Things That Lurk Studios

⋄ ◈ Manifest Creativity ◈ ⋄

• personal private commissions
• Illustration
• concept art

About

My name is Martina Scafa and I am an Italian illustrator born and raised in Turin. My passion led me to pursue educational routes in other countries: I completed the Computer Arts course at Abertay University in Scotland. I currently work as a freelancer and live in the Netherlands.I am passionate about fantasy worldbuilding and love portraying stories through painting.Past experience: Frog God Games, Onyx Path Publishing.

• Contact me
• Commission Rates
• All Links

.* PAGE CURRENTLY UNDER CONSTRUCTION *❖❖❖.

I started learning about AI by noodling about on online generators. This time helped me explore what capabilities AI generation had. It was early 2024, and anything seemed possible.

I approached learning about running stable diffusion locally by starting with Automatic1111. I was curious about the possibilities offered for what regards photobashing, and given the limitations imposed by my AMD gpu, I primarily used it as an additional tool to introduce extra noise and iterate my digital paintings in a different way.

Initially, I was leaning more on my abilities to bash and overpaint images, but eventually I grew comfortable enough with the AI generation pipeline that I branched into learning more about Stable Diffusion itself and how checkpoints and LoRAs were supposed to work optimally together.I began experimenting with images that did not focus on characters, returning once again to the worldbuilding project that had started it all.

This entailed learning the ins and outs of ControlNet and Inpainting, while also mastering the delicate art of making half a dozen Loras play nice with each other and not create an artefact hellscape - a thing Stable Diffusion was quite apt at even just a year ago, and remains fairly talented at even nowadays.
ControlNet could iterate on my images very quickly, and photobashing the results back into my painting through the use of masks and fusion layers to then continue from there offered interesting opportunities.

While I learnt the basics at the time SDXL models were getting released, I followed the developments on the AI generation scene for months, and eventually I grew more and more interested in Flux. Given yet again my self-imposed AMD curse, this led me down the path of ComfyUI.
I began the re-learning process by recreating my old pipeline focused on PonyDiffusion to begin with, but I have since been experimenting with the greater flexibility ComfyUI's nodes offer. Flux is an excellent model and the possibilities are truly endless.I am currently working on training my own Loras for my characters and exploring ControlNet's poses through the use of Openpose Editor.