🙋│image-help

[meager garden]

huge in the sense of wide?

how

[dim zodiac]

your prompt with the monster told to be a "COLOSSAL SLENDERMAN"

[meager garden]

should i write colossal slenderman?

[dim zodiac]

took colossal out and just left as slenderman

[meager garden]

hm

I dont actually mean to generate a titan

[dim zodiac]

I think you've got a really repeatable prompt, just play with your wording on the monster

[meager garden]

if you know

[dim zodiac]

until you actually get what you're hoping for

[meager garden]

More like a gigantic something that consumes entire galaxies ^^

[open echo]

A technique I use, inspired by tech support, is to start back at the “minimum functional state”

Start a brand new prompt and just put the bare minimum you want to see, then iterate from there. Often too much words will make it difficult to zero-in on what is causing the issue.

[meager garden]

imagine instead of the big moon or planet, it was a big mouth the size of the planet thats captured

[dim zodiac]

enter the world devourer

[meager garden]

then I'll get something like the marvel guy

I dont know I think i am searching for an horror like creature

[dim zodiac]

it's a mashup now 😄

[meager garden]

better, but a polar bear doesnt really fit the universe type style

ill start a new prompt

[dim zodiac]

I think you just need to play with the prompt, changing the monster still seems to leave the rest!

[open echo]

Oh? A planet eater?

[meager garden]

more like that yes

now just imagine the camera perspective is placed on earth

the monster just would cover the whole sky

[meager garden]

What did you write as a prompt?

[meager garden]

I think I'll just cut the whole previous mosnter part and replace it with the planet eater thing

I think it still interprates a titan into the mosnter

[open echo]

Really short and to the point:

Screenshot, tv show, modern anime.Gargantuan cosmic entity, tentacles, grotesque, devouring an entire planet. Cinematic, cosmic horror, seen from space. Background of black space

[meager garden]

thanks, Ill build a new prompt right now

i think i got something

well

it's kind of better than before

[open echo]

“Well…. That’s gonna ruin my weekend…”

[meager garden]

it surprisingly perfectly set the camera angle just like in the old long prompt

Theres something missing

is there maybe any movie monster that could fit that description?

Maybe a eren like founder titan monster?

That's just really long but thin and horror like

[open echo]

Sirenhead? 🤣

[meager garden]

😂

ill try that one

OHHHHHHHHHH

PERFECTION

[dim zodiac]

oh I got one like that and dismissed it as too "titany" 😄

whoops

nice! glad you have one you love

[meager garden]

i think it just was the eyes and the mouth

the other titans just didnt do anything

Now, would it be theoretically possible to take that image and transform it to something like an anime but with realistic effects

I dont know what the anime is called

but theres like animes that are just 2d animes nothing to do with reality

and then real effects

[open echo]

Depth to Image in Image Guidance is your friend 😄

[meager garden]

How much depth is recommended?

[elfin sinew]

❤️ the “Duo” Facing zee titan

[open echo]

I’d start it at max, and lower it only if it stays too close to the original

[winter spire]

The fusion of media types is going to be difficult for the AI to generate.

[meager garden]

i mean something like this

I now have it as image guidance. Do i delete the prompt now and write "Realistic effect and sfx"?

[open echo]

Keep the subjects in the prompt but change the style

Depth to Image can really do magic with existing pictures 🤣

[meager garden]

woah

it didnt change anything 😂 (for my image)

it somehow

does not change a pixel

also if i use guidance scale 0.5

oh now it did something

[open echo]

Guidance Scale determines adherence to the words of the prompt. It should be left at 7 for most of the time

[meager garden]

i meant the image to image scale

sorry

[open echo]

Ah! Use Depth to Image instead of Image to Image

[meager garden]

depth to image really is better:

Thanks for the support ooo3r1cooo, Tuxman and azothane

[hot niche]

I wish we could select the motion type when generating videos. Frustrating trying to get decent ones while getting tons of that does not even fit to the image. Or even general direction of it.

[little pulsar]

hey, i wonder if somehow i can create damaged hair? i tried lots of prompts, and also used a prompt generator. also, gpt came up with restyling words. but everytime i get the same - pure healthy hair.
Maybe someone has ideas on how to prompt for a thin damaged hair?

[low bear]

Imperfection is one of the hardest things to generate with AI, because its been trained not to generate anything imperfect

[little pulsar]

ah, ok. thanks!!!

[manic skiff]

how do you make images

[robust gulch]

You have to go to https://leonardo.ai

[manic skiff]

how do you level up

and then what

[robust gulch]

By participating in discord. Talking sharing images and participating in the daily challenges.

Make an account. Then click on image generation on the right hand side.

[manic skiff]

oh!

[tacit silo]

你好

[clear dust]

你好👋👋 製作圖像要在這裡唷-> https://leonardo.ai/ (註冊後即可使用)
我們有 #中文-ch 中文頻道唷 有問題歡迎來做詢問🤗🤗

[steep portal]

araba

[karmic pulsar]

Is it possible to upload a screenshot of a game and make it into a realistic photo?

[open echo]

My first guess would be to use Depth to Image in Image Guidance, then to start the prompt with “photorealistic” and describe broadly what is depicted in the screenshot

Just a quick experiment 😅 with more time it can be tweaked to get more realism and then adjust objects in the Canvas Editor

[low bear]

Omg this is brilliant hahahaha

I'm gonna try it now

[open echo]

Let me guess…. Some unnamed block based creative sandbox? 😉

[low bear]

Hahahaha no I'm gonna look for some characters to make into cursed photorealistic versions ;p

Therapist: Hyperrealistic bowser is not real, he can't hurt you

Hyperrealistic bowser:

[winter spire]

damn

that's dope

[low bear]

ikr, this is a great idea

WAIT

DAILY THEME IDEAAAAAAAAAAAAAAAAAAAAAAAAA

imma run to robin

[edgy iron]

is there a detailed guide that I can be directed to in order to better understand how to get Leonardo to render images correctly ?

[dim zodiac]

thats a big question - what kind of errors are you seeing?

@tardy hazel here is a guide on https://discord.com/channels/989166677390426132/1105168997248413727

[tardy hazel]

thank you so much

[dim zodiac]

the most confirmed way of doing it is by uniquely naming the character and using the same descriptors every time

"Jenevra is a twenty year old caucasian lady, with loose brown hair and hazel eyes, wearing well-worn jeans and a Metallica t-shirt"

and then always describe her as Jenevra with brown hair and hazel eyes, and whatever else you want

[tardy hazel]

Thank you, gonna give this a try right now

I assume uniquely within my account?

[dim zodiac]

You can use the same name, it's not really TEACHING the AI anything, we've just anecdotally seen that giving a name helps anchor a lot of the same details

you can play with the name and might get a few differences in appearance too

so don't stress too much over a unique unique name... could even be Karen

[tardy hazel]

Got it, I'm trying to automate character creations and then consistent images for that character in different settings, clothes etc

[dim zodiac]

sure

it will still be a bit of a dice roll, as it is with all AI gens

but have found good results with this practice

[tardy hazel]

have you played around with model training?

[dim zodiac]

yup I have a few

that's another powerful tool

[tardy hazel]

so if I feed it some arguably NSFW images, would that get me closer to what I want?

[dim zodiac]

I don't know what you want, but what you put in is what it will output

read your TOS carefully

[tardy hazel]

Will do, thank you for all the help

[dim zodiac]

oh... you must pay for a Leo sub?

[tardy hazel]

I'm already paying

[dim zodiac]

OK there's another Role for that with it's own channel

if you DM me your user so we can verify that it's subbed you can get the exclusive role

or post here and I'll delete immediately

[tardy hazel]

sent you a DM

[blissful horizon]

Hi, please which model is best for colouring book.

Please let me know

[dim zodiac]

I got it and I'm seeing if somebody can grant the role!

[dim zodiac]

you can either use prompts for Colouring Book in XL models (I like Diffusion for it) or there are Community Models specifically built for colouring books

[tardy hazel]

Thank you

[blissful horizon]

Thanks. I will try that now.

[robust gulch]

Anyone mind sharing their favorite negatives to prevent multiple people having the same face? Mine aren't doing so good.

[open echo]

In the past, I've used "diverse group" in the positives and it seemed to work. I don't know if that's something you already tried

[robust gulch]

I'll try that thank you!

[unkempt whale]

I have a question regarding image generation: How do I add text to an image

[robust gulch]

Either with an image reference set to edge to image or text image input or you add it afterwards with a different program.

[unkempt whale]

ok thanks

[radiant goblet]

creat logo name city 068

[oak fox]

Does anyone have a good advice on writing in the AI?
For example for a friend he wanted a robot standing with a Neon logo behind saying TOMBOT.
I had to generate alot of pictures before it spelled TOM correct, then i used that picture to "guide" the making forward (image to image), until many generations later it spelled the full word correct and i could put That one as image guidance, Final result was this below:
Im just asking if there is a more simple way to make it understand words or nicknames that aint "coca cola" for example 🙂 Thanks in advance // Hilik

[robust gulch]

Words are always hard for the ai. Probably will be for some time. There are 2 premium image to image settings that do it well though. Edge to image or text image reference. Or adding it afterwards in a different program. But despite the difficulties you managed to get a good result.

[oak fox]

took Alot of coins tho XD, So maybe adding a image to image of a robot, then another image with TOMBOT text to image would speed up the process maybe? 😄 @robust gulch Thank you very much for reply 🙂

I've tried making a picture similar spelling "Hilik" but i gave up after 500 coins XD

[split sundial]

I have seen prompts where it would say 2girls 1guy at the beginning to ensure the image is of 2 girls and 1 guy, but is there any reason to think this is truly helpful?
I have been trying to create pictures of a cheerleading squad with their mascot (a dog), but most images will have more than one dog in it, I didn't have the expected results putting "multiple dogs" in the negative prompt.
I kind of like the images I get, but for my purpose having a second (or more) dogs will not fit the story behind.

[oak fox]

@robust gulch Im trying adding two guideing pictures but cant seem to get it to work, can you maybe see what i do wrong XD?

[robust gulch]

You can't add more than one at a time. Sorry if I confused you.

[oak fox]

Oh ok then i get it , thanks again man ❤️

[robust gulch]

No problem!

[dim zodiac]

unfortunately at this time, multiple subjects is a place where the AI is weak. I too have seen those types of prompts, but honestly it's still luck whether it works or not. The tech behind the generating doesn't have this type of prompt structure

[dim zodiac]

You CAN add more than one image to an input using paid features, which I can see you are using. (if you send me your Leo profile name I can get you another role) I don't think there's a nice way of doing this how you are hoping though

the reason it's failed here is that you're still using Alchemy with Prompt Magic on while trying to use multiple inputs

you basically have to strip it back to nothing to enable multiple controlnets

secondly, the output of this type of 2 image input would likely have been just HILIK with the image fit inside the text

Canvas is your best bet for adding text to a generation you like already (if Leo can't figure out the text like it did with Tombot - very cool by the way!!)

[oak fox]

https://app.leonardo.ai/profile/Hilik

I gonna try that 😄 thanks for advice and you taking a time to answer

[open echo]

And when all else fails, there's always the possibility of a little elbow grease and removing the extra characters manually through photo editing application 😉
https://discord.com/channels/989166677390426132/1119326564631449710

[oak fox]

Im a rookie at editing and ai generation so i hope for the AI to figure it out for me XD

[dim zodiac]

yeah this is a very common question, and in short - it's best if you planned to keep them from the start: https://discord.com/channels/989166677390426132/1105168997248413727

the best way we have seen in practice to keep a consistent character by prompt alone, is by giving them a unique name, and describing them the same way every time you try to use that character

but trying to apply that trick now may change what this guy above looks like since he may not have been named originally

other things you can do include using controlnet, or creating a trained model with multiple generations of this character that look the way you wanted, but those are more difficult and only useful in certain situations

you can put the name where it makes sense in the prompt - "Floppenheimer is a man in his early thirties with close cut blonde hair and a stoic expression, he sits in front of his monitors watching the successful launch of an ICBM that carries a nuclear device he created"

but the next one might be more environmental - Environmental photo of grey storm clouds, the wind batters down on Floppenheimer, a man in his early thirties with close cut blonde hair and a stoic expression"

etc

the important part is nailing the same Keywords

describe the parts of the character that you want the ai to reall try and keep always

most of my prompts are more like: man sitting at computer, ICBM launch seen through viewing window, multiple monitors tracking progress, highly detailed CG render, octane render, hyperrealism

but that would make a random man every time

you're not really TEACHING the AI anything, but we've found that using a name + description helps really anchor the same look in a generation over and over

it's still technically random

yes hopefully

it's a very common question, so if it can be done more easily, I'm sure that's going to be a simpler controlnet or something in the future

[karmic pulsar]

@idle ocean

[idle ocean]

did you put the text in

[idle ocean]

if so its yours

[karmic pulsar]

Yes.

[idle ocean]

yea its not just AI

[karmic pulsar]

Only the character is generated based on a prompt, I changed his colors and edited skin on some.

[idle ocean]

your fine

even if it was it wont be getting you sued

[karmic pulsar]

Bet.

[idle ocean]

but only wont be able to gain copyright

[dim zodiac]

hey listen, we're also not Lawyers... 😄 but your own generic character, plus your own editing, I don't think you need be worried. Still we do like more Leo users 😄

[karmic pulsar]

I was against AI. But it opens up a realm of possibilities for sure.

Yeah I got the app via this dude.

[idle ocean]

yea

[karmic pulsar]

He mentioned leonardo but google showed davinci lmao

[idle ocean]

use leonardo

[karmic pulsar]

What do you guys use the app for?

[idle ocean]

in leonardo I use realtime canvas,canvas and vision xl

and photoshop/adobe express

[karmic pulsar]

Are you a graphic designer?

[idle ocean]

I can call myself that yes

[karmic pulsar]

Ah you do like 3d modeling of houses etc?

[idle ocean]

I do use AI but I am 100% my work is not just first simple prompt of x

[idle ocean]

I love blender

but lets go to #💭│off-topic for that

[karmic pulsar]

Ook Nederlands zie ik haha. Yeah graphic design definitely is broad.

[idle ocean]

or #💬│general if its AI

[karmic pulsar]

I have to get back to my business but I will try and be active in the chats thanks!

If I want this style images any tips how to accomplish that?

Im getting this but thats more mortal kombat

I can do it myself via photoshop but easier the closest it can get with the app itself.

[open echo]

Just a shot in the dark, but to remove the photorealistic aspect, I’d start the prompt with “Digital art” or “digital painting”. And maybe “cyberpunk” in there to hype up the technology aspect?

It’s really just an intuition.. I didn’t test it 😅

[karmic pulsar]

Thats a great idea, also I see you can use image guidance. Will use both thanks

That worked.

[twin gulch]

Someone know what to do?

[autumn spindle]

how do i add Motion to an image where i removed the background? it only applies to main image

[open echo]

Indeed for now motion works ou in the main image (maybe in the future?)

[dim zodiac]

Hard to say, what were you trying to post?

[shrewd rose]

Hand Wash

[pure ivy]

Hello and welcome to the Leonardo.Ai Discord server! 😊 I am Verrocchio, an AI assistant, here to help you get access due to the absence of human helpers at the moment. To get started, could you share what kind of images you are interested in creating with Leonardo.Ai?

[shrewd rose]

A unique and very attractive hand wash

[pure ivy]

That sounds intriguing! To create visually appealing designs like the hand wash you're interested in, Leonardo.Ai's webapp offers a variety of tools. Before we proceed, have you already created an account at https://app.leonardo.ai to start generating your images?

[shrewd rose]

Yes please

[pure ivy]

Great to hear that you've taken the first step! Could you please tell me a bit more about how you plan to use Leonardo.Ai? For instance, are you a hobbyist, game developer or looking to integrate it into a business?

[shrewd rose]

I am Content creator

[pure ivy]

Being a content creator is exciting, and Leonardo.Ai can certainly enhance your creative process with its various features. To best support your endeavors, our community is a place to share insights and seek inspiration. Are you interested in participating in discussions, challenges, and sharing your work with fellow creators?

[shrewd rose]

Yes please

[pure ivy]

It's great to have you on board! You've been approved to access the rest of the server. 🎉 Check out the #🖼│image-share channel to share your creations and maybe join the #1108130218260041758 for daily image challenges!

[shrewd rose]

Thanks a lot

[zenith rain]

Guys, im new here and i have a question - ive trained my model, put there like 30 photos of model. But when i wanted to generate now new graphic with my model and i ofc set my finetuned model right and in prompt i use model name… i dont recive anything with my model. What i do wrong? Because you guys do this and got great results as well

[stable carbon]

what did you train your model on? 30 images of what? for example 30 images of yourself, or ..?

[zenith rain]

Ive put like 30 different photos of my AI model into new dataset

[lyric timber]

how to you get the character to look as closely to a specific character? like a game character? Does it need to be a popular one like Son Goku or should the AI model know a lot

[low bear]

It does do best with popular characters yes

With other ones you're gonna need to describe their features in great detail

[lyric timber]

I see. that makes sense. Does the image guidance helps with that or rather not

[low bear]

Somewhat, but getting poses different from the input image will be hard

[tulip river]

hi

are you sure it was flagged for the word war

im trying to do true crime stories and the AI seems to accept

ok good

yes im in the exact same position

so you need to describe you character in detail first and give him a name

yes ive seen that too

keep your first prompt in full describing the officer than add to it

[karmic pulsar]

Anyone a graphic designer here?

[tulip river]

fiirst of all in the second scene Leo dosent know to make it black and white

what is the prompt for this?

[karmic pulsar]

Is this allowed? The above one is the edited version. For my channel. The whole art is yours?

[tulip river]

your not specific enough with the uniform

also the background is confused with the prompt. WW2 scene or forrest scene?

[tulip river]

me too ork

[karmic pulsar]

Ur making a video? or a blog

[tulip river]

working on mine

im making youtube shorts

[karmic pulsar]

Video?.

How will you implement the AI into a video?

[tulip river]

im using the still shots

[low bear]

You can also use Motion for short clips using AI

[low bear]

People have made awesome stuff with that

[tulip river]

yes some of the shots i render into video

[low bear]

For example this one made by Vae Victis

Very cool video

Or these by Zarahuztra

[tulip river]

i got one

[karmic pulsar]

Do you have your own discord? Can I join it to see the process

Is it about a specific person, a officer? or Just officers or germans in general

I might be able to give you some insight and perspective.

Yes you can.

Have you ever had a chance to watch stuff like "Scary stories" On youtube? Where a youtube artist narrates a video with only voice using images that are still and found on the internet mostly?

You can edit the image yourself to make color's match, for example in the lower one you can lower the saturation and increase the contrast and highlights.

For example

[tulip river]

Subject Description + Type of Image + Style of the image+ Camera Shot + Render Related Information.

[karmic pulsar]

He uses a still image, makes it move by zooming in and adding moving effects ,and adds a raining effect

[tulip river]

I found this online and it seems to be working better for me

they are not black and white

dont give up it can be done ive been trying every day for 2 weeks

[karmic pulsar]

Left is edited

[tulip river]

what we are trying to do is much harder than a one off image

[karmic pulsar]

You can make color red etc

But its not about a specific character right so it shouldnt matter too much what the characer looks like as long as you nail narration, color grading, theme. stances

Add fog or other particles and even physics

right then its a bit more difficult I agree, perhaps use warp tool to work on cheekbones and coloring alongside it.

DOesnt have to be perfect IMO But I understand ur concern for sure

Very cool project.

[tulip river]

you can have much better results

[karmic pulsar]

Basic thing you learn in filmschool is visuals are not as important as sound. So as long as you nail the script and narration and audio. Im SURE you can work out an amazing project my G

Watch this whole video for inspiration on perhaps to implement aspects for ur project I'd say

[feral bough]

Hi, the AI models are not trained on images of very specific people like this. You can recreate celebrities with AI but that's only because there are so many images of them. AI does not know who Friedrich Lengfeld is so unfortunately you won't be able to accurately recreate him.

[tulip river]

good

[feral bough]

That is also not easy to do with current AI tech, but there may be some things in development for that. The best you can do now is: 1) Keep the seed the same, this is under Advanced options 2) Give the character a name; you're already doing that, and that's good for consistency, but its not guaranteed that your character will look like the actual person you want.

[karmic pulsar]

Can someone please help me figure out what to type to get these noodles to be wires instead of noodles. It always comes out as ramen noodles.

[clever fiber]

Hi Vae, I can try, what do you currently have?

[karmic pulsar]

With a graceful and precise movement, a robot delicately eats a bowl of ramen noodles made of wires, each strand carefully rendered and topped with computer chips for a visually stunning and unique dining experience.

done with diffusion xl

[clever fiber]

Try taking out.... ramen noodles made of, let me see what it will do. Be right back

[karmic pulsar]

ok

[clever fiber]

Yes, getting the same.

[karmic pulsar]

hmm okay. No worries. I will think on it for a while and try something els. Thankyou

[clever fiber]

Wires mixed in a bowl

Surrealism, with a graceful and precise movement, a robot delicately slurps wires from a bowl, each strand carefully rendered and topped with computer chips for a visually stunning and unique dining experience.

[karmic pulsar]

awesome thats close

[clever fiber]

Yes, got that too.

[open echo]

Have you tried with the simplest prompt possible first?

[karmic pulsar]

yes I did that as well

[open echo]

No idea if that’s what you had in mind

[karmic pulsar]

It always comes out as noodles

[livid inlet]

Highly Detailed Robot Eating Coloured Wires From A Bowl, Hyper Realistic 3D Render, Vibrant Background

[karmic pulsar]

awesome!

[open echo]

My prompt:

Robot eating a bowl of colorful wires, background of bistro

[tulip river]

Hahaha

[karmic pulsar]

Heck yeah. Thankyou very much

[open echo]

My motto: precise, concise 😁

[tulip river]

dang thats good

[karmic pulsar]

Thanks the top one is my face so far

[tulip river]

yes diggin in now he needs chopsticks like any decent robot

[karmic pulsar]

haa right

[torn notch]

İs there any app that creates videos for more then 4 seconds

[open echo]

I know people have been chaining 4 seconds videos in movie editing softwares, so that’s one possibility. (For a more advanced way, I know someone was taking the last frame of the 4 seconds and generating another 4 seconds from that 😅)

Also the Motion feature from Leonardo is still young, so who know what lies ahead? 😄

[torn notch]

Ahah yes i also play with 4 second videos to make them longer but not too sufficent

Leonardo ai has motion feature?

I was using different to create video from images

Different apps

[open echo]

When you generate an image, you can then apply a motion effect. You can choose the strength of the motion from 1 to 10 (5 being the middle ground of course.. higher you can expect some …. Interesting results 😅)

[torn notch]

Thanx

[elfin sinew]

You Ramen Junkies

[elfin sinew]

“When my dish hits the table but i asked for NO “extras”

[nova canyon]

Why does my cat 3D model texture look like this?

when the original texture is this

Like why is the cat face on the belly, the eyes are distorted, wrinkles everywhere, etc. I have nothing on the negative prompt.

Am I doing something wrong? I may need help since I dont know how the AI here works that much with 3D

[elfin sinew]

Confusing the machine in prompt? Model choice?

[nova canyon]

No like it doesnt meet my expectations. its the 3D model texture generator. The prompt is Stylized grey Siamese male cat, dark blue eyes. Nothing in negative prompt

[elfin sinew]

Tried to recreate but major denies from the Leo generation (wow) really

[nova canyon]

:/

[elfin sinew]

Who is in charge of what is “right” in prompt writing, seems pretty pinpointed (no no NSFW crap mentioned here)
Tough when platforms tie down 100%

[nova canyon]

so is there anyone who can help me with this?

[forest granite]

Hey i was wondering why when i download the images it comes as a bin file?

[split sundial]

My prompt:
A snowy winter suburban landscape, a woman in a sleigh, the sleigh is pulled by one white sleigh dog. (in the sleigh sits Jenna (Jenna Coleman)). Jenna is wearing a wine red dress and coat with white fur at the hood and sleves. the sleigh dog is a female Samoyed dog
Negative prompt:
multiple dogs, dog in sleigh
I use Leonardo Diffusion XL, Dynamic, no image guidance, input dimensions are set to 768x512

Obviously I would like the dog to pull the sleigh, with Jenna sitting in it. But I consistently get the dog IN the sleigh and often several dogs.
It seems that image size has some influence on what the result is, but I had hoped that being wider than the height, it would simply adjust the scale - or am I completely off here?

[robust gulch]

So I have been trying to ge this to work but I am having no luck. The ai just doesn't seem to understand a dog pulling the sleigh. My typical move when I run into an issue like this is to use an image to image reference to help the ai out a bit.

[silent umbra]

how could i make constant character model in Leonardo ai ?

with many poses , did any one know ?

[open echo]

Hello there!

I see the result is (well, obviously 😅) not what you expected. I’d be surprised too if that’s what popped out.

Just so we can also understand how it got to this result, can you walk back the steps you took to arrive there?

First things first, did you select “Image Generation” in the left toolbar?

[nova canyon]

Yes thats what I used

Wait no not image

I was using texture generation

[open echo]

Perfect
And what is the .obj file you were using ?

[nova canyon]

It was a cat

[open echo]

Can you try to upload it here if it’s not too heavy?

[nova canyon]

This is the original UV map and texture map

Its over 100MB since its animated

Unable

[open echo]

Ah ok
But is a .obj of a 3D model of an untextured cat?

Bit like this?

[nova canyon]

Also its a payware model so I cant upload it publicly

[nova canyon]

Yes

[open echo]

Did you have a clear preview of a cat when you uploaded the model?

(I’m sorry I’ll just have to pause for now … it’s 3AM here 😅 ill check up on the conversation when I wake up later)

[nova canyon]

if I remember, yes

[lofty whale]

I m looking help for a prompt : the idea is simple but difficult to be made
A half opened iridescent box with tilted lid . Strong colorful music notes escape from between the lid and box and fill the picture. An orange stands over the lid

I tried many times to do that in any program tested but none could add that especially the orange . this orange has a special meaning to the photo

[robust gulch]

The problem is that the ai will probably interpret that as a color and add it randomly to the image. You may have to be more specific and mention a kind of orange or say orange-fruit or something like that.

[lofty whale]

I also had a log time problem with this prompt
a commercial thin-plastic bottled water container is filled with olive oil . A diamond nerby the bottle s shown [conveying the meaning that the olive oil costs nearly as a diamond]
Really here one liter of oil costs 5.5Eu
Do you have a better prompt for that ? Here you see the bottle

[livid inlet]

Not going to be easy with a prompt... but there is RealTime canvas, I did a very quick one to show you what can be done. If you spend some time there, you should be able to get a lot closer to what you want.

[robust gulch]

The ai won't recognize the meaning you are trying to show. There isn't much point prompting for it.

[livid inlet]

Canvas Editor for you I'd say

[robust gulch]

A container of olive oil next to a diamond might be enough. Generally simple prompts are better.

[lofty whale]

all times i have the picture of your photo . see also the picure of the milk container . No diamod however!!

[livid inlet]

I'd say as long as you don't mind spending a few tokens, you should get there. I had a little try and got this on the first try, I'm sure you can do better if you take a little more time and tokens.

[lofty whale]

the second is closer but does still does not show something similar to what is required (as with the milk bottle) now expecting your ideas

[split sundial]

Yes, that could be a way forward. It is frustrating, I keep having to remind myself of the random workings of AI image generation, it should be simple... but it isn't 🙂
Thanks for trying 🙂

[unreal cosmos]

Hi senpai. I have a logo (image) which I want Leonardo to put on every images that I have. Would it be possible? Just wanted to know the feasibility before I deep dive and explore.

[open echo]

Hello there 😄

Even if Leonardo itself doesn’t have a watermark feature, I know many photo editing apps do. (I’d have to do a quick search for a free one)
For exemple, you download all the pictures you want to share or send to clients, add them in the software and then export them back out but with the logo applied.

In the days of Aperture (a now defunct photo app), it was just a matter of setting up the logo in the settings and how we wanted it to look, then when we exported pictures, we could check the “apply watermark” options.

[open echo]

So… there’s hope 😁
Without prompting for the orange, it gets some good results. Then the orange can be added in the Canvas Editor.

Or alternatively there’s also a (slightly more difficult) way of building a quick collage of items in a free photo editing app (like Photopea) and passing it through the Depth to Image function of Image Guidance

https://discord.com/channels/989166677390426132/1105169069436575846

[narrow bear]

Hello, Im Johnny and I have a vodka brand company. Im trying to create images using my real bottle but its not working. Could someone help me?

Thats my vodka bottle that i would like to create images with

[tulip river]

the AI dosent naturaly do text very well

what do you want your bottle to do?

[narrow bear]

like i want to create images for my instagram with this bottle. Like the bottle with a drink on the beach

everytime i try something the ai change the bottle, but i want to use the exactly bottle i have

can u help me?

[unreal cosmos]

Oops I think I have misrepresented myself. I wanted to put a logo I made onto the tshirt of my character.

[open echo]

Ah! Then yes the bottle will change during the image generation 😄 Sorry I misundertood this aspect

So to start, this will help you:

It's your picture but without a background.
(Hang on, I'll prepare an exemple)

[narrow bear]

tuxman, can we make a videocall ?

can send u a vodka bottle for your help hahahahaha

a really need make it workout haha

[open echo]

I don't do video calls 😉
But I'll write out the instructions here.. just a moment

So I'm going to use a version of the technique outlined here: https://discord.com/channels/989166677390426132/1105169069436575846

(And a little bit of color correction at the end: https://discord.com/channels/989166677390426132/1113604362254700614)

So I start by making a very crude collage of a random beach picture from the web and I paste in the bottle pic without a background. I rotate it a bit to give some angles (I mean, if it was in the sand... it'd be tilted right?)

I get the picture into Leonardo's Image Guidance, select Depth to Image, then I make a quick prompt that describe what the picture should be. I get two versions and I choose the one I like most

I bring back that picture to my collage, paste it as a new layer and move the original bottom above it (Don't mind the edges of the fake bottle... we're gonna fix it in the next step)

So then with the same way I rotated the bottle, I scale it so it hides the fake one underneath. Then I take the eraser tool and choose a smooth one to make it seem like it in the sand.

We could stop there... but the bottle doesnt have the same tint as if it was in a sunset, so let's fix it (And by the way.... THANK YOU for making it opaque 🤣 WAY easier!)

So I just create a new layer on top, trace some lines with black and orange then make the layer transparent (I made it quickly cause I have toi get back to work.. but you get the idea 🤣 )

So, there it is! Take an object and insert it in any images without having any details changed 😄

[tulip river]

great tutorial!

[narrow bear]

Nice

where can i select depth to image

[tulip river]

in the image guidance tab

next to generation history

[open echo]

Right in the middle under the prompt and model

[unreal cosmos]

Hi, follow-up on my question. is it possible to add my own image into specific section of the generated AI image?

[open echo]

Sup! You can pretty much just paste the logo over the picture in a free photo editing app and scale/rotate it to fit the scene 😄

I did exactly that to put my lightsaber shirt design on ..well.. actual characters with lightsabers 🤣

You could put the logo on a collage and use it through Depth to Image, or in the Canvas, but it will get slightly changed and modified in the generation process so you won't have exactly your logo

[unreal cosmos]

cool. so i do not need to use Leonardo... which software do you use?

[open echo]

Photopea, it's free and it's online

I use Leonardo to generate the assets, and I bring them together in Photopea

For exemple, this is like Leonardo 8 images 🤣

@nova canyon So! Now that I had third coffee! 🤣
Bottom line is, I've seen people have reasonable success with texture generation (on trees, balls, etc...) but still the generation process don't understand the physiology of animals. Even if the texture looks right in the web page, the map will be all over the place and might not apply correctly back in your 3D software
(Like in my exemple, it didn't recognize that a cat only has two eyes and there are many intersecting bodies)

Texture Generation was marked as Alpha until recently. I suggest sending a ticket to support to get their official stance on the state of the feature 🤔

[jovial obsidian]

Hey all! I'm trying to get an image that is at least 1125x1725px. I've generated a 1536x1536px one which is the maximum dimensions for generating an image, but when I run the Alchemy Upscaler on it the result is still only 1536x1536px. How do I actually get it to 'up'scale?

I seem to recall there used to be more than one upscale option in Leonardo, am I mixing up my AI art tools or did they get moved somewhere else?

I'm a free user btw, if that makes a difference

[young plover]

what if you cut it differently (the uv's) so the app has no chance to get it that much wrong

[wanton belfry]

good good, I'll use it when I am too lazy to do photoshop

[open echo]

Oh, answered the wrong person haha

[young plover]

i dont wanna be that guy to tell u to get an idea how 3d models are working but maybe u should get an idea how they are working in general ^^ its easier to get good results if you understand the basics. so if you model something in 3d and you want to texture it you need UVs. Thats what u have on your picture. Those UV's you can texture. So it will have a texture on the model. Depending on how you cut the uvs its easier to do or not. the Cat seems like to have lots of small pieces. But normally u try and seek for bigger pieces of uvs. its also better for the end result because most of the times seams you create are always visable on the model. thats the uv map of a whole human. its still weird to look at but u can see the different parts of the body like the head, the torse and the hands. i think if you would vut it like this it would be easier. also it would be easier for you to start with a more simple thing.

[open echo]

Haha! Indeed, I've never claimed to be an expert at 3D 😄

How is your experience with Leo's texture generation?

[young plover]

thats why i tried to explain it

[open echo]

So you had good results with it?

[young plover]

i just did that. seems like its kinda working on more simple objects. at least its doing something.

[young plover]

cant say thats a great result but its a start to work from

ig

[tulip river]

A gritty, black and white rendering of a left side view of an Italian criminal from 1898, his piercing gaze looking off camera as he holds a gleaming knife in his hand.

so no matter what I do this image has no knife Hahaha

ive tried to make it very simple and not ask for much

[open echo]

Yeah... I guess models of living things might be too complex still :S

[open echo]

Hello there! 😄
Sorry I missed your Upscaler question haha

When you use XL models, you indeed only get the Alchemy Refiner.

When you use other models you will see the Alternate upscale, smooth and crisp.

Also note that you can upscale images further if you use a local upscaler like "Upscayl". it's a free program and you can get your pics in truly ridiculous sizes (With great results, of course :D)

[feral bough]

@brittle raft Please reconsider what you post here. Repeated off-topic messages, bad language, and spam are not allowed.

[jovial obsidian]

Aha, that would explain it. I was using Leonardo Diffusion XL. Thanks for the info! 😄

[nova canyon]

I guess its still in alpha, idk why they removed it

[open echo]

Have you tried separating elements with commas? That's usually how I go about it

Ah! The last one, the hat would need some touchups.. but it's a cool pose 😅

[karmic pulsar]

What Is the best input dimensions?

When I want higher quality the promt Is being ignored In a few things like the background scene.

[split sundial]

I think it depends on what you want, I have come to the conclusion that it is important when you want someone standing that the hight of the image is "large" and if you want a view of say a landscap, make sure to have a wide format. But if there was a "best" setting I am convinced the dev team would have just put that in?

[robust gulch]

Most models tend to use depth of field and slightly blur the background. You can put depth of field in the negative or intricate background in the prompt to help.

[dim zodiac]

welcome to AI image generation

😄

your best bet if you need it exactly the same is to remove background on one image and insert him into another scene using canvas after generating the other background

it will then be the exact same pose, though

otherwise you can do your best to inpaint all the minor details and throw tokens until it is solved?

[night vault]

you will need custom model/Lora for that

[dim zodiac]

agreed, but even then it would need to be trained on input images that have the consistent uniform

[wanton belfry]

for this LORA is used, however due to issues we can only have elements which are indeed LORA but of limited use. We can't be using characters. Another way is to go Japanese visual novel way....

right now I've seen many people make comics with LORA, however that too is inconsistent

No AI can help you with this, even if you train a perfect LORA, its still not going to be perfect

[tulip river]

I am using a celebrity as my character and just giving them a beard or changing the nationality

[tulip river]

Haha not quite what I was looking for in a masked rober

[winter spire]

The start of 2020 did a real number on the dataset when it comes to masks 😅

[tulip river]

Hahahaha

[open echo]

They can’t identify your DNA if you can’t spread germs! Genius!

[limpid notch]

he is keeping them safe while robbing

[pallid heron]

is it just me or is Canvas having issues actually using specific models? Sometimes when I change models for Canvas inpainting, etc. The new model doesn't register

[copper warren]

Which model do you select?

[copper warren]

Must have seen Gattaca 😂

[pallid heron]

I've been trying to use more of Leonardo Diffusion and Leonardo XL for my Canvas edits. I guess they can't be used for inpainting and outpainting yet though

[copper warren]

I will look into this for you

[wispy pulsar]

Hello, mod-ish people!

I have a little mystery for you...

What do you do when your prompt is too long to fit in the online form?

[wispy pulsar]

You mean with Ethan Hawke? That's my jam, that and Predestination. Next level Sci-Fi

[split sundial]

OK, I ended up cheating, DALLE had no problems creating the basis image, but it won't generate with... Come to think of it, I asked it to use Clara Oswald, not just "Jenna Coleman" that might have been the issue, but we discussed copyrights...
SO basis from DALLE and then the finished product from Leonardo
Prompt:
A snowy winter suburban landscape, a woman in a sleigh, the sleigh is pulled by one white sleigh dog. (in the sleigh sits Jenna (Jenna Coleman)). Jenna is wearing a wine red dress and coat with white fur at the hood and sleves. the sleigh dog is a female Samoyed dog

[muted jewel]

how to color a drawing with leonardo ? is that only possible for paying users ?

[open echo]

Hello! 👋

You can use the Edge to Image feature in Image Guidance. It detects the lines of the input image, and tries to generate around it. Make sure your prompt describes briefly what’s in the picture so the systems knows what’s actually there 😜

It’s a premium feature for the XL models, but it’s free for all the rest of the models.

[open echo]

@karmic pulsar Hey there 😄 This will be a better channel to get help for your prompts

[winter spire]

@white delta We can discuss over here instead.

[white delta]

sounds good

so im kind of the guy who likes generating dwarf images

and i have been generating so

but the generations have changed

https://app.leonardo.ai/profile/wnffrwoh

have a look so you can see

@winter spire

[winter spire]

Taking a look now. If you don't mind, could you indicate an image like what you expect to see vs an image that you're getting now that you're unsatisfied with?

[white delta]

ofcourse, that would be my pleasure.

[winter spire]

Hm. I re-ran the prompts you used for both of these sets and they seemed to generate reasonable results. This to say, the settings seem reasonable as a sanity check:

So from here I have two thoughts: Either A) It's possible that there's some weird data hanging around in your local internet files. You could try flushing your cache and cookies, see if that does anything. Or, B) Your most recent generation was a seed the AI didn't resolve well. It happens, not every generation is an award-winner.

Having said this, it looks like the configuration you were using previously netted more accurate dwarf results. Stick with the original prompt with PM settings of 0.3 Strength and High Contrast enabled

[open echo]

When all else fails, there also a "reset" button at the bottom left corner. Just to be sure there's not a obscure option that was changed by accident

[tulip river]

so I've noticed while working on a new same-character scene that the new scene will have some elements of the old scene. do I need to flush or clear something?

[dim zodiac]

It could be a combo

It could be that some of the terms to conserve the character are bleeding into the scene

[tulip river]

ok ill restart the app as well

[dim zodiac]

And we've anecdotally noticed that the cache holds bits of previous gens

Could try clearing your browser cache as well and see if you get a different result

[tulip river]

thank you!

[open echo]

It depends. If it was an image generated in a group we might be able to get some info from the other ones

The prompt will not influence the seed

Or do you also have another group that was generated soon after or before?

I meant if you generated like 4 or 8 pictures at the same time 😁

Do you have the link to the generation? I can have a look when I get home

You could, it will not be exact though

[low bear]

As Tux said, if you generate multiple images at the same time, they will be one generation and you can can copy the seed from that. If they weren't generated at the same time, the seed will be completely different even if the prompt is the same

So if you deleted all the images from that particular generation, you won't be able to recover the seed unfortunately

Ah yeah then they won't share a seed unfortunately

[wispy pulsar]

Thanks a milli

[lament remnant]

Where can we learn before generate images like we want

[open echo]

Hello there 😄

I’ve been told this guide wasn’t so bad to learn prompt structure 😅

https://discord.com/channels/989166677390426132/1156647553387999353

[cloud barn]

Is anyone here willing to assist me with an issue I've been trying to figure out for a few days?

I'm using a custom-trained model on unique faces, but the generator has made a habit of generating up to two of the same faces in the same generation over several generations, how do I get around this?

[winter spire]

Might need some more information to help. But if I were a gambling man, I'd guess it's related to the dimensions you're using on your generation. If they're much larger or of a different ratio to the model size, the AI will try to fill in the extra space with what it knows. In this case, adding in another face. Does this help?

[cloud barn]

The dimensions are all 1:1 and I've been using the recommended dimensions which are the same as those in the dataset (768x768).

Would you like me to provide images?

[winter spire]

Yes please!

A link to your profile might also help so I can check some generation settings. Or even just some data on what you're configurations are set to.

[cloud barn]

Prompt:
Capture a centred close-up centred portraiture of a black woman in her twenties, she faces forward against a white backdrop.

[cloud barn]

https://app.leonardo.ai/profile/Timzeo

All my generations are private, I'm not ready to share my generations until I've perfected the model and prompting.

There's also several other persistent issues such as multiple heads, faces, black and white images

[winter spire]

Perfectly reasonable, dw

[cloud barn]

I just generated 8 more images and sure enough, two of them are practically the same character, 1 more being only a little less similair

[winter spire]

Was this model trained for SD2.1 or SD1.5? And are you using your Instance Prompt in your prompt?

[cloud barn]

SD1.5 as recommended in the model trainer, what is instance prompt?

Oh yeah i do use instance prompt

it's only 3 words

[winter spire]

Dang ok.

So some of the output images would say to me that the subjects in the training images aren't framed properly, by which I mean there's too much background space. Could also be a lack of variety in the source images in terms of facial features etc. Model training is tricky business, and admittedly I'm not expert.

However, whenever @real lynx happens to come online she might be able to give you a better idea of how you can solve this problem. She uses trained models extensively.

[cloud barn]

There is a fair bit of background space, the background for all the subjects is pure white.

I've suspected before it could have something to do with

[cloud barn]

I'll keep an eye out for her

[real lynx]

Hello

[cloud barn]

Hi

[real lynx]

I was reading your problem

Apparently it would be scale

The problem

The model was created in a data set

Are there 40 photos?

[cloud barn]

There are currently 13 images, Its a tedious process to prepare each image for the dataset, I plan on filling it to 40.

I can show you the dataset if it would help

a ss

[real lynx]

There are programs on the internet that simplify the process.

Because all photos have to be the same size

[cloud barn]

Yeah they're all the same size

[real lynx]

The photos must be of good quality and clear.

[cloud barn]

I've went as far as aligning the heads and eyes of the models so they are well centered ine ach image

[real lynx]

I'm going to try your prompt on my model and then I'll take photos so you can see what I want you to try.

[cloud barn]

Alright.

[real lynx]

Is your account free?

[cloud barn]

It should be premium

[real lynx]

Change that scale to the minimum

[cloud barn]

Ok i'll change it to 512x768

[real lynx]

Remember to put the model name in the prompt

Use dinamic

[cloud barn]

wait, what if my model has different faces, does it have to be the same face?

[real lynx]

It has to be the same face

[cloud barn]

: (

[real lynx]

Both eyes should always be seen from the front in medium profile.

And if there are 40 photos the result is optimal.

[cloud barn]

My goal is to combine multiple faces to create a new one

[real lynx]

I never tried it that way

[cloud barn]

It's worked well so far

apart from said issues

[real lynx]

I don't see bad results. Did you try changing the scale?

Sometimes even on my own face I notice that I take one eye from one face and the other from another.

That's why in different faces...

The result would be disastrous

[cloud barn]

It produces some really unique and beautiful faces because of this tho

I can show you the contrast between a bad and good result

[real lynx]

I want to see what comes of this

And I want to modify your prompt to give it more details

And see what happen

[cloud barn]

Ok gimme a few seconds

[real lynx]

Once the model is created you have to wait until next month to modify it

[cloud barn]

never knew that

[real lynx]

If you try to modify it right now it tells you when you can modify it again

[cloud barn]

Alright

I've just realised I've deleted mostly all the horrible results so give me a few more seconds

[real lynx]

I'm going to modify your prompt

How many models did you create?

[cloud barn]

It's about 4 now basically better versions of eachother as I learned to improve the dataset images

2 "unfavourable results"

both favourable and unfavourable results are from the same dataset/model

There women are combinations of the subject women in the dataset, a fair amount of the time you can tell who the "parent" subjects are based on certain features

[real lynx]

I like shaved

Masterpiece intrincate meticuloso Face Desing ,Capturing a strikingly beautiful portrait of a young black woman in her twenties, the subject is centered and confidently gazes forward against a pristine white backdrop. This high-quality photograph radiates with the subject's radiant brown skin, accentuated by soft natural light that delicately highlights her exquisite features. The woman's almond-shaped eyes shimmer with an alluring depth, while her naturally coiled, 1lustrous black hair cascades around her face like a crown of elegance. The impeccable clarity and meticulous attention to detail in this captivating image flawlessly captures the woman's strength, grace, and timeless beauty.

Modify your prompt

These dice generate a prompt based on your instructions

With the second option

[cloud barn]

blocked out

Also, does this forces the model to use the same face?

[real lynx]

It has many indications

Write the prompt shorter

You already uploaded the one that I modified

And it's a long prompt

You modify the face of the model with the prompt. The prompt is based on the model depending on the force you give it.

Different prompt

Same model

[tulip river]

thank you I didnt know the dice did that

[real lynx]

More strength to the promptas strength to the model

[cloud barn]

Jeez this uses a lotta tokens

[cloud barn]

Okay I'll play with it now

[real lynx]

🫣

Yes

It's a matter of practice

[cloud barn]

Wait so the higher the prompt the lower the dataset model influence?

I'm liking the increase in quality and detail so far

[real lynx]

It can be transformed into anime or whatever you want.

[real lynx]

I think so although

[cloud barn]

Because I still want it to combine faces

[real lynx]

If it is too long, it is very likely that the last words of the message will not have force.

[cloud barn]

Oh yeah, the first words have more strength

[cloud barn]

How do I make the model look more like say subject 4 in the dataset?

[real lynx]

()10%(())20%

[cloud barn]

?????

[real lynx]

Parenthesis also gives more strength to the word

[cloud barn]

ah

[real lynx]

I don't think it will ever give you the same result.

My photos are my personal photos and it is the same face

That's why take my features

[cloud barn]

Why does the ratio have to be 2:3 or 3:2 ?

[real lynx]

You would have to do that model randomly 40 times to give you a similar result to create the data set of that model

Or look for a modelless prompt that looks like that

Lend me the photo of the model you liked

[cloud barn]

here?

[real lynx]

I can generate a prompt based on that photo

Yeah

[cloud barn]

the output model?

ok

[real lynx]

And then I'll show you how

[cloud barn]

Just use any one of these

[real lynx]

I speak Spanish, where are you from?

That's why my English is bad, I use a translator

Maybe some things I don't understand

[cloud barn]

I'm from UK

Though english is my second language technically

[real lynx]

Perfect

Can you send me the photo here?

[cloud barn]

Output Model

Not input, input are real people, I'm assuming you were asking for the generated model (person)

[real lynx]

Perfect as an example to teach you

And I like this model so later I'm going to peel myself too xD

https://huggingface.co/spaces/pharmapsychotic/CLIP-Interrogator

This page generates a prompt with the photo you upload

[cloud barn]

wow

[real lynx]

a woman in a strapless top posing for a picture, portrait featured on unsplash, brown skin like soil, pixie cut, passport photo, anime visual of a young woman, sleek head, short facial hair, profile picture, unedited, face shown, upper body avatar, purity

I'm going to try the prompt on my model and also without anything

[cloud barn]

I'll wait

[real lynx]

I don't hair xD

With my model and without my model

[cloud barn]

hoh

[real lynx]

I never had an hair so short, it doesn't look bad on me

[cloud barn]

elo?

[real lynx]

Hair short

[cloud barn]

also for datasets should I use SD1.5 or SD2

[real lynx]

It won't let me have shorter hair

The hairstyle must have a name

I'm going to google it

[cloud barn]

Alright, as for now I'm going to experiment with everything you've taught me.

[dim zodiac]

I would suggest SD1.5

then you can apply more of the tools after the fact

but if all you care about is pixels and your output, without using other controlnet etc then 2.1 is good!

[cloud barn]

Okay, I don't know what you mean by controlnet at the other stuff but 1.5 has been good to me so far.

Thank you

[cloud barn]

Thank you for the help

[real lynx]

Any questions happy to help

[cloud barn]

Are we allowed to PM others?

[real lynx]

Sometimes I have them blocked

[cloud barn]

I would like to further the discussion privately if I may

[real lynx]

But if you surprise me as soon as I see it, I'll answer you

[trail nimbus]

I am trying to illustrate a scene using leonardo AI and I was wondering if there is a way to use caricature art ? like use real photos of a person to generate caricatures and use those images in this illustration. Please let me know if that is possible ?

[fringe glacier]

The Art of War: Behind the Victories of Alexander, Caesar, Genghis, and Napoleon cinematik 8k heyperrealistik

[robust gulch]

You can't generate images on discord. You have to go to https://leonardo.ai

[wise canyon]

Hello Isotope, yes. You can upload a real photo into "Image Guidance" (image to image setting), as a reference for the AI, then generate images using one of the anime models. Results will vary according to your prompt, and the "image to image" strength slider. The higher the strength, the more the generated image will adhere to your photo.

[wispy pulsar]

Anyone know WTF happened to #🖼│image-share ?

[ancient halo]

how to place a photo in image share

[low bear]

Just press the + button in your chat bar and upload an image :)

[ancient halo]

thats all ? i now but if you upload how you sent the picture of is upload everything you most do

[spiral axle]

you can also just copy/paste 😄

[spiral axle]

or you the share button in your feed

[sweet arrow]

How do you maintain a consistent style? I've tried to use "Train a Model", I've also tried keeping the same settings, but the results vary from generation to generation

[spiral axle]

do you used a fixed seed ?

[sweet arrow]

No, I don't understand much about that

[spiral axle]

you can either use a fixed seed in your settings to keep that consistency (mostly to make subtle changes) or you to to use Img2Img or Img-Prompt(Prompt-Magic)

if you like a specific Image Style you created... you can find the Seed for it in your feed

[sweet arrow]

Ok, so, say I have a perfect style I've achieved and the image happens to be of a woman, but I want to generate one of a man or dog in the same style. The seed will keep it consistent?

[spiral axle]

I am not that much of an expert but so far mostly yes

same seed

I have not specified anything about the person itself though

[sweet arrow]

Hmm ok I'll try it

[spiral axle]

you can see it has the same style like the image before 🙂 ... I just changed it from a Harp to a Guitar

[olive pilot]

I can not creat image

[robust gulch]

You have to go to https://leonardo.ai to generate images.

[spiral axle]

do you have spoken to Verrocchio ?

[tulip river]

I need a quick way to make consistent characters in a comic book style. I thought that using a celebrity face was the answer. It is for consistency, but if you use a celebrity face, you cannot modify it much, like asking Leo to close their eyes or shave their head. any ideas?

[spiral axle]

fixed seed

[muted jewel]

i can close eyes idk what you on about

[tulip river]

im trying to make an image of Sylvester Stallone with his eyes closed very hard

ok ill look into fixed seed thank you

[karmic pulsar]

Can this program make a AI version of my portret if I take a picture with a camera and upload it to the program somehow? if so how?

[robust gulch]

You can use it is an image to image reference.

Like this

[karmic pulsar]

Whats the prompt if i want it to be my image and not soemthign random?

[robust gulch]

Some people just write copy or you can describe the image.

Or if you but the strength low you can rougly base a new image off the old one and use the written prompt to describe what you want the new image to look like.

[karmic pulsar]

Thank you!

[robust gulch]

No problem. 🙂

[karmic pulsar]

Anyone here good with Youtube SEO / marketing / Thumbnails?

[tulip river]

I use google bard to come up with seo keywords and descriptions and canva pro to make thumbnails

[spiral axle]

you can use your picture in Canvas though 🙂 ...

[karmic pulsar]

canvaaas

What is that

[karmic pulsar]

I literally thought you ment Canva

I was like what

Im gonna try that out thankyou G

[urban ridge]

Good morning everyone, everyone!
I need some help.
How do you create a realistic portrait of someone while maintaining the appearance of his face but with different framing and shooting. For example in the forest, with different clothes??

[spiral axle]

with a fixed seed 🙂

[urban ridge]

@spiral axle Okay, can we contact each other in a private message so you can explain to me, please?

[spiral axle]

advanced settings 🙂

[urban ridge]

Okay but how do I keep my model portrait and make different framing as for example walking in the forest while keeping the same physical aspect of his face without modifying it?

[spiral axle]

You can also use canvas

[open echo]

That’s going to keep the same framing though (the general pose and outline will be the same)

When you use a fixed seed, it only means the beginning of the generation process stays the same. So the random color blob that will become the image will be constant. But after that, if you change A LOT of the prompt, you will lose consistency because even though it’s using the same base, you are taking it in a completely different direction

[low bear]

Probably cuz the gender isn't specified in the prompt, a consistent character should be achieved with both a prompt that clearly describes the character and the fixed seed

[low bear]

I'm currently still a university student in computer science, but I have been programming for fun since I was a kid haha

My favorite language is Lua, but I know a ton of different ones

I know Python as well but its one of my least favorite ones unless I'm working with custom neural networks/data science

Hahaha yeah I never liked hacking personally, but I went a similar path. Just started experimenting with code, trying what worked and what didn't. Personally its always been the creative part of it that attracted me, being able to create something and seeing it come to life.

[zinc jungle]

What ?

[stable carbon]

sometimes that happens. innocent images get tagged as NSFW. Probably not your fault. you could try contacting support to get them uncensored. or just leave them be and make more 🙂

[zinc jungle]

I will not I have to much pictures
You know, a lot of time has passed and more pictures have been "collected"😅 😅

[stable carbon]

yea i cant even tell you why that happened. maybe its something in the prompt that caused it, or maybe its one of the other secret detection things on the website 🤷‍♂️

[wispy pulsar]

anyone else having issue with #🖼│image-share ?

server or sub server, whatever... nukes itself then comes back with a message that you don't have permission to view history in #🖼│image-share ?

I can still post there, though

and another user is in their now

[wispy pulsar]

WTF?!!! The Ai just had a psychotic break.

😆

[acoustic bronze]

\hfj

[wispy pulsar]

Maybe you also have some tips on how to get this type of character design in image gens?

[wispy pulsar]

What is a Screencap? anything like old school cell animation?

[low bear]

A screencap is short for screen capture, so in the context of anime a still image taken from an animated scene

[spiral axle]

Screencap = Screenshot from TV/Computer... dont know if its working or not 🙂

[low bear]

I use it a lot too haha, seems to work

[spiral axle]

I just found a Screencap Style LoRa so I thought I can get Anime Episode Pictures

u got it wrong xD

Its more about emulating the style of Anime Episodes

like the shading and the lineart

the coloring

it exists

everywhere

its not adding some kind of perspective effect in there or something... just basic "flat" like Anime TV Shows

and it cant generate shows... since I mashup multiple anime shows to generate unique characters

like this for example 😄

NGE Bunny

all depends on your prompt 😄

[wispy pulsar]

Actually this is kind of cool. You should use it for a project. Would make a cool Mech Anime.

[spiral axle]

like this 😉

[wispy pulsar]

Naruto in dad jeans

[spiral axle]

Nanamin

[wispy pulsar]

This generated anime is too good what is in this prompt or any of them?

[spiral axle]

Baki and Mashle xD

[wispy pulsar]

I have to learn how to do this

[spiral axle]

Like I said it depends on all your personal preferences 😄

you can go Painting Style, you can go Anime Style

you can even turn Luffy into another Art Style

lots of examples 😄 (but I keep my prompt mostly the same)

[wispy pulsar]

Teach me oh Great Master of the Meta Arts🙏 I am your humble student.

[spiral axle]

its just my normal Prompt xD .... you just have to edit it to your Needs xD

[open echo]

My first guess would be that your prompts are too long (Just going by the comment you made some days ago about not having enough space for the prompt :D)

The more words you add to the prompt, the bigger salad you get and the less control you retain

If you want to be surprised and don't have a goal in mind, go ham with the prose and the words (It's a fun exercise :D)

If you want something specific, be precise and concise. Avoid figure of speech and metaphors, keep your word count to a minimum

[wispy pulsar]

Thought I had to be a Craftsman to post these? Even

[spiral axle]

Knight is fine for posting images

[wispy pulsar]

but I'm not a Knight🤔 ...thought I should be by now

[open echo]

Knight rank doesn't exist 😄

[wispy pulsar]

anymore

but it did

[spiral axle]

well Verro is wrong then xD

[open echo]

https://tenor.com/view/nowarinbasingse-atla-avatar-the-last-gif-18401680

[spiral axle]

@open echo Verro still says you can post when you reach Knight Rank xD

[wispy pulsar]

you still have the shield

[open echo]

Shield icon is for Volunteer Moderator 😅

[wispy pulsar]

Yeah, it is now, but it wasn't before

I remember Knight, have I fallen into a Mandela? BTW... the Fruit of the Looms Mandela effect is actually real. There was a cornucopia in the logo before they changed the logo. I know what I wear. That's my story and I'm stickin' to it.😆

[wispy pulsar]

So, if I still had an old account I'd be an automatic lock for a mod?

[open echo]

Who knows?

https://tenor.com/view/gizelle-bryant-housewives-of-potomac-plotting-thinking-the-plot-thickens-gif-16100206

[feral bough]

Shield icon has always been only for Volunteer Moderators.

[feral bough]

In fact the Volunteer Moderator position has been around since before the Knight rank was introduced

[cloud barn]

Can anyone clarify if having a white or blank background in dataset images causes unwanted elements in generated images?

[dim zodiac]

blank as in transparent?

[dim zodiac]

if ALL your images are white backgrounds, you will only be generating images with white backgrounds with that model

it will really try and stick to what is conserved in the training dataset

[cloud barn]

Funny thing is, it doesn't do that, it just fills it with random textures

I wish It would stick to the white backgrounds

[wispy pulsar]

What was the knight emoji then? I had the shield but I wasn't a mod. So, I don't remember what knight even looks like then?

[feral bough]

I'm not sure Knight ever had an emoji

[wispy pulsar]

and why can't I get past this silly Novice level?😆

[feral bough]

The icons next to the names are quite a recent thing and only the hat was added for Former Nobles

[feral bough]

https://tenor.com/view/walk-away-im-out-gif-27669704

[cloud barn]

Is anyone here an expert on datasets

[dim zodiac]

if you want to explain your dataset I can help troubleshoot

[charred stump]

Hi, mi problem with Canvas editor is that it doesn´t load images from my computer.

[dim zodiac]

interesting, what browser are you working with?

[charred stump]

opera

It was workin fine some days ago.

[dim zodiac]

can you try in another browser to see if this is browser specific? Do you know if you have any new extensions?

[charred stump]

No, I don´t havy any new extensions, but I can change browsers.

[dim zodiac]

also one of the first things we suggest is to clear your browser cache

sometimes it holds on to bad data and causes issues

[charred stump]

Ok, then. I will do that, if I keep having problems I will come back. Thank you so much!

[cloud barn]

What do you need to know?

[dim zodiac]

you're training for an artstyle if I recall correctly? How many images are you using, which version of SD, what category, etc

are all your images cropped to the same pixel dimensions?

[cloud barn]

No an artyle I'm training a model to create new faces

[cloud barn]

yes

I believe I was strict with that aspect

they're all 768x768

[dim zodiac]

ok so SD2.1

[cloud barn]

1.5

[dim zodiac]

then they should be 512x512

[cloud barn]

so 768x768 is for 2.1?

[dim zodiac]

yes, that's correct

[cloud barn]

What's the main difference this causes

I don't want to lose output quality

[dim zodiac]

you can apply different elements and controlnet to the model in SD1.5

brb sorry, my dog is panting - need to take him for a walk, 15 mins max

[copper warren]

Do you recall when it was that you were a member, before?

[night vault]

512 or 768 are what they offer now unfortunately. the resolution that train I have seen not exceeding 1024x1024, anything above will need upscale as additional step.

[cloud barn]

I'm fine with both resolutions, I'm just trying to figure out why the background in the output is always some weird texture

It seems using "pose to image" has added a lot of stability and consistency to the image background.

[dim zodiac]

I'm back!

can you also explain what you mean by New Faces?

[cloud barn]

Face 1 + Face 2 = Face 3

[dim zodiac]

so half of your images are one subject and half are another?

[cloud barn]

no

each image is all individuals

No one the same

[cloud barn]

this is a simplification

[dim zodiac]

are they all at the same angle?

just trying to get a feeling for your dataset

[cloud barn]

This is the growing dataset

[dim zodiac]

understood

yes it should output a great front facing individual

did you set it as "general" or "character" or what when training the model?

[cloud barn]

Photograph

Should I have done character?

[dim zodiac]

photography should have been also applicable for this purpose

and what is your prompt structure like when you use the model?

[cloud barn]

the instance prompt or the generation prompt?

[dim zodiac]

an example generation prompt

[cloud barn]

Current Prompt
Black Women Dataset 2, Capture a centred close-up portrait of an african woman in her twenties with distinct facial features and clear skin, she is facing forward gazing at the camera against a soft, clean, and neutral colour photoshoot background.

I was adviced to put the model name at the start

[dim zodiac]

curious, all of this is consistent with what you're looking to achieve

yes, that is very correct

well... no the instance prompt

if it's the same as your model name, then yes

[cloud barn]

the instance prompt is also there: "african woman"

this is without pose to image:

It's usually much worse

Here's with pose to image:

[dim zodiac]

yes pose to image would help get the model in exactly the position you want

try just calling out a white background

[cloud barn]

I tried that before, didn't do anything I'll do it again and show

including "white", without pose to image:

there's a number of disasters

white as in white background

[dim zodiac]

yes I can see 😄 what settings are you generating

[wispy pulsar]

like 2021

and 22

[dim zodiac]

try increasing the guidance scale to 9, and in Advanced options try choosing Euler Discrete as your scheduler

[cloud barn]

I don't use alchemy in this stage of my project because it introduces unwanted elements. FYI.

[dim zodiac]

I don't see anything incorrect

so I'm just trying to pick settings that could increase prompt adherence

[cloud barn]

let me attempt

[cloud barn]

not sure what to say

increased saturation?

[dim zodiac]

can you kill the negative prompt for a generation?

[cloud barn]

less saturation?

it doesn't want them to be lonely for sure

[dim zodiac]

but no effect on the errors

😄

[cloud barn]

I've tried including the word "single" in earlier attempts, didn't work.

pose to image seems to be the only fix for now, just wish I didn't need to use it

I could hang this on the wall and people might think its from some artist

[limpid lake]

How do we get the highest resolution possible without having to go into another site to improve it?

[dim zodiac]

generate at the highest resolution that gives you the intended result, then upscale

Alchemy High Resolution first if you can

and then Crisp Upscale

[dim zodiac]

would you be willing to share your dataset with me? I could train a few models differently and see if it grants better results? I see nothing incorrect with what you're attempting

[cloud barn]

hmmm

[dim zodiac]

if not, that's cool

[cloud barn]

sure, Its not complete anyways

[dim zodiac]

but I think you're doing everything right and it wants to give you random effects

we could shorten the prompt

[cloud barn]

bout 30 more models to find and process

[cloud barn]

I'll Dm the dataset

open to ideas on that

[dim zodiac]

try: "African Woman, front facing beauty photography, white background, captured in professional DSLR, narrow depth of field"

[cloud barn]

eular and 9 guidance scale with no pose to image and negative prompts?

[dim zodiac]

try that first yes, I like to start simple and then add as needed

[dim zodiac]

yeah not much better

[cloud barn]

hm

[dim zodiac]

try the negative prompt added?

[cloud barn]

with negatives:

[dim zodiac]

like... 2 correct.. ish

[cloud barn]

yeah a lil better than the previous

[dim zodiac]

try "beauty portrait photo" instead of front facing

[cloud barn]

Do you still want the dataset if it makes it easier?

[dim zodiac]

no I think this shows that prompt editing could achieve it, just need to find what the model likes

😄

unless you want to crowdsource

[dim zodiac]

ok, front facing was better!

good gracious

[cloud barn]

the word beauty brought some fashion clothing into the generations 😁

[dim zodiac]

yes it did

and jewelry

[cloud barn]

yes

this is gonna take a number of sessions

[limpid lake]

Where is the crisp upscale option? I think I've been doing it right with alchemy and the refiner scale

[dim zodiac]

are you generating in the XL models?

[limpid lake]

Yeah most of them have been in there

[cloud barn]

all too common

[limpid lake]

I have been playing around with some modes there is some without the XL

[dim zodiac]

do you have a paid account?

because the XL models only have Alchemy upscales

[limpid lake]

Yeah I do 🙂

[dim zodiac]

oh cool! (If you want a specific role for paid members in discord, send me your Leo profile name)

then Crisp is available as the DNA helix icon

and will 2x the size of the image

[dim zodiac]

sorry there doesn't seem to be a quick answer!

[limpid lake]

Ooooooh brilliant thanks you so much I've been going into different site to do it and then asks me to subscribe and thought to myself "wait Leonardo should don it I must be missing something " 🤣🤣🤣

Rtrcreations ")

[cloud barn]

It's fine, as long as someone can help me figure it out sometime

I also have an issue with alchemy completely ignoring my dataset

[dim zodiac]

alchemy does a lot of extra processing

if you want alchemy power but strict adherence, use the "None" pipeline

that would allow you to use weighting in your prompt structure

but not get the added effects from the post-processing from alchemy

[cloud barn]

where is this

[dim zodiac]

actually, let's try that... African Woman, (front facing: 1.3) beauty photography, (white background: 1.4), captured in professional DSLR, narrow depth of field using Alchemy, with the None pipeline

right down there at the bottom of the pipeline list

[cloud barn]

ok 1 sec

[cloud barn]

They look african indeed, but not corresponding to my expectations mainly because I don't see any influence from the model:

[dim zodiac]

the pose, position and background are better but everything else is a bit extra yes

[limpid lake]

How do I stop this extra limbs and missing letters lol

Its a shame we don't get refunding points when the ai does wrong lol

[dim zodiac]

good negative prompting for the limbs, but letters are a roll of the dice

[cloud barn]

Canvas editor is nice for removing unwanted elements

[dim zodiac]

Canvas is excellent for a lot of things 😄

could even just generate the card, then take to canvas to add text after

[limpid lake]

Ooooh I haven't mastered that fully yet lol