📝｜prompting-help

[slow cedar]

50mb image size 😬

[atomic wigeon]

Ya. If you go in PlaygroundAI's Canvas you can drag the corner up to 10k. Largest free res you can get in the AI world. They give you 1k free images a day.

The only higher that I know of is Divinci Resolve (paid version). It goes up to 32k res.

[slow cedar]

I wonder what hardware they do have to make that, damn

[atomic wigeon]

I don't know. Most renders are 3 minutes there. StabilityAI's is much faster with their new T100's.

[slow cedar]

I will check it out for sure

[atomic wigeon]

Just know that if you do try out PlaygroundAI it has a mx image to image of 4k.

Not sure on Leonardo, Stability or the others.

[slow cedar]

Never thought of using a web since I made everything with my own pc, thanks for telling me

[atomic wigeon]

I make thousands of images a day. I do local, Playground Canvas and Playground Board all at the same time. Only local uses resources.

So you can work on an entire project and be rendering 3 images at a time for free.

[slow cedar]

Most of the renders I do is by leaving the pc run over night and such, but never thought about that

[torn cliff]

It's possible to generate the images on your pc and then scale them there too

[slow cedar]

Does playgroundAI have outpainting?

[atomic wigeon]

Yup. Even free accoutns can scale. Just upload and drag the corner. Only works in Canvas, not Board.

[atomic wigeon]

It has everything. Filters, edits, inpainting, outpainting, cropping, connecting images together, collaboration by multiple people, all free.

[torn cliff]

But no way to tell how many images you've used until you're almost out

[atomic wigeon]

Correct.

But honestly, I pay the $15 bucks a month and n3ever run out. It happens maybe once a month. But you can still generate, even after running out, but on a 30 second timer.

I'm starting to sound like a PlaygroundAI spokesman lol. They need to pay me a commission!

For the record, today I was told StabilityAI has a generator that's $7 a month for 1500 images. It has background removal.

https://clipdrop.co/pricing

3d concept art, female, medieval armor, orange and aqua spotlight, Final Fantasy style, medieval fantasy city background,

Notice how I change the spolight and it changes the entire scene. Tip.

[slow cedar]

What prompt should I have to outpaint an image? In this case, adding the rest of the body?

Im trying this as an example with "rest of armor, anime style" and it goes crazy

[atomic wigeon]

The same prompt. Use 60-70% image and let the AI fill in the rest. Or are you using a brush / quadrant brush?

Ya, same prompt

Get 60-70% of the original image in there. That's the data source.

[slow cedar]

[atomic wigeon]

😂

Why not just do "medium shot"?

[slow cedar]

Medium shot?

[atomic wigeon]

Medium shot = waist up

[slow cedar]

Again

[atomic wigeon]

Hips and up sometimes

[slow cedar]

Well, I'll try it later

[atomic wigeon]

The nice thing about making them 10k resolution is you can zoom in really tight and paint the eyes. Touchup the face.

[slow cedar]

There are many expressions that don't directly translate to their English equivalents in my language, so I'm unfamiliar with various ways of describing poses or types of photography shots

For example, the medium shot one;

[atomic wigeon]

I understand.

Watch this video. https://www.youtube.com/watch?v=Ng_GmJy_F8c&t=422s

It applies to all AI art

[slow cedar]

This kind of not knowing this expressions makes me think that many of my generations mistakes rely of myself making bad prompts because of it

[slow cedar]

Damn! I forgot I watched this video, thanks for reminding me

[atomic wigeon]

It's possible.

A South Korean company just got over 100 million dollars to develop a universal translation AI art software. In development.

[atomic wigeon]

Medium shot : https://youtu.be/Ng_GmJy_F8c?t=145

[slow cedar]

I would really like a suggestive correction type for sentences or similar expressions, like if you ask chatgpt for examples of ways to say something but based on the context of the prompt or something similar

[atomic wigeon]

Can you use google translate?

[slow cedar]

Yes, but Google Translate gives a literal translation, not a figurative one like how we speak

[atomic wigeon]

I understand. English has a lot of slang and adjectives (describe words) that dont match other languages.

Do you understand what slang is?

[slow cedar]

Its just like almost every language has its own unique vocabulary that distinguishes it from others when considering the grammatical syntax of different written languages

[slow cedar]

Yeah, those words are the best example for this case

like, most of the slangs if they are translated in a literal way they loose most of the meaning, or it becomes something totally different of what the message is supposed to be

[atomic wigeon]

This community is helpful. Just ask for best translation prompt for a certain mood. But many expressions can be obtained with two words "face pose". Or "random face pose".

[slow cedar]

I'll do, but mybe its just me

I do like to overthink what would be the best way to phrase this or that

(And mostly leads to disaster but nonetheless its a great way to learn phrasing)

[atomic wigeon]

Maybe the best way is a song, poem or book.

You can put any song in the AI and it will give you an album cover to that song

[slow cedar]

I've been trying to outpaint this image, but to no avail, it keeps generating a new one instead of outpainting the current one

[atomic wigeon]

https://www.youtube.com/watch?v=aU0jGZpDIVc Watch this.

[slow cedar]

I wont lie, I did make some prompts from parts of songs and I tought they would resemble the song's feelings, some did, mostly doesn't xD

[atomic wigeon]

Try different filters at playground. Each will give a different result. Often faster than local.

They have a lot.

I kept playing with your prompt. Gotta run. Have a good day!

3d concept art, medium shot, thin petite female, small upper chest, wild hair, windy, perfect face, medieval armor, red and black spotlight, badass, Final Fantasy style, medieval fantasy city background, stomach exposed,

[slow cedar]

Thanks for the help, I appreciate it

[torn cliff]

If that's a "small" chest I'd hate to see what it puts as large... Lol

[atomic wigeon]

That was cleaning up an image to image. It took her waist and hip down but would require 1-2 more renders to get her breasts in proportion to her body.

ReVAnimated is imfamous for sexualizing women. Mainly because part of its mechanic is to smooth and round off surfaces. It will trigger more NSFW's than any other filter. A similar one is founnd in PlaygroundAI and it's for SDXL 1.0 only. It's called "Ultra Lighting".

[torn cliff]

Yeah, I actually prefer smaller chests just because everything (like mangas) tend to go huge... v.v

[atomic wigeon]

ReVAnimated + anime = super inappropriate. Lot of us refer to that combo as "Pedophile Art". Many tried to get it removed from PlaygroundAI. They still block many images in their Rising social media thing and their Discord. Police get called now too.

[torn cliff]

Never noticed that myself, but then I have mostly been generating armor and clothes images.

[median ocean]

Hey there. New here, the options and setting here are so vast, wondering what is the Stable Diffusion equivalent to MidJourney's /blend and image referencing prompting? Done with img2img + controlNet? Any direction appreciated!

[proven jasper]

hello police someone posting lewd drawings

[atomic wigeon]

https://ojjdp.ojp.gov/programs/national-center-missing-and-exploited-children

[proven jasper]

let me call real quick can i get the name of the victim

[hollow igloo]

Anyone please

Help

[proven jasper]

it looks like cameliamix

[atomic wigeon]

The second and third image look like Ark: Survival Evolved Modkit (see Epic Games for the kit, it's free). Then he may have spun them thru AI with image to image with a prompt. The other two are fairly easy to recreate in AI. Controlnet set Edge to around 0.6 and Pose to 0.9.

[magic badger]

I must be doing something very wrong if this is the result I get with a prompt that includes the words dark-olive-brown-skin 😬

[floral arch]

Hi, can i combine both an image of mine and a prompt to create a new image or ideally an image using the pen of my first image?

[atomic wigeon]

Yes.

1. You can use a prompt alone.
2. You can use an image as your reference, combined with a prompt. This is called Image to Image.
3. You can use controlnet. This is a software that adds an edge detect line to what's inside of your image. It will lock it down so it cannot be altered. Then you add a prompt. The prompt will fill in the scene around those locked down lines. Example: Add a picture of you and your dog. Lock it down with "Edge" in controlnet. Now use "grass" in your prompt. You and your dog are now etched into grass.
4. You can scribble or accurately draw anything. The AI will attempt to build around it, based on your prompt.

[floral arch]

I would like to do something more advanced. I want to take a shot of a product and have it used in an image generated. Lets say i have a picture of a pen. I want to create a picture of a guy using a pen and the pen he uses is my pen. Is that possible?

[atomic wigeon]

Yes, but with effort. Make your pen in AI or use what image you already have. Cut out the background and turn it into a PNG.
Now make your character that's drawing. Use an eraser brush to remove the pen. Now apply the pen.
To do this well you need third party software. I recommend Davinci Resolve.

[floral arch]

But will the product be interacted with the character in the image? Like for me the real thing i want to do is this, I want to take this ops core military helmet which I have the picture of, I want to combine it with an image prompt like, human like cat that is wearing the military helmet in the picture.

[atomic wigeon]

Yes. that's much easier than what I described above.
Make a cat. Now stack your helmet on top of the cats head and save them both as a single image. Now send it into the AI as image to image, with a well written prompt.

PlaygroundAI Canvas or Davinci Resolve will allow you to merge images (flattened images). Edit: I think you can do this in Microsoft Paint as well.

Both are free.

[floral arch]

Seems i suck ahaha. Can you help me out?

[atomic wigeon]

I can assemble your images for you. Send them.

[floral arch]

Should i add you in discord or just send them here?

[atomic wigeon]

<--- Click that plus sign in the chat box.

[floral arch]

Basically want something like the cat on the right wearing the helmet on the left

is it duable?

[atomic wigeon]

Now that I cant do. You need an appropriate image to start with. Your helmet needs a front view (like the right image) to orient a cat under. You could simply use the right image with a prompt that turns the helmet black (and add features like velcro). But what you're needing is a lot of work. To incorporate your actual product (realistically) you would need a professional artist.

[floral arch]

hmm

[atomic wigeon]

However, you could try to find a near match and copy the model number and paste it into your prompt. It may produce a nearly identical helmet.

Do you manufacture these helmets or have a physical prototype?

[floral arch]

I have a physical prototype

[atomic wigeon]

Perfect! Now take a photo of the front of it. Prop it up with toilet paper rolls or anything you can find. Use a lot of light.

Maybe even a balled up towel. Just get it set to the angle and 3D view you want for your image. Then take a pic.

[floral arch]

fair

I will ask my brother who has the prototype to do so. thank you.

[atomic wigeon]

You're welcome! When you get the pic come back and we can help you with getting the background removed, getting it into the AI and spinning a good prompt for a cat to wear the helmet.

[floral arch]

Thank you so much

[karmic temple]

can someone walk me through on how to disable nsfw filters? or whatever else makes my images look black almost all the time

[atomic wigeon]

NSFW's are baked into Stable Diffusion, intentionally. As far as black and dark images go it's probably your model, prompting or version of Stable Diffusion.

In other words it's on your end. Try adding lighting conditions to your prompt.

[meager prawn]

how do i get rid of the black border on the top?

[atomic wigeon]

Cropping in third party software. Or try to adjust the camera (closeup, extrme long shot, etc.). Or a "border" and "borders" to your negative prompt.

[meager prawn]

name of 3rd party software?

just any image editing software?

[atomic wigeon]

Personally I use Davinci Resolve or PlaygroundAI.com
And ya, most video and image editing software will work.

[meager prawn]

oh alright, ill see if i can get rid of the border

[abstract gust]

How do I get the camera perspective to a view that only cuts the shins off from a character? full body is too much and cowboy shot is too less

[quiet zodiac]

In production, that is considered a faux pas, to cut someone off at the ankles. I did notice a Zoom Lora on CivitAI. It claims to allow you to zoom in and out of a composition with minimal changes.

[abstract gust]

okay... so faux pas would be the trigger word?

[languid panther]

What prompt should I use to make the eyes more realistic, it always gives it something, I don't know how to explain it but it makes it unrealistic

[quiet zodiac]

@abstract gust You could check out the theory here...
https://digital-photography-school.com/good-crop-bad-crop-how-to-crop-portraits/

[cedar gorge]

hi, i have this made this image, but i want to fix his eye. i wasnt able to do it with tutorials and now im asking here for help

[atomic wigeon]

This looks like crap (that's ok). All I did was use an eraser tool in the eye. But it's enough for the AI to fix it with one more image to image spin.

[cedar gorge]

what?

[atomic wigeon]

It's 10k resolution. If the image to image fights you lemme know and I'll send a smaller one.

[atomic wigeon]

What do you mean what? Throw it back in the AI image to image. One more time. The AI will repair the eye. 60-70 strength is optimal.

[cedar gorge]

yea playground cannot handle this

the site always dies

[atomic wigeon]

Dies? lol

[cedar gorge]

but all you did is erase the little dot?

[atomic wigeon]

Playground has a max image size of 4k resolution on image to image. Yes, in Canvas. Zoom in close and use the basic eraser tool. The AI recognizes it as a dfark color when you put it back into image to image.

[cedar gorge]

ok tomorrow i guess. i played around with it

[atomic wigeon]

Sec. I gotcha.

[cedar gorge]

its fine, i can do tomorrow

[atomic wigeon]

There. Image to image with your prompt.

That's 3k res.

[cedar gorge]

i tried it, but differently. it worked well for the dudes in the background

whats the image strength for some edit like that?

[atomic wigeon]

Typically 60-70%. Some SDXL you can go to 0%. It's weird.

Image to image is not an edit. It counts as a regular image generation.

[cedar gorge]

im not happy with that. it took alot of detail away

[atomic wigeon]

Are you doing it without a filter?

[cedar gorge]

that was the problem with everything i tried to "edit" today.

yes

image to image always messes things up for me. always turns out ugly

[atomic wigeon]

Add "sharp focus, intricate detail, flawless clarity, cinema quality, 32k resolution" to your prompt.

[cedar gorge]

or at least not how i want it to

[atomic wigeon]

Those are mechanical

[cedar gorge]

((best quality)), color illustration, ( steampunk engineer a beret on his head working on a machine), in background metal workshop with gloomy light, highly detailed, fantasy art

thats my promt

and it works fine, but for editing it doesnt

[atomic wigeon]

((( ))) stuff does not work on SD in playground

This is the best one I got. I think the "illustration" is what may be holding it back.

[cedar gorge]

but why is the original good, and all edits make it worse?

without changing the prompt

[atomic wigeon]

Any better?

[cedar gorge]

not really the style i was looking for unfortunately. also the still looks closed, but maybe i just try to give him goggles

[atomic wigeon]

I had a couple start giving him goggles. Trying other stuff.

[cedar gorge]

i mean i generated alot, and i generally liked this pose the most, some had goggles. i just thought it would be an easy fix with the eye. without generating the entire image again

[atomic wigeon]

The eye fix is easy. But will always require another spin thru image to image.

Try 90 strength if you think it's going down too much. And adjust more if needed.

[cedar gorge]

ok, ill try tomorrow again. it keeps saying that i hit the limit and im getting annoyed at this

[atomic wigeon]

Lol ok.

[cedar gorge]

but thanks for your time ive learned something still

[atomic wigeon]

Anytime bud. Have a good night.

[cedar gorge]

do you know when the daily reset it?

is there even something like that

[atomic wigeon]

It's a revolving 24 hours. So if you burned 750 images at breakfast, those 750 will be there same time tomorrow. The 250 each has a timer too.

That system drive me nuts. Still love the site tho.

[cedar gorge]

i did not burn through my images, but looks like through my edits. and its 00:25 so just past midnight. local midnight is for sure not the reset

[atomic wigeon]

Oh. I never once encountered the edits. I started my monthly sub after like 2 hours of using their site. Have been paid for the last 6-7 months.

Paid = unlimited edits.

[torn cliff]

Is there any way to change the model used, like using the X/Y/Z script or something, to generate a grid of images with different models using the same prompts to see how they differ?

[atomic wigeon]

There is "tiled" in your prompt. Other than that I honestly dont know. May "rendom genre"?

Euler has more variety

[cedar gorge]

i see. 15$ is not nothing, but its also a very cool tool. if end up using it more ill consider it. for now im just doing just for fun and use it i guess more than i realise on some days

[atomic wigeon]

Clipdrop is like $7 bucks a month. But it doesn't have the same toys as Playground. Both are good. $15 bucks or $7 bucks is better than that tokens or credits crap either way.

[torn cliff]

The idea of credits is stupid, it can take repeated attempts to get a good image, so you could technically burn through all your credits just to get one decent image

[atomic wigeon]

I concur. It's a deceptive business practice akin to Korean video games and gambling websites. In my opinion. I posted about it in the https://discord.com/channels/1002292111942635562/1008652627204132904 channel and met heavy resistance for my opinion.

[cedar gorge]

thats a cool cyberpunk version

the hat got a bit weird, but otherwise really nice

[atomic wigeon]

That's control traits with a different prompt in ReVAnimated.

alien with blue skin using a torch, realistic, space ship background,

[torn cliff]

I spent like 5 hours trying to adjust one image to get it decent, and that was on my machine... Sure it was taking almost 5 minutes per run, but that would still amount to a LOT of credits used

[cedar gorge]

lol i never used the thing i installed again after playground. i guess i can try edit there if im really desperate

[atomic wigeon]

That's why I'll only endorse sites like PlaygroundAI and Clipdrop. Monthly subs are a legit monetary model for AI art.

[atomic wigeon]

ReVAnimated? It's built into Playground.

Both Canvas and Board have it.

[next flint]

Use run diffusion or thinkdiffusion. both give 30 min free

[next flint]

and faster gen times means it wont take 5hr

[torn cliff]

Oh using SD Next on my machine it only takes 40 seconds

[atomic wigeon]

Well, Playground gives 1000 images a day for free.

[next flint]

oh damn

[atomic wigeon]

I think Clipdrop gives some free too.

[next flint]

sdxl?

[torn cliff]

Also when I said 5 hours I didn't mean to generate one image, I meant it took me 5 hours to get the image to be decent

[next flint]

yeah

[atomic wigeon]

SDXL 1.0, SD 2.0, SD 1.5 and Playground V1.

[cedar gorge]

no i mean SD that i have installed and can run on my PC. but its alot slower and playground is just much easier to use so i did not touch it after i "evolved"

[atomic wigeon]

Very true. SDXL 1.0 takes a beast of a PC to be fast. But it's free.

[next flint]

how did you get it to upscale to such high res

[torn cliff]

Not SDXL, not sure which version I'm using exactly.

[atomic wigeon]

Canvas in PlaygroundAI. Grab the corner and drag up to 10k res.

[next flint]

wow. how long does it take?

[atomic wigeon]

Instant

[next flint]

??

[atomic wigeon]

There's no timer. Drag the corner with your mouse and make it 10k. Now download. That's it.

Even if you made it local. Import it then drag.

[next flint]

is it a simple res increase or is it upscaling

[atomic wigeon]

It is an upsacle, without a timer.

[torn cliff]

You can also right click on it and select 4X Upscale which increases it's size by 4x.

[next flint]

damn

[atomic wigeon]

They found a unique way of doing it. Correct. There's a click. Board has "Face Restoration".

[next flint]

cant believe i didnt know about this. normally i just boot up a ui with faster hardware and rush to get it done before im out of time

[atomic wigeon]

Face Restoration = cleaner image
Upscale = retain more detail

[atomic wigeon]

I've been screaming it from the rooftops. LOL

[next flint]

well i know you talked about playground before when we were doing the photo thing a few days ago so i checked it out but cuz kinda just bookmarked it. never knew it could do stuff like this

[torn cliff]

I'm currenty using SD Next, generating images of 512x1024 with 50 steps, no upscaling, and it takes 2 mins 15 seconds. 30 steps was quite a bit lower.

[atomic wigeon]

Ya it's a full suite. They have a sizable team working on Canvas daily. It's their #1 focus.

[next flint]

it should make making larger images a lot easier. my workflow to make a final is already so complicated

[atomic wigeon]

Typically the max duration for PG is 3 minutes. But SDXL 1.0 is almost always faster in Canvas. Most are 60 seconds.

[atomic wigeon]

The problem with 10k - 32k tho is most image to image wont accept them. 4k max.

[torn cliff]

Yeah I think with 30 steps it was just 40 or 50 seconds. I was just testing to see if 50 gave any better results

[next flint]

i mean if u want something faster like i said rundiffusion and thinkdiffusion have a lot more models

[atomic wigeon]

Oh, my times were 150 steps.

30-40 steps less than 10 seconds or so.

[next flint]

wdym 32k. most i could get mine to was 10k x 6.5k

[atomic wigeon]

The only company in the world (that I know of) that offers 32k image and video is Blackmagic Davinci Resolve.

[next flint]

like for editing?

[atomic wigeon]

4k max on free version
32k max on paid version

Resolve is a video adn image editor. I use it daily.

[torn cliff]

I have no need for 10k or above. Very few places I'd use the images could handle the size anyway

[next flint]

i probably should start editing mine. Rn ive been just using claude for ideas and trying them

[atomic wigeon]

The cool thing about Resolve is that you can add tons of artistic stuff to oyur image and import it back into AI. That's how I make my comics.

It has edge detect. Super cool lighting effect when you do that and bring it back into image to image.

Before and after edge detect in Resolve.

[next flint]

nice

[torn cliff]

Weird, was running the same prompt through different models, and one just generated a black image. Like completely black, nothing in it... And after it finished it shows "Grayscale input: True"

[atomic wigeon]

Here's what it looks like before importing back into the AI.

[next flint]

so its just a better softedge controlnet?

[atomic wigeon]

It's meant for neon effects.

[next flint]

oh that makes sense

[atomic wigeon]

I exploit to make comics.

And lighting effects.

You can change the color of it too. And thickness. Then save it and do it again. Great for soundwaves.

[next flint]

how long does it take you to make a comic

[atomic wigeon]

5 minutes to an hour. Depends on if I'm trying to retain an actors likeness.

Or politician.

[next flint]

oh i was thinking u were doing like comic books but comics makes more sense

does playground support loras?

[atomic wigeon]

It has built in filters.

Lots of em. More added all the time.

[next flint]

thats an interesting way to make it work

instead of models it uses bases and then makes filters out of different models?

[atomic wigeon]

Pretty much ya

I'm not sure on the tech side how they do it. Fun tho.

Gotta run. Have a good evening guys.

[next flint]

im guessing its just making a batch of images from those models and making a lora from it. you can normally tell what the style of a model is by just giving it a basic negative prompt and no normal prompt and letting it do its thing

[robust pewter]

Hello! It seems crashed right now. I like to use it, but it is throwing an error.

[quick thicket]

Is there an optimal way to position loras and embeddings within a prompt? For example, having all of them at the end of the prompt?

[solar dust]

The position of your prompt does matter when generating.

[tired vigil]

how can i get this type of mouth

[magic badger]

Is there a word for this thing on the forehead I can prompt against? 😅

[quick thicket]

Does position even matter for Lora’s?

[next flint]

well a lot of the time the closer to the start the token is the more it is prioritized ( to my knowledge ) so i would assume it would still matter a bit

[next flint]

easiest thing to do would be to just making some more generations and hope it goes away

[next flint]

closed mouth?

[quick thicket]

does it matter as much as any associated tokens with a particular lora?

[magic badger]

It's such a good seeeeed thougghhh 😦

I am getting there though. I filtered out diadem, bindi, and now fringe, too 😅

[next flint]

i mean you could try a small amount of seed variation

[next flint]

i mean i assume its the same. if everything had a weight of 1 then priority would be given to tokens coming first (although some may have more effects than others)

[next flint]

i was able to remove it with the eraser in playgroundai

but im asuming you could also do something like that in the img2img tab

[magic badger]

Yeah, that strategy makes more sense 😅

[open basin]

Why does SD, and I mean all models I've tried... seem to have trouble making a kid with a scuba tank? I mean. it should not be hard. I can make baby yoda wearing shaolin monk robes, but can't strap a steel cylinder to a kids back, in fact everything scuba related with kids, the flippers look, weird too, on 10 differnt models is ... just odd. just tried and it seems even with adults it does not like to even try. even here underwater, there is no tank... lol it looks like she has a neoprene dive tank wrap on her bank, its flat and no tank inside it lol

+scuba diver, (wearing scuba tank on back), full body wetsuit, dark blonde hair, hazel eyes, neutral expression, (square face)

-nsfw, nude, bad quality, blurry, out of frame, logo, signature, mask, covered face, goggles, glasses, deformed, extra limbs, three legs, three arms

[open basin]

never really tested it. Set non random static seed, move it around the prompt, front , mid, end, see the changes it induces.

+scuba tank
by itself, as a single prompt, results in some interesting things. This the first object I've come across that I have this much trouble with lol

Anyhow, any suggestions on getting something more than this for something like:

+steel scuba tank on diver's back

[atomic wigeon]

1. Why are you adding a "+"?
2. That's a rebreather.

[open basin]

try 'neutral mouth'

@atomic wigeon I'm not, that was to denote prompt and NOT

+positive prompt
-negative prompt

[atomic wigeon]

So you're not actually putting it in your prompt then?

[open basin]

@atomic wigeon yeah it looks like armor.

no

not at all lol

I don't need THAT much help 😆

[atomic wigeon]

Ok. Because letter, number and symbol shapes in the prompt design the image.

[open basin]

ahhh oh yeah no, that was just happenstance

[atomic wigeon]

The shapes themselves.

Cool. Just mentioned it because most people dont know that.

Ok gonna try your prompt.

[open basin]

nevermind, i didnt' paste that one, another had armor looking mess like above, but with a big cross + in the embossed

aight

[atomic wigeon]

So you want a kid swimming, under the water, with a full scuba set?

[open basin]

I didn't care where, on a boat in the ocean, underwater, but basically this really was just me genning something to send to my gf, cos the little one and her at the beach and little one swam way out and she said "she needs to borrow your diving gear" so was going to just make a quick image, and now spent a while on it. lol

I usually don't really delve into people much with specific clothing / gear, so may be goofing something,

[atomic wigeon]

I'm curious too. So spinning some prompts.

[open basin]

so you having trouble getting tanks too?

I have tons of diver images, in full gear...minus tank...

maybe 200+ today already

[atomic wigeon]

Your prompt was 25% with tanks. This one was kinda hot so I saved her to my Discord. 😆

[open basin]

lol

Calling it a diving tank spits out military tanks

[open basin]

air tank, you MIGHT get one of those home propane tanks or something

[atomic wigeon]

pre-teen male scuba diver, wearing scuba tank on back, full body wetsuit, dark blonde hair, hazel eyes, neutral expression, swimming under the surface of the water in the tropics,

[open basin]

lol "sort of"

But more than I get

[atomic wigeon]

So what's mising? Deeper water? More tank? Rebreather?

[open basin]

Nah nothing It works fine for my purpose

just funny it still isnt' producing steel cylinder tanks... wait a moment..

[atomic wigeon]

Brushed aluminum for metal stuff like that. Or "brushed aluminum spotlight". Or "brushed aluminum highlights".

The word spotlight will alter specific items in a scne. Works with any color, color combo or material (like burlap for example).

So if you have a superhero and you want to change his armor or outfit, try "red and blue spotlight". Should work with your diver too.

For some reason it works much better in watercolor than it does photorealism. Not sure why. I still cant get metal tanks either.

[orchid ore]

try oil painting very detailed. 🙂

some things can be done in photos

[open basin]

ahweiird

[atomic wigeon]

I will try that. Gonna try ReVAnimated next. I'm sceptical tho. Will prolly trigger 1000 NSFW's in ReVAnimated.

[open basin]

Yeah, without the nsfw / nude NOTs it was giving me vinyl "wet suits" :\

like shiny plastic, s&m weirdness

[atomic wigeon]

lol

You're gonna get me in trouble with the FBI.

😂

[open basin]

nah, it wasn't crude, just the style of body suit, which itself is uncomfortable, but I nsfw neg'd it real quick.

what does CLIP say about the painting's tanks i wonder

no mention of the tank at all

[atomic wigeon]

My first ReVAnimated. That was image to image at 0% to help push it.

[open basin]

lol

uer

[atomic wigeon]

lol

[open basin]

impressiive.. but

[atomic wigeon]

I would save that one for politcal satires.

[open basin]

a propane tank with a black top and a white bottom and a black bottom and a gray bottom, Constant Permeke, water, a stock photo, plasticien,

that's clip's view of this

[atomic wigeon]

You could clean that up and use image to image

[open basin]

I accidentally had loopback still on :S and controlnet from my baby yodas I was doing :\

that's why it was so weird lol

[open basin]

that was what had that control net / loopback all goofy when I tried to gen in img2img lol

[atomic wigeon]

🏆

[open basin]

cos I forgot i had been making those, and then got on this. so the diver painting went kinda off

something is legit wrong with air tanks lol

so CLIP interrogate spat this out

a blue propanent on a white background with a clipping for a clipping for a clipping for a clipping for a clipping for a clipping, Aquirax Uno, blue, a stock photo, plasticien

Something with these things really goofs stuff up. lol

[atomic wigeon]

/facepalm

[open basin]

hey if you want to use a 10 gal paint bucket for air, go ahead

a blue propanent on a white background with a clipping for a clipping for a clipping for a clipping for a clipping for a clipping, Aquirax Uno, blue, a stock photo, plasticien

[atomic wigeon]

Maybe an acetaline tank for welding?

Or propane tank?

[next flint]

was there a version of stable diffusion that was popular in 2022 that had prompting similar to the stuff used today?

[open basin]

never seen CLIP interrogate spit out repetitian like a convo llm

[atomic wigeon]

SD 1.5

[next flint]

k

[atomic wigeon]

Oh wait, 2022? Wasn't SD born in 2022?

SD 1.5 prolly wan't out yet.

But I still use SD 1.5 daily. That image above in ReVAnimated is SD 1.5

[next flint]

i checked and 1.5 came out in like october i think

[open basin]

....

a small girl wearing an acetylene tank on her back

it dislikes tanks, lots

[next flint]

gotta make sure the ai im using understands the 1.5 style of prompting instead of older versions

[atomic wigeon]

Maybe "bottle"?

Diving cylinders have also been referred to as bottles or flasks, usually preceded with the word scuba, diving, air, or bailout.

google

[open basin]

ugh wtf this, even with nsfw / nude :\ it had a microbikini bottom on. why is scuba photo real so f'd

also what is with this tank? it's same kind twice, even though searching on google isn't like those

[atomic wigeon]

It associates what people wear when they visit the ocean. Most diving is saltwater.

[open basin]

I know I have 4 diff certs myself

but this tank here, I tried acetlyene, it looks like it has very low training for it

cos same tank with same lettering keeps popping up

Oh yeah

[orchid ore]

if tank is issue you can try air high pressure bottle? or something?

[atomic wigeon]

I have tons of certs too.

[open basin]

well I mean I have wetsuit in it

that was NOT wetsuit lol

lul

[open basin]

air tank on back of a subject, in diving gear,

[atomic wigeon]

Nah, I tried diving twice. Not for me.

[open basin]

If I can see in the water, okay, if not, it is pure dread

[atomic wigeon]

Turns out I cant light a cigarette in the water or near the tanks.

[open basin]

the tanks are safe, lol it isnt' o2 😛

it is room air

I also work on an ambulance THOSE tanks... don't

[atomic wigeon]

Oh. See?

[open basin]

We transported a psych pt like 6 hour triip to another hospital. Had a mask on him cos he was rec. from covid

this guy tried to light a cigarette when I was filling out the tx report, and

Shwoooooop

[atomic wigeon]

I did a little firefighting in the Army (as a punishment). We had oxygen tanks.

[open basin]

nice bloop of flame in his face

is fire SCBA not room air? I'm sure there's no mix

[atomic wigeon]

They use multiple tanks.

[open basin]

they call it Oxygen but, I mean it "is" in regards to where you are

oh dual can?

N/O2?

[atomic wigeon]

No. Seperate

[open basin]

oh, hmm II mean fire carries O2 for the vics I'm sure

never gone in on a fire, just outside

[atomic wigeon]

Some react to aircraft fuel, some dont

[open basin]

ah

[atomic wigeon]

From what I understand

[open basin]

o2 reacts all things :p

[atomic wigeon]

I'm not the guy to ask. I hated that job.

[open basin]

Nothing lol , it hates air... AI is against bottled air

[orchid ore]

o.k.

[atomic wigeon]

Appearantly. I'm trying one more. A filter on playground known for masks and tubes.

Geometrieva style

[open basin]

the hell?

a small girl wearing a large O2 tank backpack, wearing wetsuit

[atomic wigeon]

The first batch with I2I was mass murder.

[open basin]

that...turned out well

very non-eucldian

also, everytime I use control net for positioning with scuba, it puts black plastic glasses on the subjuct lol

[atomic wigeon]

The second set was terrible too. Sorry, I tried. Looks like you found the new red ball in blue box.

[atomic wigeon]

Because controlnet is SD 1.5 only

[open basin]

I was using 1.5 models

[atomic wigeon]

To my knowledge they dont have a version for SDXL yet

[open basin]

the xl did the same,

no controlnet

just no air

[atomic wigeon]

Ahh

[open basin]

1.5 models too

so went to my other models, and they have the same attitude problem about it. It must be training data

what was that site

[atomic wigeon]

There is a way to prompt this! 100%.

[open basin]

where you can search training.

like training tokens

i think it would pull up training data images and their annotation labels

[next flint]

i mean they do but its for comfy ui only

[atomic wigeon]

Sorry, don't know. The only local I have is not SD. For SD I use playgroundAI

[open basin]

my comfyUI looks like hell

like the 4th image

above

oh, I use all local

[next flint]

its a pain to use. there is also sd next which i believe also allows for contronet sdxl since it kinda uses comfy as a backend

[open basin]

it is not at all 🙂 Look how simple

bahahaha

I have a small problem..I believe, but it works.

[atomic wigeon]

Looks like my Davinci Resolve fusion sets lol

[open basin]

those are multi layerd prompt boxes on the left, eg. subject / location / misc / details / atmosphere/ and so on

[next flint]

on that topic. i tried downloading davinci and editing photos but it kept adding black bars around my media. is there a way to disable that

[open basin]

they combin into clip enc. but there's a second input set for just subject in the pregeneration, which is passed via I2I to the next set and refiner

[atomic wigeon]

Zoom in. If you're using 1024 x 1024 set your Resolve to 2160 x 2160 HD. If you use 4k widescreen use 768 x 1344.

[next flint]

do i have to add photos to the timeline to edit them?

[atomic wigeon]

Click the cog wheel in bottom right corner to set. Dropdown menu. Wont change anything if stuff is already in timeline, so do before.

[next flint]

k

[atomic wigeon]

Click these 3 dots to set default.

[next flint]

k

[open basin]

loopback is the best script in the world, FYI

I do wish there was a better step between higher noise than there is, like a tweening element of some sort

[atomic wigeon]

Nice job. ^ is why you guys want to learn Davinci Resolve. It's the future of AI.

[open basin]

This was SD :p

3

[atomic wigeon]

I know

[open basin]

Also, the end of that video. . The prompt on those images was "sunlit field of sunflowers" and it turned into some weird mayan pixel art

I have no idea WHY it went that way it was very strange

[atomic wigeon]

You can do amazing things with those clips in Resolve.

[open basin]

i have no idea what it is .

i just use command lin ffmpeg lol

[atomic wigeon]

Every movie in Hollywood goes thru Davinci Resolve. It's an image and video editor. And colorist software. And many, many other things. Upgrade to Premier Pro. Made for the film industry. Used to cost $17,000, now it's free.

Blackmagic owns it. If you know studio tech and cameras. It's them.

[open basin]

haha hrm

[atomic wigeon]

That's SD 1.5? Does it do it with SDXL 1.0 too?

[open basin]

this is all sd 1.5

[atomic wigeon]

Bummer

[open basin]

it's using a loopback script , that just refeed i2i

[next flint]

in davinci if ive made a window and made my edits how do i make a new window for different edits?

[open basin]

So what I do is I use I2I with a start image and prompt, run 12 frames (1/2 second at 24fps) then tweak a little, every 12 frames

then i switch to new subject and bump noise from .3 to .38-.42

[atomic wigeon]

Different edits to the same project?

[next flint]

i still dont understand this much. i added some images and put one in the timeline. ive made a window around clothing and am making one around the head but i want them to have different edits in the color tab

[open basin]

So I guess it is probably better than the video editor I use, Olive, which is free to, but early dev really

[atomic wigeon]

Resolve is the best editor in the world. But comes with a steep learning curve. Easy entry, hard to advance. Hundreds of thousands of Youtube videos out there.

[open basin]

I am not sure what I would do with it, aside from deflicker my videos from i2i

[atomic wigeon]

Right click image and make Compund Clip. Now Alt left click and drag to replicate.

[open basin]

if I can

lol

[atomic wigeon]

Resolve has millions of possibilites for videos. And images too. They are one in the same.

Film = fast moving still images.

[open basin]

you know what I should look into throwing together. I wonder if I could write a little blender plugin to generate batch openpose models with motion tween for frame animation

I bet I could do some sick stuff that way

I am sure something akin is out there, must be

I never think of things first

[atomic wigeon]

Try Resolve. Also Wix Studio has built in AI scripting.

[open basin]

I have downloaded it now

[next flint]

but like how do i just edit images tho cuz the timeline gives me brain damage

[open basin]

LOL

photoshop timeline will too

[next flint]

also its just an image so i cant compound anything

[open basin]

timelines just be like that.

[atomic wigeon]

Dont download Resolve Studio, that one is the paid version. Get the one on the left side.

[next flint]

ps was decent but i aint paying for that

[open basin]

yeh I got the cheapo one

[next flint]

got like a 1 month thing for free but that was like 2-3 yrs ago

[open basin]

@lord, as student, I pay 19/mo for entire suite

everything adobe

so I find that useful.

[atomic wigeon]

The timeline is the edit process. Remember, Resolves primary use is video. But videos are images and treated the same way. By default when you load an image in Resolve timeline it is a 5 second video.

[next flint]

damn

[open basin]

And lol, I'm not a typiical student lol, I am enrolled for CEU cont education for my medical license with the state, so that still is techinically a student lol

[atomic wigeon]

PlaygroundAI + Divini Resolve = no need for Adobe ever again.

[next flint]

okay but how do i separate the edit "zones"

[open basin]

I take like 10 hours a year

so it's a good price for me

Playground AI? is what.?

[next flint]

cuz these are getting the same edits and that works out rly weird

[atomic wigeon]

There's way too much for a Discord conversation bud. Youtube has hundreds of thousands of videos on Divinci Resolve. I recoomed Casey and MrAlexTech. The two best Resolve content creators.

[next flint]

lol alright ill check online and see if i can fix it

[open basin]

oh playgroundai is a remote image gen

ok..

[atomic wigeon]

It's far more than that. It's image editing and everything in one spot. Look at their Canvas.

[open basin]

Yeah I mean it works fine locally too, plus I can hack apart some extensions for webui and twist them into a little bit nicer forks

ah

[atomic wigeon]

Ya, local has its benefits

Clipdrop has some toys too.

[open basin]

https://github.com/nonnonstop/sd-webui-3d-open-pose-editor see like this kind of thing I love, if it had animation tools, even minor ones, it'd be killer

it does not yet support facemodel I don't think so,

does some kinda goofy stuff

she looks like she is on a coke bender, but okay...

wait I don't know why i did not think of this

[robust totem]

General question regarding inpaint. Let's say I prompt for a picture frame that has flowers in the corner, and let's also assume that if I use the same seed and same settings, I can remove flowers from the prompt to get a very similar looking picture frame without flowers, if I desire. How can I take an existing image with flowers in the corner, and then remove the flowers using inpainting such that the result is as close as possible to the image generated by the prompt without flowers? All inpainting techniques that I've tried significantly modify the corners to be different than the version that never had flowers to begin with.

[solar dust]

How do I remove an extra finger with inpainting?

Which one of these options should I use?

[exotic vault]

guys i have a question not for the non good but how to actually bypass this thing

[round dagger]

So here's what I was wanting. 'Stock images' with a white background. 100s of them. Of everything. Books, forks, bowls, doggos, tv remotes, flowers... everything.

And of people and animals performing basic actions.

How would you guys go about mass producing these, as high quality as possible while avoiding the stereotypical stock watermarks/white lines through it. Where would ya start with model, loras and prompts? I can make a sexy anime girl with 1.5 XD but looking for something completely different now. Realism with white BG. Any help and initial guidance would be very much appreciated.

[static surge]

how to upload our image here?

[grand crown]

I am doing product background changes via impainting models.
The issue I face is to describe the background in the prompt which highlights the product. In my generation, images look cluttered.
Can anyone point me to resources where I can get prompt inspiration for product photography in inpainting setting

[open musk]

can anythingv5 generate nsfw?

@silver valley ive tried multiple times, it always resists nsfw

and i dont know why that is

[silver valley]

Its not made for it but It can

[open musk]

do i need to assist it with something

other than prompts

[silver valley]

No, but I would recommend using other models for that

[open musk]

i heard it is the best one before

any ideas?

[silver valley]

We dont support nsfw here so I can't help but good anime models are Aom3 and Koji_v2

[open musk]

alright thanks

[proven jasper]

hasaku is good too

[hollow igloo]

I’m looking for the same answer but no one seems to reply

[marsh trellis]

If anybody can DM me for img2img game texture remastering on StableDiffusion (Automatic1111) that'll be really great. Thanks in advance

[exotic vault]

STILL LOOKING FOR HELP!

[torn cliff]

Isn't that a scam/hack that was going around?

Especially since it's posted in every channel

[hollow tapir]

yup don't worry about it.

You didn't see anything. jedi trick

[hollow igloo]

Found the solution yet or no?

[open musk]

so basically, im generating two characters and i want only one of them to have abs but both do... how do i do this

[cunning compass]

guys, help me create a similar style photo
I do everything according to the instructions, but it does not work

https://civitai.com/models/84586/aam-anylora-anime-mix-anime-screencap-style-model

[solar dust]

Someone told me that all models can.

Don't know if that's true or not.

[atomic wigeon]

Almost all of these questions would be better answered in the https://discord.com/channels/1002292111942635562/1072238304042438758 channel. This is the https://discord.com/channels/1002292111942635562/1011743094309396631 channel.

[open musk]

i just asked a question about prompts does that also go in sd chat

[open musk]

i wanna learn how to add prompts to a single character not affecting a second character in a generated image if that spossible

[atomic wigeon]

I honestly don't know if that's even possible at this stage. Typically for something like that I am forced into third party software.

I make two seperate character and then cut them out. Then merge them into a scene. But technically, inpainting can be used to achieve this. With separate prompting for each character. But I have had bad results from trying it. I still have a lot to learn about the potential of inpainting.

But to my knowledge, there is no way to prompt specific details with multiple characters in a single image, at least not in great detail. The AI might ballpark you, but inpainting or third party software will still be required for good quality results.

[solar dust]

It definitely is possible.

What I usually do is type something like this

for example

if I want a couple with different clothing I write

1boy (denim jacket, black pants, glasses, sneakers) and 1girl (china dress, necklace, heels)

and then the rest of the prompt

it may not work the first or second time but it's definitely an option.

But you're better off just using regional prompting for things like that.

[open musk]

i didnt know this was possible

dang

its like sup prompts

1boy (bla, bla, bla), 1girl (bla, bla, bla)

nice

speaking of which, what are brackets normally for?

(monochrome:1.1)

i see this for example

why is it in brackets and how does the weight after monochrome work

[tired vigil]

Hi very nice picture miku.

[cunning compass]

Yes, I want to do something like this, but it doesn't work.

[solar dust]

The brackets are for weight

just like the value in the prompt.

I don't remember the exact weight for each bracket tho but I know that the more you put in it, the higher the weight

((prompt)) has more weight than (prompt)

(((prompt))) has more weight than the previous two, and so on

[cunning compass]

also works with unwanted elements?

[solar dust]

so in that case the values are added to add yet another layer of weight to the prompt

(prompt:1.2) has more weight than (prompt).

[solar dust]

Afaik, yes.

If you're talking about negative prompts.

[tired vigil]

Ah, and you tried to write the same program there that you wrote there by switching to that anime . @cunning compass

[cunning compass]

yes

[tired vigil]

Aha and it worked.? @cunning compass

[solar dust]

I also suspect that the weight in prompts can be used as a sort of slider

like

[cunning compass]

Nope, I didn't get it

[solar dust]

if you put (long neck), it will give your character a long neck

but if you put

(((((long neck))))) or (long neck:1.5) instead

I think the size of the neck will be even longer.

I didn't do enough tests to be sure but I've seen people prompting like that to get bigger proportions.

[tired vigil]

Ah.Now I thought you tried to put there a similar program , the same thing written just a little bit something written there differently. @cunning compass

[cunning compass]

can you help with this?

[solar dust]

I wish someone could teach me how to use regional prompting.

I'm tired of using the 1person (x y z) and 1person (x y z) method, it's not as reliable as the former.

[tired vigil]

I use a different site. Callsasa Prodia.

https://app.prodia.com/

[cunning compass]

oh

ok

[tired vigil]

On this one, I just write a girl or a boy there and describe the hair color, eyes and background. Only before this must be given a number One or two that's up to each

[solar dust]

I generate images locally.

So I don't think I can use that.

[cunning compass]

guys, who can help me create the character I need, I can’t (

pls

[solar dust]

??

Just describe the character to the AI.

[tired vigil]

Ah, because I use this site and it works there, so nice picture was written there, and then I found out how it works. Just when he's doing anime . They are only there, Aniting V3 is used, for example, there are others, when it is clicked on it can be vibrated . and that image is vigenerated.

https://app.prodia.com/

And what is the name of the site you are using @cunning compass

[cunning compass]

i try but it comes out soapy and poor quality and also has problems with hands

[solar dust]

You have 0 negative prompts

of course you'll get bad results.

Also where's your VAE?

[cunning compass]

what are they needed?

vae?

what is it?

[solar dust]

Negative prompts are needed so the AI can know what you wish to avoid in your images

in this case, you want to avoid poor quality and soapy textures.

And about the hands

that's to be expected

AI struggles a lot when it comes to hands and feet.

You gotta use LORA's, prompts and plugins to try and circumvent that, even if just a little.

[tired vigil]

That page must be downloaded @cunning compass

[solar dust]

And you must keep trying

AI is all about trial and error

[cunning compass]

can you recommend good lora

?

[solar dust]

never assume that you're gonna get the perfect results after trying once, even when your prompts are perfect.

[solar dust]

I'd advise you first get yourself a good VAE for anime images

I recommend klf8Anime2

[cunning compass]

https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt

this?

[solar dust]

or vae-ft-mse-840000-ema-pruned

both are really good for 2D anime images.

Yes.

[cunning compass]

where to store it?

[solar dust]

Download it and put inside stable-diffusion-webui/models/VAE.

[cunning compass]

I downloaded

how to run it?

[solar dust]

There should be an option for you to select the VAE

on the right.

[cunning compass]

I don't have it for some reason

[solar dust]

That's strange.

Does anyone else know the reason for this?

[cunning compass]

i restart program

wait

[solar dust]

I know

go to settings

on the left, click on user interface

[cunning compass]

Nothing showed up(

[solar dust]

Go here first

[cunning compass]

yes

[solar dust]

then go here, on the left

then look for this thing right here

[cunning compass]

нуы

[solar dust]

You should have this written on the line

if you do not, then that is the reason why you cannot select a VAE

[cunning compass]

right?

[solar dust]

just write

yes, that's it

now see if the option shows up.

It should appear now.

[cunning compass]

no(

[tired vigil]

And what you wrote there, I will try to see if it does not blame me so similar, I will write to you whether it worked or not.You just have to send me what you put there.

[cunning compass]

oooo

i seee

right?

[silver valley]

Yes thats right

[cunning compass]

what's next?

@solar dust

[solar dust]

Just start generating for now, also try using better negative prompts.

And don't forget to select the VAE you downloaded.

You can use another sampling method if you want

I use either Euler a or DPM++ 2M Karras for anime.

[cunning compass]

ok

[silver valley]

Also try a resolution of 512x768 at start

[tired vigil]

Can you also send me the page pleaseBut no photo, please send me the booth because it shows me otherwise.

[cunning compass]

that's what happened

@solar dust

[solar dust]

It did turn out way better.

[cunning compass]

what else can be done to improve the quality of the photo and add details?

[solar dust]

Try using a different sampling method

this is the one I go for when it comes to anime

you can also increase the resolution of the image a little

try adding (masterpiece, best quality, ultra detailed) at the end of the positive prompt. It should help but do not expect it to have a large impact on the results.

[solar dust]

Keep in mind that, depending on your GPU, it may take a long amount of time to generate images when you do that

or even fail if it does not have enough VRAM.

[cunning compass]

you can tell how to be sure that you will not lose quality

[solar dust]

First I need to know how much VRAM your GPU has.

[cunning compass]

i dont now😅

[solar dust]

You need to find out then, that's a very important information. If you have a low amount of VRAM then your options for improving the quality of the image will be limited.

[cunning compass]

ok

wait

6GB

as I understand

[solar dust]

Okay, that's not much

I recommend you put the --medvram command in your webui

[next flint]

how about adetailer

[cunning compass]

how to do it?

[next flint]

i mean it doesnt cost any vram and should help with overall quality a little bit

[solar dust]

Yeah, but 3ntry wants to improve the image quality as a whole for now

Adetailer works for specific things like face and hands

[next flint]

i get that but adetailer is a step in the right direction for at least improving the subject

[solar dust]

I agree, I just wanted to start with the basics fist

first*

like putting commands in webui such as --medvram and --xformers.

[next flint]

makes sense

are you planning on having medvram and xformers for hires fix?

[cunning compass]

how to do it?

[solar dust]

That's for 3ntry to decide.

@cunning compass First you will need to close stable diffusion.

[cunning compass]

ok

i close

[solar dust]

Now go to the stable-diffusion-web-ui folder

look for a file named webui-user.bat

right click the file, select 'edit' and show me what is written on it.

[cunning compass]

this?

[solar dust]

That file here

right click it, select 'edit' and then show me what is written on it.

[cunning compass]

Which one of these?

[solar dust]

The one that has this written on it.

[cunning compass]

ok

can't find click button

this

[solar dust]

There it is

now

I'm gonna give you some commands for you to write on the notepad you opened. When you're done, it should look like this:

[cunning compass]

ok

[elder saffron]

do anyone know if you can save prompt-style presets in comfyui?

[solar dust]

I'm gonna explain the commands:

--autolaunch is so you don't have to type the IP in your web browser to open stable diffusion anymore. When you have this command, all you have to do is just click the webui-user file you edited and stable diffusion will open automatically

--xformers and --nohalf are for improving performance and make you generate images faster. --xformers is exclusive to NVidia cards.

[tired vigil]

Mislel made such a character similar. @cunning compass

[solar dust]

Since you have 6GB of VRAM, I think --medvram will be enough It makes your GPU use less VRAM, which can make you avoid overload and open up the possibility of higher resolutions.

[cunning compass]

right?

[solar dust]

Yes, now just save it and close the notepad.

From now on, you don't need to type the IP anymore when opening stable diffusion

all you have to do is click the webui-user file you edited and it will open automatically.

[cunning compass]

ok

So, how do you improve the quality of the photo?

[solar dust]

Now open stable diffusion again.

[cunning compass]

really liked this photo

[solar dust]

It can get even better.

You can use hires fix to upscale the resolution while simultaneously fixing some things like face and such.

[solar dust]

Images will take longer to generate but the result will be noticeably better.

Okay, so

for anime, I don't recommend using the latent upscaler

Use either 4xAnimeSharp or 4xFatal Anime 500000G

[silver valley]

--no-half is not helping for performance. Its worse

[solar dust]

What?

But I used it before

in my 1660

[cunning compass]

where to put it?

[silver valley]

Yea its not good to use it. But some cards like the 1660 needs it to work

[solar dust]

to be honest, if I didn't use it I would get many errors

and my previous card had the same amount of VRAM as 3ntry's.

[silver valley]

With a 2060 I would first try using --no-half-vae

If he then gets errors he needs --no-half

But xformers and medvram are needed

[solar dust]

I can't believe I forgor where to put upscalers.

[tired vigil]

But what thou hast vived and sent the defendants are pretty angels.

[solar dust]

Found it

[solar dust]

stable-diffusion-webui/models/ESRGAN

[silver valley]

Your missing the vae file

[solar dust]

also adjust the sliders for hires steps and denoising strengh.

The values you use in both are really up to preference though, but for anime images I mostly see people using 15~20 hires steps with 0.4~05 denoising strengh.

[cunning compass]

folder is empty

[solar dust]

Remember that this process will upscale the resolution so don't go overboard and try to use it while generating a 1024x1024 image.

It can lead to memory errors.

[cunning compass]

bro, I'm already confused, let's start from the very beginning and in order, what needs to be done to improve the quality of the video

[solar dust]

Yes, you need to download 1 of the upscalers I said and put it inside the folder.

[cunning compass]

ok

can you give a link please?

[tired vigil]

I guess yes, I don't know and what kind of file it is.

[solar dust]

I had a link before but didn't save it.

Let me search for it.

[cunning compass]

ok

[tired vigil]

Not bad, you managed wings. You've also managed to get bigger. @cunning compass

[solar dust]

https://openmodeldb.info/models/4x-AnimeSharp

Hope I'm not breaking any server rules by posting this here.

[tired vigil]

And sorry stim I can't help you, I'm using another site not the one you use.

[cunning compass]

right?

[solar dust]

No need to put Animesharp folder in there

just the file inside.

so it goes like this:
stable-diffusion-webui/models/ESRGAN/

and the .pth file inside.

[cunning compass]

right?

[solar dust]

Yes

I don't know what those other folders are tho.

You just need the .pth file inside the ESRGAN folder.

[cunning compass]

ok

what's next

[solar dust]

Now you have to select it here.

[cunning compass]

processed

[solar dust]

Now

as I was saying before

Adjust the sliders for hires steps and denoising strengh.
The values you use in both are really up to preference though, but for anime images I mostly see people using 15~20 hires steps with 0.4~05 denoising strengh.

[cunning compass]

right?

[solar dust]

No

https://cdn.discordapp.com/attachments/1011743094309396631/1143967924730736691/image.png

you did not adjust the sliders.

Nothing will change besides the resolution.

[cunning compass]

I mean, the sliders are on the same values as yours

[solar dust]

Yeah but that was just an example

as I said before

the values you use in both are really up to preference though, but for anime images I mostly see people using 15~20 hires steps with 0.4~05 denoising strengh.

so in this case the value here should be from 15 to 20

[cunning compass]

Can you please send a screenshot of what values should be set?

[solar dust]

and the one right here should be 0.4 to 0.5.

These values are up to preference, there's no objective answer to it

it's all a matter of you preferring some results over others.

[cunning compass]

ok

[solar dust]

Denoising strengh determines how much you allow hires fix to influence the content of your images

0.4 to 0.5 are medium values, it will try to adjust some things but without changing the content of the image a lot.

Hires steps are how many times it will conduct samplings for the image.

Hires fix is a really good start for generating good faces.

[cunning compass]

I'll post the result soon

[solar dust]

I'm going to the gym, will be back in a few hours.

[cunning compass]

that's what happened

[orchid ore]

@elder saffron it is all saved in image, if you drag it in workflow. All is saved. Probably you mean something different?

oh i see my bad...

[hollow igloo]

somone help this man

[next flint]

to put it simply its hard

[hollow igloo]

i watched this video of somone and he said that he trained a lora or something like that

https://www.youtube.com/watch?v=GxvGBw-YdRk

that bypasses that

[next flint]

well i mean he just said they base it on noise

at least thats likely the case

[hollow igloo]

yeah i tried adding noise

nothing really worked

[next flint]

like manually adding noise or prompting it

okay im going to experiment with this. i threw the transcript of the video into a prompt generator i made telling it make a prompt that can bypass it

[hollow igloo]

here is the website

https://hivemoderation.com/ai-generated-content-detection?gclid=Cj0KCQjwuZGnBhD1ARIsACxbAVi4jzZn9C--czqCpKXD--GznSDi_wmU8yL7PX-IEnd4xH0tghtVDg8aAlgXEALw_wcB

[solar dust]

It's a clear improvement in resolution, but things like the extra wing can and will happen from time to time

that's why you gotta keep generating and also use more negative prompts.

[next flint]

okay so i kinda get it. so adding noise and or artifacts does help with bypassing it but noise alone needs a VERY high amount. i added 100% noise and still got 87% ai to an image. in addtion he mentions pixel variation from a lora he trained but i havent been able to find

[hollow igloo]

Hey man, Thank you so much for researching there is no one here that has talked about this topic or bypassed it all i can do is thank you man ❤️

[next flint]

i just tweeted at him asking about the model. i doubt he would response but ill let you know if he does since that might be the easiest way to bypass the detectors

although these are some ways to manually achieve pixel variation

[hollow igloo]

Wait

I have something intersting

That might help us

@next flint https://imgur.com/a/08eDCKB

[hollow igloo]

Thats really nice man Your a great guy ❤️

[next flint]

getting lower

[hollow igloo]

Check this post its intersting

Even though I didn't understand what it really meant

[hollow igloo]

It would be so great if we get it under 49%

[exotic vault]

Hello did you bypass it?

[next flint]

Away from computer right now but I’ll keep working on it later

My next approach is manual edits in davinci resolve

[sharp ember]

how do you force a color?

for example asking for a white car but you get a red one every time u generate

[silver valley]

(White car) or (white car:1.3)

[earnest garnet]

I use this configuration to take a photo of myself and output a suit on me in a professional portrait, but it keeps returning an image with just my body altered when i only want to alter the background and clothing. Any tips? This is using stable-diffusion-xl-1024-v1-0.

formData.append('init_image', fs.readFileSync('init_image.jpeg'));
formData.append('init_image_mode', "IMAGE_STRENGTH");
formData.append('image_strength', 0.8);
formData.append('steps', 80);
formData.append('seed', 0);
formData.append('cfg_scale', 35);
formData.append('samples', 1);
formData.append(`text_prompts[${index}][text]`, 'A business man wearing a suit in a professional portrait')
formData.append(`text_prompts[${index}][weight]`, 0.5)

[random citrus]

Why it send the pics as attachment?

[lavish trench]

I like to play with the prompt but... dang, is hard to make Medusa (from Mythology)

Greek woman with (golden wings:1.1) with a round face framed by a (mass of hair snakes:1.3), a wide mouth with swine fangs

[tired vigil]

Making exactly a character from something can be quite hard, usually I just use some lora for them >.>

[lavish trench]

I know but I'm too lazy 🤣

[orchid ore]

i know medusa medusa worked, not sure in sdxl. But yes if you want play with prompt 🙂

[haughty thicket]

hey guys, any prompting guide recommendations?

[orchid ore]

@haughty thicket what UI are you using? Also probably you can check what users are using in bots channels

[next flint]

i was able to get good results with

Medusa, legendary beautiful gorgon with long flowing emerald green snakes for hair, piercing golden snake eyes with slit pupils, full red lips, flawless light skin, elegant facial features, wearing golden jewelry on her arms, neck, ears and head, standing in an ancient greek temple illuminated by sunlight streaming through tall columns, highly detailed digital painting, cinematic lighting, by Arthur Rackham and Waterhouse

definitely a different style but i think its a bit closer to medusa in terms of how she is depicted

this also worked well

Medusa, legendary beautiful gorgon, with long, flowing emerald green snakes as hair that cascade down her back and shoulders, hundreds of individual snakes with flickering tongues and glittering emerald scales for hair, piercing intense golden eyes with narrow slit pupils like a serpent, full blood red pouty lips, flawless porcelain light skin, elegant and regal facial features, wearing intricate golden snakes as jewelry coiled around her arms, neck, ears and head, standing tall and powerful in an ancient greek temple illuminated by bright sunlight streaming through enormous marble columns, highly detailed digital painting, dynamic and cinematic lighting, by Arthur Rackham and Waterhouse

[quiet zodiac]

@lavish trench I recently had some luck with the term...
as a medusa
Using model: galaxytimemachines/GTM_semirealV11

[atomic wigeon]

White spotlight. All I changed in these images is the color of spotlight.

[cunning compass]

Guys, please help me, I need to save this photo in 3000x3000 format, how can I do it?

This is for my cover.

[gentle lantern]

If you really want high quality, use low denoising upscale, tiled is probably a good idea if you don't have a ton of VRAM. Last time I used a1111, the ultimate upscaler extension was best at this.

[cunning compass]

can you explain what needs to be done?

[gentle lantern]

Click send to img2img, then either set your denoising to 0.3~0.45 and either set image size to 3000x4500 or set image size to 1500x2250 and animesharp at 2x. You'll have to experiment a bit. If it crashes, you'll need to install a tiled upscaler script.

Lower denoising means less change in the image, but also means less detail for the high res image.

[cunning compass]

ok

ty bro

[gentle lantern]

Upscaler extension is at https://github.com/Coyote-A/ultimate-upscale-for-automatic1111, but if you prefer YouTube instructions you'll have to find that--i personally can't stand video tutorials, so idk where to find one.

[cunning compass]

can you help me with the installation, I can not figure it out

?

[tired vigil]

HI, I am sure its probably a simple question but it is driving me nuts. Is there a way to avoid hands like this? Like, melted into one big hand. I have not figured it out yet. I know I can, and most likely have to, inpaint those, but maybe there are some tricks within the prompts? Any help is much appreciated.

[orchid ore]

are you using sdxl? A1111? @tired vigil

[cunning compass]

I downloaded

@gentle lantern

[tired vigil]

SDXL, but it is more of a general question about hands, if you will. I know they suck a lot, kind of, but those melted ones I get constantly. I can do whatever I want, more or less, but I cannot do hands. What I am basically asking is if there are some catch phrases that SD recognizes to avoid those? If there is a style guide I am happy to read shit up on my own 🙂

[orchid ore]

it is always problem, biggest problem i see in spiderman hands. Try download Bad hands 5 from civitai.com. Not sure it will help you in future but should a bit.
This picture for sure you have to solve manual.
Hands are simply difficult for ai, i mean fingers.

[tired vigil]

Thanks. So no way to prompt it properly, if I understand correctly, but SD in general need some outside help? I think my main problem might be that I try to do religious / orthodox / devotional art, in many different ways, so I guess the AI algorithm reads this as praying and/or connects it to praying, then those folded aka melted hands would make sense. Anyway, thanks for the pointer in the right direction. Very much appreciated 🙂

[orchid ore]

It is common problem. It will happen even no hands and any pose mentioned mentioned. 🙂 Same thing as 6+ more fingers or thumbs 🙂

[burnt crown]

Hey, I am trying to generate backgrounds for various products with stable diffusion, but I am not sure how I can then place the product on the image and make it look like it was actually always part of the image. ( I already have a way of generating the position of a similar product using controlnet ) The challenge is the harmonious blend of the images

[atomic wigeon]

After you setup the image just save it and spin it in image to image at around 70-85 strength. Most of the time this will fix most eraser edits and add ons.
If that doesn't work you can use Davinci Resolve (free) to soften your products edges. There are two techniques in Resolve. One uses a soft well blended outside perimeter of an image (your product). The other is hard shadows with the words "motivated light" in your later AI prompt. Motivated light is just a term for synthetic light that has no real world source (like a lamp or window). Try to put them both on your background and spin it again. Good luck.

[burnt crown]

thanks for the ideas

[quiet zodiac]

@tired vigil I recently added "connected fingers" to my negative prompt for hands. Here's the full negative I use for hands...
(extra fingers, deformed hands, polydactyl:1.5),fingers wrong length,six fingers,three fingers, connected fingers

[sleek apex]

I find placing hands in the positive prompt helps quite a bit as well.

[manic wolf]

Alright I've been trying to solve this one myself, but I just can't seem to get what I want. I'm hoping someone might be able to offer some new suggestions for getting images in the following line art style.
I'm using the sdxl base 1.0 0.9 vae, and the sdxl vae fp16 fix. I usually use Euler a or KDPM2 a for the Sampler but have tested the others as well.
For prompting, i have tried : line art, minimalist, simple lines, clean lines, monoline, single-weight line, vector art, flat art, flat illustration, graphic design, cover art, bezier curves, poster art, print, adobe illustrator, and many combinations therein with both simpler prompts and complex prompts. I have the traditional negative prompts like photographic, realistic, ugly, deformed, etc.
I can't find a LORA to add to it to help, and I unfortunately can't train one on my PC.
Any thoughts someone could suggest I try to get this style? Thank you!

[orchid ore]

i think you need higher cfg as well say 12-15 maybe

[quiet zodiac]

Try using an Euler scheduler if SDXL provides the option.

[lofty pagoda]

yeah sdxl failing raelly hard on this one.

could not do it myself too

1.5 with loras better for this type of things atm

[quiet zodiac]

@manic wolf Here's an attempt at revising your prompt...
clean line drawing of a lighthouse by the sea, monoline, single-weight line, vector art, in style of flat art illustration graphic design
Using SD RealisticVision 1.4 DDIM Scheduler
Supplying your reference image as a guide allows for variation on that theme.

[solid bramble]

I'm trying to do one of two different things with a propt each involving hand action. i know SD and AI generation on general have issues with hand stuff but what i'm trying to do is fairly simple. i either want to try and have ducks eat out of hands or hold a duck. is this possible with SDXL alone?

[next flint]

ik its still far off but is this closer to the style you are looking for

[next flint]

i mean yeah it can definitely do that. i tried the concept of a duck being held and worked well and eating the breadcrumbs is a bit more annoying but doable. with both cases you would likely have to do a bit of inpainting to get the hands correct if you really cared about the hands

[solid bramble]

interesting you specified breadcrumbs, i didn't specify any food. can you get them to work without inpainting? i ask because i'm using comfyUI and haven't touched inpainting yet.

[next flint]

i mean if your using comfy ui you might have some luck with openpose. i use a1111 so i havent touched sdxl openpose but i mean it helped a lot for hands in 1.5

[next flint]

well i still couldnt get it to work properly for eating out of hands but i think specifying breadcrumbs might help a bit since most people would feed ducks with breadcrumbs

[next flint]

but if you dont wanna touch that ( i wouldnt want to considering more nodes = more pain) i mean its just luck and maybe a decent negative

[solid bramble]

i apologize i didn't give enough background. i'm basically a noob who has struggled to prompt for 2.5 months. i only use comfy for the lesser vram usage i don't add or touch nodes.

trying to see if this prompt idea is worth burning a day on or not. (gtx 1070 takes lil over a min per image)

[next flint]

for sdxl only a minute

damn

takes me 10-15

[solid bramble]

i don't use refiner, it's a huge waste of time

[next flint]

same

i mean you could always just try using playgroundai.com

its got sdxl

and free

[solid bramble]

eh i'll look but i like to keep everything on my system and private.

[next flint]

yeah i get that

[solid bramble]

would you mind if i PM'd you what i'm trying to do?

[next flint]

just saying if your trying to get a general idea quickly before spending a lot of time its about 20s per image with sdxl on their site

[next flint]

sure

[maiden geyser]

#1072015001029722182 message
Three paid prompt engineers are offering a very inexpensive beginner to intermediate Stable Diffusion 6 week masterclass, going over things like prompting, inpainting, ControlNet, Deforum, and more. Class is 25% full, first session is Sunday at Noon. (Yes, an admin approved the advert a few days ago, I think someone named Bunny I cannot recall lol sorry)

[finite kiln]

Could you please help me create a new prompt to break the moon in half? I have tried but have been unsuccessful.
prompt : a moon with a crack in the middle, earthquake, super detailed image, night! width:1216 height:768 style:Photographic

[tight pasture]

I was wondering if anyone could help me. I've been working on a game and I'm trying to get it to create outdoor backgrounds of houses but it ALWAYS does an interior no matter what prompt I put into it. Can anyone tell me what the best background checkpoint is or how to word my prompts better?

[maiden geyser]

For prompting, a good syntax is medium topic details background style actor, like this

full body, walking pose, slow motion, (human archmage:1.2) 50-year old male human archmage wearing black and gold robes (expensive cloth clothing:1.2), (fine intricate fillagree black and gold cloak), (cloth mage robe outfit:1.2), (insanely detailed, bloom:1.5), (analog:1.2), (high sharpness), (detailed pupils:1.1), (painting:1.1), (digital painting:1), detailed face and eyes, Masterpiece, best quality, (highly detailed photo:1.1), 8k, photorealistic, long dark grey hair, long grey beard, (grey hair highlights:1.3), ecstatic:1.1), (small stubble:1.1), (50-year old french male:1.1), sharp, (perfect body:1.1), realistic, real shadow 3d, (highest quality), (concept art, 4k), (fantasy metropolis in backgound:1.2), by Michelangelo and Alessandro Casagrande and Greg Rutkowski and Sally Mann and jeremy mann and sandra chevrier and maciej kuciara, inspired by (ian mckellen:1.101) and (tom cruise:1.001) and johnny depp:1.001)

Always put “in the background” after it, and specify outside of house in the background

Also put door, window, things that might be seen outside

[tight pasture]

This is what I have been using...

((elaborate atmosphere:1.4)), ((best quality)), ((masterpiece)), ((professional)), ((rich colors)), ((natural lighting)), ((high quality)), ((cinematic lighting)), ((medieval fantasy cityscape)), ((street view landscape)), ((outside view)), dark blue door, aged wooden sign, the building appears just as old, constructed from brick and clay that has been bleach blasted by the sun, a very large window beside the door, inside you can see numerous cages, aquariums, terrariums and kennels for some of the larger animals, several plant varieties for atmosphere and ambience too, even the windowsill contains several of them

*NEG
((ugly, ((malformed)), ((low resolution)), ((unprofessional)), ((blurry)), ((no people)), ((no animals)), ((no characters)), ((low quality:1.3)), watermark, author signature, text, cropped, bad proportions, jpeg artifacts, render)), ((inside building)), ((inside)), ((no fish lens)), ((no warped scenery)), ((Not indoors))

But every single time it's an indoor shot of a door.

[maiden geyser]

For this I would probably use some sort of image to image, or do something with photoshop to split a real moon pic the. Denotes it with low denotes

@tight pasture Change not indoors to indoors

Raise interior phrases to 1.3

That’s how I tune prompts, raise or lower weights slowly

Also you have inside in the prompt

Forget that section, remove, inpaint stuff in windows later

[tight pasture]

Thanks 🙂 I'll fix those things and see what happens

[maiden geyser]

👍🏻

[manic wolf]

thanks, I hadn't though of adding "drawing in there, that seems like it might help. I'm testing out adding some negative prompts to it as well

[manic wolf]

That's decent, it's nice and flat. I'm trying to get it to focus on thick simple lines without added detail. There doesn't seem to be a good consensus on what to call that style in graphic design, so it's difficult to describe it to the AI.

[quiet zodiac]

This might lead you a little too kid-like, but try coloring book style.

[solar dust]

How do I get something like this?

[azure plinth]

anyone suggest a good prompt gen for SDXL? perhaps even using a LLM to assist as well?

[solar dust]

What is a prompt gen?

[azure plinth]

prompt generator

I have seen some using chatgpt to produce random prompts which is great but something a little more guided

[azure plinth]

colored pencil sketch of sonic walking
Negative: multiple, multiple sonic, pencils, red

[solar dust]

What model is that? I'm not getting results like that at all.

[next flint]

oh yeah i use ai all the time for my prompts ( when im not getting rate limited )

[azure plinth]

Any suggestions ?

[azure plinth]

I am using playgroundai

[next flint]

a lot of the time when people need help and arent super specific it helps get a better idea

[azure plinth]

I don't have a local SDXL

[azure plinth]

Yeah i have found reddit templates but often produce weak results