📝｜prompting-help

[red slate]

Hi! Does anyone know how to go about making a functioning QR code with ControlNet?

[next flint]

I believe all you would do is get a qr code from an online generator and then put that in using canny

[unique wind]

how do I make puffy cheeks expression?

like this

[next flint]

has just putting puffy cheeks into the prompt not worked well?

[tired ravine]

hello, i found this prompt online: priest, blue robes, 68 year old man, national geographic, portrait, photo, photography –s 625 –q 2 –iw 3
does anyone know what the parameters at the end mean? the ' –s 625 –q 2 –iw 3 ' they do seem to make the image good so i'm wondering what they do / mean

[atomic flume]

most of hte image here keeps coming out blurred any suggestions on how I fix that ?

(Rembrandt lighting:1.2), (from_bellow:1.3), (a character portrait:1.1), art by artgerm and greg rutkowski and magali villeneuve, analog style, (best hands:1.2), (full body shot:1.4), (milf:1.2), sultry woman in a (white night gown:1.3), posing on a bed, red high heels, (auburn hair:1.4), red lips, (hourglass figure:1.2), (big boobs:1.2), (left leg on top:1.2), thick thighs, (stockings:1.4), (seductive Pin up pose:1.4), sultry face, sultry pose, highly detailed, sharp focus, Art Frahm, comic cover art, (pin up art:1.4), (art by alex ross:1.4), Altichiero, concept art, realistic character concept, realism, character design, character portrait, masterpiece, detailed face, painting by Ed Blinkey, Atey Ghailan, Studio Ghibli, by Jeremy Mann, Greg Manchess, Antonio Moro, alphonse mucha, trending on ArtStation, trending on CGSociety, Intricate, High Detail, Sharp focus, dramatic, photorealistic painting art by midjourney and greg rutkowski, (3d, unreal engine:1.2), <lora:PerfectFullBreasts-fCV3:1>

Negative prompt: (bad-image-v2-39000, bad_prompt_version2, bad-hands-5, EasyNegative, NG_DeepNegative_V1_4T, bad-artist-anime:1.2), (cleavage:1.5), canvas frame, ms paint, ((disfigured)), ((bad art)), ((deformed)),((extra limbs)),((close up)),((b&w)), weird colors, blurry, (((duplicate))), ((morbid)), ((mutilated)), out of frame, extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), ((ugly)), ((bad anatomy)), (((bad proportions))), cloned face, (gross proportions:1.2), (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), (fused fingers), (too many fingers), (((long neck))), Photoshop, video game, tiling, poorly drawn feet, (cross-eye:1.3), body out of frame, bad art, 3d render, watermark, letterbox, lowres, (error body), error hair, ((error arm)), ((error hands)), ((bad hands)), error fingers, bad fingers, missing fingers, error legs, bad legs, multiple legs, error lighting, error shadow, error reflection, text, error, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, username, ((error eyes)), ((bug eyes)), ((bad eyes)), bad mouth, error mouth, (error face), (nsfw:1.2), to many legs, (three legs:1.2), logo, water mark, black and white,
Steps: 60, Sampler: DPM++ 2S a Karras, CFG scale: 7, Seed: 944608653, Face restoration: CodeFormer, Size: 564x760, Model hash: b76cc78ad9, Model: dreamshaper_6BakedVae, Denoising strength: 0.6, Clip skip: 2, ENSD: -1, 

[spice orchid]

I'm trying to generate just an upper body portrait but can't seem to get it to work. these are the prompts i'm using and then the image i get

Positive Prompts: cheerful cute anime catgirl, beautiful face, cat ears, long pink hair, detailed blue eyes, black crop top and booty shorts, black collar, curvy body, cleavage, Anime, 2D, Masterpiece, Digital Art, Colorful, highly detailed, high quality, 4k, white background, symmetrical, facing forward, upper body,

Negative Prompts:3d, b&w, bad anatomy, bad art, bad hands, blurry, child, children, cropped, deformed, disfigured, disgusting, error, evil, extra digit, extra limb, fewer digits, glitch, gross, horrible, jpeg artifacts, kid, low quality, lowres, missing digits, missing fingers, morphed, mutated, mutation, mutilated, normal quality, old, poorly drawn, poorly drawn face, poorly drawn hands, scary, signature, text, two heads, ugly, watermark, weird colors, worst quality, poorly drawn arms, weird arms, duplicate arms, tail, lower body, full body,

i'm trynna get just an upper body shot kinda like this

[next flint]

If adding tokens relating to showing more of the upper body to the prompt isnt working you could always use controlnet with either a high weight for an exact frame you want to copy or a lower weight to just tell it to include the upper body

[spice orchid]

thanks. any idea what to add to the prompt so i get consistent images with the arms down? most generations have the arms up in the air and i don't want that

any good tutorial links for it?

i think i can get thru an install ok

ty i'll check them all out once i'm done instaling it

i'm downloading all the .pth files from here https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main do you recommend anything else?

[next flint]

yeah the open pose editor has a website that does the same thing but it runs much better in the ui

[obtuse torrent]

@spice orchid for anime tagged models ,,, controlnet, is the best option, ...the tag "long hair" frequently leads to an "S pose" and a "side view", "Curvy" also influences that sense,...... "arms at sides", "v arms", + a lucky seed, can give you a litle help... but a guided generation is better..

[spice orchid]

ok i installed it and got it to work but it like overrides my prompts.

it generates full bodies now, even when i tell it not to

these are my control net settings

[spice orchid]

nvm i fixed it, i just had to remove the lower part of the body in control net thingy

[quiet zodiac]

Try...

facing forward well lit shoulder shot...

The square aspect ratio helps with closer shots, use the taller aspect for full body.

[silver valley]

The trick is sometimes to say portrait of ...
And changing the resolution to 512x768 helps a lot

[river trellis]

hello everyone,
I'm basicall trying to make a person "hug" a giant star. I made a lora for the mario star but all my attempts end up in either hugging like weird objects or the character hugging nothing and the star is floating around elsewhere.
Increasing the weight doesn't seem to help anymore. Not sure what I could do.
Can anyone help me out here?

[quiet zodiac]

Try a different tense, instead of hugging, try hugs. Also look up synonyms. Embraces or embracing might work. holding..?
Sentence-like prompts might also help instead of lists.

... sits on a bed embracing a giant myra star...
... sits on a bed with her arms wrapped around a giant myra star...

[leaden socket]

Suggestions on getting a perfectly sphere ball?

[quiet zodiac]

Ask for a perfect sphere and add negatives if it doesn't render correctly.

[tardy cedar]

Hey, would anyone be able to assist in getting a prompt to produce similar results to this? (including background and overall atmosphere. I'm relatively new to SD and I'm learning I gotta be quite specific compared to MJ

https://www.tiktok.com/@pxl.pshr/video/7237729089002687790?q=pxlpshr&t=1685762664308

Also can I add negative prompts with deforum?

[fervent swan]

Prompt

#imagine

/ imagine prompt

[leaden socket]

Dudes IM bot got lost.

[real juniper]

hey guys, I'm generating an anime character with stable diffusion but lighting always creates shadows on thighs which I want to be visible. How can I fix it?

[tardy cedar]

Hey could anyone help me with prompting? I'm at a total loss

[obtuse torrent]

?maybe

[tardy cedar]

I'm trying to get results at least similar to the video I posted but not much I change in the prompts affects the the frames, This is the current prompt im using:

cond_prompt: vampire::5 with glowing red eyes::3 spinning fire, world bourning around him, hellish landscape, Dark clouds and moon in the sky,dark, ominous, threatening, haunting, forbidding, gloomy, stormy, doom, apocalyptic, sinister, shadowy, ghostly, unnerving, harrowing, dreadful, frightful, shocking, terror, hideous, ghastly, terrifying hyper realistic, 8k, trending on artstation

and this is what i tried before:

"0": "Bald vampire with glowing red eyes spinning fire while the world burns around him, style of Anato Finnstark, hyper realistic, 8k, trending on artstation"

(Left is original frame and right is run though SD)

Both prompts are giving pretty much the same result, I have absolutely no idea how to move towards the direction of this video
https://www.tiktok.com/@pxl.pshr/video/7237729089002687790?q=pxlpshr&t=1685762664308

any tips or insights or suggestions?

[jolly wolf]

hi everyone, i intend to generate landscapes without any people/women. despite having detailed negative prompt, i tend to get people or waifus on images. please would someone have some suggestion how to get rid of it?
Here is my prompt: (((masterpiece, best quality, ultra detailed:1.3))), (anime illustration of tropical landscape, scenery), sunrise, rainforest, tropical, jungle, wild flowers, palm trees, ferns, lush foliage, green leafy plants, psychadelic, (iridescent color palette, rainbow colors), ((colorful:1.3)), makoto shinkai, beeple, rhads, (bright, airy)
Negative: (((1girl, solo, girl, woman, person, people, crowd, waifu, human, character, female:2))), (bad quality, low quality, worst quality:1.3), ((dark, dark colors, black)), bad-artist-anime, bad-picture-chill-75v, EasyNegative, FastNegativeEmbedding, verybadimagenegative_v1.3

[quiet zodiac]

Try using the Redshift model. It makes beautiful, people-free landscapes with minimal effort.
https://huggingface.co/nitrosocke/redshift-diffusion-768
POS:

NEG:

fog, blurry, soft, ugly, mangled, bad, disfigured, worst quality, low quality, normal quality, jpeg artifacts

[orchid ore]

@jolly wolf try use just one system. () (()) or (something:1.5)
This (((1girl, solo, girl, woman, person, people, crowd, waifu, human, character, female:2))) must really confuse SD 🙂

[jolly wolf]

oh okay! thanks for the tip, i wanted to really emphasize that i dont want any woman 😄 i didnt know that both systems cannot be combined

[orchid ore]

it probably can be, but for sure it is too strong :2))) that it can behave weird imo. @jolly wolf

[simple finch]

hi all, so I need a quick help, I'm trying to generate some pictures of a whole planet in space, I'm using A1111, and writing a Whole planet in the prompts isn't working

any tips?

I'm getting something like this

my Prompts;
draw a picture of a ((full planet)) floating in space

I know it's a short prompt but I tend start with few sentences and build up

[simple finch]

I got around it with controlnet and an image file with just a circle

[obtuse torrent]

try "no humans" "no characters" in positive, .. (IE...start prompt whit "Landscape, scenary, no humans") in positive (sometime it works...tagged models) ...you reference ... makoto shinkai ...the IA may be paying attention to the heroines and not to the landscape photography

[coarse karma]

Hi,
Is there a list of artists that are used in prompts, along with the art styles?

[coarse karma]

try writing the description of the planet you want. For example A giant red planet. or a dwarf planet. Something like that

[obtuse torrent]

@simple finch .. add tags "full view", "solo" ... add to negative anything that appears and you do not want (I.E. buildings, constructions, objets, spaceships, satellites, orbital stations, etc)... deep space .. can help in some models.. check

[noble adder]

Hello Every one, I'm working on a personal project I hope to achieve. I've been out of the community for the past 6 months so things have moved fast and I'm just catching myself back up to speed.

I'm working on a series of starcrafts or starfighters that are merged with classic sports cars, super cars, and brands, rendered in isometric concept design style. What you see above is what I've achieved after one evening of play. I'm getting decent results but not great. I have a few questions on how to fix a few things.

1. keeping the vehicle one piece and in frame, I don't want extra secondary designs. I've tried various negative prompts to no avail.

2. I got rid of wheels with negative prompts but in positive prompts I can't get wings.

3. the car model influence is't as strong as I would want it to be.

4. A pure white background.

5. the entire vehicle in frame.

my Prompts examples

A starship racer modeled after the Porsche 919, X-wing, tie fighter, fins, tails fin, race paint job, starship, isometric illustration, detailed, concept design, Detailed and Intricate, Line Art, Geometric, Concept Art, vector art, white back ground.

Negative
no color in background, only wings, only on subject, no wheels

A starfigher designed by Lamborghini, tail fin, attached wings, race paint job, isometric illustration, Detailed and Intricate, Line Art, Concept Art, vector art, background white, single view, one starship, 3d render,

First one is bugatti, second two are the porsche 919. Results have potential but I can't seem to get further than this or get to far. It's a tough on to wrangle.

[simple finch]

I actually tried and kept getting similar failures, I'll keep trying in negative prompts

[simple finch]

Thanks I'll try more with the negative prompts

[jolly wolf]

thanks so much, will try!

[jolly wolf]

ive ran XYZ plot with several models with this prompt and some of them tends to produce women some not

please does someone know if this image stacking can be somehow avoided with the prompt?

[quiet zodiac]

@noble adder Generally when you see lots of double stamping results, your target resolution has exceeded your model's source resolution. Try matching your render size to the model size. (i.e. 512x512 with many models). Try supplying an image guide of a crudely photoshopped car with wings on a black background. That solves a couple of your problems, such as keep everything in frame and eliminating background noise.

@jolly wolf Stacking can also be a symptom of your output resolution exceed the input models sample size. Try reducing the size.

[quiet zodiac]

Here's a few spaceship/station guides I generated at the 768x576 resolution. I wanted my result to appear on the right-hand side of the image output. Matchnig your image guide size to your output size helps a lot.

[jolly wolf]

will try, thanks

[brave leaf]

im trying to make toph from the last airbender and idk how to fix the eyes

im using these negative prompts "ugly,tiling,disfigured,deformed,low quality,pixelated,blurry,grains,grainy,text,watermark,signature,out of frame,disproportioned,bad proportions,gross proportions,bad anatomy,body out of frame,duplicate,cropped,cut off,draft,extra hands,extra arms,extra legs,extra fingers,extra limbs,long neck,mutation,mutilated,mutated hands,poorly drawn face,poorly drawn feet,poorly drawn hands,missing hands,missing arms,missing legs,missing fingers,fused fingers,unnatural pose,out of frame,low resolution,morbid,blank background,boring background,render,unreal engine"

[obtuse torrent]

using Img2Img to scale by 2, based on your image and prompt ....+ small correction (you omitted an "s" ... eye ---> eyes) + adding "detailed" tag to prompt..

[brave leaf]

can you send the negative prompt cause I'm confused with the "slight correction (you omitted an "s" eye for eyes)"

[obtuse torrent]

@brave leaf it have metadata ...use png info tab ...

[brave leaf]

Alright 👌

[brave leaf]

@obtuse torrent no parameters

[obtuse torrent]

@brave leaf ..nop they have.... right click ...choice open link .. go to browser tab ... dowload ...use png info tab in downloaded file .....

[brave leaf]

still confused@obtuse torrent

[nimble fox]

hello, I need help with a prompt, I want to create a faceless doll, like in this photo.

I try many prompt and can't get a empty face like in the photo

[obtuse torrent]

...mmm.... man not download of discord ,,,, open the image in your browser and download there ..

@brave leaf

[nimble fox]

@obtuse torrent veo que hablas español si puedes ayudarme o tienes idea donde tengo que buscar, te lo agradeciria

[obtuse torrent]

@nimble fox shows what you have so far, prompts and the image that is generated or the image with metadata...

[nimble fox]

ok

[obtuse torrent]

@nimble fox
this way if I can't help you maybe someone else from the community will answer

[nimble fox]

this is the image, but I want with no face, like the second picture.

@obtuse torrent

[obtuse torrent]

@nimble fox ?? it is ceramic? you know?

[nimble fox]

yes

@obtuse torrent

I just need to have a clear face and that would be

[obtuse torrent]

@nimble fox

[nimble fox]

no me salen lo prompts

I opened on the browser

and download it from there, but there is no meta data

[obtuse torrent]

@nimble fox corrupted?

@nimble fox

parameters
a statue in the shape of a woman without a face, without features, without eyes, without a mouth, without a nose, in a blue dress and hat holding a pink flower in her hand and a red hat on her head, Negative prompt: ((Face, eyes, mouth, nose:1.8)), long body, lowers, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality ((((ugly)))), (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))). out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))) Steps: 20, Sampler: Euler a, CFG scale: 12, Seed: 3238185597, Size: 512x512, Model hash: db2c51c333, Model: dreamshaper_4BakedVaeFp16, Version: v1.3.2

@nimble fox
there is a possibility that some tag in the negative prompt will be detrimental...like bad anatomy, deformed, muted... you should check the effect one by one

@nimble fox try too: a statue, woman shape, brown ceramic, ((no facial features)), ((without a face, without features, without eyes, without a mouth, without a nose)), full body, frontal view
wearing a blue dress, red hat,
holding flower, pink flower

[tacit valve]

Sorry for late reply and thank you for your advice.
I'll try to improve my prompt from your advice. Thanks again!

[nimble fox]

Thanks a lot I will try later today!

[spice orchid]

how do i fix stuff like the hands and stuff?

[tired vigil]

I'm trying to generate a sorcerer character, and in my prompts i specify the color of the fair but it just really doesn't seem to want to get the right hair color "which is gray by the way" does anyone know how to get hair color right?

[gilded parcel]

How do I get dark skin tones?

[orchid ore]

@tired vigil have you tried put more attention on it? (((gray hair))) ?

[quiet zodiac]

@gilded parcel Specify ethnicity for your character or simply state it near the front of your prompt...

...beautiful brown skin...

[gilded parcel]

Thanks, yeah I get brown skin a lot, I've tried African, skin colour/tone etc etc but can't get the black skin tone that I'm looking for. Something like the attached.

[quiet zodiac]

Maybe one of your negative prompts is working against you? Try eliminating all negatives and building it up one token at a time.

[gilded parcel]

Good shout. I'll give it a go. Cheers

[quiet zodiac]

Google tells us...

From his research, he's found that people from Senegal and the islands of Micronesia have some of the darkest skin tones in the world. 

Perhaps add that keyword as well...

[polar glade]

One trick is to find the name of a celebrity with the correct skintone

[ripe rain]

guys if you wanted to do a shot of jason statham looking straight into camera and hos upper body from chest upwards visible, what would be the prompt?

[quasi skiff]

What does a double comma do in a prompt? I've seen it a few times, <lora:VampiricTech-20:0.8>, vamptech, blood,scifi,vampiric, ,utopia,city, urban area, over clouds.
Notice the vampiric, ,utopia.

[obtuse torrent]

nothing..is only a error..

[quasi skiff]

hahaha thanks

[obtuse torrent]

@quasi skiff space" " is neutral "," same... normally no affect .

@gilded parcel it depend of models ... in realistic models ..ethnic groups can work, nationality too, if is a mix ..or anime models, tags can help,, IE "Darkskin" , in open clip models ,,any descriptive sentence can work ..

[gilded parcel]

Have you come across any Loras for skin tones?

[obtuse torrent]

mmmm @gilded parcel civitai have near 12.. i think... lora, Lycos, etc (never checked).. but Lora (specific trained models) are a help ...and a restriction ... at same time

[gilded parcel]

Yeah I have mixed luck with them for sure. Thanks 🙏🏽

[obtuse torrent]

@tired vigil Normally the colors are filtered and affect the whole composition, if hair color is really important to you, declare it first (at the beginning of the character description list, then mention characteristics that are not related to color, and then mention the following thing with color, and repeat the process) in the new descriptive clip models it is not so necessary... if the model you use recognizes anime tags you can use : white-grey-silver (+/- same visual effect), and use () ... IE ,,,,(grey hair:1.2) or ((grey hair))...

[leaden socket]

It can only output what it was trained on, you could make a train checkpoint and release it?

[polar glade]

Hey, so I noticed that art by artgerm and ruan jia and greg rutkowski surreal painting makes it almost impossible to get paintings of babies. How might I get a similar art style, but with a baby consistently?

the rest of the prompt, for example:
close up solo portrait of baby infant 1 year old female, intricate, elegant, highly detailed, majestic, digital photography, plain white background

^ this works fine, but if it's in the style of those other artists, the subject becomes older

I'm guessing Greg Rutkowski doesn't paint many babies, or something

[gilded parcel]

I thought about that. As well as the 120 other ai things I think about daily. I’m gonna look into it this week for sure. Thanks for the reminder

[crude flume]

How do I prevent this image split?

Nvm I've found it in the chat

[leaden socket]

Do 512x512 then upscale

[crude flume]

Yeah, but I need 9:16 aspect

[obtuse torrent]

@polar glade (very large :D) ...
It's an interesting topic that you touch on, basically many of the artists that are commonly cited as a reference in the prompts have nothing in common either with the "styles" or with the "themes", if you use google and find that some of them lived more than 200 years ago and there are things they never knew (IE anime)... however if a "critical" amount of AI creations, which are tagged with their names... start to form...it will end up becoming a real influence on generations?... its original sense of reference can be "renowned oil painter", "renowned portrait painter", but people copying prompts end up using them to generate "a landscape with cat girls running through it with vivid colors and 8K quality".... 😄

@crude flume ...432x768 .... or 288x512

[polar glade]

I guess the issue is that these painters don't paint babies/children very much. And so "adult" is sort of programmed in. Are there painters known for painting children perhaps?

[obtuse torrent]

@polar glade IDK ...unfortunately I must admit that in painting and photography I am only one step ahead of a complete ignorant... however, if any author is referred to this subject by the search engines, it is likely that it will serve as a guide... alternatively you can search for similar images generated by IAs, and observe which prompts were used..

[vale sand]

what do you guys think is the best way to make an OC using SD? been thinking on maybe mixing gear and characteristics of other characters, it wouldn't create anything new but i guess it's the closest thing i can think of.

[humble ether]

Hello, I'm trying to generate an image with two characters however, their prompt keeps mixing with each other ie their clothes or hairstyle swapping with each other. Anyway to work around this?

[fringe scroll]

interested in this too. specifically when doing img2img, i'd like to be able to maintain certain things about the people like gender & ethnicity

[ripe rain]

guys if you wanted to do a celebrity that looks like the real one, how would you do it. my generations look close but not 100%

[quiet zodiac]

@humble ether That's a tough one, but try explicitly defining each character, separating them by an action. Don't use just man/woman, give them a title or a role. I had some success with a prompt like this...

a void with flowing purple (dripping:40) slimy tentacles surround a muscular 36 y.o. italian husband wearing a suit (kissing:20) a 26 y.o. african mother wearing a sequin dress

[spice orchid]

use a lora trained on that specific celebrity

guys how do i achieve this type of pose? i can't seem to get it right

maybe with controlnet? but what preprocessor and model would i use?

[silver valley]

Did you tried arms behind back, side view, ?

[hollow tapir]

controlnet + open pose.

[spice orchid]

how do i make sure the character's hair is fully in frame?

[tired vigil]

looking to the sky at the sunset with water

[quiet zodiac]

@spice orchid Try something like...

facing forward well lit shoulder shot

at the start of your prompt.

[spice orchid]

ok will do

[spice orchid]

how do i get straight hair and not this mess

current prompt

positive: facing forward well lit shoulder shot, portrait, full body portrait, curvy body, thighs, wide hips, 2D, Masterpiece, Digital Art, Colorful, highly detailed, high quality, 4k, highres, white background, simple background, simple white background, symmetrical, lower body, 1girl, single person, solo, girl, breasts, cleavage, pink hair, straight hair, cat ears, no deformed legs, black top and jeans, sideview, left hand behind back, left hand grabbing right arm, perfect hands, perfect face, high quality hands, full body, hair in frame, standing up, side shot, looking out of frame, shoulder length hair, lora:Trush-A:0,

negative: 3d, b&w, bad anatomy, bad art, bad hands, blurry, child, children, cropped, deformed, disfigured, disgusting, error, evil, extra digit, extra limb, fewer digits, glitch, gross, horrible, jpeg artifacts, kid, low quality, lowres, missing digits, missing fingers, morphed, mutated, mutation, mutilated, normal quality, old, poorly drawn, poorly drawn face, poorly drawn hands, scary, signature, text, two heads, ugly, watermark, weird colors, worst quality, poorly drawn arms, weird arms, duplicate arms, tail, blushing, blush, multiple people, weird shoulders, poorly drawn shoulders, 2girls, deformed legs, poorly drawn legs, complex background, multi colored background, out of frame hair, flowing hair,

[obtuse torrent]

@spice orchid add () (straight hair, hair in frame)... try adding croped,messy hair,waving hair,floating hair to negative prompt..... (try whit others models too)...
(these images are generated with your positive prompt, unchanged, and reordered, but using another model.. and without the Lora)..

[spice orchid]

what model are you using?

[obtuse torrent]

@spice orchid Steps: 36, Sampler: DPM++ 2M Karras, CFG scale: 7.5, Seed: -1, Size: 504x896, Model hash: 0a03c339c7, Model: divineelegancemix_V5

[polar glade]

Pretty much no

[silver valley]

A regional prompter extension can solve this.
Regional prompter or Multidiffusion

[noble charm]

any good prompts for archaic "ancient" attire that is stronger then just putting ancient linen tunic?

something along the lines of this

[tidal dock]

Help with the promt. I want to make a noble coat of arms on which will be a grail with blood. can not do as I want. I need about this result (2 photo)

[noble charm]

I reccomend trying to generate a blank coat of arms then using inpainting for the grail

[leaden socket]

Best upscaler?

[bright grove]

hi guys any1 knows how to create a guy with slim body, whenever i input prompt
1guy age 21, ((slim body)), skinny body,
it still creates muscular body or muscle definitions. I am using cardos Anime model. Any help ??

[quiet zodiac]

Look up synonyms. Try emaciated or slender perhaps..?

[gloomy kelp]

Is there any way to adjust the "camera"/viewpoint of the image? Like where you'd be looking from above or below. I've tried "view from below" and "low angle view" but I can't find the right prompt to make it consistent

[gusty coyote]

how important is spacing for prompting?

nevermind

[orchid ore]

@gloomy kelp i noticed it also vary by models.

[gloomy kelp]

That's unfortunate, guess I'll just experiment

[gusty coyote]

I have a lora for obsidian black skin, but have noticed that only some models are able to understand it. They keep giving me brown skin, no matter what I put in the negative. Any ideas on which models are best for alternate skin tones that dont include anime

[quiet zodiac]

Here's one using SG161222/Realistic_Vision_V1.4. Try supplying an image guide with the skin tone close to your goal. Then you can get away with a simple prompt and no negatives.

facing forward, a well lit shot of a fashion model wearing a yellow dress

[rose canyon]

does anyone know how to make a prompt use 1 prompt for steps 1-x of the sampling steps, use a different prompt steps x-y and a thrid prompt y-end? i found some guys talking about it on youtube but they dont show the syntax

[silver valley]

Thats doable with the Prompt editing feature of auto1111 webui:
All info here:
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#prompt-editing

[rose canyon]

cool. the conversation was how you can use thiis to create compositions early on from goofy prompts that match exactly what composition you want. e.g. "a top down view of a bowl of soup" then switching it to a bath, or a hole in the ground, or anything really

and you get a circular centered main subject

[silver valley]

Okay interesting. If its doable with this show me an example 🙂
I didnt tested it much but for stuff like that it would be cool for sure

[rose canyon]

sure thing ill try to grab an image or 2

the prompt on ALL of these was very simple, and the exact same
[top down a bowl of soup : : a beautiful woman taking a bath while wearing a swimsuit , top down view, night time, warm lighting, candlelight :6]
Negative prompt: easynegative

@silver valley

[a beautiful woman in a city, 1girl, yellow sundress, full body, portrait : a close up of a tripple scoop ice cream cone :4]
Negative prompt: easynegative
Steps: 28, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 610140200, Size: 512x512, Model hash: 7c01e48033, Model: meinahentai_v3, Denoising strength: 0.7, Hires upscale: 2, Hires steps: 14, Hires upscaler: R-ESRGAN 4x+ Anime6B,

[leaden socket]

When training, what's the census on training 1024x1024??

[quick sparrow]

Can anyone suggest how to tune my prompt; I'm trying to get an image of a man fishing from a data lake. The system gets the man fishing right, but I can't seem to get it to make the surface of the lake look digital. I've tried data lake, digital lake, lake surface made of binary data. They all just end up as normal water style lakes.

[noble charm]

inpainting might help?

[silver valley]

Ah very nice!

[normal aurora]

Has anyone had luck with negative prompts that eliminate the “double” person output? In doing some research online, it sounds like at 1024 you have a greater likelihood to getting it, which I don’t understand, but for the sake of quality, I want a larger output. Thoughts?

[silver valley]

SD 1.5 based models are trained on a resolution of 512x512 and SD 2.1 based ones on 768x768
Going higher as that and you get doublicates and artifacts more likely

The way to get high resolution images is to generate at lower Resolution and then upscale

This can be easily done with the Highres Fix or the sd upscale script

[normal aurora]

Ahhh, that makes sense. Now, if I’m using a private model I trained (on 512), can I retrain my model with 1024x1024 to fix this? Or would the rest of the trained data still impact it?

[quiet zodiac]

Another approach is to use an image guide with just one person placed in the scene where you want them. Match the image guide to the same resolution as your output. Certain seeds will always give you doubles, however, so don't forget to change that when you're stuck.

[normal aurora]

768 seems to be working right now

[quiet zodiac]

That's great. Recently, I've been able to push a 768 model as high as 1280x768.

[normal aurora]

Coming from midjourney I’ve been spoiled with higher res haha

[silver valley]

You can train it on higher res sure. Would suggest 2.1 as base model then

[silver valley]

Try 512x768 and then activate highres fix.
Hires steps to 10.
Denois to 0.4
Upscaler Esrgan4x+
Upscale by 2

For example

[normal aurora]

High res fix is a script?

[silver valley]

Its a setting

In the automatic1111 webui

In txt2img its a checkbox

[normal aurora]

Okay, I’ll check it out, thanks

[pearl belfry]

Does anyone know how I might go about consistently generating this cloak on the character? I tried ControlNet Reference_only but even that will scramble the details. Any ideas?

(the rest of the outfit is quite consistent to generate. It's really just the cloak)

[solar dust]

Where exactly do I put the nohalf argument?

This is what it looks like rn.

It doesn't tell me where the nohalf argument should be so I tried putting in different places but that led to errors.

[silver valley]

After --medvram and also add --xformers --autolaunch

[solar dust]

Like this?

[silver valley]

No seperate them with a space

[silver valley]

Copy this:
--medvram --no-half --xformers --autolaunch

[silver valley]

You typed nohalf instead of no-half

[solar dust]

It's working so far, thanks a lot.

[silver valley]

No Problem

[normal aurora]

LoRA model question. If I’m using the v4 DreamBooth API, is there a way to find all the LoRA models available? If I filter on LoRA in the models page for stablediffusionapi.com, most of them come back as invalid lora_model. I’ve found a few that work. Also, any way to use the Civitai.com models through that the DreamBooth v4 API?

[solar dust]

@silver valley For some reason stable is generating the images but not allowing me to view them once it's finished

why is that?

It wasn't acting like that a few minutes ago.

Nevermind, I think I fixed that.

[silver valley]

Okay what fixed it?

[solar dust]

Enabling image save.

[pearl needle]

What's the best way to get darker skin? Like tan to dark brown.

[silver valley]

Depends on the model if dark skin or tanned work.
There are also embeddings and loras for getting dark skin

[pearl needle]

Ok. I'll hunt for some in Civitai. Thx!

[grizzled steppe]

anyone know this model? been looking for something similar

[solar dust]

I'm trying to use img2img to generate new results

but instead of doing that based on the image I give stable, it just repeats the same image over and over

why is that?

[spiral belfry]

Did you happen to set demonising strength to 0

[crude crystal]

Not sure if this goes here but because of the Reddit shutdown I cannot find guidance on frame to frame consistency. Can anyone please advise me how to do this better? I managed to get slightly better results than that but they might be too nsfw (revealing dress, no nudity)

[hollow tapir]

depends entirely on your workflow, pipeline. Did you use deforum, converted video to img and batch img2img, etc

[wheat harbor]

hello i want to get rid of this shadow or even the lighting so my image will look like just 2d art without light source, what prompt can i use for it? i've tried dim lighting, backlighting, sidelighting, perfect lighting, but the light source still persist 😦

[blissful sapphire]

quick question, does the placement of loras matter? and should they be separated by comma. Or does it not matter?

[crude crystal]

Batch img2img, is there a better way? I'm watching lots of tutorials but there isn't infinite time and I'm hoping someone has an answer. Thanks for the response

[hollow tapir]

Here's a non exhaustive list of way/extension/cript to work on videos :

• deforum
• SD-CN-Animation
• ebsynth
• mov2mov
• vid2vid

[wheat harbor]

Erm is this for someone else?

[hollow tapir]

yup I messed up ^^"

[hollow tapir]

It was meant for you.

using the batch img2img approach I'd say the easiest way to increase coherency is to play around with the denoiser value.

[crude crystal]

Cheers I'll try that when i wake up, I'm also thinking that the type of prompt helps. Woman in front of blank background is easier than in a garden or street

[solar dust]

Demonizing?

🤔

[spiral belfry]

Typo I meant denoising strengty

[silver valley]

Do you want something like this?

[midnight sluice]

I want to create a character for a visual novel using A111 + RevAnimated. I already have a fullbody portrait I like and now I want to create variations of this character in different poses and even more important: different facial expressions. Later on I'm intrested in different clothings, but the facial expressions are most important. Can someone help me to achieve that? Thanks in advance 🙂

[finite ermine]

all I want is this image staying the same, just wider to fill a 3440x1440 monitor.

I cannot get inpainting to work and anything else just changes the image to different character

[orchid ore]

Realesrgan 4x + Anime6B for example do not change content. Unfortunately keep AR it cropped it automaticaly bad, but i think it is posible that way.

so wrongly cropped but i think with more patience it can work.

[finite ermine]

I am trying, but I still need it to impaint properly, I am trying to get openoutpaint to run

but openoutpaint I am not sure it can do it with image that is not ai generated

[orchid ore]

you can open it, and it should work

[silver valley]

Openoutpaint extension needs --api to work

[finite ermine]

done

[finite ermine]

what model would match the style of the image I sent

it is like waterpaint

acrylic

[orchid ore]

watercolor i think

[noble charm]

anyone know some good tricks to get photorealistic animal ears to work?

like catboys/girls etc, since they are a canon fantasy race in one of the worldbuilding things I am working on

there was a final fantasy miqote lora I found but the one for generating male ones seems to only generate one specific character

nvm just remembered inpainting exists

anyways, anyone know any good loras/embeddings for spears/swords that actually look good?

[golden cove]

Is there a way to neg prompt faces and humans in general? Been trying the usual keywords with no luck.

[shut falcon]

Looking for models/loras good for creating "item card" art in the style of (or at least close to) The Griffin's Saddlebag. Anyone have any good suggestions?

[atomic flume]

Anyone know how I can make a bulletproof vest without the pouches ?

[narrow bane]

How do I prompt an image to have 2 of the same characters in it without it just generating one character?

[atomic flume]

Try 2 Charcaters, 2 girls or 2 boys something like that

[narrow bane]

Ok

How do I specify one look for one character and another for the other

Because I'm doing a pic for Zelda from BotW and TotK

[atomic flume]

just describe both and let the AI pick which is which or use control nets

[crude crystal]

That's deforum right? I'm not so interested in that zooming in to an image effect, I'm very interested in flawlessly replacing an actor with something else. Take video of my friend DJing and replace him with a robot, or like that girl at the psytrance party replaced with a vampire version of herself.

[silver valley]

Nope thats not with deforum.
But what you then need is video2video or mov2mov extension

Thats for replacing

[crude crystal]

Thanks, I tried img2img batch so far, I'll find vid2vid once I'm ready today.

[crude crystal]

This... capability. Its so cool. I can't wait to see what movies get made like... next year

[silver valley]

Yea maybe there are even easier ways to do this than now 😄

[barren spire]

Has anyone found a way to plot the generated character twice in one image? Trying to see if I can make a comic panel

[crude crystal]

I wanted to replace these women with Gundam (the male fantasy is disgusting), but I can't figure it out. I've tried inpainting and inpaint sketch. I've tried high and low denoise, cfg, etc. I've set it to latent, inpaint whole image or just masked area. I've followed through this https://stable-diffusion-art.com/inpainting_basics/ and a few youtube videos. I'm misunderstanding something foundational I think...

[crude crystal]

Yeah the easy way is coming did you see this?! https://www.reddit.com/r/singularity/comments/148yl49/rerender_a_video_zeroshot_textguided_videotovideo/

[near hull]

Does anyone know a good prompt to get hand gestures next to person/faces? (like the above picture)

[obtuse torrent]

Nop... Img2Img, controlnet, or Depth ... guided generation, is the only reliable, libraries for openpose hand, or gestures in Depth.@near hull

for tag anime models ..the tag is "victory sign", sometimes "v" or "v sign" .. but it doesn't work always or at all well

[near hull]

Ah I see thanks! How would I combine depth with controlnet reference?

Basically I‘m generating a similar looking person off an image with controlnet reference and then how would I get her to make the gestures but keep the same face etc?

[undone rivet]

Hello everyone, I would like to request your help

I've been trying to make some portraits like in a professional photo studio

I want to make the effects of "gel color" lights behind of the person, but I can't get it

I've tried using these words within the prompt: beam lights, rim lights, ray trace but it doesn't work

Does anyone have an idea how to get that effect....?

I want something like the attached photo (I took the photo and is real 😁)

[lavish surge]

what are some effective prompts to make your character put down their hands?

[orchid ore]

how about hands in pockets? Somebody suggested it in past.

[brazen roost]

i think you can use controlnet for this, use the depth map.

1. get roughly the background you want
2. go to img2img and place the background from step 1.
3. go to controlnet and place your subject there with the depth model
4. place prompts and click generate

i've done that before, i think these were the steps i took

[undone rivet]

Oh ok, I will try, I am not an expert using control net yet....😅

[brazen roost]

watch this video: https://youtu.be/MDHC7E6G1RA?t=871

also, the timestamp is at the current time of what im talking about

you can see there the technique i used, very cool

[undone rivet]

🤩perfect, thanks a lot 🙌🏻

[brazen roost]

no problem G, keep it up

[lavish surge]

but what if I am not making the character to wear a clothes with pocket?

[orchid ore]

one have to try anything. I dont know, not need it 🙂

[lavish surge]

the momento when you tried anything and decided to ask someone else 😭

[restive mesa]

Okay hello

managed to create two working QR codes back to back but they are too much QR code basically

What should I change to make them be... well more image?

these are my current settings

second control net unit has a different model but tile, and has a weight of 0.7
shoudl i reduce those weights?

I would try bunch of options myself but it takes like 20 minutes for one image so

[silver valley]

20 minutes? Did you tried the different Commandline args?

[restive mesa]

no the

normal images are fine just taking few minutes etc

control net just takes too long

though I have to say that I noticed I was not able to install xformers

[silver valley]

Yea xformers only works for nvidia gpus

[restive mesa]

gotcha

got another one

It's... kind of better I guess?

But still too much QR code

[silver valley]

Do you use 2 controlnets ?

[restive mesa]

Used two till now,

[silver valley]

Would suggest to try only with one

Also did you tried the qr code controlnet model?

[restive mesa]

Got it, but it crashes at %35 for osme reason

I got the 2.1 model, and upgraded my model to 2.1 as well so shouldn't be a problem but will try it with the 1.5(or whatever the other version of the qr code control net model was)

[silver valley]

Yea try the 1.5

[restive mesa]

I already started another one with one control net but

god damn it is gonna take a bit lol

[silver valley]

When im at home i will try again some qr codes

Tile resample was the best so far. But i heard that controlnet brightness worked too

[restive mesa]

I have it

used it for all the pics I sent

now usin only tile though

[restive mesa]

doesn't work, but it is closer to what I want

[restive mesa]

did two or three with one, they didn't work

last one was made with two and works

[quiet zodiac]

@atomic flume Try another word, perhaps. Instead of bullet proof vest, try "seamless armored vest"? If you've already cast the image in steampunk, try specify another genre just for the vest..."modern seamless armored vest" or "medieval knights brest plate"

[quiet zodiac]

I'm making some progress with the term 'bas relief' but I'm still not sure how to prompt the tunneling z-space effect shown in the clean source. My farm scene offers only one level of relief.

[orchid ore]

@quiet zodiac have you iterrogate it?
And i think your relief got several levels but smoothly one to another.

[quiet zodiac]

What does that mean, "interrogate it?"
Here's me prompt:

facing forward, a well lit bas relief stacked holiday scene sculpture, by Charles Fazzino, vibrant pastel colors, highly detailed, cinema4d extrusion in a clean white space, soft shadows

And a new output. I'm really struggling to get that carved into the wall look.

[orchid ore]

i mean put that jpg into img2img and click on interrogate clip. My english is bad.

i got those images. but faaar from original. Probably when from top could look bit better.

[quiet zodiac]

Those are nice. I'm using Stable Diffusion through the MLOPs Houdini addon, so I don't think I have that option. It really needs to be flat carved into the paper for a book cover.

[orchid ore]

yes it reminds me book art. Art made of books.

[rich crow]

Bas relief art might be one of the rare times the controlnet for normals might come in handy

[quiet zodiac]

I have access to a control net, normals, z-depth in Houdini. But that requires a source image, right? Maybe I'll try supplying some colorama after effects still of descending circles...
I'm still getting interesting results, but I can't seem to nail down the prompt for that carved into z-space look.

[orchid ore]

i know it started with paper cut. will post it once again in A1111

[quiet zodiac]

Getting there by using an guide image.

[quiet zodiac]

A bit of a kaleidoscopic guide instead of solid colors.

[obtuse torrent]

@quiet zodiac .. I don't speak English... and I don't understand the concept well, does any of this sound like what you're looking for?

[quiet zodiac]

Thanks for the reply. Some of those hold promise. I'm thinking less graphic art colors, however. My image guide is working. The white area around the concept is most important to me. After that, happy accidents are welcome.

[obtuse torrent]

@quiet zodiac ... That's good, those last ones look like cold porcelain works..

[tired vigil]

A man

[restive mesa]

in all the images I create with QR codes, the white-space always stays as white, instead of becoming one with the image

I want them to be like the image above,

not this
Which setting should I tweak/?

[solar dust]

What's the best model for realistic full body portraits?

[solar dust]

I feel you.

[uneven pelican]

I really need help making a cover with a tow truck with weapons mounted all over it

[verbal shore]

what negative promp can i put to not have half of their weapon somewhere them?

[quiet breach]

Can anyone help me generate a prompt to generate a true human hand

[tranquil shard]

what works better in your opinion art by artist or in the style of artist?

[tranquil shard]

That the problemt with neural networks - they dont really have a concept of lines, perspective and consistency. I don't think there are negative prompts that can cure it. You can try negative embeddings like EasyNegative, FastNegative, etc - there really a lot of them now. Using Control net can help getting image in composition and consistency you need, but I'm not very familiar with them and can't recommend anything. I'm just doing manual edition of images to fix such problems in my project.

[obtuse torrent]

@verbal shore ask on the anime channel, I think one of them made a LoRA or a complete model thinking of generating better hands holding things, there are also LoRA or Lycos specialized to hold weapons... another option is guided generation with a reference (maybe a scene from a movie or something like that)

[quiet zodiac]

@rocky coral Siddiqui Take a picture of your hand from the angle you want your final. Supply that as an image guide and type a simple prompt like a human hand

@verbal shore You may not need any negative prompt. Try a non-square aspect ratio. Try a size like 1024x768 or 1024x512, or 576x768.
Also consider adding some camera instructions to your prompt like...
camera rule of thirds or golden ratio

[wicked sinew]

Is there a list of prompts that describe angles of the camera and the distance aswell?

[orchid ore]

You can try flavors.txt. If you are using A1111 then once you interrogate image it appears in stable-diffusion-webui\interrogate folder, there are most frequently used words. All sort of word as well some related to camera. But i think depends on model used too. @wicked sinew

[wicked sinew]

Thanks I'll check that out

[visual gust]

Sorry if this is the wrong channel.

Is there a way to quickly edit Controlnet's preprocessed image canvas? Kind of like how NovelAI has it set up.

Say I use controlnet to turn a picture into a scribble, I then I use that scribble to make new generations with. Problem is, there are a few white lines that I'd love to get rid of because they keep messing it up. Or maybe I want to add a line somewhere. You get the idea.

Back when I was still using NovelAI, editing the scribbles was easy and straightforward but I can't figure out how to do it with Stable Diffusion. I'd have to download the scribble, edit it with paint, and then reupload it which is a pain considering I tend to do tons of minor adjustments constantly.

Are there any plugins or some such that could help with this? Thank you.

[crude crystal]

Deforum, human mask, many variations of CFG and seed behaviour and stuff. Going to try these settings next https://www.reddit.com/r/StableDiffusion/comments/127y2rm/finally_grasping_deforum_settings_in_comments/, unless anyone can look at the video and magically know what I should do differently. CFG15, fixed seed. 40steps. 576x1024. Not sure what else is relevant. Goal is frame to frame consistency. Cheers - my best effort so far: #📝｜prompting-help message

[blissful shuttle]

I need to pose this character (Made with midjourney) in different ways, with consistent style and look. Any idea how? Maybe using controlnet somehow?

[undone rivet]

Hello everyone, I would like to request your help

I'm trying to make a portrait of a woman using a makeup like "la catrina"

I'm using Realistic Vision and epic realism

But I'm not able to get what I want, the terms that I use does nothing, the woman does not have any makeup on her face

Prompt (partially):
(La Catrina makeup, Pale, Skull facial details, Dramatic eyes and eyebrows, Striking lips, Floral details:1.2)

Do you know how could I make the prompt works...?

[quiet zodiac]

Try supplying a reference image of a woman with the makeup you have in mind. StableDiffusion doesn't understands brands.

[rich crow]

with make-up, you have to describe specifically what type it is that you want. mascara, eye-shadow, blush, lipstick, etc

[silver valley]

you need to prompt different. la catrina is not known for the model.
Here is what i get with my prompt with realisticVision:

[undone rivet]

fantastic, do you mind share the prompt please....?!

[silver valley]

sure:
close up portrait, 1girl, dia de muertos, skull makeup, skeleton makeup, mexican clothes, face paint, focus, depth of field, hair rings, hair intakes, streaked hair, limited palette, hyperrealistic pupil, papercut, masterpiece, dynamic light, dynamic pose, volumetric light, hyper detailed, deep eyes, looking at viewer, shiny hair, glossy skin, soft skin, dramatic, cowbowshot, cleavage, modelshoot,

[undone rivet]

thanks a lot buddy...!!!!!

[undone rivet]

question, why do you use "papercut".......? I have used that with images of paper craft.......

[silver valley]

haha i thought you would ask that, and i dont know i made this prompt last year xD

[undone rivet]

ok, thanks

[silver valley]

i tested without and it looks good too

[undone rivet]

perfect....🙌

[tired vigil]

NansException: A tensor with all NaNs was produced in VAE. This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

whats this mean

[tired vigil]

i think the thing im tryna render is frozen at 8 secs remaining

Does this usually happen?

[leaden socket]

Best prompt for plant?

too much with plant

[trail sphinx]

Can anyone please tell me how to change aspect ratio?

[chilly karma]

Hi guys

I need some help with some prompts to make something like this:

[next flint]

There is an extension that can easily do that but its just the image sliders if thats what you mean

[next flint]

do you happen to have any of the generation data

[chilly karma]

ya that data seems to be wrong

[next flint]

could you send that

[chilly karma]

This is what I was able to create

[chilly karma]

1 Sec

Model: meinamix_meinaV10
detailed eyes, (Perfect features:1.1), (masterpeace, best quality, good quality:1.4), masterpeace, intricate details, smile, smile, smile,
Negative prompt: (EasyNegative:1.2), (worst quality, low quality:1.4), (blurry:1.4), novel,
Size: 576x1024, Seed: 1602425008, Model: meinamix_meinaV10, Steps: 30, Version: v1.3.2, Sampler: DPM++ 2M Karras, CFG scale: 7, Model hash: 77b7dc4ef0, Denoising strength: 0.2, SD upscale overlap: 64, SD upscale upscaler: R-ESRGAN 4x+ Anime6B

[chilly karma]

@next flint here

[next flint]

alr let me check and see what comes up

[chilly karma]

☺️ ok

[next flint]

since my gfx card is bad its taking a while to get something but i see what your getting at. the annoying thing with stable diffusion is that the generated image can be different even for reasons that seem unrelated. even if given all the same gen data two different cards will render different results.

if your looking to get something similar you could try using the gen data and using the shuffle mode in controlnet v1.1

[chilly karma]

ya I get that part, but with some line art and stuff, you are able to get almost the same stuff, what I requre the most is the artstyl

[chilly karma]

what does shuffle mode does?

btw what card are you using?

[next flint]

2060 super

[chilly karma]

Thats better than mine 😄

1060

[next flint]

oh

damn

[next flint]

its hard to explain. ill send the github doc in a sec

[chilly karma]

ok I got a message from the original artist, says he's using a lora for the character I guess

or maybe the artstyle

do you got anything similar effect generating lora?

[next flint]

This is what shuffle does

[next flint]

wdym

[honest cave]

What does 'bokeh' do and why do I see it so often?

[chilly karma]

for the artstyle

[next flint]

photography term

[next flint]

well there isnt a lora listed so idk what you mean

[chilly karma]

I meant that maybe you stumbled upon something that gives the similar effect, nvm 😄

[next flint]

if there is enough of the artists work in that style you could make a lora of it

[chilly karma]

just a couple

https://civitai.com/models/5737/akemi-takada-1980s-style-lora?modelVersionId=6677

seemed to be using this lora

with very low multiplier

[next flint]

╘Rajul╛ʕ•́ᴥ•̀ʔ this is what happens with shuffle

granted im using the v9 instead of v10 of the model so its kinda different but you get the idea

[chilly karma]

damn you did a far more better job at it than me

[next flint]

controlnet is a wonderful thing

[chilly karma]

that it is

[next flint]

im going to try to mimic the gen parameters more and see what happens

[chilly karma]

hey one more thing, when using openpose of the control net, I used a heat model with well defiend fingers, but some Models seems to still mess it up

[chilly karma]

yes please 😄

[next flint]

can you send an example

[chilly karma]

This is what I was using

As you can see the fingers still are all messed up

[next flint]

are you using ti or lora in the negative prompt

[chilly karma]

I am using lora for the character in the +ve prompt, and badhandV4 in -ve

what's a ti?

[next flint]

textual inversion. as much of a blessing as controlnet

[chilly karma]

ya I got some textual inversions

which ones should I use though

These are some I got

[next flint]

look up "neg" and "bad" into civit. provides a ot of good ones

[next flint]

good start. bad hand, easy, and fast are good ones

[chilly karma]

thanks

should I increase their multiplier?

[next flint]

you can experiment with that but like with normal tokens increasing it too much can yeild weird results

[chilly karma]

ya I tried that and had to delete the photos, I do keep my generations if it's just hands or little bit of texture and saturations

but the negative prompts really does something else

[next flint]

they are very powerful. i like to just stack them on until i run out of things to add

i did this thing a while back where i just used the top 1000 negative tokens

[chilly karma]

1000, damn, how much time did a generation took

[next flint]

same time as normal

it was just a rly long prompt

[chilly karma]

I did something around 400 and it took me more than 4hrs, and the end photo was something weird

[chilly karma]

oh ok it's fine then

I don't know what happened with these

[next flint]

like americans with guns, when one doesnt work just add more

[chilly karma]

seems like a thesis, is that what it created?

the picture at bottom

[next flint]

ive though about just putting books in there

[chilly karma]

ya you should try that and show me the results

[next flint]

can you give a passage or something

i dont read very often

[chilly karma]

wait

In a realm where magic danced freely through the air and fantastical wonders awaited at every turn, there stood a breathtaking scene nestled within the embrace of nature's majesty. Sunlight filtered through the ancient, towering trees, casting ethereal rays upon a serene meadow, dappled with vibrant wildflowers of every hue.

Amidst this enchanting landscape stood a young woman of ethereal beauty, her porcelain skin radiant under the golden light. Her hair, a mesmerizing cascade of pale orange, seemed to capture the essence of a fading sunset, as if a fragment of the heavens had been woven into her very being. Flowing in gentle waves around her delicate shoulders, it shimmered like strands of pure magic.

As the breeze caressed her cheeks, the young woman's striking eyes, the color of emerald, gazed with a mix of wonder and longing toward the distant mountains. Those ancient peaks, crowned with snow and veiled in mystery, seemed to hold secrets of forgotten ages within their towering embrace. Their majestic silhouettes reached toward the heavens, touching the very edge of the ethereal realm, as if beckoning her to embark on a grand adventure.

With an air of determination, the woman stood tall, her graceful form clad in a flowing gown of pale greens and blues that mirrored the tranquil hues of the surrounding meadow. She had an aura of quiet strength about her, a spirit untamed and unyielding, ready to face the challenges that lay ahead.

As her gaze remained fixed upon the distant mountains, anticipation danced in her eyes. Her heart whispered with dreams of exploration and discovery, for beyond those rugged peaks lay a world untamed, where ancient ruins guarded by mythical creatures awaited her touch, and where the echoes of forgotten spells reverberated through the wind.

In that fleeting moment, the young woman embodied the essence of adventure, her pale orange hair catching the sunlight like flickering flames, as if her spirit burned with an unquenchable desire for the extraordinary. And as she stood there, on the cusp of the unknown, she could almost hear the mountains whispering tales of courage and destiny, urging her onward into the realm of limitless possibilities.

Here

[next flint]

All i added was badhandv4, badv5, BadNegAnatomyV1-neg, lora:easynegative:1

[chilly karma]

Damn that realy looks nice

wait lemme try thay

[uneven pelican]

Can I get help creating an image. New to this and I have an image in mind.

[next flint]

ofc. whats the main idea

[uneven pelican]

Well I wanted the background to look like a derby arena. The surrounding landscape is fiery. The arena floor has trap holes that shoot our hot jets of flame.

Have more to add

[next flint]

?

[uneven pelican]

Adding the main vehicle description I want to add

[next flint]

k

[uneven pelican]

But that image is actually good for the background I like it

Although I prefer the the hole shooting flame to be level with the arena floor and the other circular platform sloped at an angle if possible

[next flint]

for some reason this idea is rly hard for it to get right

[uneven pelican]

The vehicle description is a 1960s neon orange tow truck. The truck on tge door has the words Rambling Wrecker. The trucks windows are tinted black. Instead of a tow hook it has a magnetic tow cable that can grab things. The rear also has a total of six barrel drums on the rear carrying oil in then. 3 on each side in the bed. Near the back bumper the truck there is a dispenser for land mines. On the roof of the tow truck is a missile launcher with a rack of 6 missiles sitting flat across the top. On tge hood of the truck on either side are getting rifles. On the front of the truck, there is a train plow. The plow has 4 holes in it. 2 that shoot out flames from flame throwers. The other two shoot out beams of purple plasma. I would like to have the gattling rifles firing along with the plasma beams.

The plow on front the truck looks like one of the modern cow cathers

Something like that is on front of tge tow truck

@next flint I tried getting the image oc the tow truck myself but it was just a mess

@next flint there is the tow truck description not sure how long that would take

Not sure how long this will take but just dm when loads. I honestly need to get some sleep but have no idea how this thing works...

I'll mention you in my book for creating the AI art cover

[tribal talon]

PSA for anyone struggling with something I did for ages: The word that you use to get hair that transitions from one color to another is ombre. Nothing I typed ever got that result, then I learned that word and now it gets it right almost every time.

[chilly karma]

Ya I was struggling with the same too, and what about the colors, can you give an example for orange that transitions to yellow?

and do you know what to do for 2 toned hairs, black on the outside and yellow on the inside

[tribal talon]

I'm not sure what you mean by on the outside and inside

[next flint]

Eldrias I’ll try tomorrow but with my experience I don’t think I can get something as detailed as you mentioned

[uneven pelican]

Hmm, is there anyone who has that I guess and I know it is a lot to ask for.

[tribal talon]

Eldrias, what you're looking for is likely best done with inpainting

[chilly karma]

@next flint

this is what I got from the paragraph

[next flint]

How do you do that thing w the models

I’ve always wanted to try

[silver valley]

nice to know, thanks! you need a vae for your model it seems

[chilly karma]

Something like this

[uneven pelican]

Well I know some who paints for fun in vr, I'll just ask them

[tribal talon]

I have one, I just had a lora set a bit too high which messed with the style.

[next flint]

Rajul how did you do the thing where it compares different models

[chilly karma]

it's the script

goto x/y/z plot

select checkpoint names in the x

and select all the checkpoints in the slider next to it

leave y and z empty

[next flint]

Thanks. Always wondered

[silver valley]

ah okay 👍

[chilly karma]

u can use different vae and Promts too for a grid comparision too

[chilly karma]

mention not

[chilly karma]

@tribal talon can you see the inner side of the hair is orange and outer is all black, I can get this effect with only a couple of character loras, but if I specify it a lot it just makes the whole hair 2 colored

Like this

And what I want is this

[tribal talon]

I've never tried to do that, though I have got it on accident before. I might be able to try

[chilly karma]

ya please let me know when you get it right and thanks for the ombre prompt, I wondered that too and used to use hair tip for different colors at the tip

[trail sphinx]

I'm just using the discord bot-1 to bot-10 I don't what sliders you're referring to

[silver valley]

ill get this effect with some models. i use the tag hair intakes, but i dont know if it causes this.
Model: Cetusmix

[chilly karma]

that's exactly what I'm talking about, you got it perfectly right, wait lemme try that with prompts

[tribal talon]

Yeah I can't get it to do that, unfortunately.

[hollow tapir]

two tone hair color can trigger that effect too for some models.

[chilly karma]

ya it was working in some models and not in some

[chilly karma]

I think intake makes the color blue by default

[hollow tapir]

it works great mostly for models trained with booru tags, two toned hair too and all sort of variations of it.

[chilly karma]

@silver valley @hollow tapir

this is the result I got for using intakes and underlights

intakes had a constant under color but was blue in all

[fallow roost]

Can stable diffusion take image as a prompt?

[hollow tapir]

yes that's img2img

[fallow roost]

How can i do that

[onyx marsh]

does this have anything to do with ai video effects?

[hollow tapir]

Using Automatic1111 stable-diffusion-webui you simply go the the img2img tab

[hollow tapir]

What are you talking about ?

[serene wasp]

I need to stop getting legs

I'm trying to make something like this

with what I have so far in negative prompting no luck

but I keep getting imminent tentacle rape which is not what I am going for

it put more legs on to spite me

[gusty coyote]

well I unintendedly made a lora of a character that will only produce naked pics of them not matter what I put in the prompt. Any workarounds for this? The fantasy skin tone I created for it is perfect though

[quiet zodiac]

If you don't have a negative prompt, that's the place to reduce nudity.
Try adding...

nude, naked, NSFW

[gusty coyote]

yea I have all those and more. No avail

[noble charm]

prompt for nun clothing?

see what it does with that

[gusty coyote]

ok sure

[noble charm]

but yeah I don't see any clothing in that prompt

that may be why

seperator idk
any idea why the controlnet openpose face feature keeps making multiple people???

any idea how to fix this?

[quiet zodiac]

@gusty coyote Try dropping cinematic sensual. You might want to start your prompt with a sentence instead of all those conditionals.

A long cool woman in a black dress

[chilly karma]

How do I get these lighting effects

the shining thing effects

[bitter trout]

what model is that?

[chilly karma]

Counterfeit-V3.0

[chilly karma]

https://civitai.com/user/Radish_/images

this is the guy, nothing more is provided

[bitter trout]

damn

[chilly karma]

exactly

[bitter trout]

what are the best general use quality tags for anime? also negative tags and embeddings, i see people with a really long list of quality tags and negative prompts that they use in every prompt so what are the ones yall use?

[wise slate]

This can't be posted because it contains content blocked by this server. This may also be viewed by server owners.
Only you can see this • Dismiss message• Learn more

why do i keep getting this?

[tender cape]

I'm experimenting with isometric dungeon landscape generation and I wonder if somebody could share some tips and tricks to improve prompts / perhaps use better models?

For starter, I created a simple isometric picture in 3D editor and tried to generate something using it as a base image and using the following prompt:

isometric view from above of a throne room in a dark dungeon, bricks floor, simple stone pillars, intricate stone walls, masterpiece, volumetric light, eerie, realistic shadows

I tried 0.5 and 0.75 denoising strength. The former is to try to stick to the source as much as possible, the latter is to let neural network be more creative.
In all cases I tried DPM++ 2M Karras with 40 steps and set image resolution to 768x512

Source:

[tender cape]

Here are the best results I got so far (picture 01 is for 0.5 denoising strength, 02 - for 0.75 denoising strength).

[tender cape]

...And a few more models

[tender cape]

Tried models: A-to-Z RPG Artists Tools, Break Domain Anime, Chillout Mix, Dreamlike Diffusion, Openjourney, Paragon, revAnimate. After a few runs I think revAnimate is the most creative, but it seems it'd require a lot of additional inpainting as well.

[next flint]

Likely because discord is identifying what you are posting as nsfw

[glacial creek]

I want to recreate this image face, any tips?

[bitter trout]

Should i put quality tags first of after my prompt?

[worn dome]

@orchid ore I'm here again

I actually want something like this. The cat's standing still, but a full body portrait

Realistic, 4k, Full body view, full body portrait, Russian cat-animal agent standing still dressed like business man, a cat-animal is wearing hat and suit, white background, studio lighting, 50mm

That's the prompt I'm using now, but I'm still having issues

[quiet zodiac]

@tender cape I played around with your prompt and reference image. I resized your reference to 1024x768 so it matches my output size before upresing. I think the terms "isometric" and "masterpiece" might be limiting some of your look. Once I removed them and added a few more detailing terms I got these results. My theory is "isometric" is only trained on game-like images, but the refence angle helps with the final output.
a view from above of ((larger-than-life extra slimy tentacles)) covering the floor and slithering up to wrap around every column destroying the dimly lit ancient crumbling dungeon throne room overflowing with treasure, broken doors, cracked walls, dust and debris

[gusty coyote]

can anyone point me towards what a "break" is in prompt making and its benefits?

[orchid ore]

i think it is residue from Regional prompter. I got those where addcol or addrow are.

[pearl needle]

What's the best way to stop getting more then one navel on a pregnant woman?

[hushed monolith]

you only get 75 words in a prompt, if you use more it combines words into bigger tokens to keep your limit of 75, break stops the current 'prompt' and starts a new prompt and words at start of a prompt get more weight so it lets you choose when the next 'block' of tokens start or something like that

[hushed monolith]

inpainting or perhap photoshop with a 'clone' tool to remove the extra button

[tired vigil]

What is the easiest way to get a logo on a tee shirt to be a specific thing?

I've tried:
logo on chest of the number 8 in a white circle

...and things like it with little success.

[hushed monolith]

Again, photoshop 😉 would be the easiest way

SD can do alot but to make the image perfect the way you want it consider other editing techs

[tired vigil]

Shame, but that's kind of what I figured people would say.

Thank you!

[quiet zodiac]

@pearl needle Try lowering your output resolution. Doubles are often presented when you exceed the model's training resolution.

[worn dome]

I've rendered an image with Stable diffusion, then I've Photoshopped it. Now I get The Photoshopped image into png info and I don't get it parameters that i used to generate the image. Is there any way to get that information into the Photoshopped image?

[sterile widget]

How do I use negative prompt?

Like can I get an example ‘/dream’ command?

[gusty coyote]

All hail the power of cutoff

[obtuse torrent]

if you can answer... what is the reason why you want the parameters included, in the image altered by an editing program?

[hoary prairie]

how would i recreate this image with stable diffusion?

is it even possible to do that type of 'hair'

[rich crow]

specifically that one might need a LORA

but you could always simply create the composition and then just edit the hair afterwards

[hoary prairie]

ok thank you

could you edit the hair by using something like inpainting or would you need to manually draw it?

[dusky schooner]

I've been trying to create a shattered, blown up moon using I painting, but I just can't seem to get it to work. Any recommendations?

[terse cloud]

I'm trying to use image2image to make colouring in art. (the kind you get from a colouring in book). my base image is a flower, what prompt would you guys use to get it to be a blank image with just black lines?

[rich crow]

yep you could inpaint it. for the editing route, i would personally just photoshop/krita a pic of the hair that you want (could even be the one above) and just transparency filter it in, then img2img that to blend it in

[rich crow]

idk if the AI is trained enough or is smart enough to do this currently, i would personally just make anything you want then go into an editor (photoshop/krita) and set the colors to 0/0/0 (makes everything black if not blank)

[high hamlet]

how much can interrogate CLIP take? it has been like 10 minutes already and it says processing

nothing downloading on the terminal

[worn dome]

Because I want to remove the background of the image to use It in a video, but I don't want to have two versions of the image. So I want to delete the original and have the information in the version that is useful for me

[worn dome]

Is there a way?

[cinder sky]

That is art of a Vtuber named Bao

And there indeed is a Lora

https://civitai.com/models/81567?modelVersionId=86557

[cinder sky]

Takes around 14 seconds for me

[cinder sky]

I do not believe so. That would require editing the metadata and transfering it from the original file

But why would you even do that? If you edit the image in anyway post AI Generation, meta data is kinda pointless since that metadata will not match the photo

[worn dome]

Is it posible to get a transparent bg in Stable Diffusion?

[cinder sky]

AFAIK it isn't possible because the way image generation works is using rgb values of 3 dimensions, and alpha channel, which is required for transparancy, is not implemented into it.

If you meant generating image with transparent bg like using txt2img

Limitation of Stable Diffusion at the moment

¯_(ツ)_/¯

[worn dome]

And that's why I want to use Photoshop to remove bg

[worn dome]

But you say I can't do that

[cinder sky]

You will have to learn about meta data

I personnaly do not know how to edit meta data, and I am not even sure how to transfer SD specific metadata structure

I don't think it is really worth your time

[onyx cosmos]

Hi there...🙂
Is there a list of authoritative terms that circumscribes setting sizes as in photography and film? For example: full shot, long shot, distance shot, etc.
Most of the common terms I've tried don't work at all or only sporadically.

[edgy tundra]

If you're using AUTOMATIC1111, there is an extension for removing backgrounds
https://github.com/AUTOMATIC1111/stable-diffusion-webui-rembg

There is a bg remover built into the Vlad fork, under the "Process Image" tab.

[worn dome]

That is amazing!

Thank you

[edgy tundra]

Examples from Vlad

[edgy tundra]

Something like this?

[worn dome]

Love it

[worn dome]

Also preferrable If It has paws instado of hands

But that's what I want

[edgy tundra]

Those last 4 above were from

detailed full body concept art illustration soft focus oil painting on canvas of an anthropomorphic overweight (cat:1.2) businessman in full business suit, Russian oligarch, biomutant, dystopian, micro detail, octane render, 4K, by Carl Larsson, (shallow depth of field, bokeh:0.4)

[worn dome]

How did you do this?

[edgy tundra]

Using my "SemiReal" model from
https://civitai.com/user/galaxytimemachine

[edgy tundra]

The prompt is above your question 🙂

[worn dome]

Is it necessary to use the model?

I had a headache tryng yo install them

[edgy tundra]

It should work with most, but you'll get different results from different models.

[worn dome]

I searched for tutorials to use diferent models but I had problems

[edgy tundra]

What are you using? Local install? A1111?

It's as simple as putting them into a directory for that.

[worn dome]

So I just download It and get in the directory?

In The tutorial the models were downloaded in Google Drive, so that I don't need to have It in my computer

I don't have too much space, so I would prefer to have the models in Google Drive

But I couldn't be able to follow the tutorial due yo problems

[silver valley]

If you run the webui localy you also need the models downloaded on the PC.
If you use Google collab, then they can be stored in gdrive

[worn dome]

I'm using Automatic1111

[silver valley]

Yea then you need to download the models.

You need to put them into the models/stable-diffusion folder

You also can put them onto another driver and link it in Stable-diffusion.

[worn dome]

So, I can't just have them in GDrive?

[silver valley]

Nope

[glacial agate]

hey guys! Disclamer: i am a total noob. So i installed automatic1111 on Google colab and trained the model with headshots of a person (around 30 photos) using the 1.5v and hassanblend/hassanblend1.4. Now if i use the session name only in my prompt i get the image i want, but if i add anything else like "{session_name}, wearing dress i get the image of a totally different person. What am i doing wrong?

[tired vigil]

i'm wondering
what would yall recommend to use for the Sampling method

i see Euler A and DPM++ 2S a Karras the most

but idk what really distinguishes them

if it even really matters

[glacial rapids]

been getting some problems with the hands of my art, what do y`all recommend to add to my prompt to make the hands look better? beautiful, (masterpiece:1.2), (best quality:1.2), (ultra highres:1.0), (perfect face), detailed face, perfect eyes, perfect lighting, 1girl, dark hair, curvy, shirt, pants, cleavage, looking at viewer, wide hips, thighs, outdoors, lora:FemLaw:0.9, lora:add_detail:1, lyco:GoodHands-beta2:1.0,

[silver valley]

some are slower, some faster, most used are Euler a, DPM++ 2M Karras and SDE
Here is a good information on them:
https://stable-diffusion-art.com/samplers/#Samplers_overview

[silver valley]

do you use highres fix?

[glacial rapids]

yes but I think that my gpu is too weak for me to be able to use it or something

but im gonna try to use it anyway

[silver valley]

you can try to use the tiled diffusion extension with tiled vae. that could help with highres fix.
but what normaly should work for your card is the sd upscale script in img2img

[glacial rapids]

ok then how do I use the upscale script?

[silver valley]

example:
put your image in img2img, set the resolution to 512x512,
denois to 0.3
activate the sd upscale script at the bottom,
choose an upscaler like esrgan4x+ for example,
set upscale by 2
then press generate
if you get artifacts then you need to remove person specific tags from your prompt in img2img

[glacial rapids]

ok thank you im gonna try

[glacial rapids]

it worked man

thank you bro

[silver valley]

nice, no problem 🙂

[rich crow]

Use the xyz grid script with different samplers to see what you like best. There's going to be people who prefer each different one. What matters in art is what you prefer 😏

[lyric goblet]

How do you write in an artist as a style in prompts? Will all artists work or has there been some restrictions put in after the Twitter protests?

[rich crow]

You could simply do "art by artistnamehere"
If that's not giving you similar results, you can use controlnet's reference model. It does a good job of copying style

[worn dome]

I don't finf the way to download your model

[edgy tundra]

1. Click on the link.
2. Select the model you want to download.
3. Click on the "Download" button.

[worn dome]

I want your semireal model

But I don't find it on the page

[edgy tundra]

It's there, 3rd one along.
https://civitai.com/models/61734/galaxytimemachines-gtm-foryou-semireal-fantasyai

[spark trench]

how do you choose certain characters to be generated in the prompt for instance people irl and or anime characters etc

[orchid ore]

Simply type their names, but depends on models as well. Some are training more some less and some not at all.

[worn dome]

Is this the file name?: galaxytimemachinesGTM_semirealV11.safetensors

[edgy tundra]

Looks like the one

[glacial rapids]

hey man

Ive been getting this problem when I use highres

[silver valley]

You can try to add --no-half in the webui-user.bat and try again

[glacial rapids]

I already have --xformers --medvram --no-half-vae

[jovial phoenix]

is there a list or something off all the "modifiers" i see people using with the "--" and stuff? or is it just infinite and contextual like the rest of the prompt?

[silver valley]

Yea then change --no-half-vae to --no-half

[silver valley]

These arguments starting with -- are for the webui-user.bat and not for the prompts.
Here is a complete list:
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings

[jovial phoenix]

hmmm ok i misunderstood i think cause thats all Star Trek to me lmao

how does one specify like the canvas format for example? like if i wanted a 1920x1080 output and not the default square format

[orchid ore]

You are using A1111 webui?

[jovial phoenix]

i am using this discord in the bot channels, i havnt tried any other client

(quite new to AI image gen)

im really impressed with the output from SD so far but want to up my prompt game and get a better understanding of how to do it besides "say stuff and stuff do pop out" lol

[orchid ore]

anyway 1920x1080 is too large for native size. Better 768x432 and resize it with some advanced resize algorithms

[silver valley]

If you use the dream Command in the bot channels you can select the aspect ratio (size) by clicking on aspect and then change the 1:1 to 3:4 for example

[jovial phoenix]

ah there we go.

[gray scarab]

is there a way to tell SD to generate images without bokeh and shallow depth of field? Looks like everything is taken with very expensive cameras and fast lenses

[tired vigil]

any way to grab one guys art style and kinda translate it onto the generated images

[orchid ore]

@gray scarab have you used photo in your prompt?
Also you can use all f.20 50mm and so on.

@tired vigil have you somebody famous on your mind?

[gray scarab]

i'll try f.20 thx

[orchid ore]

o.k. hope it helps

but if you mentioned photo earlier it will behave as photo, maybe delete word photo from prompt.

[gray scarab]

i use the option style:photographic. So far I got good result using "wide angle"

[orchid ore]

great

[tired vigil]

kinda

[orchid ore]

@tired vigil then try what you want and by yourartistname

you can try - park by bottero

@tired vigil can you tell me his/her name, so i can check it if in list at least?

[tired vigil]

sunhyun

[orchid ore]

one word? Going to check

Sun Junze fount this

[tired vigil]

?

[orchid ore]

that i found that name.
You can type it as already said, things you want by Sun hyun or what is exactly his name

@tired vigil
check this
https://civitai.com/models/91849

[tired vigil]

ah yes

wait how do u use the - ... by ...

is it in the prompts

[orchid ore]

just as ordinary word

[tired vigil]

oki ty

[orchid ore]

but using loras i send link, there will have to help you other ppl using loras, i dont

[tired vigil]

ye, ik how loras work dw

ty

[orchid ore]

it is lycoris, but should be same

[drowsy gazelle]

Lycoris requires an extra extension

[lyric goblet]

How do I prompt if I want the the person looking at viewer, while walking towards the viewer, but the angle is slightly up like the person is shorter than the viewer?

[quiet zodiac]

For super specific framings try using an image guide, otherwise try.

facing forward and looking down on subject

[worn dome]

@edgy tundra your model's amazing, but is it useful for doing landscapes?

[edgy tundra]

It is very versatile and can do a lot of things. Landscapes are also good. Just try it and see 🙂

[edgy tundra]

Some sci-fi ones here #1072017962531307540 message

[worn dome]

I think fighting with computer will be over with this model

now I will be able to work :D

[gusty coyote]

How would one prompt someone sitting behind a desk? What model has the best chance of nailing it?

[rich crow]

im guessing "sitting behind desk" not work when you try it?

[gusty coyote]

correct lol

after several hundreds of generations and model switching

[orchid ore]

@quiet zodiac i think something similar or probably same in some cases is called kirigami, going to test it

[kindred needle]

Guys, What prompts and model should I use to generate hyper-realistic images?

[orchid ore]

@quiet zodiac Shoji-paper art. Probably

[orchid ore]

i think any photorealistic model. Hyper-realistic is thing to write in prompt.
this is simple prompt with hyper-realistic tag. Realistic vision v20 <-model

[quiet zodiac]

@edgy tundra Trying out your model. It looks great!

[cobalt osprey]

Does anyone know the best propts for 2d mascot creation like Duolingo or Hootsuite

[ruby lake]

Hey! I'm new to this and i want to create some ai art from one of my mains in league of legends, how do i get the model of that character in the style of league of legends, im looking to make a wallpaper out of it as well!

[tired vigil]

use civitai and find some lora / lycoris that has the/a trained model for it

[ruby lake]

thanks 😄

[unique lodge]

Never tried that, cheers!

Hey! Check this out:

ACTIVITY = [
  "looking at the camera",
  "attempting to set a new world record",
  "recreating a scene from a favorite movie",
  "attempting to cook a gourmet meal with random pantry items",
  "building a pillow fort",
  "starting a dance-off in the living room",
  "holding a karaoke competition",
  "learning a magic trick",
  "conducting a DIY science experiment",
  "participating in a home fashion show",
  "hosting a virtual game night",
  "trying to speak a new language",
  "making a homemade music video",
  "attempting a world record for the most cups of coffee consumed in a day",
  "inventing a new type of dance",
  "trying to make their pet internet-famous"
]

def outfit_based_on_gender(gender)
  if gender == "male"
    [
      "streetwear: baggy jeans, oversized graphic tee, and chunky sneakers",
      "luxury casual: designer sweatpants, high-end logo t-shirt, and limited-edition sneakers",
      "edgy: leather pants, graphic tank top, layered chain necklaces, and combat boots",
      "flashy: colorful suit, diamond-encrusted watch, and patent leather loafers",
      "athleisure: branded tracksuit, matching cap, and sporty sneakers",
      "minimalist: black slim-fit jeans, plain white tee, and black leather high-tops",
      "vintage: retro sports jersey, baggy denim jeans, and classic basketball shoes"
    ]
  else
    [
      "edgy: black leather mini skirt, studded crop top, and knee-high boots",
      "glam: sequin jumpsuit, silver stiletto heels, and oversized sunglasses",
      "punk: ripped black jeans, band t-shirt, leather jacket, and combat boots",
      "retro: polka dot midi dress, red patent heels, and cat-eye sunglasses",
      "festival: neon fringe crop top, denim cutoff shorts, and glitter ankle boots",
      "avant-garde: deconstructed blazer, asymmetric skirt, and platform shoes",
      "streetwear: oversized graphic sweatshirt, bike shorts, and chunky sneakers"
    ]
  end.sample
end

[...]

inputs = {
  "prompt": "photo of #{ current_site.nationality } #{ gender } #{ ACTIVITY.sample }, #{ ['blonde', 'dark blonde', 'brown', 'black', 'red'].sample } hair, wearing #{ outfit_based_on_gender(gender) }, #{ LIGHTING.sample }, DSLR, ultra quality, sharp focus, tack sharp, DOF, film grain, Fujifilm XT3, crystal clear, 8K UHD, highly detailed glossy eyes, high detailed skin, skin pores",
  "negative_prompt": "disfigured, ugly, bad, immature, cartoon, anime, 3d, painting, b&w, nude",
  "width": 512,
  "height": 768,
  "num_outputs": rand(1..4),
  "num_inference_steps": 50,
  "guidance_scale": 7,
  "scheduler": "EulerAncestralDiscrete"
}

prediction = version.predict(inputs)
fetch_prediction(prediction)

Ruby's so gorgeous and fun to work with that I don't need no UI 🙏

[modern bridge]

Hello, i've a simple question, how to normal basic anime style please ?

[orchid ore]

just download some anime model you prefer from civitai. @modern bridge
Or ask #🍥｜anime here

studio Ghibli for example somehow works in most models i believe, if your taste

[modern bridge]

i'll try

(if i find a good model i'd try)

[tired viper]

Hi I cant get this character smiling using reference_only in controlnet. It always spits out a different guy ;(. Any ideas? Its my first time using controlnet

Should I even expect reference_only to work this way?

[rich crow]

Reference should correctly spit out a new guy in the same art style. I would try controlnet tile with a denoise of 1 as a starting point alongside an openpose controlnet of a smiling face and experiment from there

[lunar osprey]

Any idea of how to get a person without legs? I've been trying to do this for ages with all sorts of models but every one wants to make people/Androids with legs and won't replace them with anything else.

Here's latest attempts but have tried just about everything with other bots and in my own automatic 1111
#1101178506530013284 message

[rich crow]

Might require some image bashing because even I can't think up of how a girl with tank tread legs would look like. Mabey throw a girl on top of a tank, edit out the lower part of her body, then img2img

[lunar osprey]

I can't even get a girl without legs though. Even without attaching to anything. Something like this doesn't seem possible ATM

[rich crow]

I'd use variants of amputee to get that, but for bashing images, you could simply just cut the bottom part of the person off in an image editor is what I mean

Transparency masks help but literally just cutting the bottom off works as well

Img2img to blend, 👨‍🍳👌

[fringe jewel]

any suggestions on getting the ai to give people different colors of skin? Like, unnormal colors like teal or cyan?

I am trying to get the ai to make a person with teal skin but it keeps making their outfit teal

[orchid ore]

@fringe jewel not sure it helps, try teal-skin.

i am genius 😄

[fringe jewel]

Oh that's a good idea, I'll give it a try

[orchid ore]

then add blonde hair or what. But somethimes it has glitch, like one hand pink etc. But in general should work

[bright thistle]

Hey all, I'm using controlnet - very new to this 😅 .

Was wondering what kind of model/settings would be best for preserving the "pupils" in this image in txt2img output. Currently getting less than desireable results using canny.

[lyric goblet]

@fringe jewel There is a Lycoris called Colored Skin - Loha on Civitai for that purpose, I won't link it because the preview images are a bit nsfw ^^

[hoary prairie]

so ive been trying controlnet and cant figure out what im doing wrong

i feel like it has to be obvious. it worked a little before, but i changed the image to a lower quality one and added a second controlnet unit and now it doesnt want to follow controlnet

[obtuse torrent]

@hoary prairie ...

check models ...

[rich crow]

yeah youll need to select the model on the right handside as well

[fringe jewel]

thank you it worked

[fringe jewel]

is it supposed to only make barely clothed people?

nvm

[main vortex]

hi i need some help in prompts, i am trying to generate a pure white background but stable diffusion keeps generating a background with some stuff behind for me these are the prompts i use: look at side,white background, girl, elf. Negative prompts: nsfw,(worst quality:1.2), (low quality:1.2), (normal quality:1.2), lowres,watermark, monochrome, background

[honest cave]

What kind of prompt can I do to get my character laying on their back on the ground? No matter what I do, my character doesn't want to lay on the ground. They want to sit or show their back to the viewer lol

[obtuse torrent]

@main vortex ...

[main vortex]

how did u get the white background?

[obtuse torrent]

@main vortex metadata is in the image,,, depend of models of course... look at side,white background, girl, elf, solid color background, simple background, transparent background

[main vortex]

how do i view the metadata?

[obtuse torrent]

@main vortex mouse ..right click on the image, in the dropdown menu choose open link, (this opens the image in the browser).. in the browser right click on the image and save to your disk, finally use the PNG info tab and open the saved image ...... but the prompt is "look at side,white background, girl, elf, solid color background, simple background, transparent background".

[main vortex]

Right click in SD or output folder?

[austere raptor]

Hello guys, can you help me with a prompt maybe ?

I'm looking for a ""similar"" lofi girl composition
But one of my friend tell me to add something like that

girl wearing amazing good looking Cyber suit, studying on desktop, sitting on a gaming chair with headphone, screens and bunny plush on the background ,adult big breast, made in the style of wakfu, dofus, ankama, Half Body portrait, Focused on the Face, perfect composition, hyperdetailed, insanely intricately detailed, wlstyle, WLOP, Makoto Shinkai, Ilya Kuvshinov, Rossdraws, 8k resolution, unreal engine, octane render, dynamic lighting, triadic colors, Greg Rutkowski, global illumination, CGI, 3D Character, Highly detailed and intricate environment, sharp focus, studio photo, intricate details

But all the artist name, style, ect break my actual check point
I edit into this one, it take the anime classic style, but i think the HUGE words about high details break some point

girl wearing amazing good looking Cyber suit, studying on desktop, sitting on a gaming chair with headphone, screens and bunny plush on the background ,adult big breast Half Body portrait, Focused on the Face, perfect composition, hyperdetailed, insanely intricately detailed, 8k resolution, unreal engine, octane render, dynamic lighting, triadic colors, global illumination, CGI, 3D Character, Highly detailed and intricate environment, sharp focus, studio photo, intricate details

Idk if it's really usefull to add too much "high details" words inside

[silver valley]

You need to set the controlnet model too

[obtuse torrent]

I really don't think the AI ​​could consistently render the pupils (unless it recognized the image as an animated object).. maybe highly descriptive prompts might help on some models + a high CFG ..... try adding "single colored glasses frame, single colored lenses",it will raise the consistency a bit

[main vortex]

manage to get white background but the character is not looking at side. Prompt used: look at side, white background, girl, elf, solid color background, simple background, transparent background

but when i use this prompt i get a background instead: look at side, white background, girl, elf

but its looking at the side

[obtuse torrent]

that doesn't have a standard solution... it happens a lot with anime models and it's a matter of the seeds, the way in which they were assigned the tags, and the cross attention that is caused... try adding "side view" and /o removing "solid background", or generating new images until a suitable seed appears for your composition.... if all fails....controlnet...(there are extensions in A1111 and external ones that can remove the background, I don't remember the names right now, but if you ask for that someone else can give them to you)

[obtuse torrent]

in SD only you know if a result works for you or not, and commonly we all fall into testing and comparison to decide, here are some examples for given parameters... from left to right... your first prompt, your second prompt, and the third is one to which I also delete everything related to details and CGI, and add "Absurdres" and "professional art" as a quality indicator ...
In everything else the same were generated (seed, sampler, model, etc.)

[austere raptor]

The first problem i have is , the high res

First pic = high res x2 Euler a
Second pic : no high res

Now i'm working on making perfect face

[obtuse torrent]

@austere raptor SD 2.1 model?

[austere raptor]

https://civitai.com/models/67994?modelVersionId=88143
Dis one

[obtuse torrent]

ok

[austere raptor]

🙂 i test the prompt of the first picture, to improve my skills inside of SD

[obtuse torrent]

ah but the first image does not match the description... it has a post treatment in the upscaling it was somehow modified the aspect ratio does not match

[austere raptor]

Ahh yes, i just edit that part and make 1080p pic i get this first picture

[main vortex]

can controlnet be used in sd?

cause my A1111 has graphic card issue installing

[obtuse torrent]

@austere raptor
I think the author generated the succubuss and the background independently and then used inmpaint to add it... or he added the data in an upscaling process ...

[austere raptor]

I can't send the pic i get, but one person help me to improve the art, my before after "trying" is really huge, that's just what i'm looking for

[obtuse torrent]

..Controlnet work for me at low res generations ....(6 GB VRAM RTX)

[main vortex]

what i meant is A1111 gave me graphic error even when i set my vram to low with set COMMANDLINE_ARGS=--lowvram

at the cmd side

[obtuse torrent]

@main vortex oh ...It's complicated, the operation of SD in A111 depends on your graphics card, how much VRAM it has and the OS too, the arguments vary depending on that, as well as performance, and the possibility of using extensions..

@main vortex Do you know the name and how much VRAM your GCard has? ..and your OS ?

[main vortex]

no idea, how do i check?

OS is windows 10

[obtuse torrent]

@main vortex serach for ... device manager....Display adapters ?...

[main vortex]

no idea why i have 2, im using a laptop lol

[silver valley]

Whats the problem with highres?

[austere raptor]

I'm just noobi trying to understand everything, to make better art !

[obtuse torrent]

@main vortex 1050 are 2 or 4 DDR5 ...
that must be the cause of the problems with Controlnet... I don't know if there is a resolution at which you can use it with that VRAM

[main vortex]

ya, so in summary the problem is low vram right?

[obtuse torrent]

@main vortex yes ...
although it may not be of any use for controlnet, I recommend installing the extension ... multidiffusion-upscaler-for-automatic1111 ... to use Tiled VAE...

@main vortex and for your original problem of getting a side view, if you can ... try the prompts on other models, there are differences between the models and some understand some things better

[main vortex]

Any recommended model for side view?

[obtuse torrent]

@main vortex really never tested it .... I think that the anime mix models, which have incorporated a photorealistic model... will have a little better understanding of these concepts (views, anatomy positions, etc.)... but I don't know which one is better... you should read the descriptions of how they were made... or you can check on the anime channel, or use a photo realistic model directly.. https://discord.com/channels/1002292111942635562/1091193032273043516

[main vortex]

We have an anime channel?

[obtuse torrent]

@main vortex https://discord.com/channels/1002292111942635562/1091193032273043516

[main vortex]

That is for posting anime images?

[obtuse torrent]

yes

@main vortex there are many of them who really know ... although not all have time to teach

[main vortex]

Guess i am alone on this side view quest for sd, i searched on this discord and not much topic on this, maybe what i am searching is not side view

[obtuse torrent]

@main vortex tags looking away ... looking at distance .... looking out of frame ... sometime work ... (for gaze direction)

[main vortex]

Im looking for something similar to azuki nft arthttps://opensea.io/collection/azuki

Not much people in discord mention about achieving a pose azuki, guess there is a reason why the artists are making millions lol

[obtuse torrent]

@main vortex ?

[main vortex]

close enough but i want a side view

like a fixed side view pose to mass generate

[obtuse torrent]

Agano have long hair ..SD wants to show it ..

[main vortex]

sd has an extension of control net, not sure if i could use it or not, i still do not understand the difference of sd and A1111

is it possible to ask SD to crop?

crop and zoom that generated image?

[obtuse torrent]

yes is a controlnet feature o Img2Img ..work ..kepp the pose change the character ..

[main vortex]

i can do a side view pose in control net?

never try it before, i have been using prompt so far

[obtuse torrent]

you need a reference .. controlnet are many models fpr diferents proposits ..

[main vortex]

is there a control net channel here?

[obtuse torrent]

@main vortex no specificly ... is a tool add on in A1111

@main vortex https://github.com/Mikubill/sd-webui-controlnet

[main vortex]

Oh, i installed that extension in my SD

[obtuse torrent]

@main vortex 🙂 yes but it need VRAM and VRAM for generations.. so is bad news for GCard ,,,

@main vortex
I just borrowed the one on the right... and with controlnet... I made the one on the left (the same one as before)

[main vortex]

Close but i need white background

Also how do u describe the art style at the right?

Like some of the lines look like ink

[obtuse torrent]

@main vortex
I have no idea of ​​art, that's why I use this AI, everything is formulas and experimentation... 😄 if I knew, it would be enough to identify the color palette and the general style ...and the bacground is white.... the red are clothes ..

[silver valley]

the word you search is profile:
1girl, white background, profile, sideview, modelshoot, masterpiece,

same prompt, different model:

[wary lance]

so here's the image i'm trying to replicate with the exact same settings within its civitai settings:https://civitai.com/images/456209?modelVersionId=41383&prioritizedUserIds=611798&period=AllTime&sort=Most+Reactions&limit=20

same model, CFG scale, everything, but my issue is that my image looks not very similar to the one that i'm trying to copy. my prompt differs in the way that my lora weight is lora:kivye_19:1, not the same as the original picture's prompt of lora:kivye_19-000018:1 (changed because the lora wouldn't work if i typed it in the same way as the original prompt)

copied image: https://cdn.discordapp.com/attachments/798626801140236288/1122523422715543682/image.png
original image: https://cdn.discordapp.com/attachments/798626801140236288/1122523524800708628/00072-4135477966.png

is there something i'm doing wrong? i'm also using the same model as the original picture

[orchid ore]

it seems you have not Vae, therefore washed colors. Also you can have everything same, but size and you get different result.

[digital echo]

Any tips for getting sheer/semi-transparent clothes?

[rich crow]

If I had to guess, your sampler is Euler a?

I'd not Euler a then it's one of the SDEs?

[stable valley]

Any good tutorials on how to use 2 or more colours on an object, i.e. a red and green hoody, or yellow and black sofa? i cant seem to get items to be more than 1 colour at a time.TY

[rich crow]

There was an extension that supposedly perfected that, wish I was at my computer so I could provide the link or name

It was pretty popular at the time that I learned about it tho so a search for SD/a1111 colors extension or something similar should have it come up

[stable valley]

was it a regional prompter?

https://github.com/hako-mikan/sd-webui-regional-prompter is the way i do it now

but in midjourney you can simply specify "a red and green striped cushion" for example

[stable valley]

you may just need to use a LoRa

https://civitai.com/models/26806/seethru

[digital echo]

Ok ty. I'll try it out! I imagine I'd need something like this for transparent bodies like ghosts?

[rich crow]

Oh I saw your reddit post with the generation info, ya besides the mentioned vae and image resolution, ancestral samplers (any that ends in a) don't converge, meaning you probably won't ever get the exact same picture. https://stable-diffusion-art.com/samplers/#Ancestral_samplers has a good demonstration of this

[obtuse torrent]

Possibly you are not doing anything wrong, except for the following..... the ExpMix_line model (base of that particular image) has 3 versions, each one differs from the previous one, this would be ExpMix_line V2...... all the versions of that model need a VAE.. the author of the model recommends this VAE: BerrysMix =** (kl-f8-anime2) **, ...also it is very possible that the author of the image would have used another similar one... finally the metadata information makes think that the image was generated at low resolution + Hires.fix x2 as upscaling .. possibly (512 x 616) x2 Hires.fix upscaler : R-ESRGAN 4x+Anime6B, DenStr:0.51

[obtuse torrent]

,,in anime models ..tag "wet" "wet clothes" "see-through"

[obtuse torrent]

..mmmm ghost... search for models models specialized in horror, those models have a better understanding of transparent bodies and/or translucency ghosts and specters.... https://civitai.com/tag/horror ....... as an alternative..... [girl: ghost:.5] ..... https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#prompt-editing

[wary lance]

i'm using a VAE and the image still looks pretty much the same, doesn't have the same style or thick outlines like any of the expmix_line models

[obtuse torrent]

@wary lance Although your image does not have metadata and you consult without attaching prompts or parameters, I am very sure that you do not use the aspect ratio, nor the upscaling recommendation that I gave you, even because of the name of the file, possibly not even the seed itself.. It is also normal not to get the same results because you can never know if you have replicated the same hardware or operating system, versions of SD, gradio, xformers and etc, plus random generation factors ..... Finally, I think that image was generated as an example of the Lora "Character", not the model expmix ....

[wary lance]

i forgot to update, i fixed it now

thanks for helping me

finally got the image to look similar in terms of style

[obtuse torrent]

@wary lance Comparing the generation capabilities of our local version of SD, copying example images is good, but trying to understand how the prompts are formed to obtain this or that result and how changes in the parameters or the models themselves affect the generation is even better 👍

[digital echo]

Brilliant, thx. I don’t have any anime models but that modifier trick is helpful!

[obtuse torrent]

@digital echo
..really I don't have any credit, I saw that example somewhere else where they explained how to use that feature of A1111....

[tropic yacht]

how to prompt img2img in bot channels?

[stoic wave]

i need some help making a good prompt, i am wanting to do a realistic photo of a women for the 1950's in a satin dress. and i woukld love to know what models you guys use for realistic stuff

also what models do you guys use for not human things like th inside of a car or house

[stoic wave]

iam looking to do something like this

or this

[stable valley]

you may need to supplement your models with Lora's for specifics like the dress
https://civitai.com/models/57214/haute-couture-or-ball-gown
and for insides try
https://civitai.com/models/28112/xsarchitectural-interiordesign-forxslora

[edgy tundra]

Like these?

@stoic wave

a scene from the movie "Gone with the Wind", 1950s, a beautiful woman from the deep south, wearing a billowing pink satin dress

[kindred needle]

why does some people do this : ((photorealism)) (masterpiece 1.3). Like why thay do this () ?

[orchid ore]

it is more attention on that words. (()) or (something:1.5)

[kindred needle]

and what is the 1.5 ?

[orchid ore]

it will have higher value than ordinary something
i dont know for example
woman with (yellow) skirt
will put more attention on yellow, so it should gave yellow skirt not a red or blue one.
You can read about it i believe in settings () is 1.1 (()) is 1.1x1.1 Better is used to (:1.4)

[kindred needle]

oh okkk

[rich crow]

fair warning you probably dont wanna go higher than 1.9 since it tends to (at least in my experience) give some wildly distorted results

[fickle idol]

anyone had luck doing a prompt with two people in it? eg. a picture with elon musk standing next to mark zuckerberg

I always end up getting two twins that are a merge of both people

[rich crow]

https://github.com/hako-mikan/sd-webui-regional-prompter

you can use this extension

you essentially make a box 1 and a box 2 and prompt for mark in box1 and zuck in 2

[fickle idol]

thank you!

[hexed forum]

How would you go on about creating a background for this doodle?
I want to create a Background in a similar style. But Im only able to create either realistic, anime or heavily stylized images... Either i can't find the right model or don't find the right words to use in a promt

[snow musk]

im trying to make a character look to the side with img2img, but i cant make it work, any prompts i could use?

[drowsy gazelle]

profile or profile view

[carmine epoch]

👋🏻 What do you feel is the expression shown here would be? please use "reply", so I get pinged
I didn't set anything specific in my prompt

[steep fossil]

A toothbrush next to a wrinkled fabric bag on a beautiful wooden table. Photo, hyperrealistic. Steampunk.

this is really disturbing. halp!

[carmine epoch]

I know this doesn't help any, but the wooden toothbrush is kinda cool

[drowsy gazelle]

try getting rid of the wrinkles on the bag sketches

[quiet zodiac]

@steep fossil What is that node based interface you are using called?

[steep fossil]

@quiet zodiac comfyui

[steep fossil]

having problems running my docker container again. :/ (i'm new to it). But I first tried without the lines anyway; it still gets confused.

[steep fossil]

I'm using https://github.com/AbdBarho/stable-diffusion-webui-docker/ for it

[steep fossil]

a bit better.. but still awful.

it just confuses all my words

I think I just don't use things right :/

Here's what it looks like if I just use a standard SD node layout (no controlnet):

I don't know how people make such amazing things in SD

[wary lance]

hey guys, how do i just use img2img but just change none of the contents of the original image? i just want to make it into the same style as the model's artstyle

[rich crow]

you could try using the image as a controlnet, try various ones. mabey combine it with a controlnet tile of that image as well

[crude crystal]

This is an attempt to teach myself the deforum settings by running through all of them one at a time, iteratively. It becomes NSFW on some of the later settings. https://youtu.be/Tf0Jj4wFzFs

[last hill]

any tips for getting dark or tan-skinned anime characters

[tired vigil]

Hello, I am beeing quite desperade. I am using Controlnet and Segmentation model (colormap, where each object has it's color code). The problem is, next to bed I need to have two mirrors, but SD is always putting windows there 😅 I think the problem is, that window and mirror have vary similar color in the colormap legend. I am trying now Latent couple, but no luck so far. Thanks for any help.

[steep fossil]

what happens without the orange things in the middle of the mirrors? (I can't find a label for that color in the color map)

https://docs.google.com/spreadsheets/d/1se8YEtb2detS7OuPE86fXGyD269pMycAWe2mtKUj2W8/edit#gid=0

[honest perch]

what width and weight did you use?

[tired vigil]

those are lamps. But without them nothing happens, still there are windows. But now, I finally managed to "convert" them to mirrors. Thank you for help.

[true shore]

Une lampe

[edgy tundra]

Just download it and look at the size. I usually use 512x768.

Size an weight is really irrelevant for this though.

[quiet zodiac]

Does anyone have any tips on prompting for spaceships/starships? While I get output with machines in space, they don't really make sense as functional crafts.
Is there some model that might accel for space scenes?

[fleet vine]

【Stable-diffusionのエラーに関しての質問です】
僕はStable-diffusionで画像生成しています。
昨日、いつもの様に「Google Colaboratory」からStable-diffusion web UIを起動させようとしていました。

添付させて頂いた画像の赤い丸で囲んだ「command line」の左の所の黒い丸をクリックして実行すると、いつもですと3つほどリンクが出てくるのですが、昨日は、青い丸で囲ったようなものが出て来て、もう訳が分かりません｡ﾟ(ﾟ´Д｀ﾟ)ﾟ｡

ネットで調べてみたのですが、何かが足りない？何かを構築し直す？様な感じみたいですが、よく分かりません.°(ಗдಗ。)°.

元々パソコン自体苦手と言いますか、難しく感じながらも日々頑張っている所です(๑•̀ㅂ•́)و✧

この添付させて頂いた画像を見て、
「あ〜、あれを直せばイイじゃん」とか分かる方がいらしたら、何をどうすれば良いのか教えて下さいm(_ _)m

なにとぞ宜しくお願い致しますm(_ _)m

[Questions about Stable-diffusion errors]
I generate images with Stable-diffusion.
Yesterday, I was trying to launch Stable-diffusion web UI from "Google Colaboratory" as usual.

If you click the black circle on the left side of the "command line" surrounded by the red circle in the attached image and execute it, usually about three links will appear, but yesterday, the blue circle I can't understand the reason anymore.ﾟ(ﾟ´Д｀ﾟ)ﾟ｡

I've searched on the internet, but am I missing something? rebuild something? It seems like something like that, but I'm not sure.°(ಗдಗ.)°.

You might say that I'm not good at computers, but I'm working hard every day even though I feel it's difficult (๑•̀ㅂ•́)و✧

Look at this attached image
If there are people who know, "Ah, it would be nice to fix that," please tell me what to do m(_ _)m

Thank you very much m(_ _)m

[crude crystal]

I've watched so many videos, tried so many settings combinations, can anyone tell me the current best way to get a consistent conversion of video to video? I want to take a video of someone dancing and, for example, replace them with Darth Vader, but I want it to look consistent and realistic.

[quiet zodiac]

No matter what prompt I use, I can't get a decent spaceship out of Stable Diffusion. It doesn't seem to understand an asteriod field/belt either.

a spaceship in an asteroid field by all artists ever born in the past present and future and their unborn children who are trending on websites yet to be invented

[lethal pawn]

ai/ml doesnt understand spatial relationships, or depth

[lyric goblet]

How do I get the Inpainting Sketch to only generate withing the masked area? No matter how big I make it there is always a little that is outside/unfinished.