Thread by @emmanuel_2m

How game artists can explore more creative options with #AI, using their own art, in their own style.

A short (and yet powerful) example with @DanielPlaychain 🧵

#AI #gaming #game #artist #StableDiffusion

.@DanielPlaychain is the Chief Creative Officer @playchain_ (Web3 gaming).

He has posted some awesome game art on his @ArtStationHQ - check it out: artstation.com/danieljohnsen

Daniel and I worked on a specific dataset of stylized characters (a style exploration, made a few years ago).

I focused on the characters below (removed the weapons, beast, and badges) and trained a finetune using @Scenario_gg on 7 images.

It started really simple, and I kept a basic prompt for most of the exploration.

"A character". Just this. And I ran a dozen batches.

Most of the output were OK (80%-ish) but then I removed the weird ones.

And here you go. 81 characters, AI-generated from @DanielPlaychain's art.

Quick comparison of the closest original asset, and a randomly AI-generated asset:

Some other close-up views (AI-generated too)

Then I tried making other variation by "forcing" the AI to draw a female character (elf-like)

"Character, elf, female"

Or an orc ("Character, orc")

Or a wizard 🧙

I generated variations around the "gremlin/little creature" using img2img

Made more orc-like creatures, also with img2img.

And even tried getting other shapes/silhouettes, still using img2img.

This was a very quick - and yet quite exciting experiment.

In that particular case, the output might not be final art (i.e. production-ready). Yet, it's a simple demonstration of how artists can use their own work to explore more creative options for their customers.