I can't send a picture of the body because it's naked x)

[coral sinew]

Maybe try img2img with an alpha mask?

[charred wigeon]

so if I paint it all black, it will generate based on the shape and prompts?

[coral sinew]

Mileage will vary. A lot.

Using inpaint option with transparency in the png might be better option.

wahahaha this makes me smile

[charred wigeon]

hmm I am still getting results that are too close to the original in texture S:

[tired vigil]

hey i'm trying to generate an interior inspiration for my office but run into the issue weird looking objects anyone have any luck getting realistic results with img2img or depth2img, i keep getting outputs that look like this with weird backdrops/furniture:

[wanton girder]

**what was your best tips you get for making an promt right ? mine was **
"when you eyes of your generatet character look not good and its not an anime character, just use (((anime eyes))) as negative promt"

[north lion]

hahaha don't do that again xD

[charred wigeon]

are there really not a simple prompt related way to remove gloss from skin ?

[wanton girder]

there is

1. DPM2a It's probably the slowest sampler, but it makes the most realistic skin

2. Sand and mud in the negative promt because "trust me bro"

3.((faint)) and (dull) in the positive promt

@charred wigeon

test it with and without "face fix"

option

and pls post you "before and after" results

[charred wigeon]

I can't because it's a nude texture :S

[wanton girder]

oh

ok

XD

[charred wigeon]

but faint and dull helped a bit!

[wanton girder]

test the others too

[charred wigeon]

tried with sand and mud, that just made it look weird

should have specified that I am making a 90s style rpg body texture into an anime texture through img2img and no matter what I do, I can't get rid of shadows and highlights and gloss ect on the body xD

all I want is a nice simple flat skin texture

[wanton girder]

ah ok when you want i can help

i just need the original and an example what style yopu want

and no i dont use SD for it

i need to see the pictures to understand the problem in detail

do you want to make it as an 90 oixelart too ?

[coral sinew]

curious what you mean by 90's style RPG. when I think "90's style RPG" I think of star ocean and breath of fire.

[charred wigeon]

the 90s rpg is Daggerfall xD

I'm making a weeb sprite overhaul for it haha

[rancid shore]

The polical correctness hinders result's.i just wanted to build the scene of a first class service in a Pan Am flight. It completely ignore my Promt.what is wrong with the prompt asking for a blonde flight Attendant in the cabin service of 1980

[brittle geyser]

I'm really bad at promptwriting. I want to generate images of freefloating crystals in space, but I can only really get generated images like this, where it's sorta on the ground. What could I maybe add to make it generate it floating in space?

[inland spear]

maybe like this?

[brittle geyser]

Yea

this is the closest I've gotten so far

it also really does not want to generate space

[inland spear]

well, you could just img to img, would be easier... But I that one using just prompts, like you said...

centered view, single giant pink colored amethyst crystal floating in space
negatives: flower, sphere, ground, many, multiple, lots of, lots, blurred, poor quality, person, people, cluttered, deformed, frame, circle

DPM++ SDE Karras, 20 steps, deepSpaceDiffusion_v1.cpkt

[inland spear]

I think you can use somethink like dreamlike-diffusion for more artistic images, like this

lemme try

[brittle geyser]

Yea I'm using dreamlikeart

[inland spear]

noice that sure looks like it... Then I think it would be easier to just inpaint the crystal there

[brittle geyser]

not bad

[inland spear]

yeah dreamlike is a good model

[north sapphire]

Is there any way to improve the coherence of colors in a prompt? If I enter "photo of a person with blue hair, wearing a pink and gray tshirt" it generates a person with blue, pink and gray hair wearing a blue, pink, and gray tshirt.

[desert walrus]

The prompt is oil painting. Any tips on getting an oil painting without it looking like a photograph of an oil painting?

[dense gorge]

If content of the image is right, just crop it. stable diffusion can get you like 90% there. manual editing is inevitable right now.

other than that, be more descriptive of the style and you probably won't get photos of a gallery wrapped painting, or frames.

[ebon jackal]

i guess its to do with how oil paintings or other stuff textured find their way online. as in light is going to catch on them quite often. I guess you could maybe play with ambient lighting, but you're likely to still get that kind of sheen to it

[desert walrus]

I'll try and find some known artists that do digital oil paintings, might help. For now, doing it in midjourney and then inpainting seems to be the best way.

[ebon jackal]

yeah its just that, well if you scanned in an oil painting, the museum would be very angry. midjourney would have lots of digital oil stuff, so probably a better choice for this 😄

[dense gorge]

oh no. not angry museums.

besides, almost half of all paintings in museums are fakes anyway. Jean-Baptiste-Camille Corot painted 3000 paintings. there are 5000 of them in just America. The art industry has been full of gatekeeping d-bags since forever.

[ebon jackal]

you know what i mean though, its a textured medium, would be very hard to scan without any reflection at all

i dont mean angry as in gatekeeping, i was joking, as in feeding one through a big xerox style scanner 🙂 not saying its impossible, just that you'll get a lot of glossy look photos of oil paintings

[elder dew]

Trying to put the territories of the Risk classic map (territories on it have different borders to countries in real life) onto a globe logo.

Any tips on how to get the AI to get the colour scheme and borders of the Risk map right? From what I’ve tried, it keeps changing the borders to the ones of real countries, presumably since it’s associating it with a normal world map

Example of it changing the borders of the Risk map, to the borders of a real world map

[solar storm]

If anyone happens to see this and is decent with prompting, I have been trying to generate a light-hearted, brightly colored snowball fight scene. I have tried 100s of different prompt ideas in various forms, but no joy. I am starting to wonder if there are not enough well labeled snow scenes in the training set, especially in the genre I am aiming for.

My best results so far (I started using a base image to try to force the scene and let it get the art style right):

[ebon jackal]

try using "hallmark" or "christmas card" or something. its the composition is the issue. try my ol' faithful "informal composition" close to the description of one subject in it. it might work or might give you a closeup and bokeh out everyone else

you might need to inpaint though, it can do big scapes, good portraits, but blurs the two.

also its subtle but i can see it tiling a bit so if you change the aspect ratio to a little smaller and try that, then maybe outpaint it after to look longer

[autumn kite]

Hello everyone 🙂

So I have a question, am I going to insane with my prompt?

Like I am unsure if I am doing this correct, and after much experimentation I have come to a conclusion with my prompts. But I don't know if I have just gone nuts lol. Here is the prompt for this Pos: "--ar 9:16
[(((<Dr. Phil: 1.0>|<Book:Charlie Sheen> =0.5))-(<At a party: 0.4>|<During a show: 0.3>= 0.01>)=1.1)]
(Realistic:0.4-<Photo realism: 1.0> = 0.6)
(Colorize:1.5)"

Neg: "(<low quality = 1>-<low detail = 1>-<medium quality = 1>-<medium detail = 1>-<blurry = 1>-<scribble = 1>-<artifacts = 1>-<signatures = 1>-<watermarks = 1>-<mask = 1>-<logos = 1>-<pixelated = 1> = 1.0)
(<scribble = 1>-<cartoon = 1>-<anime = 1>-< Low quality = 1>-< medium quality = 1>-< low detail = 1>-< medium detail = 1>-< blurry = 1>-< jpeg artifacts = 1>-<signatures = 0>-< watermarks = 1>-<mask = 1>-<logos = 0>-<pixelated = 1> = 1.0)" Is there a guide to show me what symbols do what, because I think I get it... But... I don't have a clue if I'm right...

[ebon jackal]

is everyone trying to break the bot now?

[autumn kite]

Lol, I have been messing with it for a bit through Auto1111 watching different vids. So I may have over engineered it a little bit.

[coral sinew]

Anyone know of prompt uses cases that include characters *, #, and %?

Working on a prompt extension, and using these for tags.

• for wildcard terms, # for random number in a range, for example

[ebon jackal]

i know that # annoys naming files, but if you want wildcard stuff, yeah there are extensions for it, v good ones

[coral sinew]

The dynamic prompts one works great. Really powerful. I'm just extending the simple wildcards one to include negative prompts and rng from a range for weights. The characters are filtered or wildcarded out before the naming happens so I'm not really worried there. More like "is there some subject or prompt syntax that needs an asterisk??"

[hushed tide]

Is there a way to comment out part of a prompt, so that the words remain but are ignored in generation? It would help me to tinker with a prompt

[tired vigil]

anything beyond 77th token works like that I guess

[harsh vale]

Been using "Alex colville painting" for my prompts to get some pretty goofy looking stuff, definitely recommend having a play with it

[tired vigil]

a way to avoid combined two heads/faces - what's the best prompt/negative prompt?
two heads or two faces gives a similar result, 10-15% of pictures going instantly to trash

looking for something to be more rare

[charred wigeon]

anyone know how to do sprite animation? I've put together the sprites into one picture, any ideas what I have to prompt to get consistent results?

[tired vigil]

what exactly do you need? I understand this as "cut picture into 4 parts, save as 4 image files and go to online gif generator"

[rain sundial]

A man and lion talking each other on mobile phone

[silver valley]

The trick is to use portrait Format like 512x768

[dire warren]

fun prompt in jinja2 using dynamic prompts {% for i in range(5) %} {% prompt %}[singularity:infinity:{{ i/4 }}], microscopic view of a cell, symmetrical, [motion blur:shallow depth of field:{{ i/4 }}], HDR, 4K, gigafractal artstyle {% endprompt %}{% endfor %}

[silent marsh]

consistency isn't really SD's domain yet. no amount of prompting will help unfortunately

[tranquil folio]

how can i force a prompt away from head & shoulders? i've tried medium photo in the positive prompt and close-up in the negative prompt. i suspect maybe i overtrained my dreambooth subject on too many headshots (but my data was like 7 closeups and 5 mediums)

[stiff bolt]

maybe -- full shot body photo Parallax wide angle , aerial view, landscape painting, aerial photography, massive scale, street level view photo lens wide-angle up to as much as 300mm.

negative Out of frame

[tranquil folio]

thanks, i'll give it a shot

[stiff bolt]

stuff i got from search havent tried

full body rendering

[scarlet beacon]

how can i create a vatar from a image?

[frank moat]

if the problem is the low quality people dm me i have a secret way that not everyone knows

[frank moat]

instead of negative prompts just use inpainting

[tired vigil]

Inpainting resizes my photo from 1024x1024 to 512x512, so it's some kind of downgrade... well it's not a bad idea if I cut a 512px part of it to keep the original scale, this case should let me do it.
Or that's just a stupid pixel limit in individual colab link, because elsewhere it doesn't resize? Is it normal that inpaiting doesn't output images larger than 512 px?

[dire warren]

@tired vigil if you select the proper image size in the inpaint tab, you can choose the Resize Mode: Just resize, Crop and resize, Resize and fill, Just resize (latent upscale) - no it usually keeps the same image size

[tired vigil]

Okay, that's just limitations of the used source, thanks!

[hardy dew]

very cool

[dire warren]

@hardy dew I'm writing a tutorial on how to use Jinja2 in automatic1111, should be on reddit later today

the video that generated is in #🎥｜animation and its really trippy

[dire warren]

Using Dynamic Prompts with Jinja2 Templates in Automatic 1111 - https://www.reddit.com/r/StableDiffusion/comments/10jgmtk/dynamic_prompts_and_jinja2_templates_in_automatic/

[shy peak]

Imagine

a fighter black man,4k

[stiff bolt]

funny was generating some images that had several animals --- certain animals will dominate over others ie elephant macaw parrots dominate over cat dog

[tired vigil]

imagine a cat with red eyes

bruh

[high terrace]

What works best, putting all negative prompts in the same square brackets or word by word?

[shy peak]

/dremboot

[ocean ivy]

Imagine sun rise in pluto

[tired vigil]

tolio

[fervent vine]

@silver valley working with waifudiffusion rn

anime artstyle, girl eating hamburger, cardboard burger king crown, masterpiece, modern, cute, inside, eating at table, high quality, beautiful

I put it on 83 steps and upscale so its gonna take a lil bit

[silver valley]

xD maybe start with 60 for WD

Which WD Version did you Downloaded?

[fervent vine]

https://rentry.org/voldy

the one from this

WD Final

[silver valley]

Okay you should take the float32 variant, but maybe the 16 works for you

Is the image done? ^^

[fervent vine]

86%

ETA 1:25

done

[silver valley]

Okay xD we got the burger right

What you need is also negative tags.

[fervent vine]

I just want like

something like this lol

[silver valley]

Try to add:
Blurry, lowres, deformed, ugly, cropped, low resolution.

These go into the negative prompt field

Also you need to get Quality tags, try to add:
Highres, masterpiece, highly Detailed, intricate, best quality

Also when you test dont use the highres fix

You just test and want to generate fast

[fervent vine]

ok

[silver valley]

Much better right?

[fervent vine]

yeah

adding "bad anatomy" to negatives

and "life sized" to positives

to fix the burger size, and her lip

[silver valley]

Also add Bad hands and multiple limbs, to negative

[fervent vine]

how can I make the style more consistant?

[silver valley]

You mean artstyle?

[fervent vine]

yeah

like this

[silver valley]

Depends a lot on the model.

You should try the one i send you and compare it with WaifuDiffusion

[fervent vine]

ok I will do that tomorrow

4am for me lol

[silver valley]

Also the Rest is just tags

Like short hair, streaked hair, glasses, evening light

Autumn would be good too

[silver valley]

1pm here xD lunchtime

[fervent vine]

nice

thank you for your help

[silver valley]

No problem and good night!

[cyan wraith]

"a single fire elemental magical rose sitting on a wooden table, love, a raytraced image, epic, romanticism, romantic, ray tracing, painting style, fantasy, high detailed, ultra realistic"
neg:"face, low detailed"

i'm trying to create a magical rose, but i'm struggling

already tried using this image as a base:

"a single fire elemental magical rose with petals made of fire sitting on a wooden table, love, a raytraced image, epic, romanticism, romantic, ray tracing, painting style, fantasy, high detailed, ultra realistic, dream-like"
neg "low detailed"
cfg scale 22,5
denoising strength 0,6

i was thinking in something more like this:

[tired vigil]

creating very specific images is like trying to create that burning flower as your first project in photoshop. The ai has no idea what you want, because it doesn't really "know" anything, it just add the stuff you write. and using commas helps to put one thing from another. But the question is, are the amount of commas enough? Do you need more? Less? There are so many details to know to get a very detailed and specific image. :(

[cyan wraith]

it's hard 🤣

[tired vigil]

just how we like it! ;P

[cyan wraith]

but I think there is a way to generate these images.
We just need some engineering.
Maybe using some specific text embedding, using different models, mixing images or something like that

[tired vigil]

these were the results in the normal 1.5 model using your prompts

but the 2.1 model is very different, and from my knowledge, also weaker when it comes to none realistic stuff. That might be the reason why your tests look more "real?"

[cyan wraith]

i'm using 1.5

[cyan wraith]

cfg scale?
Eueler a?

That's what i get

[tired vigil]

ah, you used the img2img?
Then it's hard to get the result as, depending on your settings, it tries to keep the color, style and details from the image used

[tired vigil]

lower the cfg to 7 or less

[cyan wraith]

i think you put this as reference, right?

[tired vigil]

I just wrote the prompts a single fire elemental magical rose with petals made of fire sitting on a wooden table, love, a raytraced image, epic, romanticism, romantic, ray tracing, painting style, fantasy, high detailed, ultra realistic, dream-like into the normal txt2img

[cyan wraith]

🤔

ow

it was the negative prompt making things bad

now i got what I wanted

[tired vigil]

yeah, I only use the negative prompt field to fix errors I don't want from the prompt words. Most often it's just the color purple I remove

[cyan wraith]

thank you 😄

[tired vigil]

I didn't do anything, you did all the work :)

[tired vigil]

To get emphasis on a word is it ((word+)) / ((word))+ or something else?

[wanton patio]

hi guys, im trying to use some black and white sketches to create a model, but everytime i do, my resulting image is too dark

any ideas on what i can put in my prompt

[wheat cedar]

trying to get dark blue skin; putting in "navy blue skin" puts the subject in the navy (where you can sail the seven seas), dark blue skin does nothing, tanned blue skin (and blue tanned skin) gives normal-tanned skin, so i've run out of ideas

anyone got any ideas?

[tranquil shard]

Where are prompts coming from? Are they build in the model? Is it possible to retrieve all of prompts from the model?

[untold crane]

did you end up finding a guide on symbols like -, --, ( ), (( )), <>=, and the numbers you're using?

[autumn kite]

Nope, just been goofing about buys gpu time on colab to figure this out. I want to know if I am heading in the right direction. Been experimenting like it was a game I have to get achievements for lol. Here is another one I am working on figuring out. I think I have a color prompt working kinda the way I want.

--ar 9:16
(Skyrim-Cyberpunk)
(<--i(10)><[Black|White:{i/5}]:0.9>-<[Red|Gray:{i/3}]:0.3>)
(<Masterwork>-<Extreme Detail>-<High Quality>=1)
(<Wide Angle>-<In Shot>-<Dynamic Angle>=1)
(<Perfect Anatomy>-<Aesthetic Face>-<Symmetrical>=1)
Concept Art, 4k

I need to rewrite the first part to (<Skyrim>-<Cyberpunk>). I am experimenting with setting my own intergers there with --i() which I don't know if it is working like I think or not.

I have theories... (<>-<>) is like a equalizer, like take a look at both equally. It's why I use it in my negative prompts. I try to use it for certain categories like (<High Quality>-<High Detail>=1). Anyone else got anything like this they've come up with?

[true violet]

Please i know only about () chars! What are all other symbols? How to use them?

[signal jewel]

puts greg rutkowski in negative prompts master race reporting in

you don't really need prompts that look like algebraic formulas to get good results lamo. seems like a huge waste of time

the key is the model.

In fact, most blends will put out worse and worse results the longer the prompt

[true violet]

How I can force the subject to stay into frame? Any negative tips?

[signal jewel]

most models are trained with centered subjects

but if urs isnt, try (centered in frame, photo studio)

[true violet]

Ok

Sd 1.5 model

Animals with the body in the centre and head out of frame

[signal jewel]

1.5 model base?

I dont recommend using base model for image generation

[true violet]

Why? My final goal is to create coloring sheets

Any advice on models?

[autumn kite]

I have various models I use, and yeah the longer the prompt it get confused. It looks like a formula, but it seems to work. That's why I love about this, their are so many ways of doing it. This is just the one I have been trying to work out for myself.

[true violet]

Experimenting I've found the way.. But maybe testing other models can result easyest

[autumn kite]

Well I have about a 150gb in different models lol

[true violet]

Any post or tutorial on symbols use in prompt, other than () []

[autumn kite]

I try to imagine a slider on a rpg, and that's what I've been doing. (<>|<>:0.0) is just one of the ways I've found to put two concepts in one area and to slide between them depending on the weight of the words themselves. Also I switch between models often to see different results. My negative prompt seems to fit for everyone of them. Though I need to expand it with another group.

[silent marsh]

if you're using the a1111 ui, the syntax for that blending is [ A:B:0.1 ] to change to b after 10% of the steps, or [A|B] to alternate between A n B each step. (A|B:0.1) is just going to make the A|B text deemphasized to 10% it's strength. You'll see an effect with that syntax, but for different reasons

[autumn kite]

Okay, I thought that is how it worked. I'll use the brackets for steps, and the other strength. So can internal parameter like this work? (<--i(10)><[Black|White:{i/1}]:0.5>-<[Red|Gray:{i/9}]:0.3>)

[dusty urchin]

Heya guys, is there other alternative for clip interrogator ?

tryin to find some good interrogator here

thanks in advance

[autumn kite]

Can I also think of this mostly like {Important}(Pay Attention) basic [Kinda here?]

[silent marsh]

{} is an alternate form of emphasis for deemphasizing. examples. (example) is the same as (example:1.1), and ((example)) is the same as (e:1.21) as it multiples by 1.1 each time. {} multiplys by 0.9 each time, so its the same as (e:0.9) and {{e}} is the same as (e:0.82) , i think i got the math right there. point is, it's just alternate emphasis syntax. i prefer doing the (e:1.1) form because you can click on the word and hold ctrl+arrows

i'm not sure what < > i think they're just tags

it's sort of like xml rules. you have to close tags before you open new ones. so something like ([example):0.5] wouldn't work because those brackets are closing like a venn diagram

but ([example|alternate]:1.1) would flip between ex and alt each step, and they'd both have emphasis

or [ex:alt:0.5] would do ex for half the steps, alt for the next half, while [ex:(alt:1.1):0.5] would do ex for half, then alt with emphasis for the next half

automatic1111 also has a little bracket counter that tells you if there's a closure problem in your prompt. it'll turn red next to the prompt box

[autumn kite]

Yeah that's what I've been using to kinda guide what works or not. I've watched some stuff. Yeah I've been using <> alot, I don't know but if it works for me to see that part easier too. Okay so the {Example:1.0} is like (Example:1.9)?

[silent marsh]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#attentionemphasis and
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#prompt-editing are the two major sections about syntax cues. other extensions might add features as well

[autumn kite]

Thank you!!!

[silent marsh]

{example:1.0} shouldn't work. the {} are just simple and don't take that extra : far as i can tell

[autumn kite]

I can try here

[silent marsh]

wait.. my bad. i'm totally wrong here. [] decrease attention lol

as always, the manpages are the god here

not some dumb rando on a chatroom << me apparantly

the {} stuff we see all over the place in public prompts are from other AI systems like NAI or openart

[autumn kite]

Well I have whole folders of me just goofing around with different things to see what looked good at least lol.

[silent marsh]

in A1111 {example} will try to pull a style you have saved named "example" into the prompt

[autumn kite]

Oh

So that would come in handy for somethings.

[fervent vine]

Trying to generate an image similar to this with just prompts

Positive: Wilderness Nature Landscape Mountains Trees Rocks Water Primitive Indigenous Hunter-gatherer Sublime Untamed Rough Rugged Unspoiled Wild Savage Natural landscapes Historical settings Detailed architecture Vibrant colors Romanticism Symbolism 1800s ruined roman civilization Catastrophe Devastation Ruin Collapse Demolition Wreckage Disaster Decay Degradation Deterioration Depletion Armageddon Annihilation Obliteration Extinction Desolation Desecration Destruction Erosion Deterioration Dilapidation Disintegration Dissolution Fragmentation Breakdown Decay Deviance Corruption Deterioration Deterioration Destruction Demolition Devastation Decimation Ruin Raze Level Wipe out Eradicate Exterminate Annihilate Obliterate Eliminate Extinguish Wipe out Erase Destroy Demolish Devour Consume roman civilization statues detail {oil painting} art painting Roman Republic Roman Architecture roman society city town zoomed out wide shot

Negative:
Modern elements Abstract elements Minimalism Pop Art Futuristic elements blur blurry realistic photorealistic

getting pictures like this

sorry for flooding chat

tldr: want more images like picture 1, getting images like pic 2

[silent marsh]

i need to rebuild my style catalog still. i'm running script on a txt file and so far caught 3gb of example photos of different styles. "in the style of {style}" over 1000s of lines from a txt file like, "hyper" or "8 k photos"

gotta start looking through that and saving certain successes as prompt styles

renaissance master oil painting?

[fervent vine]

ill try that

tbh im using tags generated by chatgpt lol

[silent marsh]

your tag with mine at top. 2.1 eular a

[autumn kite]

I'll try in my style to produce it.

[fervent vine]

changed some things up and getting this

amazing

how can I take this output and upsacale it?

[silent marsh]

very nice. there are a variety of ways to upscale. if you using a1111, you can "send to extras" which has the upscaler

or there are scripts that spliit it into chunks and use SD to upscale each grid section independantly, in img2img scripts

[fervent vine]

im using a11111 yeah

what are the best settings

[silent marsh]

you only need upscaler 1 generally. people use a second one and sometimes blend the results by transparencying them over each other. its another effect its okay.

some work better than others. i like realesrgan models, the default 4x works good

[fervent vine]

what about the slider bars?

should i use scale by or scale to

[silent marsh]

up to you. i always go for broke and just use a 4x. but if you wont a specific output size, thats your choice

scale by 4 means it multiplies it by 4. 512 becomes 2048

[fervent vine]

thanks!

[untold crane]

wow super helpful thanks. is this interpreted at the SD model level, or by automatic1111 upstream? I'm running a local SD instance with my own prompt writing UI

[autumn kite]

Getting there

[silent marsh]

all the prompt syntax is stuff at the webui level. SD just wants clip targets at the end of the day

[dusty urchin]

any prompt ideas to get flat colors like this

[autumn kayak]

Prompt: beautiful woman shocked suprised

[limpid hill]

I am an absolute noob starting today, but I have an art background. I would try something with "illustration, flat colours" seems to work ok-ish, I experimented a bit with different media of illustration prompts "acrylic" "digital" with different results

[novel bane]

How do you get stable diffusion to generate something it's very stubborn about? Should I describe it in more detail? For example, I wanted it to generate a menacing eye looking through a large crack in the ceiling, but it was incredibly stubborn even with inpainting. It always generated a person with the eye or no eye at all. For some reason, it kept generating windows too. I'm not that bad at prompt creation, but this one stumped me.

[limpid hill]

where is a good place to learn the basics about the prompts? I will give that menacing eye a try now anyways 🙂

[fresh loom]

If you want something to look at and see how their prompts look like (there're prompts come along with generated images), then prompthero.com or lexica.art are some of places to take a look. Other hand, if you want to dig into text explanation/comparision/tip tricks then I think just google/reddit is way to go.

[minor quarry]

imagine Afrofuturistic child ultra realistic

[mystic snow]

beautiful woman shocked suprised

[dusty urchin]

Thanks a bunch for the answer Labeardi!

[marsh pulsar]

here is how i do it.....i do some very rough sketch of what i am looking for and then use that for image to image instead of text to image....for something like this.

[limpid hill]

Pleasure, hope you got the results you were looking for.

[mellow summit]

any ideas for generating iphone mobile phone camera quality style photos

not professional bokeh camera style

[autumn kite]

This is the best I could do. Well so far at least.

[fervent vine]

That's crazy! How did you get the people to look so good

[autumn kite]

Messing around with different prompts till I find a decent perspective.

[fervent vine]

Mind sharing them?

I'm assuming you're using SDV1.5

[autumn kite]

Yeah, AUTO1111

[fervent vine]

Any vae?

[autumn kite]

I am not sure which model, what's a VAE?

I used Huggyface clip interrogator to find the right style, than just messed around from there. Here is my positive prompt

[fervent vine]

Tysm

[autumn kite]

Negative prompt: Scribble, cartoon, anime, Low quality, medium quality, low detail, medium detail, blurry, jpeg artifacts, signatures, watermarks, mask, logos pixelated, low quality, low detail, medium quality, medium detail, blurry, scribble, artifacts, signatures watermarks, mask, logos, pixelated, duplicate heads,

Welcome! 🙂

Model was GalaxyTimeMachine3 I believe

Seed: 3456470619

Well after chatting on here, my work is improving, just would like to figure out how to organize the wording a bit into categories so I can edit more easily.

https://www.deviantart.com/labyrinthgateomega Made a DeviantArt, where I'll upscale some of my favorites that are unique

[ornate cairn]

IMPORTANT WARNING: Don't type in "only fans" into stable diffusion: https://stablediffusionweb.com/#demo It contains hardcore pornography. I only did this for SCIENTIFIC PURPOSES. And now have to wash out my eyes and pray to Jesus. I hope this is not the new "safe" version they are using. PLEASE somebody fix this!

What if someone wanted a plain photo of just electric fans and that's what they saw 😫

[azure horizon]

#🏞｜general-with-images Rodrigo, um homem comum, descobre que tem o poder de saltar na linha do tempo. Ele começa a experimentar sua habilidade, mas rapidamente percebe que cada segundo que ele passa nas dimensões do salto, atrasa 25 segundos da vida das pessoas na Terra.

Rodrigo tenta lidar com as consequências de sua habilidade. Ele vê como suas ações afetam as vidas de seus amigos e familiares, incluindo a perda de empregos, atrasos em casamentos e até mesmo o nascimento de filhos. Ele se sente responsável pelo sofrimento das pessoas e tenta encontrar uma maneira de consertar as coisas.

Rodrigo descobre que o único jeito de consertar as coisas é desistir de seu poder. Ele faz um último salto, voltando no tempo para o momento em que ele descobriu sua habilidade, e decide não usá-la novamente.

Rodrigo vê como a vida das pessoas volta ao normal e como ele é capaz de seguir em frente sem o poder do salto. Ele aprende a valorizar o tempo e a importância das escolhas que faz.

Rodrigo segue com sua vida, agora sabendo que o poder de mudar o passado vem com grandes responsabilidades e consequências. Ele segue sua vida normal, como se nada tivesse acontecido, mas ele nunca esquece a lição que aprendeu.

[glad crescent]

Hi, Im still quite new. Trying to generate an image of luke skywalker holding a lightsaber. Prompt

photo of ((luke skywalker)), holding a lightsaber, full body, beautiful, epic pose, cinematic, photorealistic

Result, I get sometimes a good results, sometimes I get an extra lightsaber...

I tried "holding 1lightsaber", same result, sometimes good, sometimes extra lightsaber...

I tried to use negative prompts but I don't know exactly what to type to avoid the extra lightsaber... I tried multiple lightsaers, 2lightsaber... no luck. Ideas?

[heavy smelt]

anyone having any luck with openjourney?

I am noob at prompting with SD

I had only experience with MJ

Is there some catalog/gallery to look? is very good to learn

[old dagger]

does anybody know of a good prompt to get quarter turns, when combined with things like 'front view', 'back view', 'side view', etc, just to indicate it's not a direct front/side/back view? I've had a lot of success training those terms into a multi-concept model, but using 'turned' to indicate a non-direct view doesn't seem to be something which the model ever learns

[warm quail]

Only luck I have had is with 3/4 2/4 and even then it is hit and miss.

[old dagger]

Thanks, anything SD already knows might be a better starting point for training

[warm quail]

Have a look at how professional photos are labelled and imitate that

[coarse cliff]

This should show an error before you can do a prompt like this

[sonic dagger]

Is it true that this symbol '|' finds a halfway point between 2 words on both sides?

Like asian|caucasian

[tranquil shard]

How do I use concepts from the sd-concepts-library? Trying to use it as embedded not works for me. Tried renaming .bin to .pt. Are they intended for a specific model only or they not meant to be used as embedded?

[rocky nebula]

hey all, does anyone know of a good prompt database? Or some place where a lot of prompts and associated images can be searched?

[tired vigil]

how do I "learn" prompt-engineering ?

I don't want to just copy a prompt I want to actually learn how to make a prompt that gets the kind of image I want

I've been searching on google for artists and terms related to art etc. also asked chatgpt but haven't seemed to have made any progress

thesaurus doesn't help much either

[silver valley]

@tired vigil The Best way is just to try and Experiment

Use negative Quality tags for general better images Like blurry, lowres, deformed

Also use good Quality tags like masterpiece, highres, highly detailed or intricate

For Anime models you should try tagging like: word, word, word, but for other models you can also try: a fish flying trough space, as a prompt

Also take a look here:
https://openart.ai/promptbook

[silver valley]

https://lexica.art/

[tired vigil]

I see thanks, I'll try experimenting more

[tired vigil]

I already have this

[jagged tulip]

How do I exaggerate facial emotions? Like prompt for a person so surprised, his jaw dropped to the floor

[tired vigil]

Someone has suggested Lexica already but I found one called Krea too: https://krea.ai

[vague stone]

Is lexica prompts only valid for their custom SD model ?

[silver valley]

if you want to get the same image then yes, i would just take the prompts as inspiration. they also work with every other model but you will get different images

[tired vigil]

Any tips on replicating these images that I made in Midjourney using Stable Diffusion? model / setting recommendations?
Prompt Example 1:
Beautiful Impressionist painting of a Final Fantasy male, short silver-haired warrior, blue cloak, ancient ruins background, fantasy, bright, Dramatic, Beautiful Lighting, Fusion between Jeremy Mann and Jean-Baptiste Monge and Darek Zabrocki, aaron griffin, Long shot, characters on the right side. --ar 3:2 --c 30 --v 4

Prompt example 2:
Beautiful Impressionist painting of Final Fantasy White-haired princess overlooking city. wearing Armor, fantasy, bright, Dramatic, Beautiful Lighting, Fusion between Jeremy Mann and Jean-Baptiste Monge and Darek Zabrocki, aaron griffin, Long shot --ar 3:2 --c 30 --v 4

Prompt example 3:
Beautiful Impressionist painting of a possessed DnD Orc warlord standing in front of burning village, army in background, Medieval city environment, fantasy, dark, Dramatic, Beautiful Lighting, Fusion between Jeremy Mann and Jean-Baptiste Monge and Darek Zabrocki, aaron griffin, Long shot --ar 3:2 --c 30 --v 4

[dense junco]

Can we create this type of optical illusions with Stable Diffusion? I have tried various prompts but could not get this type of optical illusion

[warm quail]

Not really unless you train such things specifically. Doubt that AI saw many of them. And even then it generalised them to an average representation. These illusions rely of very specific patterns, and the model doesn't learn such thigns.

Like it doesn't even know what "face" is. It knows that a face is something that has two eyes, nose and mouth. The earlier models 1.4 and before kept absolutely cocking up faces. And 2.1 still fucks up hands. Because it doesn't know what hand is, it is average idea about what hands are supposed to be. But whether it is open hands, closed, fist, holding on to something or even correct amount of fingers.

[hard zealot]

is there any reliable way to generate an object and put a greenscreen behind it? or any color solid background. i can do it with input images but i havent been able to prompt it or find an embedding or anything

[hollow surge]

Is there anyway to stop getting multiple subjects in one image ive tried the negative prompts as well it keeps generating multiple for one

[sonic dirge]

sharing a random thing i typed:

i asked another discord (bluewillow) about prompt tips, they had no prompting 101 for me, so i just speculated how i image prompt engineering might be, just that i dont know yet

[tired vigil]

can you share the link ?

[static valve]

Simple monocolor background? Or just black background.

[hard zealot]

something that can be chroma keyed out

so either green, red, maybe blue depending on what the object is

dont have any image editing software and basically want to make youtube thumbnails by tacking together different things sd creates. i can sort of get it when i give a reference image that's greenscreened, but it would be cool to always be able to do it.

[tired vigil]

either use remove.bg or InstructPix2Pix

SD doesn't get the background color right

[vital leaf]

Hello, what exactly clip_guidance_preset does? is it better to use it with or without?

[dapper rune]

/

[worldly falcon]

I have an question for perspectives: I want an orthographic perspective, I always get one where I could see of an cube 3 sides, but how would I get an orthographic perspective from the side so that I get basically a plane?

[near cosmos]

anyone found good negative prompts for dreamlike model?
Currently using this: boring, low quality, not detailed, watermarks, logo
It tends to keep more details, but I figure it could be better.

[lone oriole]

anyone know how to make characters look at different directions or look at something? having a hard time generating characters reading books, the face always look at the camera

[dense junco]

Got it ..thank you

[dense junco]

So seems like generating optical illusions it challenging unless we especially train model for these types of patterns

[warm quail]

Still probably wouldn't work. Namely because you train it too much and itll just keep remake those illusions exactly. And somewhere in the middle it knows how to make average representation of them

It would frankly be easier to solve a mathematical equation to how to make them, the ask an I make images based on that.

[hushed sage]

Hi what does "--ar 3:2 --c 30 --v 4" do and is there a list of such arguments?

[remote orbit]

These are midjourney specific parameters I suppose

[remote orbit]

The closest I got so far:

[hushed sage]

Not sure what C is but AR is aspect ratio and V is version

[polar pulsar]

hiii

thank you mr.@crude brook

i problem with my negative prompt

Its my negative prompt: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

i how to fix hands problem

[warm quail]

You can't fix bad hands with negative. Because there is no dataset of "good hands" and "bad hands" in LAION

you can try adding terms for unwanted hand poses like "holding" "fist" "open" or whatever.

because keep in mind that negative tokens only remove things that AI has learned which had that token in the description.

so generally you should add in to negatives medical terms.

[frank oak]

is it possible to apply emotions to character's face? I'v trained embedding of previously generated character with six images of him. This embedding works good and creates character with consistent appearance but I barely can make him smile and can't make him cry at all. Is there any workaround for this problem?

[stark lichen]

If anyone wanna start learn more about online income, or wanna start get a passiv income with crypto for free.. then dem me. I will help everyone.

[dire warren]

please don't...

[orchid scaffold]

hey guys! wondering if anyone has ideas for improving the historical accuracy of images? For example, I am trying to get historically accurate depictions of gladiators.

Is there a good model to use for this?

[dusty urchin]

what models have you tried many you could share prompts and the results?

have u tried to find some settings on lexica and other sites?

[pulsar canopy]

Hello guys 🙋‍♂️ is there a way/certain keyword to add to the prompt to generate one-line art similar to this pic ?

[toxic shell]

dream/

[tired vigil]

Is there an actual difference between 8K and 8 K also other things like 1920 and 1 9 2 0 as I see a lot of apps seem to add spaces between numbers?

[tired vigil]

Maybe line-art? There is a model on Civtai but the site appears to be down currently. https://civitai.com/models/211/line-art

[pulsar canopy]

yeah i tried line-art but it's more of a realistic drawing (which is pretty good), but i can't get a simple drawing like the plane...

[tired vigil]

This uses one line drawing, but again like your result it doesnt look like it much https://www.krea.ai/prompt/132bf9f5-0181-4c34-b15e-43186a4879a4

[pulsar canopy]

sometimes an "," might affect the outcome

[pulsar canopy]

well thanks for the help 🙌

[true cloud]

anyone got a good negative prompt to stop getting multiple heads when making portraits?

[silent marsh]

i've been running a the inspiration extension's script on text file of different styles. a lot of them are 8k, 8 k, 8 k photography, stunning 8k, etc..

all of them seem to pull up the same 8k style promotional material that seems to be the target aimed at. photos of vistas and high grained detail that look great on high resolution displays or convey "high resolution" when in print media.

i used a simple prompt, "character in a scene, perfect composition, {style}" and it pastes the next line of the text file into the block. You can create your own x/y grid to s/r the 8k word with 8 k, 8k photo, and try a bunch of different ones on the same seed, to see what the result is. I always recommend doing personal explorations of the latent space, instead of just trusting someone's word. it's only 6 month old tech. none of us are experts

[ebon plover]

cat on a computer

[tired vigil]

Ok, it's just some apps like StarryAi for example have spaces between numbers. Even Lexica does so I wondered if it made any difference.

[silent marsh]

i think it's just meta, people copy pasting prompts

[ebon plover]

||cat||

cat

imran

[copper sun]

im trying to create a prompt to recreate V1 from ULTRAKILL. he looks like this:

however the AI keeps making stuff like this:

here is my prompt: a picture of a [blue] thinly built robot. it has a round (security-camera-like head) with a single flat ((yellow)) eye. it has [eight] wing-like protrusions coming from its back which look like thin (yellow) shards. it is holding a pistol in one hand and a shotgun in the other. it is [humanoid]

and my settings

is there anything im doing wrong in the prompt?

i tried removing the weight brackets but it wasnt any better. changing the number of steps, the sampler, the CFG, nothing

[shy oasis]

Image showing how short form content is eye catching

[fresh loom]

Many SD apps sees number as separated, just you don't see that in details. Even official Dreambots on this discord or other UI see the number (in prompt input field) separated. You type them as join number but on later part of processing they separated, as I read it somewhere, not very sure though. But at least it's true for official dreambots and other UI I used.

There's an independent developer said this about the matter.

[silver valley]

Your using SD 1.4 model, would suggest 1.5 or a robot specific model

[tired vigil]

a simple question more many, please let me know how I can make different variation of the same character (could be animal, person, or persons) with different background or characteristics(smiling, crying, laughing, and so on) , thx people 🙂

[serene meteor]

can someone prompt only in the negative part and get a avocado picture ?

[queen osprey]

/prompt

[merry jetty]

Hello. I often have clones of characters, that is, I enter a prompt in which there is not a single word that I want to get several characters, but as a result, 2 almost identical characters are formed on my picture. If there is any solution in the form of a negative / positive prompt, then please share it. (Model: Deliberate, but on other models the situation is identical)

[charred mantle]

Is any tool or best practice out there, which can manipulate a given prompt in the form, that every element of the prompt is present in the picture?The generated picture does not need to be good looking. It's just about recognizing the given prompt within the image. Currently using the stability ai api to generate my images. Any suggestions? 🙂

[woven canopy]

Make a proper description of the character, including gender, body type, hair colour and origin...or reference a celebrity as a reference 🤓

[royal yoke]

❓❓❓
Anyone have any well-rounded universal negative prompts for landscapes? Something that doesn't break the tokenizer bank.

An example of my landscapes: Photo of a lush dreamlike marsh, Gothic castle in the distant background, lake in the midground flanked by weeping willow trees, celtic mystics in the foreground sitting by a campfire, sky is a starry night with green aroura borealis.
Res: 1920 x 1088

I have issues with tree branches floating around.

[tight hare]

game art assets tips?

[carmine spear]

I haven't tried it yet, but try emotional puppeteer on civitai.com

[mystic crater]

#⭐｜pow-info

[silent marsh]

I haven't found too many prompts with it yet but i'm still skeptical that i came up with it on my own. something i've had a lot of great results with, instead of "portrait of" yadayada, i'll lead with "scene with a character" then describe the rest of it. thought i'd share

not to say "portrait" is bad. if i want a portrait i'll use that still

[tired vigil]

High resolution, felted doll cowboy with hot pink boots and a pirate at a tea party in a tree

[orchid scaffold]

wondering if someone could help me understand exactly what highres fix does? And is it helpful for deform? So far in my tests I don't understand the different between having it on and off

[blazing marlin]

us 'single line drawing' or 'one line art'

[gusty flicker]

does anyone know why my deforum video only stays true to the prompt on the first frames and then just disintegrates into neon chaos?

[novel flint]

hello how would I go about generating art based on some of my own ?

[cerulean pulsar]

make your own model trained it with your art

[heavy smelt]

can someone helpme? trying to make more artistic urban scenes. But the texture looks quite bleh

I would like a texture more impressionism/foggy/paint

[lyric badge]

whats your current prompt?

[vapid bloom]

How might I go prompting something like the works of Tyler Hobbs Fidenza? My current prompt is something like a colorful abstract painting on solid background with a pattern of squares, Alfred Manessier, triadic color scheme, an abstract drawing, generative art but it's not quite producing a similar result...

would it be best to train a model or embedding on these works and try something like that?

[wintry token]

#📝｜prompting-help jyuytti

[vapid bloom]

I succesfully built an embedding trained on 1.5 this data and was able to get results I liked

[tired vigil]

best model to use for making emojis/ simple cartoons? looking to make some custom discord reactions using img to img with some imgs I have

[wheat dirge]

/imagie casa

[soft hawk]

Dog

/dog

[barren spire]

😨 jesus christ that was quick

[hoary oar]

Yo guys help would be appreciated im new to this, what do the (( )) brackets do when entering prompts and also for negative prompts?

[old sorrel]

They say stable is way better for creating illustrated story books due to the ability to recreate the same character on a different page in a new pose, can someone send me a link detailing the best way to illustrate stories using Stable and other tools ?

[mint orchid]

is there anyway i can improve my prompt building? previously before this every generation looked like a doll, now they look like realistic freaks

here i just had 14 cfg instead of 12, and i didnt have the "doll, blocky" negative prompts

[silver valley]

Maybe try other models, or generate an image with txt2img an use this as input for img2img

[mint orchid]

for img2img-ing drawn sketches, do i need to improve my crude input drawing to get better results?

[silver valley]

It mostly depends on the model

[hasty jasper]

Can anyone give me advice on mitigating mismatched or just simply bad eyes? I use a mix of detailed face/eyes with bad eyes set as a negative prompt. And generally I get nice faces when it's a close up or portrait but full body shots... Not so much. One eye will look good, while the other looks unfinished.

[languid mist]

You just got started, feed the bot more info. It's on the right path

[cerulean pulsar]

use "bad anatomy" in negative

[hasty jasper]

Thanks a bunch. This improved it a lot actually, putting that in negative and good anatomy in positive

[cerulean pulsar]

👍

[eager pier]

cross eyed, heterochromia, phased out eyes, washed out eyes, poor detail, bad eyes, less than two eyes, more than two eyes, droopy eyes, elastic eyes, melting eyes, color bleed, poor coloration <-- on negative
golden ratio, extremely detailed eyes with symmetric composition and xdetailsherex, <--- on prompt

golden ratio should be weighed highly, symmetric composition could also be given a weight

maybe add something like exquisite precision line art at start or end of entire prompt

[hasty jasper]

Thank you for the response ❤️
I'll try these out too

[tired vigil]

Im trying to generate some pokemon/digimon style things and I have the word yellow in the negative prompt box of Auto1111 and yet its still generating yellow things. What else should I be adding, please?

[open ravine]

/dreans

[hoary pecan]

area composition should be able to do that: https://comfyanonymous.github.io/ComfyUI_examples/area_composition/

[hoary pecan]

having 2 different subjects is pretty much impossible with regular prompting

well you can get multiple subjects with regular prompting but you'll have to generate a lot of images before it randomly gives you one that's acceptable

yeah it assumes you know a bit how SD works

definitively not for newbies

[coarse walrus]

This is really cool. It reminds me of blender nodes. Does this work with the Automatic1111 install?

[hoary pecan]

it's a completely separate thing but you can use the same python venv as the auto1111 to run it if you don't want to install anything

[pure charm]

hi, what does it mean when the preview of the image gets cartoonish (high contrast skin reflections and no texture on skin) on the last step before finishing? It looks better 3/4 of the way through

[tired vigil]

Hi anybody has an idea how can i place multiple persons on the same image?

I mean like a man and a woman looking at each other

[jagged tulip]

Where can I do a MSc. in prompt engineering?

[civic chasm]

Any suggestions on fixing fingers/toes anomaly in the image?

[keen valley]

Hi all! How add bot in my server?

[mighty zinc]

I mostly use the parameter (anatomically correct), as well as several negative parameters like 'bad anatomy, (bad hands), missing fingers, extra fingers, mutated hands, poorly drawn hands, mutation, (deformed), (bad proportions), (extra limbs), (disfigured), gross proportions, (malformed limbs), missing arms, missing legs, extra arms, (extra legs), fused fingers, too many fingers'

Depends on the model, though. Some are better at hands/feet than others. Also not every model 'understands' every parameter. If you are still not happy, then apply weights to the parameters thru single or even multiple brackets (((fused fingers))) or as decimals like (fused fingers:1.5). Could also use inpainting when masking just the hands and let SD imagine new hands...

[silver valley]

Person, human, in negative prompt

[silver valley]

It also depends on your prompt. If you type yellow shirt you will mostly get a Person wearing it

You can share you prompt and result maybe we find an issue

ohh then maybe it depends on the model

getting this

yea your right its not easy

thats made with pencil line embedding i think

the best i could get:
with: stratosphere, floating islands in the sky and waterfalls, trending on artstation

😄 its a bit tricky to get it floating, you have to add sky, haven, or stratosphere in that case

me neither

example?

colors are a bit difficult

you can try emojis, they really help for some poses, but hugging is very difficult

ah you use invoke

i dont think xformers do anything bad with the results

ah yes

[royal verge]

Hey, my friend is a "loot pig" in Cod/Apex Legends, so I'm trying to make an img2img of him with a pig nose... I get some decent ones but nothing great yet. Anyone know if there's a better prompt besides just puttting "pig" there?

Oh, wrong channel eh...

My bad.

[civic chasm]

Thanks for the great tips, they actually make a difference!

[weak hornet]

does anyone have good experience with fine tuning the shape of a person's face? I'm trying to recreate a specific video game character. any good docs or blogs to read, vids, etc...would be greatly appreciated

[stone garden]

https://old.reddit.com/r/StableDiffusion/comments/zv5mbq/simple_trick_i_use_to_get_consistent_characters/

this might help

[weak hornet]

thanks!

[cerulean pulsar]

sauce?

need that embed

[cerulean pulsar]

I have 4 thousand words negative prompt if you want me to send you that?

[weak hornet]

sure

[cerulean pulsar]

here you go

did I say to use the whole thing? it was just list of all the good negative prompt

if you mixed ice cream with miso it'd taste like shit too

what model are you even using?

[coral socket]

I keep wondering who tags an image "worst quality". Like, you look at an image and go "Well, this is as bad as it gets. Nothing is worse than this"?

[cerulean pulsar]

what if you put all that negative prompt into positive and see what it give

[minor field]

Does "lowquality:1.4" still work like that?

Keep getting images like this even with "((canvas)), ((signature))" in the negative. Anyone have any other ideas?

Ahh you see that's the problem

Trying to maintain single step process

Inpainting is super powerful though I agree

It's not the mindset, it's because I'm automating these generations and want to get rid of stuff like this :/ I mean here and there it's okay

The other 3 from that set are perfect

I guess it'll do 😆

[cerulean pulsar]

this's what I got when I tried

which actually doesn't look that bad

except the pixelation

[minor field]

Generated 50 desktop backgrounds earlier and only one was unusable because it had the photoshop UI

[minor field]

Looks good

[cerulean pulsar]

thanks but it was supposed to look bad

[minor field]

oh

Yeah "low quality"

hah

[cerulean pulsar]

I was using all the negative prompt as positive

[minor field]

interesting

See I wonder how many of the negative prompts people are using are actually effective, it's all about the datasets tagging in the end

What was the prompt?

Flag?

[cerulean pulsar]

clearly FOGE FOVEAR

I'm surprised it actually wrote coherent text

[minor field]

Probably some of the best text i've seen

Hah

[cerulean pulsar]

I had 1 accidentally write out word in different language before

1 sec let me find it

[minor field]

That's the thing, the reason for scribble text is probably the mixing of languages.
I wonder if you can get semi decent looking ( still not full sentences or even words ) lettering if you prompt for "English" and a specific font

Next you'll be generating every sign around the country, just wait.

[cerulean pulsar]

"Dry massage"

[minor field]

What

Weird af

But impressive

Well I had something cuter in mind tbh

[rancid lodge]

#shahrukh khan

[cerulean pulsar]

resurgent of annoying orange and his mutated fruit army

[silver valley]

maybe this:
https://huggingface.co/yehiaserag/anime-pencil-diffusion

[cerulean pulsar]

thanks

[humble wharf]

does anyone know if an image-to-text app exists on the web?
You know how blind people can surf the web and images are described with speech accessibility
I wonder how to access those descriptions google has
Would be useful to enter an image an see how google describes it, for better prompting

[silver valley]

something like this?:
https://huggingface.co/spaces/fffiloni/CLIP-Interrogator-2

[humble wharf]

Exactly!! Thanks!

Have you used it yourself?

[silver valley]

i testet it but not much, there are other sites than can do this aswell, but need to find the links

[humble wharf]

that would be great thanks. Do you understand what the "best mode flavors" slider does?

It's a good start, but it also included emojis in the description hehe

[silver valley]

here is one for Anime:
https://huggingface.co/spaces/saltacc/anime-ai-detect
Here is one for Danbooru image boards (Anime):
https://huggingface.co/spaces/NoCrypt/DeepDanbooru_string

[silver valley]

emojis work in Automatic1111 webui, maybe someone used it

[humble wharf]

ive mainly played around with specific SD models so far, is the Automatic1111 SD version 1.5?

[silver valley]

Automatic1111 is the webui for SD, so its the tool not the model, it supports all models

[runic osprey]

any tips for avoiding characters that are cropped off? I'd like a full body shot.
I have full body in my prompt, cropped/cut off in my negative prompt

[fresh loom]

Sometimes, has more vertical ratio help

[silver valley]

try 512x768 as resolution

[acoustic hatch]

Hello, I would like to explore changing what something is made of. This photo is a good example of what I am talking about, where the leaves have been replaced with the smoke. I've taken this picture off of lexica.art https://lexica.art/prompt/6ce6fe70-5e85-49d9-99cf-313bf31d639c
First, I'm wondering how to reproduce that image. I don't believe I can get the Lexica check point, and the same prompt doesn't even produce remotely the same thing in any model I've tried (v1.5, v2.1, dreamshaper_33, openjourney, probably some more)

It seems no matter what I do, there is a high chance I end up with leaves in my work. Negative prompts with things like (leaves:1.3), (leaf:1.3) don't seem to get rid of them, sometimes working better than others.

Is there any guide or advice on how to do things like this? Can anyone actually reproduce the picture above and tell me how they did it?

I do have a beefy graphics card, so I can do some training, and am willing to do that as well.

[warm quail]

The thing with negative prompts is that they only do something if the generation process ends up encountering them in the "travel through latent space". It is like "Do not go to McDonalds while driving cross country". Well there is a chance that you wont encounter McD along your path, but if you do, then that instruction is valid.

I have tested this a plenty, by just doing variations on the negatives and whether they do something can't be determined without testing. Meaning that you should write the negatives with the attitude of "These are things that I for sure don't want". Then as you find a good output, interating that by adding and removing of negatives.

[polar pulsar]

i how to find "R-ESRGAN General WDN 4xV3" this upscaler (for Automatic1111)

please helpp

@silver valley

[burnt bloom]

We're do I create an image?

And how

?

[quaint coral]

the discord bot is not available anymore, if you wanna generate images and have a good (medium-high) gpu you can use it locally installing InvokeAI

or dreamstudio but its paid

i have an 1050 and i generate 1 image in 35 seconds

[warm quail]

Just go use any of numerous colabs

And buy tokens if you must.

They are cheaper per/hour than electricity in europe

[granite coyote]

Hello, i'm using Anything V3 and am trying to make an image with multiple characters. My question is how would I specify little details for each character without changing the whole image? For example if I wanted to add blonde hair to the one on the left, or if I wanted sparks of fire to surround the one in the middle.

[warm quail]

You can't. The AI can't do that. The best you can do is to specify like "... something, man with red jacket, man with blue shirt, boy wearing a hoodie, ..." because it will try to pull these individually.

But you can't change a scene that includes 3 people like that already

Since the AI doesn't know that there are 3 subjects. It knows that there are scenes that are composed like this.

[granite coyote]

Alr

[warm quail]

Keep in mind that the models limitations are that it has been described pictures, not elements that make pictures.

However... You can try using Comfy UI and the segment the diffusion by diffusing areas individually.

This is not Auto1111 systems but a seperate repo.

And you need to define coordinates for blocks, with individual prompts.

[prime mantle]

Hi all, sorry simple question from new player. 🙂 What prompt can I use to ensure that my subject remains within the frame? I am trying to generate buildings, but they are getting "cutoff" at the top. I would like the whole building to be within the frame. thanks a lot.

[warm quail]

basically what you do here is that you make a prompt, set Condition area, then feed that to Ksampler, then to encoder

However I have Absolutely no idea whether it works with custom models.

[granite coyote]

I have no idea how to do any of that yet so I guess i'll just manually edit it lol

[warm quail]

There is no 100% way to do that. Other than describing the head, face and hair. But then you can get portraits. The AI was trained with 1:1 images which autocropped so generally heads got left out for being in the upper 3rd

[warm quail]

It is simple node systems. Really intuitive.

[unique jay]

Anyone have the syntax for SD 2

[warm quail]

What you mean?

It is GDT3, it prompts with same words as any other GPT3 based system.

However it is way more responsibe to negative prompts than 1.x models.

Basically you can prompt with single token positive then only negatives - if you want.

Also the tokens are much more... how would I say this? Pure? Fundamental? They aren't as polluted as in 1.x models because they use much stricter filtering for quality of images.

[summer cloak]

Hello, is SD able to create characters in different angles? I want to create full body reference sheets.

[stone garden]

probably inpainting

[granite coyote]

How do you use that?

[thin temple]

Hi

I am having difficulty generating the image of a spiral from the side view. The AI keeps giving me top view or not a complete cone spiral. This is what I want to generate....Any ideas for a prompt?

The cone should be viewed from a side perspective, as if the viewer is standing on the ground and looking up at the cone. The bottom of the cone should be drawn with tightly coiled lines, while the top should be depicted with longer, flowing lines that appear to be reaching towards the sky. The overall effect should be a cone shape made up of spiraling lines that get larger as they move further away from the ground

[thin temple]

When I used this prompt, white spiral made of white lines that get larger as they move further away from the black ground, black and white colour only, side view, black and white painting, intricate detail,
Step 35,CFG 14, seed 1263306784

It gives me this instead, haha...

Any idea what prompt I should use to create the right point of view ie. the side view of a spiral seen by an observer on the ground?

[eager pier]

Sorry to intrude, may I ask what this is and where to get it?

[warm quail]

https://github.com/comfyanonymous/ComfyUI

It is not Auto1111 so... Keep that in mind. It has none of the QoL or optimisations.

[eager pier]

but it does provide a visual representation of what is happening and ability to change the information route?

to a degree

[warm quail]

Ehh.. It is literlally just the stable diffusion system as node based.

No imge previews or changing during previews.

It is raw deal basic stable diffusion just node based.

[brazen valley]

how can i create smth like this?

[cerulean pulsar]

By doing it? I mean that’s quite a broad question

You need to be more specific

[brazen valley]

i want to create image that looks like it but im getting bad results. like really bad using same model

[cerulean pulsar]

What model are you using?

[brazen valley]

openjourney

[cerulean pulsar]

Ah

[brazen valley]

what should i be using?

[cerulean pulsar]

Maybe try “night city background” and “neon light” in prompt

Might help

[brazen valley]

will try that

this is what im getting

[brazen valley]

its good but its not it

like something is missing idk how to make it better

[cerulean pulsar]

Resolution maybe

You used different resolution so result are different

[brazen valley]

ok

thanks! it looks good now

[cerulean pulsar]

Yw

[fading oxide]

hello, i am using stability ai but my generated pictures came out blurred, what can i do to make them visible?

[final dome]

I have an easy question: if I want a “woman with green eyes and a leather jacket”. How can I force the AI to use green only on the eyes and not the jacket? I have yet to figure this out, all my attempts have green applied randomly to other descriptives.

[silver valley]

try woman, green eyes, wearing black leather jacket, if your only color is green in the prompts it will aplly it randomly

[cerulean pulsar]

try putting "blur" in negative prompt

[cerulean pulsar]

Copy and paste it in file name when you rename it

Ahh

Well I mean when they’re too many it just have difficulty judging which’s more important

So you can make certain thing more pronounced with number

1.2,1.4 etc

Next to the prompt you want

[vapid moat]

also higher cfg scale could possibly help with that I think

[cursive token]

noob question: can I upload a photo and apply a cartoon or videogame style to it?

[vapid moat]

Yep! Just use img2img

[cerulean pulsar]

yes

you can even use your own pic

[cursive token]

Using dream studio, correct? I upload the image, then what do I enter in the prompt field?

Example: this image of Tom hanks, in the style of, say, Minecraft

https://www.cinemascomics.com/wp-content/uploads/2021/08/tom-hanks-2065547-e1628845289206.jpg

Or in the style of the Heavy Metal magazines illustrators

[shadow spade]

using Anything, is there a way to tell it to not give waifus to the character i'm trying to prompt? adding (1boy) just aggravates the issue

wait, i can use solo?

oh hahaha i'm dumb,

[slim obsidian]

/dream:

[tired vigil]

Anybody here who knows how to use Animator v6? I don't know how the syntax is for Keyframes and the prompt templates

[silver valley]

pls send your prompt maybe we can find the cause for that

yes sitting requires a person

what object?

maybe try standing

or just on ground or on object

can you send the prompt so i can test?

[warm quail]

You can't. Because the dataset for landscapes and images could have had people without having them labelled. It just generalised them in. The AI doesn't know it has people in it..

[crimson bay]

does anyone know how to make tanlines consistent instead of a random tribal tattoo?

[crude brook]

Any idea why I can't get wildcards on auto's webui working?

I'm using suggested syntax __filename__ and have the identically named .txt file with a list in wildcards folder

[granite coyote]

How would I make a fight scene between two characters?

[green hollow]

hello newbie here, i was directed here for advice on prompting, not sure what I do *scratches head

[celest ice]

Not sure if this is the right channel for this question, just getting into SD after moving away from MJ and I have found some models accept () and weighting and others seem to simply break and output garbage when you use parenthesis? Any ideas why?

Case in point dreamlike diffusion I can use brackets and weighting, but seekartMegaMidJrny same prompt is utter garbage until you remove the brackets in it

[patent fox]

Hello folks, I'm new to SD and trying to learn the program by generating a pair of holy boots like for RPG games. I'd like something similar to these:

But everything just isn't coming out quite right. What would be a good prompt here?

I suppose this is the best one I've gotten so far but it's a little too cowboy boot. I want it to look more like armor, big pieces of gold and stuff on it

[celest ice]

Maybe try using words like medieval, low fantasy, high fantasy, knights armor (that sort of thing?)

[patent fox]

Ok cool I'll try that thanks!

[outer plaza]

Is there a way to specify a color more precisely? Using just "green", for example, isn't generating the green I want

[still jolt]

Any idea on how I might get this type of aesthetic. I was hoping that I might get something like a plains rather than grass lands. I'm using the web version of Stable Diffusion if it makes any difference.

[green hollow]

how to fix hands anatomy, my negative prompt is..i think 50+ words long

[thin escarp]

anyone know how to improve this technoblade prompt?

this was the best one, others were very bad, actual pigs instead of just a pig mask on human and bad fur coats

[cerulean pulsar]

For hands you really just need to hope for luck negative prompt can only help so much

Hands are the biggest problem AI has

[candid solar]

Hello, do you know how to avoid multiple head or multiple character, everything is fine in 512x512 but when I try with higher resolution, SD makes double or triple heads or characters to fill the picture

[tired vigil]

Anybody here who knows how to use Animator v6? I don't know how the syntax is for Keyframes and the prompt templates. I want to be able to have different prompts for different times, like one prompt at 0 secs and another at 15 secs and another at 30 secs and so on... Here’s the doc: https://github.com/Animator-Anon/Animator/blob/main/README.md

Can anyone please write an example of how that could be written in the right syntax?

[cobalt mountain]

Use highres fix. It will always screw up composition generation if you go above the res model was trained at without this

[candid solar]

Oh OK make sens now, I'll give it a try, thanks

[inner dune]

Is there a reference sheet of prompt db to learn more about good prompts. Also prompt rules/order ect?

[ashen hare]

I'm trying to make character with non human skin (in this case, slime body and furry body) as fanart for Rimworld characters. All of my attempt to make anything result in at best character with non skin faces while the rest is human, or at worst human skin all over and furry or slimy clothes.
How can I more regularly have a body of a different texture?

[silver valley]

you need an other model than can do this

[ashen hare]

So the issue is model. Alright. How can I find appropriate models for my purpose?

[silver valley]

On Huggingface or Civitai.com you find a lot of models

for your purpose i would recommend an anime model

Anything v4.5 or Melty Mix Monsters

[ashen hare]

I am trying anything v4.5, I'll check Melty Mix monsters.

[velvet pecan]

I'm attempting to generate a humanoid squash. I keep ending up with a human holding a squash. Any suggestions for anthropomorphizing veggies? 😄

[royal yoke]

❓ ❓ ❓ Anyone have a good, well-rounded set of positive and negative prompts for 2.1? Something that's been honed to work with various genres and sizes of pictures? ❓ ❓ ❓

Please respond directly to my message here so I can see it later in case I don't get to it right away.

[sterile hound]

anyone know if anything 4.5 can generate slime anime girls? I'm having trouble with that

[compact forum]

probably can. just might need the right prompts.

if not there's probably a LORa for it somewhere

[sterile hound]

after a bunch of trial and error I finally got good prompts with good settings, saved the seed and sent it right over to my friend with 3090 to chug em out lol

2734701502
25cfg
cute, slime girl, transparent body, slime body, blue skin plus whatever you'd like, have fun slime enjoyers

[crisp badge]

25 cfg?

https://tenor.com/view/elmo-burning-hell-let-it-burn-hands-up-gif-17806161

[sterile hound]

it really didn't want to create slime girls on it's own

had to nudge it just the right amount

lol

[crisp badge]

just change the model. civitai.com they are plenty of them

[compact forum]

prompt weights confuse me

tweet one thing a point too far and everyone's broken into little chunks and turned inside out or looks like hamburger

tweak*

[royal yoke]

❓ ❓ ❓ Anyone have a good, well-rounded set of positive and negative prompts for 2.1? Something that's been honed to work with various genres and sizes of pictures? ❓ ❓ ❓

Please respond directly to my message here so I can see it later in case I don't get to it right away.

[slate rose]

Hello guys. There is some syntax in the prompt to specify imaging with "A" or "B" (or set some percentage for certain term to be selected)? I'm asking this because I would like to specify some prompts if I leave the PC running while I sleep.

[slate rose]

Ty. I'll try.

[brittle geyser]

Trying to generate something like this (first pic) but a "warrior" type, but I'm just getting second pic. Trying with prompts like chitin, war form, wings, claws, etc

[waxen moss]

can someone explain to me what does it mean when the prompt (word:1.5) happen? why 1.5? i have no idea what this ":" is used for

[tidal minnow]

I think it's emphasizing the word(s) between parenthesis, giving it more weight in your prompt, here setting it at 150%.

[waxen moss]

aahh thank you thank u

[burnt notch]

Hello! Is here anyone who can help me with creating minimalism art or art for nursey or childrens?

[humble jewel]

is there a "OR" keyword? can I prompt something like "man with red OR blue hat"?

[woven canopy]

If you are using Automatic1111 there are a lot of ways to do this...did you read the documentation?

[tidal dragon]

u can tell by his question he most likely has not or only partially

[candid solar]

do you guys know what to say in prompt in order to get a full shot (like from head to toes) I try "((full shot))" (duh ) but SD keep on giving me some close-up medium shots

[burnt notch]

tried full body shot maybe?

full body portrait, full body shot, head to toes

"A full body shot of ..." like this maybe try out

maybe tell if some of those words worked 🙂

[candid solar]

yeah i'll try those

nope it doesn't help

[silk crest]

Hey how do i give prompts??

I m just new to all these things

[burnt notch]

@candid solar which model do you use and what you try to create?

and which version of sd

[serene robin]

I’m making a character in stable diffusion and for a particular scene, I want them to look like their really struggling to lift something heavy, sort of like the picture above. No matter what I’ve tried, nothing seems to work because I want to include the gritting of the teeth, the strain, struggle, etc. Does anyone know how I could prompt stable diffusion to get something the above picture to pop out. No image to prompt generators have worked with it so I was wondering if anyone here could help.

[spare forge]

asian girls

[loud pumice]

Hello:) Can anyone help me? Trying to understand prompting in SD 2, but ineffectually. Suggested the more prompts you type, the worse result you get. Is this true? Also would be grateful for any tips on prompting in SD 2

[fair elbow]

i am wondering the same thing

[cerulean pulsar]

Not really

But there’s moderation to everything

Too much or too little is bad

[lofty gazelle]

i just downloaded anything v3 but it has a .safetensors extensions and not .ckpt

what do i do to get this running on my automatic1111?

oops, wrong channel 😛

but since I'm here, any tips for hands?

[fair elbow]

safetensors runs just as good as ckpt

[lofty gazelle]

Thanks!

I come and go, playing with it for weeks and then going away for weeks so I tend to forget and relearn stuff ><

here's my Vampire Survivors anime poppea built from img2img starting w/ the original sprite

"red hair girl white shirt green skirt leather corset holding a piece of paper with magical symbols, concentric waves of magic in the air, dark forest"

had trouble w/ the "magic"

[stable vale]

Hey Everyone! I'm a lot new to this and I'm just trying to get SD(v2 768 v ema / euler a ) to make me a book on an altar. Would anyone know of any negative prompts or tricks for not drawing merged books or books on top of other books? Most of the images look like the attached picture here. I've tried quite a few different seeds.

[burnt notch]

Anyone knows a model for stable diffusion or prompts to create art like this, is this even possible? tried different but not work really good

[icy ravine]

After reading this article https://towardsdatascience.com/how-does-ai-see-your-country-3899e4057735 on a bias AI models may on different countries, I wanted to do similar analysis on Stable diffusion. The author used this prompt on Midjourney Illustration for an animated show set in [country] about science and progress, taking place in the future to get images like these

any Idea how to create something similar on SD?

[silver valley]

If you use the Automatic1111 webui you can do this With Script X/Y/Z Plot and then choose Prompt S/R for X

Then you make a prompt like (City in country1), and for the X You then insert country1, England, america, turkey, China. For example

[icy ravine]

Thanks but I wanted know what sort of prompts I should try to capture the essence of a given country from the model's perspective. For iterating through the prompt and logging the images I think weights and biases is the best option.

[silver valley]

Oh okay, yea that would work

[burnt notch]

anyone may have an idea?

[warped lake]

@stable vale find a similar picture online, paste into img2img, save as a style, go to txt2img, play around (keeping your original prompt)

[rapid ermine]

one man walking alone ¯_(ツ)_/¯

[brazen valley]

what is a good model for photorealistic images?

[potent dirge]

Hi everyone, we've put out a new algorithm to discover prompts from images (so another way to do img2prompt, if you've seen those tools before!), we have a demo here: https://huggingface.co/spaces/tomg-group-umd/pez-dispenser. We hope this helps in finding stable diffusion words for stuff that is hard to describe :>

[terse veldt]

Guys, is there anybody who know how to type prompt in right way? I need to get prompt with angle of cam which looks from ground and up and high fov

[blazing quartz]

There are several on civitai. I'm using realisticVisionV13 as my default. Keep in mind many are nsfw and you'll need to provide negative prompts to stop nudity if that bothers you.

[upper bane]

Any tips for getting your trained model to have it's head in the shot?

I've tried loads - like Full body, full body shot, head, head in shot, and negs with cropped, out of frame etc. but maybe 1 in 30 images have the head in shot.

[frosty bison]

Is there a way to help define words like hobgoblin (having a myriad of sources and meanings) to have a more singular origin or focus? I'm getting neat output but the hobgoblin is always crazy off.

[blazing quartz]

simplest way is probably train it with your own reference images and call it "myhobs" or some such

I suppose you can try playing with style or origin prompts and see if they make a difference

[soft mural]

Girl

[compact forum]

is there any good way to increase the 'prompt tokens' or whatever it is. Something to make it acknowledge and force it to generate content using all prompts?

[blazing quartz]

That's usually the CFG Scale, although setting it too high will likely cause undesirable effects

different systems have different ways of stressing a particular prompt

[compact forum]

i see. hmm

I noticed it likes to ignore ones further down the list in the prompt order

and having a lot of issues keeping a set style for one character. changing prompts not related to the character itself seems to be altering how they look. by a lot. might look into going back, using the same prompts as then to generate a ton of photos and then using just the ones that have them looking exactly how i want to try to train a LoRA to do that 'style'

and what's "model keyword" extention for the webUI? it seems to have an assortment of choices to pick from for how it handles prompts. but i don;t understand them

keyword prompt. prompt keyword. keyword, prompt. prompt, keyword.

[smoky totem]

Does anyone know if quotation marks work in prompting to specify a subject? I tried to issue the prompt "cauliflower jellyfish" for a specific type of jellyfish, and on negative prompt I wrote cauliflowers. The images that come out all feature jellyfish with some cauliflowers attached to them, not how the specie is supposed to look like.

I also tried cauliflower-jellyfish on positive prompt but it either gives me a typical jellyfish or again cauliflowers stuck on jellyfish, completely misses the subject

CFG scale tried from 8-15-2x(way too high)

using auto1111

standard SD1.5 and SD2.1 models. I think I tried realvision model too which is 1.5 based.

[slate prairie]

You could try different names for it. If you ever want to find hints on what to prompt for, you can search the LAION dataset

[slate prairie]

https://laion-aesthetic.datasette.io/laion-aesthetic-6pls/images?_search=crown+jellyfish&_sort=rowid

[hollow shale]

Why are the channels locked?

[slate prairie]

what channels? the bot ones?