pan-pushkin

[pan-pushkin]

When logo designers will lose their jobs

Quick answer: not so soon.

I've tried to create a nice logo for our Czech LGBT community. I don't have much experience still. And my AI started to give me a hard time. Apart from that, I was experimenting with some prompts for at least creating an idea of a logo. Well. Let's see how it works.

a colorful and whimsical logo for the LGBT community, a tilted rainbow umbrella, symbolizing diversity and unity. include the word 'Vmeste' in an elegant font, white background

a colorful and whimsical logo for the LGBT community, a tilted rainbow umbrella, symbolizing diversity and unity. include the word 'Vmeste' in an elegant font, white background

a colorful and whimsical logo for the LGBT community, a tilted stretched rainbow umbrella, symbolizing diversity and unity. include the word 'Vmeste' in an elegant font, white background, a minimalistic, abstract graphic

*AI broked several times in the process. Spend around an hour to make it work again*

a colorful logo for the LGBT community, a tilted stretched rainbow umbrella, symbolizing diversity and unity, include letter 'V' and letter 'M' in an elegant font, white background, a minimalistic, abstract graphic, clean composition

a colorful simple linear logo design for the LGBT community, a tilted stretched rainbow umbrella, symbolizing diversity and unity, white background, a minimalistic, abstract graphic, clean composition

And now it is totally broken. I'm frustrated. And you say that AI art is simple and requires no effort. For a start, try to do smth yourself and then tell me about simplicity

[pan-pushkin]

Amazing candlesticks! No AI involved.

Сandlesticks 🕯️

[pan-pushkin]

I'm up to no good prompts

A good prompt is an art they told. Well, that is true.

Here, we observe the outcome of my sloppy prompt on the left and the result of a normal prompt written by a normal person.

But what they almost failed to mention is that you should not only have a good prompt but also a reliable checkpoint model. Otherwise, despite all the detailed writing, you will end up with images like this.

What I didn't understand at that time was why, for Merlin's sake, the official base model, which had been enriched with tons of images, worked like this. And so began my journey to install an actual good model and an extension for correcting faces and hands.

I sincerely believed that working with image-generating AI meant loading photos, images, and actually generating something. However, I ended up scrolling through and correcting miles of Python code just to persuade my AI to function properly: loading models into the right folders, installing correct versions on the virtual environment (venv), correcting rights and access to extensions. And given my limited knowledge of Python and virtual environments, I couldn't help but wonder how people without such knowledge at all manage Stable Diffusion. Or is it just me who can damage everything just by looking at it?

Anyway, after hours of a coding nightmare, I finally achieved some truly magical images of my best boys.

And some unexpected result where I asked for magic David Tennant and Michael Sheen and AI happily gave me this 😁

I couldn't persuade it to generate two separate, distinct persons, so I gave up and asked it to create a magically blended mix. This is what a son of David and Michael could look like. Pretty awesome, isn't it?

And of course, I attempted to generate images of Daniel Brühl and Luke Evans, but I suppose this model has a rather vague representation of them.

So I have my next goal: create the best model that would know the best people

[pan-pushkin]

Stable Diffusion for Idiots (me)

I'm no good at drawing. Trying this thing for over a decade but I've never had enough patience to perfect the skill. So. I can draw sloppy portraits and still somehow I want to create a comic book.

I have an idea, a plot, characters, all pack, except the crucial part where I can draw. It would be a very tragic story but I thought wow, wait a minute. Thank Merlin, we have AI now! That can do things for us.

So when I dropped the idea of actually drawing myself I started to explore Stable Diffusion. It turned out that that shitty stubborn thing understands concept art even worse than me. But I really hope it's just a start and we will find the right level of connection and understanding.

Currently, I barely understand how to launch SD, what models are and how to install them, and how to write simple prompts. You have no idea how long I read all the code and tried to grasp it. So when I got my first worst image in the world I was freakingly happy.

Here it is. Just look how perfectly dreadful these men are but they are holding hands 🥰

Today I learned useful settings, how to get more accurate images, how to fix faces merlin this part was killing me. But look! This young man already looks like a human being after just one try!

And so I continued to prompt something more interesting just for the sake of experimenting with the model. That's how it sees da Vinci portraits of a blond-haired man.

This was a pretty one. We also have something that looks like Igor Nikolaev by da Vinci and it wasn't intentional I swear 😂 (on the right is actual Igor Nikolaev's photo)

The last funny beat for today, I was trying to do something with Daniel Bruhl without any understanding of what I was doing (I had 3 minutes till the end of my VM session so I had to work quickly. Do not repeat this at home, kids 😆)

Thank you for reading this for some reason.

I'm really excited to continue my journey with Stable Diffusion and I have huge hopes on what I can eventually get with some amount of determination. And so here will be The History of this journey.

Wish me luck and inspiration 🤞🙏

[pan-pushkin]

I just leave it here

[pan-pushkin]

And that's how an incredibly complicated universe started in our minds.

Luke Evans and Daniel César Martín Brühl González Domingo look perfectly splendid together and amazingly inspiring