Day one of this devlog, day 42 of Pixel Mold.

Today I spent around 10 hours testing and debugging.

I didn’t add anything new. I decided to keep the legacy membrane as a separate working system rather than deleting it, so the whole day went into making the program more stable.

I built a test harness for crowded scenes, different numbers of organisms, different tissue types, nested membranes and long simulation runs.
Codex can spend 30–50 minutes running tests, finding a cause and making a fix. Then I open the program, watch it for a minute, notice another bug, describe it, and the cycle starts again.

Test, fix, run, find another failure, repeat.
Nothing visually new, but the program is becoming more stable. I now have 22% of my weekly Codex limit left.

In parallel, I made sweet cherry syrup from a ChatGPT recipe, drank a few beers, and invented a cocktail with the syrup and Gorki List.

Day 44 of Pixel Mold

I spent another two days testing and debugging. Still nothing new added

LLM are obviously very good at writing code. You can understand absolutely nothing about programming and avoid looking into the small details of how your own program works. But when it comes to architecture and basic principles, you still need to understand them yourself

A language model runs into roughly the same problem as a human. While focusing deeply on one specific task, it can stop seeing the bigger picture and miss a simple solution that has been right in front of it the whole time. This is where human control is still needed

During testing, I found that with many organisms and long simulation runs, the nesting of regions could break. An internal membrane or tissue would gradually escape outside its parent membrane

A lot of workarounds were added, and a lot of tokens were spent testing and debugging them. In the end, the solution was very simple

I suggested limiting the living area of every internal membrane to the area inside its parent membrane. This way, the internal membrane does not even know that the outside world exists and only uses the space inside its parent for movement

And it worked

In three days, I used my entire two-week token limit. For now, I’m continuing the research

Today I dreamed about 3d spirall cellular automaton

Day 51 of Pixel Mold

I spent a few more days auditing the code and restructuring the project

I split the Rust engine into separate modules and made a performance patch. After that, I finally moved on to new mechanics and interface changes

The right panel now works a bit like Blender

The upper section is an Outliner where you can select any organism or tissue type. Tissues belonging to the same organism appear as one expandable group

The lower section is the parameters panel. The settings are now grouped into vertical tabs such as Information and Form

I also added a new tool — Stamp

You can choose any PNG file and place it on the canvas. Tissue then starts growing from the image pixels. It can be used to grow text or other shapes

The next step is a VJ panel for quickly controlling scenes, presets, and audio reactivity

I need it for an illegal open-air festival outside Belgrade, where I’m taking my pixel mold

And special thanks to OpenAI for resetting my weekly limits two days in a row. I spend each new limit in one day

Added classic cellular automatons.
And a lot of other stuff, but I’m too lazy to explain it, sorry:)

New type of tissue

Procedural map generator for Pixel Mold Game Prototype

Procedural map generator

Beograd / Kalemegdan
03/10/26