💬│general

[analog nebula]

I haven't tested this method however. Just something I thought about

[slender ore]

This AI tool is really cool! I am really digging the prompt editor.

[analog nebula]

Glad you're enjoying it!

[devout bridge]

I just noticed my invite already came in a while ago, and registered my account.
I'm trying to figure things out now.

[twilit pumice]

IM SO SORRY

I did not mean to say that

My brother got ahold of my phone

[deep shoal]

Lower your init strength - it’s probably too high. The higher it is, the more the output will conform to the image via image to image

[soft bolt]

omg this thing is AMAZING

[analog nebula]

❤️

You're amazing

(I think?)

[vague silo]

hey guys, I'm new here... does it take a long time to release access to the site?

[analog nebula]

Depending on where you've signed up from. Though our website, it'll take maybe 2 weeks or so. You could make an introduction in #👑│introductions and then apply for faster access though #✨│priority-early-access . It will still take several hours+ but should be much faster. 🙂

This 2nd option is something special for our discord community members essentially

[vague silo]

ok, got it...thanks a lot.

[analog nebula]

No worries!

[vague silo]

How long has the project been on the air? I discovered it at dawn and found it very promising haha

[analog nebula]

Late December

[gleaming viper]

Morning all, I am looking forward to getting my early accsess!

[analog nebula]

We actually started with an really small test group the week after Christmas

[analog nebula]

Welcome!

And good morning

By the way if you make an intro in #👑│introductions , that unlocks #✨│priority-early-access . Much faster than signing up though our website. Just a little something for our discord community 😉

[trail arch]

Hello im introducing myself to get some awesome perks

[hidden lily]

What is your start prompts? How do you write it? With "commas" between promts? I got a lot of hands with 6-8 fingers 😅 And bodies are difficult it seems

[hidden lily]

#👑│introductions

[trail arch]

Tack fröken

[hidden lily]

Bare hyggelig 😉

[analog nebula]

Don't add too many commas

Can break things sometimes

Especially with prompt magic

Try prompt generator if needed

Also pro tip, you can study what others are using as prompts by clicking on the images in the feed, or go to Finetuned models > Platform Models and study the examples done by our team 🙂

By the way commas sometimes work and sometimes don't

But with prompt magic it sometimes does break things so just be aware 🙂

[hidden lily]

Thank you, I did that. It helped a lot, except the fingers 😅 but getting better and better results. Can't expect to get it perfect the first time. Have to learn to "speak" like Leonardo first 😏

[analog nebula]

Our platform models does okay with ,

[trail arch]

Can i make some awesome pixel art assets for 2d game!?

[analog nebula]

Just be aware to try changing it if the AI gives you words instead

[analog nebula]

Yes

[trail arch]

Very poggers

[analog nebula]

We have an pixel art model but you can try your own too

[analog nebula]

Yeah

Basically generate a bunch

And pick from the best

Then image to image if needed to create more variations of one generated image

[fluid viper]

This AI really has no idea what a mole is, does it?

[analog nebula]

Possibly not

Time to train it on moles

We are talking about the animal right?

[fluid viper]

Yes.

[analog nebula]

Hmm interesting

@mild walrus seems like Leonardo doesn't know what moles are lol

Is this close enough @fluid viper

Experimenting with prompts

[fluid viper]

I mean, compare it with real moles. It seems to be getting the idea of "small, furry creature" (in absence of other words in the prompt, which completely kill the mole) but that's it.

[analog nebula]

Yeah

It's kind of a pain to get it to generate an actual mole

Now that I look at it

[dull crag]

Hi everybody, if I generate an environment, is it possible to extract the assets from it and to have multiple images of the same environment from multiple perspectives or angles?

[analog nebula]

Seems a bit complicated for now to be honest. You could try image to image and then adding the view point you want into the prompt maybe but I don't think it's quite possible at the moment

[dull crag]

I will try thank you

is it possible to import an existing image of a castle, and ask to keep the shape, but change the style? I would like to enter an existing girl cancel picture, and and have it as a as an illustration.

? a-t the end 🙂

[analog nebula]

Possible I believe

With image to image feature, though you will have to experiment with the init value

[fluid viper]

It's also either ignoring the words "antlers" and "mane", or it doesn't know what that is too.

[vocal notch]

Oooh

Cool Art

Ok, Thanks!

[hidden lily]

Question; Haven't quite got the hang of the "image to image" tool. Can anybody explain and tell/show what they do and use it for? 🤔 Maybe I'm doing it wrong...

[shadow lantern]

Gm!!

guys, how to create something?

[hidden lily]

You should take the time to read information available to you here, read all channels, and follow instructions then you might get access 🙂

[shadow lantern]

Never mind

[analog nebula]

Good morning! Leonardo.ai is an web based application so you will have to wait for an invite to be sent to your inbox. 🙂

[shadow lantern]

I will wait,
I think I do everything right

[shadow lantern]

[valid junco]

hi all. prob a noob question but is there a way to impose text onto the images that are generating? it seems to give garbled words and letters even when i say "include the text XYZ"

[analog nebula]

Not really. Image generation ais currently do not have a concept of language

Only what visually looks like language

Now if an hybrid model/solution was to be developed

That may be possible

[red kindle]

Question, do I need to enter my actual name in the form or the name associated with my email? (Email was used to create a youtube channel hence the name)

[gilded galleon]

Hey everyone! Filled in the form, looking forward! 🙂

[thin yew]

Yo

[random fossil]

Hey, just wondering how to weight words when writing image prompts?

[bronze crag]

Question: thoughts about a SD model specifically designed to produce pixel art?

[analog nebula]

Am actually unsure about this, @mild walrus might be able to help 🙂

[mild walrus]

Got that ready to go, we’ve just been priori zing other things first :)

[rancid harness]

Hi Guys how can I give suggestions?

[analog nebula]

Type /

/suggestion suggest

Then enter your suggestion

[rancid harness]

Ah perfect

thanks!

[analog nebula]

👍

[trim basin]

Does this work similarly to midjourney AI?

[analog nebula]

Our service operates differently from Midjourney. Leonardo.ai is currently a web based application 🙂

[trim basin]

So i ll be able to create 3D game assets? what formats will be the output format?

[analog nebula]

2D only at the moment, we plan to expand into 3D as the tech becomes more viable

Outputs are png

[trim basin]

mmm right okay, so how is it different from Midjourney AI atm?

[analog nebula]

Custom models would be the most notable difference right now along with other smaller features

Basically you get to train your own models on your own datasets

or use community shared models

[wide lark]

I wonder who did this prompt?

"hyper-realistic giant floating head of nicolas cage attacking new york city"

hahahhaha

that's really creative

[amber wyvern]

Oh no 😂

[wide lark]

not tagging anybody on purpose... just wanna say this is super creative

[grand torrent]

Hi guys - need help. Is there a way how to convert graphic style to another for example like here?

how to set leonardo to get similar results?

[amber wyvern]

Not out of the box...
But what worked for me (although only tried it in very specific cases) is to train a model on the style that you want, preferably with the subject matter you intend to convert and then use img2img with a high init strength on that self-trained model

[analog nebula]

train a model on a specific style

then image to image with high init strength as Neprau mentioned 🙂

kinda works

Have tried it but it sometimes deviates too much from source for sake of the style

[amber wyvern]

^

[analog nebula]

depending on how much you want to preserve source image, adjust init accordingly

I actually ran an selfie of myself though leonardo for my profile picture

Had to bump up the init a lot 😅

[grand torrent]

I see

thank you

I will try

[analog nebula]

No worries!

Hopefully you'll get a decent result!

[grand torrent]

another question when I train a model I use same style, for example:

but results is:

not really good and dont know how to set proms or general setting

[wide lark]

@distant tide may have a more specific answer with future options in the pipeline

[analog nebula]

hmm

could try adding white frame to the prompt

see what happens

using prompts takes quite a bit of experimentation to get the results you want

so can be tricky initially

there's no fixed way to get results since everyone wants something different

[grand torrent]

okay, thank you for your time :)

[analog nebula]

also bottom row seems more in line with what you want

so it's kind of getting there? but also a hit and miss thing

typically I run several generations and modify the prompt if needed till I get a result I'm happy with

[chilly bloom]

Hi am new here

[analog nebula]

Welcome to our discord! 🙂

[chilly bloom]

Thanks!

[viral igloo]

Hello people!

I'm in!

[analog nebula]

Welcome, thanks for joining us!

Also @chilly bloom , @viral igloo , do stop by #👑│introductions if you want to get early access invites faster. It'll unlock #✨│priority-early-access which allows you to do that 🙂

[chilly bloom]

Is it gonna be completely a web based service or u have plans to make it in discord like midjourney?

[analog nebula]

There might be one in the future

[chilly bloom]

Nice

[analog nebula]

What is certain though is that we will have mobile apps for both android and iOS

[chilly bloom]

Awesome!

[river holly]

I've never noticed the "unzoom" button before. Is that a new feature?

It's a shame I can only do it on images I generated, not ones I uploaded. It also doesn't seem to work on what's been relegated to my feed.

[analog nebula]

Never noticed it

probably is new

[wet scaffold]

How long did you guys wait until you're whitelisted? the suspense is killing me... 😂

[analog nebula]

might take up to a day or a little more if done though the priority access channel

they're still manually processed but at a faster turnaround

though currently the team members who are able to mail them out are very busy so that's a bit of backlog

should be all sent out when they get around to it next time 🙂

[analog nebula]

I don't see any changes

maybe I'm using it wrong or its just not fully implemented

oh wait

it's just slow

amazing

@mild walrus , @marble moat much love to you guys if this was your handiwork

[wet scaffold]

sigh...guess i'll have to wait. Thank you!

[analog nebula]

No worries. And apologies for the wait 🙇‍♀️

[analog nebula]

By the way @river holly

They seem to have added search function

to both feed and models

[river holly]

That's good. You know what I could really use, though? Upscaling and "unzooming," i.e. outpainting on pictures I upload, not just generate. Otherwise the only way I can apply it is by running an old image from my feed through a model again at the highest init strength, but even that changes it a little bit.

Or is there a model that outputs an identical, unchanged picture? That would solve it.

[analog nebula]

Yeah I think that'll be one for the suggestions 🙂

Am a bit bothered too

maybe it only does it because of trace data that's not been purged

don't really know how it works

[rancid harness]

Working on a miniseries rithgt now and using all types of AI with my team to create the concept art. Just wanting to let ya'll know. We are actually spoling quite a lot so my apologies for that if you ever see the series in the future.

[analog nebula]

Well I would love to see them anyway

[rancid harness]

😅

[river holly]

Oh, right, there's a suggestion channel. My bad!

[analog nebula]

All good!

[rancid harness]

By the way is there a way to bulk download?

[analog nebula]

Not at the moment but is an feature that will be implemented I believe

[rancid harness]

Ok nice!

[analog nebula]

but it's more of a download by prompt kind of thing it seems

[rancid harness]

aaaah I see

[analog nebula]

hopefully multi select will be a thing in the future

[river holly]

Oh, wait, I just went to "suggestions" and it told me, "you do not have permission to send messages to this channel."

[analog nebula]

I've requested for such a feature myself

[analog nebula]

/suggestion suggest "your suggestion here"

press / to open the command menu then type the rest in

🙂

also my discord is borked so its not showing them for me personally but it is there

[river holly]

Wait, do I type "/suggestion" here and it makes it show up in the other channel? Because I cannot write anything in that channel.

[analog nebula]

yeah

basically

[fluid viper]

How are you all getting these awesome results and all I get is these eldritch abominations?

[analog nebula]

you mean mine?

or just generally

[fluid viper]

Just generally, but yours prompted it.

[analog nebula]

mine's a custom model just meant for humans and as for the prompts, I just used the prompt generator and a theme

so I really do not know

abominations are not as frequent with this model

[halcyon ore]

/suggestion command isn't working for me... anyone else?

I've waited and it only shows:

[analog nebula]

restart discord if needed

I think it's bugged out for all of us

It's not showing up for me

probably a discord issue

[halcyon ore]

Alright - I tried restarting, so probably discord issue. Thanks for confirming it's not just me

[analog nebula]

I was explaining it to another person on here and it was there when I checked 😅

yeah most likely discord's end

[halcyon ore]

I saw that just up above!

[analog nebula]

so restarting may or may not fix it

[river holly]

Does it make a difference to the A.I. if a person is upside down? Because I had a picture like that and it just wouldn't recognize the face and tried to do weird things with it.

[analog nebula]

Possibly. Considering Ai is trained to recognize objects in a certain way I believe. So an upside down human isn't visually recognizable to an AI. At least that is my guess. I'm not very versed on the machine learning side of things

[sand wren]

.

[mild walrus]

Mostly lachlans I believe !

[fast tiger]

Welcome to Leonardo. Feel free to make an intro in #👑│introductions, and that way you can unlock the #✨│priority-early-access channel, fill out a form and have a quicker access to Leonardo 🙂

[sand wren]

Thanks, eta on release?

[undone wolf]

@fast tiger When will we be able to use Leonardo ai ?

(I just joined and introduced myself)

[fast tiger]

It is now available, and for now it is free. Plans are not yet known 🙂

[sand wren]

the 3D model version?

[fast tiger]

You must wait for your invitation to use it. The process is fast, if you filled out the form that is in: #✨│priority-early-access 🙂

[sand wren]

Does it include 3d model version? Also i cannot see that channel

[undone wolf]

Done !

[fast tiger]

For now Leo is working on 2D assets, the team is working on that

[sand wren]

ok

[sand wren]

I cannot see that channel ?

[fast tiger]

I mentioned you in #👑│introductions 🙂

[river holly]

Oops! Something's wrong. Rendering seems to be down. "Failed Generation - [...]"

Now it's working again.

[cloud plinth]

Have you guys had success with other AIs for assets and sprite sheets?

I'm still finding it a bit difficult for sprite sheets specifically to have decent detail without distortion

[shut raven]

are the invites still given?

[undone wolf]

@mild walrus

I have this i verified my mail but still cannot sign up

[mild walrus]

@marble moat

[marble moat]

Click log in

[halcyon ore]

If I remove the background from one picture and want to reupload it to use it as image to image, is there a way I can ad a new background only by prompt?

[fast tiger]

I tried it, but it didn't work. I don't know if @mild walrus might have an idea if it is possible. 🙂

[little glacier]

Press forgot pass

Worked for me

[shadow lantern]

woah I just got the invite thank you, amazing results, WOW!!!!!!!!

[little glacier]

Same here, this ai is absolutely ground breaking

It feels like it shld be forbidden

I can make so many actual good quality assets in minuts

[ebon yacht]

hey guys

[mild walrus]

once we get the canvas feature going i imagine that's something we could do, for now i believe the transparent parts are turned black

[neat axle]

@stable solstice is the paper model yours?

[halcyon ore]

Thank you for the help. Amazing work, by the way!

[trim basin]

that is excellent

[wet scaffold]

i keep getting kind of messed up eyes in my portraits... is there a prompt or setting to mitigate this?

(astonishing results though)

[fast tiger]

This will serve you well: 🙂
#1058409094421155870 message

[valid junco]

If I delete a prompt and its results, will it delete it in publicly as well?

[fast tiger]

That's right!

[jaunty ore]

Any clue on how scheduler works?

[spark cosmos]

Blademort is an amazing contributor to this community.

[mild walrus]

Some slight differences to how an image form, euler ancestral might be better when working with your own image (init image)

[untold fern]

I'm curious if the product on offer here is the source platform, or only an interface to it?

[analog nebula]

Interface to it

We do plan to have an API available later on but no source.

[analog nebula]

@marble moat

Actually

@mild walrus

[untold fern]

Would it be possible to run the API in real time applications? (e.g. generating objects in realtime)

[analog nebula]

I'm currently not too certain what capabilities will be available for the API at the moment. @distant tide or @deep shoal might be able to give an reply 🙂

[marble moat]

API is yet to be finalised, in theory it would be possible to generate images and models through the API

[undone wolf]

Which password?

[marble moat]

did you set a password?

[marble moat]

your screenshot was on the sign up page, click here to log in

[undone wolf]

well yes on the account creation

Oh yeah it worked !

Nice

[marble moat]

good to hear! 🙂

[analog nebula]

Yes. But it'll be much faster if you made an introduction in #👑│introductions and then apply for #✨│priority-early-access after it is unlocked. The turnaround rate is much faster than though our website. Just something special for our discord community members 🙂

[shut raven]

https://tenor.com/view/boom-gif-25557098

[shadow lantern]

This is my fav ai so far

[analog nebula]

Glad you're liking it 😄

[dapper shale]

I agree infinati it's stellar!

[river holly]

What does the "tiling" switch do?

[analog nebula]

makes things easily repeatable

like say if you generated an ground texture

tiling essentially makes it so you can repeat the texture over an surface and make it look cohesive

hides the seams

[wintry nova]

Hey group

Anyone tried to create character from scratch in Leonardo, so that you can use the same one in different images after?

[dapper shale]

Like generating from a seed?

[wintry nova]

Probably. Not sure what exactly it means.

On practice, I need to create a comic book. Meaning pictures in the same style, with exactly the same character, but in different situations and environments

[bright haven]

Hi, it seems that i have forgot my password, I tried the forgot password form on the leonardo website and it doesn't send me the reset link on my email 😦

[wintry nova]

I'm trying the following way: creating a model, feeding it 3 images from our 3d model (different angles/poses). Not sure it is the right approach (I cannot test for now as I get server error and waiting for support to get back to me on that)

[analog nebula]

Hello, is this your first time logging into the application?

[bright haven]

no, the second

[analog nebula]

Could you open an ticket though #989166677910519850 and include your email address? I'll get Peter to investigate when he returns

[analog nebula]

Could try retrying anyway, see if it works now

Still waiting on team response at the moment

[wintry nova]

yeah, doing this every 5 minutes 🙂

but anyways I'm not sure if my approach is the right way to solve my specific task 🙂

[dapper shale]

I haven't had much experience with this tool but I think you could accomplish that using a combination of a custom dataset, seeds and slightly altering your prompts as well as photoshop.

I have a resource that explains how to do this in a sort of anime setting where the same character appears through different backgrounds but it's for a different Ai, I think the same principles should apply though. 🤔 (I could be talking out of my butt here I'm not an expert)

[wintry nova]

sent you a friend request 🙂

Btw, do the images fed to train the model have to be the same resolution I choose in the parameters when setting up training? (512x512 / 768x768) or can be any?

[dapper shale]

👍 Accepted!

[analog nebula]

https://app.leonardo.ai/faq

we have a guide on training 🙂

[wintry nova]

It doesn't tell if all images should be of the same size or not though

[analog nebula]

true

I think optimally yes

but I have ignored that

and even not square crop

just threw in a bunch of old paintings

[wintry nova]

ok 🙂

[chilly burrow]

Hi everyone, I was wondering if there's a way to add more than 40 images to a dataset

[coarse wolf]

Hi, say my mail has not invited yet

[wintry nova]

So, my model was trained. Pretty badly (obviously what can you do with just 3 images?). But it seems like when you train a character model, it then can create different characters of the same style. But how to create a model that will be generating exactly the same character in different situations?

[analog nebula]

perhaps image to image feature might work

[wintry nova]

interesting, will try it out!

[raw furnace]

Hi, why i dont recive the email?

it like 1 week ago

[analog nebula]

It's a tad slower though the website. If you make an introduction in #👑│introductions , it will unlock #✨│priority-early-access which is a much faster way to get an invite 🙂

Thank you for tweeting about us!

[void dune]

Just have to say this is amazing, I’ve spent 2 hours making everything that has popped in my head 😂 incredible job 🥳

[analog nebula]

Thank you for the compliments on behalf of the team! ❤️

[wet scaffold]

Upscale feature is great for faces, but it loses detail for everything else... right now i merge the 2 images together to get a ideal result. Is there a way to mitigate this?

Ex.:

[analog nebula]

Not at the moment on Leonardo. I've actually talked to our developers on this and they will be looking to implement an alternative upscaling solution that preserves more details

[royal timber]

Are you able to create sprite sheets? Can you generate a character in different poses or will it change a bit in every generation?

[analog nebula]

possible but not easily done at the moment, though one of our developers has been working on a solution that hopefully will be implemented in the future 🙂

[grim spire]

yeah you just need to put model sheet of...

in the prompt

[shut raven]

guys accept me pleaz

https://tenor.com/view/speechbubble-speech-bubble-please-meme-gif-25693113

[analog nebula]

Will happen soon

Probably within the next day

The team members with access to the mailing list are rather busy lately

[shut raven]

https://tenor.com/view/impressive-very-nice-patrick-bateman-gif-15247511

[analog nebula]

lol

https://tenor.com/view/meme-american-psycho-business-card-gif-25561759

[neat axle]

Anyone uses reddit here? https://www.reddit.com/r/leonardoai/

[river holly]

I've noticed that the more often I apply one of my self-made models over time the redder and redder and redder the result becomes. Eventually I have to download the image, do an automatic color correction on it, and reupload it so it doesn't look like a scene lit by a darkroom lamp.

[spark cosmos]

would love for it to grow, subreddits work great to collect art and knowledge

[neat axle]

https://twitter.com/i/spaces/1OdJrzYWjwyJX

We’re going to be in this Twitter space today

[vestal hinge]

i have to return some video tapes. 🙂

[sacred rose]

Hello

Is the Ai already fuctionnal and can be tested already?

[neat axle]

yeah, you can check in #🖼│image-share some of the pictures the community has created

and you can register in https://leonardo.ai

[sacred rose]

I already registered

, i want to create a youtube video on it

[fast tiger]

Hi, @sacred rose
Welcome to Leonardo. Feel free to make an intro in #👑│introductions, and that way you can unlock the #✨│priority-early-access channel, fill out a form and have a quicker access to Leonardo 🙂

[analog nebula]

^ I'll process any addresses added recently soon and pass it along to my team mates to add to the invitation mailing list before I go off 🙏

[vague silo]

hi how are you? I have a doubt.... in order for me to use Leonardo's art (there are some that I'm in love with, they turned out exactly as I wanted.) do I need to download them in jpg and then put them in a modeler? blender for example...Or can I play directly in the engine and animate?

[analog nebula]

Depends on what you're after, if it's a 2D production, it's much easier to get things done

If it's 3D you would have to model it out yourself and maybe use the results as textures if possible

Much love for the tweet

🙂❤️

[vague silo]

yes it's in 2d anyway

[analog nebula]

Might be possible to just animate in engine if it's 2d only but I'm not familiar enough with pipelines to recommended

[vague silo]

I understand, thank you... I had this doubt because I don't understand much about 2d animation

[analog nebula]

Definitely worth researching into. There are software out there as well that are designed for animating 2D illustrations

[uneven fern]

Love seeing Leonardo.Ai being a part of Kanpai Panda spaces 🤝

[ripe anchor]

https://i.reddituploads.com/c97e410d416e4ed29acb5fb86371b89b?fit=max&h=1536&w=1536&s=a2cdc38f465e560e5218550199225896

[warped sleet]

what should i do to get you added me faster on leonardo.ai?

[fast tiger]

Welcome to the community!

If you want to speed things up, make a little presentation on #👑│introductions and the #✨│priority-early-access channel will be unlocked, fill in the form and you will receive the invitation as soon as possible. 🙂

[fast tiger]

Welcome! Glad you're here

[warped sleet]

ok thanks

[clever oxide]

Do you provide transparent background feature? It's what is missing from Midjourney to make 2D game assets like icons

[fast tiger]

Yes, we have this function 🙂

[clever oxide]

Omg! Can I try

I filled the form

[fast tiger]

Yeah!!

You will have to wait for your invitation, which is a very fast process.

[clever oxide]

Okay

[fast tiger]

To give you an example:

[clever oxide]

Is it just a background remover?

Or the picture is generated directly without background?

[fast tiger]

No, there is the function to remove background. 🙂

[clever oxide]

Oh ... 😞

[calm comet]

one suggestion i would have... please use paypal, or something like that, credit cards are pretty rare by people of europe 😄

[sacred rose]

where can i see examples of 3d object create dby leonardo??

[fast tiger]

For now Leonardo makes 2D assets, the team is working on it. 🙂

[sacred rose]

ok

[fast tiger]

@neat axle when the payment plan is launched, will PayPal be comtenplied? Any thoughts on this?

[steady cobalt]

How do you get white listed 🤔

[fast tiger]

Welcome to the community!

If you want to speed things up, make a little presentation on #👑│introductions and the #✨│priority-early-access channel will be unlocked, fill in the form and you will receive the invitation as soon as possible. 🙂

[steady cobalt]

Well alright thx

[neat axle]

not that i know of, most probably it will be Stripe

[steady cobalt]

Thanks for the help, i really appreciate it. Idk why I had such a hard time figuring all this out 😁

[fast tiger]

@calm comet just in case

[calm comet]

ty

[fast tiger]

A pleasure!

[gusty grove]

Hey

Help me pls 😁

A wanna make picture how?

[fast tiger]

Welcome to the community!

If you want to speed things up, make a little presentation on #👑│introductions and the #✨│priority-early-access channel will be unlocked, fill in the form and you will receive the invitation as soon as possible. 🙂

[hot stone]

i am here to make Playstation 9 games

[gusty grove]

How long i need waiting? for whitelist?

[fast tiger]

I already answered you in #👑│introductions

[jaunty ore]

Does Leonardo.ai learn from prompt and their results VS the action of the user?

[analog nebula]

Not from what I understand at the moment, but @mild walrus would be the person for this question

[mild walrus]

Not sure I understand the question, could you explain a bit more?

[analog nebula]

Might be something like the user "liking" certain results which then creates a feedback loop

[mild walrus]

ah we do not do reinforcement learning

[jaunty ore]

okok thx

[analog nebula]

Any possibility of such an training in the future?

Just wondering myself

[mild walrus]

ehh doesn't seem super useful, more can be conveyed by providing reference images and such rather than having an algorithm try to guess what a user likes

plus reinforcement learning is a big WIP in ml stuff

[analog nebula]

Ah

[jaunty ore]

Ah do you know how scheduler works?

[analog nebula]

I actually would need to know as well

Need to explain to others

[mild walrus]

yes, the short explanation is that it has slight effects on the outcome, particularly schedulers like euler ancesteral are a bit different

some are able to handle less steps better than others

most of the schedulers are "deterministic" though meaning that they should provide more/less same outcome given enough steps

[jaunty ore]

I'm working mainly in 60 counts. Is all scheduler support it?

[mild walrus]

yep!

[jaunty ore]

Nice, if I understand well, scheduler will give me different outcome from the same prompts and options I select? Is this right?

[mild walrus]

yes, although very slight differences

we will allow fixing the seed of the generation soon so you can get a real comparisson

[jaunty ore]

Ok and when it comes to faces. I see often the same face pattern or character coming back generation after generation within a close time laps. Is it normal?

[analog nebula]

Yeah can happen

Could try varying up the prompt and adding more specifics if needed

Or add negative prompts

[jaunty ore]

Do you have a list of prompt working better than others?

[analog nebula]

Not personally but you can study what others have used by going to either community feed and clicking on the images and seeing the prompts they've used

Or going to Finetuned models > Platform models and looking at the examples our team has done for the default models

[teal brook]

gm

[analog nebula]

morning 🙂

[sharp zenith]

Anyone wanna try?

[analog nebula]

Is this a prompt generator?

[sharp zenith]

I made it

Mhmm

[analog nebula]

Amazing

[sharp zenith]

I still gor cleaning up to so with it

[analog nebula]

I won't mind a copy 🙂

[sharp zenith]

Also got to add more prompts to the actual prompt list

But the main thing for me was the other selections xD

[analog nebula]

It's clever

Hopefully we can expand our prompt generator to include such features

[sharp zenith]

Yeah I'm still adding a lot so belive it or not I worked 50/50 with chat gpt

When you download it it'll probably warn you but just run the .exe in the build folder

I need to make a web page to host my file haha, but my main focus was a lot of people miss out on using these types of things

I also want to make it so I can push live updates to the people who have it installed

The last three options are a little wonky I have to move the code to reposition them below prompts but yeah haha

Means a lot to get feedback

Whenever you test it haha I'm hoping you like it as much as the other two testers

[analog nebula]

👍

That's quite a bit of features you got going on

[sharp zenith]

Yeah it was a headache 🤣

[analog nebula]

live updates, never thought you'd include that

[analog nebula]

seems like it 😅

Impressive though!

[sharp zenith]

Yeah live updates because I want to share my creativity lol so each section besides ABC which was going to day quality but python had a fit, has about 25-100 options

The last three I added tonight have 100 options each

[analog nebula]

hmm speaking of python

it would be cool if once the API for Leo is out, we have a blender addon

that generates textures in app

well not in-app but just faster than pulling up an browser window and then downloading and applying textures

[sharp zenith]

So that's interesting and I agree it would be awesome , chat gpt told me to learn more before it would help me generate 3d renders of the images 🤣🤣

[analog nebula]

chatgpt seems a bit witty

[sharp zenith]

Yeah lmao, so I seen this post where someone convinced it that it stole their wallet, lied about, used their credit card, agreed to pay interest and bought drugs 🤣🤣🤣🤣

[analog nebula]

Time for some research

[sharp zenith]

The guys title was "when chat gpt becomes sentient I'm dead first"

Yeah so I guess you can command it into scwnario by being specific with you want from it then proceeding eith your questions

This prompt generator is terrifying because of how much it writes

[analog nebula]

pixar's renderman

lol

typical

pixar style so it makes sense

I'd use redshift though

[sharp zenith]

If I'm being honest if that guys gpt script makes prompts all that large its an abominable

Abomination

[analog nebula]

there is something about being too specific yeah

can you imagine art directors going up to their art team

and then telling them a bunch of specifications (about 50)

😬

[sharp zenith]

🤢🤮

[analog nebula]

I wonder how limiting on deviations it would be

the more lengthy it gets

[sharp zenith]

My guess is after the first 25 words the rest is noise

[analog nebula]

essentially I believe yeah

less is more

[sharp zenith]

Yes yes it is

[analog nebula]

overused but it works

so I say it all the time 😂

[sharp zenith]

I say it all the time too

[analog nebula]

We're on the same page then

[sharp zenith]

Oh yeah

[analog nebula]

✋

[sharp zenith]

✋️

Wow mines tiny

🤣

[analog nebula]

https://tenor.com/view/tales-of-zestiria-bros-fist-bump-gif-13804889

random gif go

lol

[analog nebula]

that's actually interesting

[analog nebula]

I wonder if it has to do with specific emoji keyboard

[sharp zenith]

This pictures not insane but I really enjoy it

[sharp zenith]

Lmfai

Maybe I think it prolly is

So do you do coding?

[analog nebula]

Actually no, I do art 🙂

[sharp zenith]

Oh that's right you told me that

And then showed those awesome pics from your personal data set xD

[analog nebula]

oh hey it actually works

[analog nebula]

yeah they're kind of to my taste 😄

I like the painterly stuff more

it feels more human in a way

I actually really like this one

[sharp zenith]

Yeah I agree

[analog nebula]

the way the background directs into the figure

[sharp zenith]

The texturing is epic it looks like you could run your finger over it and feel the image

[analog nebula]

yeah I love that textual quality of paints

its why I hardly use the current upscaling feature

it removes that too much for my liking

[sharp zenith]

I always forget it has the upscaling

They just come out so good

[analog nebula]

yeah its great when it works

just not for most of my results 😅

something about the originals has that charm

this is the upscale

[sharp zenith]

Idk most ai generators lack depth in their images but like Leo is mangeficto

[analog nebula]

oh yes, really nice job, but I really love the ability for custom models

[sharp zenith]

I still Minda like it

Kinda

[analog nebula]

oh it is great

just not to my taste in a sense 😄

[sharp zenith]

Haha

The custom models are awesome I've trained a few data sets xD

[analog nebula]

also the more details makes that insanely long fretboard look weird

since you can see how many frets it has

[sharp zenith]

Oh I zoomed in to look and like I like how the strings are waving

[analog nebula]

yeah

its funny though how AI doesn't have any idea how much frets to put

[sharp zenith]

Like he's in the middle of playing

Guess tou should train it on guitars haha

[analog nebula]

perhaps, though this was just my portraiture model so it definitely doesn't have enough guitar data

guitar model for designing guitars 😄

[sharp zenith]

So that's the biggest thing for me once the training cap gets increased is the fact that I wanna teach it everything haha

[analog nebula]

Yeah I wonder how much is too much, we'll find out I suppose

[sharp zenith]

Wrll the Leo fine tuned or signature has like 20k and ot works great

Well*

[analog nebula]

Guess I'll make an monstrosity trained on all the old painters I like

[sharp zenith]

Oh my bad

40k+

Hahha

[analog nebula]

How many art styles do you want to emulate?
Yes

[sharp zenith]

I'm not entirely sure

Probably something g close to this

I love the depth of this image

[analog nebula]

Oh that one is actually nice

in a graphical way

[sharp zenith]

I gave it motion with leia.pix

But even the still image carries the same depth

[analog nebula]

Speaking of image animation

I would so love if we can integrate some sort of 2D animation solution

its quite a thing over in Japan

to animate 2D illustrations

[sharp zenith]

Oh man

I'd love that

[analog nebula]

https://www.live2d.com/

[sharp zenith]

No crap chat gpt can code 2d animation

[analog nebula]

yeah there are software like live2d

[sharp zenith]

Thays cool as hwck

[analog nebula]

basically you can animate an single illustration without having to draw too much

[sharp zenith]

Thats awesome

I've been looking for that forever

Honestly

[analog nebula]

Yeah, I wonder if it can be automated more

currently its a lot of keyframing and manual tweaking

it's a rather complex piece of software though

at least to me

[unique rivet]

ChatGPT can code complete softwares, but all it requires is rechecking as it sometimes can make mistakes and you will have a hard time searching all the code. Also, it can comment the code if you ask for it.

[fluid viper]

I'm really getting desperate here... can the AI really not understand "glowing golden eyes" and "evil smirk"?

[analog nebula]

Yeah I've some issues with specifics like that

@mild walrus if you've the time, any advice on this?

[fluid viper]

I've got issues with every specific.

The only time I got anything useful out of this was when I had really, really low expectations.

[analog nebula]

An occasional bug we are aware of

For now just ignore it

Should be able to do other generate tasks regardless

[barren hollow]

[shut raven]

Yo im in letsgoooo

[analog nebula]

Congrats 😄

[shut raven]

https://tenor.com/view/cristiano-ronaldo-soccer-footy-football-futbol-gif-22763381

best way to find out is to just practice with all of em?

[analog nebula]

try em all

and explore other models too

also you can study the prompts of others

either by browsing the models examples (when you click on a model in Finetuned Models) or in Community Feed when you click on a image

[little glacier]

Just wanting to let 3d modelers know: you can give your model just a color texture, make a screenshot from an isometric camera angle, upload to leonardo, generate a nice texture, like so:

then, use the image, UV unwrap at with Project from View (blender):

Then rotate it 180 and do the same thing

[analog nebula]

brilliant 😄

[little glacier]

😄 ty

leonardo is just amazing

[little glacier]

It completely simplified creating my mod. And more fun too

Other ai generators never came close to what leonardo can do

[balmy bolt]

Hello people, i’m new to all this and I want to learn to design :)

[analog nebula]

Welcome to our community! If you have yet to receive an invitation, do check out #👑│introductions , making an introduction in there will unlock #✨│priority-early-access , which allows you to get an invitation much faster than though our website 🙂

[river holly]

I've noticed that whenever I have the model work on people it seems to have a lot of artists' renditions of them and tends to turn them into paintings. While if I have it work on some form of exotic animal it seems to go by National Geographic type material and leaves them photographic looking. So then I try to combine the two and one looks like a painting and the other looks like a photo...

[analog nebula]

That's definitely a bit of a complicated situation

maybe image to image and run it though an stylized pass

(by adding prompt keywords like paintery / a painting etc)

[hot stone]

yay im in. Playing around with it right now. There are impressive feature sets!!

Hope we can do some photo compositing with this thing. Was trying to do that with MJ and SD but couldnt get it right. Seems like this thing can help me with Unreal5Engine assets ?

[balmy bolt]

Thanks for the help Ayu, I filled in the form. Fingers crossed!

[analog nebula]

Not 100%, especially if it's 3D

Still some manual work required depending on your desired production type

[analog nebula]

You're very much welcome!

[hot stone]

i have a lot of cool ideas. I do a lot of photogrammetry in unreal5 also. Was doing alot of camera match perspective for photos to do as prints

and building out a world with it as well. So i have alot i want to do. WOuld love to see where the future for this is.

[analog nebula]

you could throw an screenshot of your unreal scene blockout

[hot stone]

so far im very impressed by the interface

[analog nebula]

and see if Leonardo can help with concepting

image to image but on a low init value

[hot stone]

you mean image2image right

yeah

[analog nebula]

yeah

We're on the same page 😄

@hot stone apologies

just trying to test things out for the team 😅

[hot stone]

lol no worries

[shy beacon]

Great project

[analog nebula]

Thank you on behalf of the team 🙂

[barren hollow]

Ayu you're too nice

[analog nebula]

Yes I've been told a lot 😅

[bronze crag]

Question: anyway i can train a model with multiple images, with each image getting its own prompt

[blazing spire]

Stayed up all night playing with Leonardo. Did some learning. BUT - it seems impossible to get the models to generate a horse with four eyes. More specifically, a portrait of a horse in which there is an extra eye close to the normal one. ... and why does it generate Adolf Hitler with a broad, normal moustache? I mean, a square moustache is even called a "hitler-moustache"? Takes the fun out of Hitler riding a unicorn waving a pride flag.

[analog nebula]

Not possible I believe. @mild walrus any additional input. Apologies again for the ping. 😅

[mild walrus]

not at the moment, its a harder thing to set up but can be done i believe

potentially on the todo list

[analog nebula]

Most likely because the base SD model has been trained on actual horses. Also specific prompts sometimes do not work as intended

[analog nebula]

Thank you for the quick reply Ethan! ❤️

Ah

I thought you used the metal gesture

🤘 this is better 🙂

[raw token]

Is there a fine tuned model for avatar generation?

[analog nebula]

Not at the moment, at least for our official platform models

possible to create an model for it however

depending on the type of outcome you're after, just add a bunch of examples (preferably of similar style) to a dataset and train an model

[topaz siren]

Any feedback channels for disscussion/techniques/examples, do enjoy taking notes =]

Image-share channel ok for such?

[analog nebula]

#🖼│image-share , #1053969672699129856

🙂

Assigned you the artist role

Though artist-x-ai is more of a forum where artists start their own individual threads

We really ought to create an community tips forum

[halcyon ore]

Ooh! I want to try

[analog nebula]

Okay added #1068246518412361808

if anyone would love to post any in an more organized space, please feel free to do so 🙂

[topaz siren]

Oh excellent! Looking forward to sharing my knowhows, creation of tips, and helpful info as exploration into the Leo origin times. Great!

[pliant gust]

Excited about this!

[analog nebula]

Welcome to the server 🙂

[stable solstice]

not sure in which treat to post this question, so I can't download upscaled transparent image right? will this option be available in near future?

[analog nebula]

Would have to redirect this question to @mild walrus , hopefully he can give an reply to it 🙂

[pliant gust]

Just put this project on the radar for a bunch of gaming projects! It looks soooooooo awesome

[analog nebula]

Thank you so much! If you'd like, making an intro in #👑│introductions and then going though #✨│priority-early-access when it's unlocked gets you an invitation much faster if you haven't received one 😄

Just an thank you for joining our community and taking the time to do an intro essentially

[pliant gust]

@analog nebula Perfect! I just submitted a msg to the website about possibly getting an AMA / Chat going with you guys!

[analog nebula]

Oh actually, let me open an private chat between you and our COO Jachin 🙂

[pliant gust]

Perfect that would be great!

[halcyon ore]

This is super clever. Nice work!

[analog nebula]

Oh by the way about this, if you ever would like to write an short guide in an more organized space for this workflow, I've created #1068246518412361808 for such things 🙂

[hasty jay]

👀 I filled it out

[analog nebula]

Should be sent within several hours to 2-3 days 🙂

Sometimes there's a bit of a bottleneck, hoping to get the turnaround time shortened in the close future

[hasty jay]

awesome 🙂 really looking forward to trying this out, I have some others I have invited already

[hasty jay]

Im just excited to be here

[analog nebula]

Thank you so much for sharing 😄

[jaunty sphinx]

Hi there, is there a guide for using starting images? I cannot get my image to apply affects or be modified. Eg turn into stained glass, XYZ made out of ABC

also finding that Leonardo struggles with abstract concepts; I can add a comparison to this blog entry re dalle and midjoruney because the results are so far off

https://saipien.org/dalle2-vs-midjourney-art-prompts/

[analog nebula]

@mild walrus might be the best person to answer your questions if needed 🙂

he works mainly on our AI side of things

[jaunty sphinx]

please and thanks

[mild walrus]

there it a setting on the side bar called "init strength"

[jaunty sphinx]

https://saipien.org/leonardoai/ having a difficult time completing this

i tried various levels of that

not new to prompting

[mild walrus]

reducing the value will allow the image to change more

[jaunty sphinx]

at lower settings i does become stained glass but completely modifies the subject

so either there is a very precise init weight or some prompt phrasing

[mild walrus]

i've been using it to morph some of my images from older ai programs

[scenic zinc]

Hi is there a way for me to upload a selfie and get an ai image of my face.

[fast tiger]

Of course, you upload your photo where I mark you in the image, and a selfie of you will be generated.

[scenic zinc]

Thanks, tried that and gave me 4 versions some of which were male lmao

[analog nebula]

In addition: experiment with the init slider , higher values means the AI will tend to preserve the source image more

also use negative prompts and see if you can negate some of the unwanted results

[scenic zinc]

Ok cool thanks

[analog nebula]

No worries! Hopefully you'll eventually get something you like 🙂

[fast tiger]

This is very important to take into account. 🙂

[scenic zinc]

Thanks hopefully 😁

[vestal hinge]

Welp. I dumped a bunch of sprite sheets in to train. Maybe it'll do some stuff I want.

[urban wasp]

Hi, i'm hoping someone can help me.
What prompt would I use to remake this graphic, so instead of X it would say v4? If that's possible

[fast tiger]

mmm... @mild walrus is it possible? 🙏

[mild walrus]

unlikely with current models

well,

if you can draw a template for v4 though it could stylize it similarly maybe

but from scratch not so much

[fast tiger]

Oh thank you, @mild walrus for your quick response. 🙂

@urban wasp just in case, that's the way to do it.
What Ethan wrote 👆

[urban wasp]

Gotcha. Thanks guys

[mild walrus]

from prompting its not so easy just as models are trained on general concepts/ideas rather than exact text, but we'll have a canvas feature soon where you can draw a sketch of what you want and feed it through, i imagine theres some potential there!

[urban wasp]

that sounds amazing

[barren hollow]

If you could also draw something over an existing image that would be cool

[mild walrus]

i believe that will be there too

[twin adder]

how do I generate 3d models

or is this only for images?

[fast tiger]

For now they are 2D assets, the team is working on it. 🙂

[twin adder]

oh ok

thanks

[fast tiger]

A pleasure 🙂

[vagrant wolf]

Hey guys can I ask for some assistance I see some of the frankly amazing renderings but I can't seem to word my prompt thoroughly enough to get that level of badassery any assistance would be most appreciated

[fast tiger]

Hi, maybe this will help you:
https://app.leonardo.ai/faq

And the #1068246518412361808 channel has also been launched for users to post tips or guides on how to use Leonardo and get the most out of it.

[shadow lantern]

Hey everyone!!!!

[vagrant wolf]

Well thanks a million

[tribal cloak]

Anyone else get so caught up in prompting they forget to check in on Discord once in a while? 😂

[hot dome]

ai is very cool

sugoi

[rain shuttle]

Hello 🙂

[fast tiger]

If you would like to get an invitation to the web application much faster, just drop by #👑│introductions, that will unlock #✨│priority-early-access, where you can apply for an faster invitation 🙂

[sharp zenith]

@analog nebula I did a little upgrade to my generator

[analog nebula]

Very nice! 😄

[sharp zenith]

😊

[analog nebula]

Just saw this, best advice I can give is going though the community feed and clicking on the images you like and seeing what prompts they have used, copy that and play around with them. Delete and omit stuff, append etc.

Also check out our platform examples by going to Finetuned Models > Platform models > clicking on the models and looking at the images displayed in the initial popup, the first bunch is generated by our team 🙂

Example for the model thing

[sharp zenith]

So my prompt tool has over a Trillion possibilities. Kinda hurts my brain

[analog nebula]

that is... concerning 😔

[sharp zenith]

Idk I mean it's pretty much endless usage I guess but I'd rather it add value.

[analog nebula]

good intentions 😄

[sharp zenith]

I haven't even used it myself 😭 I keep getting caught up in what my brain says I have to fix cause I'm paranoid about putting a crap product out

[analog nebula]

QA testing required 😅

throw a bunch of prompts into Leo and see how it fairs

[sharp zenith]

I'm definitely going to.

I honestly can't wait to use it xD now I'm excited

[analog nebula]

Do it! Got to test what comes out of it 😄

It's one thing to get a bunch of prompts and another to see it all visualized by AI

[sharp zenith]

Giving me goosebumps hahaha

Alright I suppose I'll make like a collage of my favorite ones to share.

I despise random discord scammers

Its like go crawl back in the hole you came from

[analog nebula]

yeah I've disabled dms for a lot of servers

it's just annoying to deal with especially since I am managing an community for an startup 😅

[barren hollow]

Do you get paid?

[analog nebula]

Yes. I am the actual community manager

I wasn't joking around

I do backend administrative work as well so it's not just discord but spreadsheets 🙂

[barren hollow]

[analog nebula]

That does mean I am not as qualified to answer some technical related questions though. I would love to learn more but I definitely am not an machine learning expert 😅

[sharp zenith]

That's still pretty cool

How did you end up linking with leo

And this is just a different kind of wtf from discord

[analog nebula]

I'm sorry I was thinking someone decided to run chatgpt

on their account

[sharp zenith]

I have no idea it's weird

Anyways

[analog nebula]

Connections, I was already working with the founder of Leonardo beforehand 🙂

He liked how I managed stuff community wise so here we are

[sharp zenith]

Oh that's pretty sweet

[barren hollow]

Is ethan the founder?

[analog nebula]

No, but he's our resident machine learning guru

[sharp zenith]

I just noticed I got a title for Leo I feel special 🙃

[analog nebula]

mindspheres is the founder but he isn't around at the moment till next Monday I believe

[analog nebula]

because you've been working on a tool that seems really nifty 🙂

[sharp zenith]

🙂

[analog nebula]

I'm still not very good at figuring who to give it to, and someone else has been doing it 😅

So I really do not know the actual criteria, also I do believe that the choices will be biased...

(feels horrid so I hope we can do it by vote amongst the early contributors)

[sharp zenith]

:o! Voting?

[analog nebula]

Yeah but informally

I just feel taking a slightly more democratic approach would be better than relying on my own personal judgement

[sharp zenith]

Yeah I can respect that,

It involves the community too so definitely makes for good rapport

[analog nebula]

By the way I really ought to have my lunch

I have the bad habit of jumping right into my work roles right after waking up 😂

[sharp zenith]

I have a bad habit of getting sucked into doing something and sleeping 4 hours before work 🤣

[analog nebula]

This definitely, we aim to build an community that is of an collaborative space, so getting people involved in some (administrative) decisions is definitely helpful for that 🙂

[sharp zenith]

I love that I really do.

[analog nebula]

By the way you now have a smaller group to share your tool if you'd like first 🙂

with that other channel

Though I will of course be taking the responsibility of scanning the files (I'm paranoid like that but it is my job after all)

[sharp zenith]

Oh its cool

I appreciate that haha

I dmed it to you already anywas hahaha

Amen for fake names on the internet, I feel like that random dm is trying to curse me

[elder current]

What do you think is the best generation model for tileable textures?

[analog nebula]

Best bet would be to train your own with an dataset of the type of textures you prefer, that way you have greater control 🙂

We do have an community made texture generator but I haven't tested it personally

Shoutout to @hard galleon

Apologies for the ping (other Lou)!

[elder current]

ohhh seems cool, I will try it! Thanks!

[analog nebula]

A pleasure! Hopefully you'll get what you need out of it 😄

[untold fern]

Does this app create game ready 3D assets?

[analog nebula]

Not at the moment. 3D is something we do plan to expand into when the tech becomes more viable in the future.

[untold fern]

A lot of your promotional materials seem to imply 3D art assets - such as saying
"Create stunning game assets with AI." With pictures of 3D models behind the writing.

Also "Create models of various items from your existing work or use art from mood boards or references." Sounds like it means 3D models, but it isn't about that. Some could find it a bit misleading. perhaps being specific about it being concept art or technical art or something might be clearer?

[analog nebula]

Yeah it is understandable considering the results do look very much like 3D models (they're actually just 2D images). I'll have to pass this feedback to the team as I'm not part of the promotional side of things 🙂

[sharp zenith]

You'll love it though once you try it 😁 it's a innovative take on Ai very awesome

[midnight cloud]

Wow Leonardo came up on my Facebook ads, got super hype seeing it!

[analog nebula]

Send pics

[timid glacier]

i'm intrigue 🙂

until then i came across this. https://openart.ai/promptbook might be a good resource for some.

[midnight cloud]

And now I can't find it lol if it pops up again, which I'm sure it will, I'll post.

[analog nebula]

Time to join as an Leonardo user

😎

Very useful site by the way, thank you for sharing it 🙂

Considering the current models are built on top of SD, this will be very helpful

[timid glacier]

yea, there is a lot of good info especially on writing the prompts.

i would've never believed on how important wording in a particular order would be. I liked there example the a pink ice cream tuck with a mounted machine gun may not properly convey the image but a machine gun mounted on a pink ice cream truck would be better. so crazy.

[analog nebula]

🤔

I need to generate this

[timid glacier]

haha

[analog nebula]

whoops

Forgot to clear my image to image attachment

[analog nebula]

always clear your image to image before generating

mental note

[analog nebula]

nice

[timid glacier]

darn. guess i need to outpaint 🙂

[analog nebula]

this is pretty cool

but not a gun

[timid glacier]

haha looks like a 50 cal lol.

[analog nebula]

lol

this is so gangster

ice cream truck rolling up the hood

[timid glacier]

hah

[analog nebula]

still the best result so far

[analog nebula]

really loving the look of this one though

[timid glacier]

yea, better have correct change or else

[timid glacier]

if you can have it on a ice cream truck then why not on a baby stroller

[analog nebula]

probably should have instagram ones

Those reactions though 😦

[analog nebula]

perfect

[pulsar musk]

[analog nebula]

Welcome 😄

[pulsar musk]

Thanks!

[sharp zenith]

I rather much like vangph compared to tintoretto

Vangoh*

[timid glacier]

very intetesting 🙂

[sharp zenith]

I love leonardos freedom

Hands down the absolute best thing about it only second to the data training

[fluid viper]

Why am I getting "error invalid height" when I try to do image to image?

Does it need to be some very specific height?

The question mark tooltip is not very helpful.

[analog nebula]

@mild walrus any ideas?

[fluid viper]

I figured it out.

Uploading a picture changed the sliders up above from regular image sizes.

Don't know why, but it did.

[analog nebula]

Ah

[fluid viper]

Once I pressed 512x512 again, it worked.

[analog nebula]

Well that should probably be fixed

[fluid viper]

Okay, maybe I'm doing this wrong. Let's say I want to make a VTT token for a nightstalker from Dougal Dixon's Man After Man. I have a model trained on my other VTT tokens that I made individually. But when I input the prompt for "nightstalker (from Dougal Dixon's)* Man After Man" all I get is gibberish. So... am I doing something wrong?

• sometimes I input that part, sometimes not.

[analog nebula]

what part of it is gibberish?

[fluid viper]

All the parts.

It seems to have gotten the idea it's supposed to be seen from above but that's it.

[analog nebula]

I was reading this by the way https://openart.ai/promptbook

not too sure if it helps

[halcyon ore]

I'm learning that getting the prompt right is SO key to the success of the ai. I'm refining my approach each time I try. The Prompt Suggestions helps, for sure. Now I'll need to read that post from @analog nebula

[analog nebula]

its tricky yeah, I got to re-read the guide myself

[analog nebula]

though I would argue that it really depends on what is defined as success, it's more about meeting the user's expectations. Typically works fine if they don't have much expectations or just have a general idea, but if they have something specific, getting the prompt right is really essential

[halcyon ore]

That's right. I've been doing a lot with specific characters from books or from RPGs that already are defined a certain way, so I'm looking to capture the vision I already have in my head. For general ideas, the prompts matter, but not as much and you get amazing results

[fluid viper]

Upon reviewing other models, it just doesn't understand what the nightstalker is and grabs any nightstalker, regardless of putting the book title in the prompt.

[analog nebula]

Checked with MJ

same issue

And yes I looked up how it's supposed to look like

[pulsar musk]

welcome yall

[tranquil crypt]

anyone making 3D art?

[halcyon ore]

Anyone found a way to use a character you generate and put them into another scene? I'd love to preserve my character but place them in various locations, etc. I'm assuming the AI doesn't do that... yet...

[fast tiger]

As far as I know, it is not yet possible.
It is possible that @mild walrus has an idea of what he wants to achieve. 🙂

[halcyon ore]

Thanks for the reply - that would really take this to the next level for my purposes. For example, I've generated some of my RPG characters, and placing them in specific environment that I create would be really neat. If it's not a suggestion, I can add one

[fast tiger]

Yes, of course, it would be amazing, the change of scenery. If Ethan knows anything about it, it would be good to wait for it. If not, go ahead and post it as a suggestion. 🙂

I think we would all love this option.

[mild walrus]

It is possible, we’ll have a thread on it soon

[halcyon ore]

Oooh! Looking forward to this - thank you!

I just finished creating the suggestion when you posted

[fast tiger]

Oh great, thanks for the prompt reply Ethan 🙂

[untold fern]

I did enjoy the interface. It kind of felt like a UI built for gamers and game devs, which is on point! It's also innovated on the simple text to image design by letting you train your own models and more easily set your conditions. But all of this is doable in code if you download the stable diffusion or dream diffusion repository (which I assume this is based on?). So essentially it is just a convenient UI wrapper on top of open source code. I could be wrong though

[mild walrus]

That’s just our alpha, just making sure we’re hitting all the basics first, but we have some other things like image prompts and 3D texturing around the corner

[untold fern]

That's awesome! I wonder how far away real time generation of game assets is (as in, you are playing the game and assets are being generated in the game world)

[mild walrus]

i am excited but also my gpu is crying at the thought 🥲

[heavy swallow]

In quantum computers we trust 🫡

[sharp zenith]

Is it possible the umage is larger than 1024x1024?

[fluid viper]

It was smaller than 512x512 in fact. Is the uploaded image size supposed to change the output size?

[sharp zenith]

Maybe it's too small? Try resizing it eith a ai resizer?

Lmk if it works if you try it

I know when I was building my gui that wad an issue

Nope it's not the dimensions

I resized an image to 500 360 and 100 it still worked

Try a different browser

[fluid viper]

I just manually changed the output size back to normal and it worked.

And by worked I mean, the AI proceeded to create images.

[dapper shale]

I've had some success in getting the same character to appear with different poses and fairly consistent design on different abstract backgrounds through the use of magic prompt.

The work flow was a little convoluted but it evolves generating the specific character over iterations to the point of consistently getting that result, turning on img2img, magic prompt and slightly tweaking the prompt every so often from there.

Hair pulled back, helmet and no helmet. Is this kind of what you're after?

[hot stone]

you guys seen GP3 integration for stableDiffusion? @analog nebula Looks amazing. Its what I was trying to do. Hope you guys can pop that in this app 😉

[drifting yarrow]

How does it work?

[fast tiger]

You must wait for your invitation to access Leonardo's web application.

At the moment there is no bot on Discord, it's just an app, eventually it will also be on mobile, first iOS and then Android.

[drifting yarrow]

Oh. Via email? I think I got that. I was asking @deft roost about what he said about gpt

[fast tiger]

Ah ok, hahaha. Sorry I got ahead of myself...

[drifting yarrow]

No worries, thanks for the info

[fast tiger]

My pleasure...
Quick instinct of a common question 😅

[drifting yarrow]

😂

[analog nebula]

Might be something we'd likely add but I'm not too sure as I'm on the community management side of things. 🙂

I'll ask in the team chat because I'm curious anyway

[gusty grove]

Good evening, unfortunately I still haven't received anything by email!

[analog nebula]

Check your spam inbox

[gusty grove]

Yeah i check :/

[analog nebula]

Seems like we might have had mailing issues. I'll dm you an text copy instead 🙂

[gusty grove]

Okey thank u

[analog nebula]

No worries 👍

[analog nebula]

DM sent by the way, check your message requests if it's not showing up

[topaz siren]

Oh my, I really enjoy this direction ark, and colors ;] Good show!

[analog nebula]

Oh I scrolled pass these too quickly but they're so pretty to look at

Honestly needs to be featured on our twitter

[analog nebula]

@neat axle 🙇‍♀️

[dapper shale]

I'm hoping to develop it into an actual game, digital card game or physical, but that's a bit out of reach at the moment. If anyone has any experience in this space I'd love to talk to you! But thanks so much Areleous and Ayu, I'm absolutely stoked with the direction it's going and the consistent style results.

I'm hoping to replicate it again with a different style then I'll know the method for sure and can let everyone else know how to achieve it with their own works!

[analog nebula]

🙂❤️

[queen plaza]

wheres this from?

[analog nebula]

I don't think she knows either

Arkose mentioned trying to Google reverse search but it brings up nothing

Which is really odd but nice that it's not traceable in that sense

[analog nebula]

By the way were you talking about prompt generation? Because we do already have an gpt prompt generator right in the AI Image Generation section

[dapper shale]

For that specific example, for instance It brings up similar elements like anime characters with headphones of which there are many, and similar colours but nothing like that I can see was a direct call back to if that makes sense 🤔

[neat axle]

is it ok if we feature it in our Twitter?

[dapper shale]

Sure! It would be an honour!

[analog nebula]

Much thanks Midas!

[hot stone]

you're a wesome! 🙂

[hot stone]

I was talking aboutg img2img using GPT. But I'm not sure what you're referring to

are you saying we can already prompt in GPT3 style?

on you're right, it does by asking questions

was that what you are talking about ?

[analog nebula]

Ah, @mild walrus might know

[tired summit]

@analog nebula hello 🙂

[gusty grove]

This is my first ai pic, I just wanna ask how change style for horror?

[analog nebula]

Hey there

[analog nebula]

You'd have to figure out the right prompts to use essentially

[tired summit]

i need To ReCreate this photo in leonardo app , its digital art paint " the Magnificent seven " but in Egyptian style , its famous movie and i'd like to re-create the characters again to use it in a Stealth mode game like " Commandos Games " but ehen i tried it here it gives me bad results , also when i put beards in negative prompt and type shaved in prompt it still gives me unshaved persons

[analog nebula]

For one, daylight like that is not horror like

[analog nebula]

Have you got a bunch of images of this particular artstyle?

Or are you going after another style

[tired summit]

here is my prompt
A 4K full cinematic epic scene of a modern Charismatic Egyptian Heroes standing , middle eastern races, ultra high details, desert background, epic cinematic lighting, 35 years old, His light skin face features are serious and strict , shaved beard, mercenaries, face visible, khaki green outfit, tattered layered clothing, holding Combat Knife , ammo pouches across the chest, action pose

[analog nebula]

Also try image to image

[tired summit]

did i missed something or missed up with something ?

i'd like to have it like high resolution 3d characters

[analog nebula]

Yeah you have a lot of specific terms, but it won't give you the exact characters since it's just a general description of the group and you are generating an group of characters .

Again, image to image may be worth an try.

Its on the left sidebar, the bottom section

But you may already have tried it

[tired summit]

i need same style but in 3d , in Egyptian look style " as the digital pint " above

[tired summit]

i already tried img2img since the normal mode didn't help me much

[analog nebula]

So this

But in the style of the image above?

[analog nebula]

Which is this sample image?

[analog nebula]

Right so an photo to this sort of style

Am I correct?

[tired summit]

this photo is for the Egyptian characters which i want to transfer it to 3d high resolution characters

check this please https://www.behance.net/gallery/20307467/Shams-El-Zanati

[analog nebula]

So is this the 3d high resolution style you are referring to? Apologies for the confusion

[tired summit]

yes

[analog nebula]

Honestly it does seem like a pretty demanding task for any AI image generation service including midjourney, considering it is a group of characters with specific characteristics. And considering we are converting an stylized image to something extremely photo real (the sample you have provided is mostly very likely not 3D)