I've released a beta version of my SDF Model editor project. This is my first major project that I'm trying to "push past MVP".
How it all began
I've long been interested in real-time rendering. I started learning glsl. I came across this article by Inigo Quilez and realized that my mathematical knowledge...
Long story short, I decided to create my own full-fledged 3D editor that would allow artists to create 3D models using simple shapes. The main idea was a fresh approach (not really all that new, to be honest) to traditional modeling – to ditch all that sculpting and polygonal editing, and instead move spheres and boxes around like a little kid.
Process
| Initial, December 2025 | Beta, January 2026 |
|---|---|
 |
 |
I read a lot of articles and used a fair number of AI models to learn. I used the AI not as a co-developer, but as a "teacher." I didn't ask questions like "how to do it", but rather "how and why it works." And after I understood "how and why", I started asking copilot to write code based on the AI's thinking.
And what happened?
It was terrible. But it worked. I started studying the code. It was a far cry from what it is now. It was written in Pygame and consisted of 500 lines of code. I couldn't do anything except rotate the camera, which didn't even do that correctly 🤦♂️.
I got bored. I abandoned it. But I got bored, so I continued. It took me a month to move away from Pygame and switch to Imgui, as well as fix the code. Everything worked well, and I was thrilled with the result.
Cursor
I decided to download cursor ai. Everyone was talking about how much it could do. I prompted it to completely refactor the code and add functionality, adding primitives and operations. And it was a miracle. I started studying the new code and customizing it.
That was good, but I still had to do the input. I turned to cursor and... It turned out my limit had been reached 👎. I went back to copilot. And it was strange - for some reason, copilot kept putting spaces after periods! It took me half an hour to fix the code. And what happened? It didn't work. I repeated this until I reached my limit. I'm already tired and barely got everything working. I posted the project on GitHub.
Publish
Publishing on GitHub motivated me to develop the project. Even though views were low, I still wanted to finish the project. Step by step, I improved the bugs slightly, sometimes using AI for refactoring, testing, and bug detection (even though it often created these bugs itself). Overall, I realized how bad AI is for large projects.
Now
24 days have passed since then.
It felt like months had passed. I can't believe I've accomplished so much. I can't believe my code is 80,000 characters long. Even though I still haven't achieved all of my MVP goals, I'm very happy with the result.
And by the way, I changed my device (Windows 7 was better than 10...)
Final
I learned a lot, made a ton of mistakes, corrected them, and didn't sleep a single night. I think I realized that AI is like a casino: you either hit the jackpot and get a fantastic app, or you waste all your limits, lol. I'm tired of writing this post...
Top comments (13)
I've been building something similar in JS - Though I use cursor.
Very frustrated trying to build complex SDF compositions in code only (glsl)
So I built this prototype: sleditor.com/lab/sdfeditor/
or see video here: youtube.com/watch?v=PZ3V-6J80Qo
It's very clunky, node based, has bugs but I've used it for actual work now.
Next step is integrating it with a full featured shader editor I'm working on so it's being completely rebuilt.
It looks good for now, but yes, there's still some work to be done. I've been developing my app for months now and have made more than a few architectural and logical mistakes, so I decided to rewrite it from scratch. Even though it was already functional, I'm very unhappy with the code itself. My advice: keep your code clean. Sometimes less functional but clean code is worth more than any written incomprehensibly.
Yes, I started building a shader editor in October - I had to completely rebuild it from the ground up with a dedicated UI library and a much better architecture.
Also the first SDFeditor was a single file prototype so that needs a complete overhaul and rebuilding so it can integrate with the main app.
I actually think its worth throwing yourself fast at the task and doing it all wrong first time, you know so much more about how to put it together the second time.
I guess if you want to compare V1: sleditor.com with V2 sledit.xyz
It's interestign you're focused on this whereas I'm building it as a component for a bigger app, so you're probably way ahead of me on how this should work.
Will follow with interest.
This is a really inspiring journey—pushing a complex idea past MVP while actually learning the why behind it is impressive. I love your approach to AI as a teacher, and I’m excited to see how this SDF editor evolves as you refine the core ideas and usability.
Thanks!
Looking forward to your next post!
But you said Cursor was great, however you reached your limit ... so then you went with Copliot (cheaper), but it sucked, lol ... isn't the point then that AI as such doesn't suck, but that you just need to use the better (but more expensive) tool?
I installed and used Cursor too, and it was a miracle, but I hit the limit on my (free) plan within an hour :-)
So how can we have our cake and eat it too? Maybe for "hobby" projects old-fashioned manual coding is still the way to go, companies with enough $$$ can use AI tools (the best tools, that is) ...
As Phillip II said, “A donkey loaded with gold will open the gates to any fortress.” Here also: "A company loaded with 1000 bucks will open the gates to any AI". Unfortunately, all the sweetest things are too expensive.
On the other hand, if Cursor Pro for $20 per month does get you 80% of the way, with good quality, then maybe it's the answer to this puzzle - but I've read MANY stories and comments on forums and so on that you end up paying through the nose with Cursor ...
Spending a fortune on Cursor doesn't guarantee success. Therefore, recouping your development investment will be difficult, and even if you do, you'll still lose $20 of your earnings.
No, it absolutely doesn't guarantee success, and I'm certainly not advocating spending a fortune on it - for certain tasks, Cursor (or AI coding tools) works well, and I'd be curious to see how far the $20 a month for the Pro subscription would get me, when I attempt to use it "with judgement" ...
But why Python for this project?
There are a lot of reasons for this. It would take too long to list them all, but basically:
Using other languages would have required me to relearn, and besides, the code quality would have been lower, and development would have taken longer and been less enjoyable (but I would have developed my typing skills by using a lot of unnecessary words, public private variables, and a lot of symbols "{}; {}; {};"