🖼️ I made an avatar with Stable Diffusion 🤓

What to do when you have an illustration idea and no artistic skill at all : use Stable Diffusion.

The idea 💡

Facts

• Every collaborative tool legitimately ask its users for an avatar
• I don't really want to see my face everywhere all day long
• I have a couple of ideas for an illustration but no artistic skills whatsoever

Fortunately

• Stable Diffusion exists!
• There's a github project : https://github.com/cmdr2/stable-diffusion-ui that makes it very easy to use.

The image

There's a very little known animal in the south of France: a genet. They live at night, nobody sees them and I find them very cool. So I wanted an anthropomorphized version of it.

Tips & tricks to use Stable Diffusion

I reckon there was a intense use of my CPUs for a couple of days. Trying different prompts and settings to try to get it right.

Some good ideas :

• Don't hesitate to generate multiple pictures for the same prompt. Then you can use the least worse version as an input for the next batch.
• Use a specific model, trained with the type of images you want to achieve.
• If you have some drawing, painting or graphic skills it'd be much faster to do it by hand.
• It's hard to have both foreground and background right at the same time so I did them separately.

Here is the result 🎉

Things I never managed to get right

• I drink mate and there's no way Stable Diffusion has ever been train with images featuring people from Argentina or Uruguay!
• I drink coffee too but the mug always appears in a very surrealistic way