I am a nerd. In so many ways, and one of those is the enjoyment of miniature boardgames (Frostgrave, Warhammer Quest, Trench Crusade and many more). As part of this is the zen like discipline of painting those small minis. And if you are like me, a bit technical, but not very good at color theory and selection of matching color palettes the results sometimes look, let us call it awkward.
Technology to the rescue – I know what color combinations I like when I see them in paintings, art work and other mini painters work. So what if I made something that could transfer the color palette of any image onto my mini? At least in-silico, the real job of the actual painting would still need to be done.
The back of the envelope solution looked something like this:
- An interface where you could upload a 3D image (STL) of the mini, and some other piece of artwork with the color palette.
- The colors from the image would be extracted and their respective ratio in the image calculated (this could be done easily with GCP Vision AI)
- The 3D image would be projected from a given angle into a 2D image, then using some sort of interactive click based segmentation would be used to partition the mini into the different components like armor, face, cape, weapon etc. And then the segmentation would be projected back on the 3D volume
- Now I would calculate the relative ratios of the different segmented volumes, map that to the ratios on the color palette and transfer the colors.
- Finally I would allow for playing with the light source to show highlights and shadows instead of bland flat colors.
All that would require is a bit of Next.js / Typescript for the frontend, backend running on GCP Cloud Run in Python with some additional calls to standard GCP APIs. The 2D mapping to 3D volume would be particularly interesting to implement.
Total implementation time? Who cares, this would be fun almost every step of the way. A good project to take me over the Danish winter and the many hours of greyish darkness.
The Google went ahead and released nano banana, also known as Gemini Flash Image 2.5. And they showed some pretty wild image editing abilities. OK, headed over to aistudio where you can test it for free and did the following prompt
I want this image recolored in the style of John Blanche, specifically I want heavy use of orche and terracotta with a few pops of an accent color
The result:
Got re-colored into this
In all of 20 seconds. A few more tests and iterations with prompts and it was pretty clear that my entire project was no longer needed.
Build it anyway – with AI
So as it was clear Google had already upended me and my idea, a new experiment formed. How fast would I be able to go from idea to full product? Granted I have some experience in programming, GCP infrastructure and AI but still. I fired up VS code together Claude code, settled for Next.js / Typescript as the frontend, GCP Cloud Run Service as the backend (in Python) and away we went. Around 4 hours across a Saturday and Sunday afternoon and it was completed. What I got was this as the core functionality.
The full site looks something like this:
It got a community section for showcasing recent or upvoted work done on the platform, a coming soon feature section with email signup and suggestion inbox, stripe hookup as payment method – complete with firebase auth etc.
Not in production yet?
Is it rough, yes quite. And I would likely spend 8 hours more on it to make sure everything is working correctly, especially on security and stability.
Is it what I thought about? Not completely, but it is like 80% there and maybe that is good enough.
What I am most fascinated with is the idea of how quickly I could get upended by a general purpose AI model. Imagine you are a company that is invested hard capital and time in building a niche product and from one day to another anyone with access to a general purpose AI system can do exactly the same in a matter of hours? It really makes you wonder what type of SaaS tools will be obsolete. It also makes you wonder what the new moat needs to be to protect your business. If everyone can build a clone of your idea in a matter of minutes, the moat of engineering hours is gone and you need to look at better distribution, better value proposition etc.
This has been a really fun little hobby project, and it might go live someday. At least I have purchased the domain myminipainter.com just for the fun of it.