Firstly the moodboard and reference sheet to maintain every input and output of my process converging to aimed art direction

Then, a colab notebook to run a remote repo of stable diffusion (since Google's GPUs are freaking faster than my RTX3080)

Download the right version of stable diffusion (2.0 and 2.1 sucks when not working with photorealistic images)

After half a hour making the colab notebook mount the tool with no error messages and fixing evey single error message with the very first prompts, we design the prompt keywords and then add the importance tags

Since it's nearly impossible to output readable coherent text we are going to photobash a very simple initial guidance image within the desired art direction of our moodboard

Many thanks to Nicolas Raymond for his / her awesome anarchy flag under CC0 licensing

Thanks Wikimedia commons for that awesome icon!
Also, that's just looking like Adobe is already giving you a spoiler about the future of illustration ">.>

Since I'm a technical artist, and I came from a technical support background, I will use my "swiss army knife" tool for photobashing ... MICROSOFT's WORD

No need for something REALY AMAZING, all that matters is the big picture of the idea and a lot of noise textures spread everywhere to make the AI aim for these regions

Back to our tool on the image to image tab, with subtle prompt tweaks and after 16 samples we may notice that a "layer of noise" is missing over the whole image to enhance the AI perception of the unchanged areas

Back to my editing tool, I added a noise layer over the image

Now the AI is reaching every pixel of the guide image and it's just a matter of tweaking the strength

With the lowest denoise strength (just to glue up the whole composition into a seamless icon), checking the moodboard through a thumbnail test, I notice that a darker background could "bump" the content of the image and stand out on the front page

Ok, back to my editing tool, addded some dark shapes and another layer of noise

yeah... not good enough, i should surely soften the edges on microsoft word

Editing the shape points with bezier curves for round corners and soft edges for better blending with the edges of the icon

Yeah.... should work ...

Yeah... much better when compared to the previous one, the contrast really bumps the content and has better readability when compared side by side against the reference

Then, attempt to "converge / bend" the Ai towards that awesome inspiration behind the idea

GlueIT for a grid of 24 samples

The generated 24 samples with 512x512 pixels each!
From left to right each sample has the same parameters with ascending seed value!

Chose the one with the best readability, but also with the desired "artstyle".

I did not use any of the samples replacing the "I" vogal by a "posca pen" because of the additional cognitve layer of iconography and probably "too cliché"

settings for "prompt prototyping"

settings for "mining" the samples

Increased resolution with 33% of denoising strength (not such a big deal)

The settings for upscaling, with tweaked prompt parameters

Don't forget to add the correct tags for your AWESOME ARTWORK =D

Check if it works as expected on the previous thumbnail testing phase