Back to blogs
Nov 29, 2025
05
A small frustration in Goa that became a 20-artwork experiment
When I landed in Goa, I had one quiet goal: to capture the place through the lens of my drawings. I carried my sketchbook everywhere — beaches, cafés, winding lanes, random corners that felt like they had room for me to just be. Every space that sparked meaning became a sketch. And without realizing it, I had built a ritual: sit, observe, sketch… repeat.
Sketching My Way Through Goa
The more I sketched, the more personal the process became. Each scene held a mood, a memory, or a moment I didn’t want to forget. These drawings started as quick impressions, but somewhere along the way I felt a shift — I wanted them to live beyond the raw pencil lines.
The Realisation: Bring Them to Life Through Illustration
Halfway through the trip, I looked at the growing pile of sketches and thought: “What if I illustrated all of these?” Illustration meant I could bring in color, depth, texture, emotion — all the things I felt while drawing but couldn’t always convey on paper. It wasn’t just documentation anymore; it was storytelling.

The Need: Automate Sketch → Vector
Turning 20+ sketches into editable illustrations by hand felt impossible. So I started wondering whether I could automate the sketch-to-vector pipeline in a way that still gave me control to edit and tweak. Something that preserved the essence of my lines but elevated them into digital art. This curiosity became the start of a much larger experiment.

Enter Lovable: My First Exploration
Since I had some experience with Lovable, it became my starting point. With a combination of ChatGPT + Lovable, I eventually got a version of the tool that produced decent sketch-to-illustration outputs and these were the results:

But there were major issues with these outputs:
Output quality wasn’t consistent
Layer structures were messy
Vector formations weren’t clean enough for Figma or Illustrator
If I wanted this to be usable, the AI needed to understand what a correct output should look like.
So I Built a Golden Data Set
To train the AI properly, I created a set of 20 illustrations — one for each sketch.
I finished them over four days.
Funnily enough, I have never illustrated at this speed or quality before. Maybe it was Goa, maybe it was the challenge, maybe it was Figma’s amazing stroke features — but I loved every minute of that sprint.

The Next Roadblock
Just when things were flowing, I hit a technical wall:
Lovable doesn’t allow direct AI model testing on the platform.
So I moved to third-party APIs to train models using my golden set. More tools, more trial-and-error, more learning curves. But at this point, I was all in.
Where I Am Today
This journey has been surprisingly rewarding. What began as a small creative frustration has turned into:
An experiment in automation
A deep dive into AI workflows
A personal illustration project
A crash course in building toward a usable product
My bigger hypothesis is simple:
How much work does it actually take to turn a small MVP into something the market can genuinely use?
I’m still exploring that answer.
So stay tuned — I’m continuing to refine, train, and build until this little frustration becomes something concrete and, hopefully, something useful for many more creators like me.

