← sources

Programmed Art can be the new (and better) Programmer Art, and AI is enabling it

source: reddit (reddit:gamedev)published 2026-10-24 09:01ingested 2026-10-24 10:57

retained text

EDIT: I don't care how much hate this is getting. This space is for everyone making games, and I will not give up space to miserable haters, ever.

*This post is NOT about AI-generated art (boring! 🥱)*

*Yes, I know what* ***procedural art*** *is. Programmed art CAN be procedural, but it doesn't need to be, and when it is, then there is still more to it*

*This post, like all my posts, is NOT written by AI, and I'm not promoting a product here. The tools I built for myself that are mentioned here I built for my own game, and I'm not currently sharing them, or their outputs.*

I am seeing a trend of what others are doing and what I have done to make art for our games, and I believe that this trend is being powered by AI making it much much easier to create custom workflows.

Programmed art is to **make art like a programmer**, but not in a bad way (shitty pixel art and even shittier 3D models), but in a way that leverages all the things programmers do best.

Some examples first of what I have seen people shared (I don't have the original links right now)

* A renderer for pixel art characters that can produce the same character in different art styles and render it in different poses
* A DSL for mesh operations to build procedural meshes

Some examples of what I have built:

* A real time synthesizer for retro sound effects
* An extensible format for data-driven VFX
* Something similar for data-driven NPC behaviours and movesets
* A workflow for vector icons
* I am also working on procedural meshes and skinning

The basic characterisrics of programmed art:

* Small, usually text-based source files
* A mechanical workflow to produce the actual artwork at build time, load time or in real-time

(Mechanical is a word I use for classic "if/else" code, not involving LLMs)

So where does AI come into this?

1. It can mass-produce the source files rapidly and at low token cost
2. It makes all those custom workflows viable, because you can vibe code them in much less time than it would take to produce programmer art by hand.

Some obvious benefits of proframmed art over handmade and purely AI-generated art:

* You can keep your artwork in GIT
* You can change your art style cheaply and try different styles rapidly
* You can refactor your artwork, e.g., separate style, colors and shape

The last point is why I call it "programmed art". You can shamelessly apply programming concepts to the process of making art. Programmers making art by doing what they do best, rather than being wannabe artists.

Let me lay out the details of the least interesting but most representative example: Vector icons. This is most likely not even new and has been done many times, but... My workflow is as follows:

1. Start with an SVG file
2. At build time, bake it into an MSDF atlas (texture plus metadata)
3. At load time, optionally bake it into a bitmap texture, or alternatively, render the MSDF icons in real-time, if you need scalable icons.

**The SVG files are produced by AI**. They are just text (XML). Making dozens of icons and re-making them many times is cheap. And I keep the SVG files in GIT.

I'd go even further: What if SVG had a way to include other files or set and expand variables (does it? It would take me a few seconds to fine out, but I'm not doing that now). Then I could refactor my source icons into

* Color palettes
* Style parameters
* Recurring backgrounds (e.g., buff/debuff icons)
* Reusable shapes/symbols
* Composite icons that put all of the above together

If SVG can't do it, then bastardising or wrapping SVG or making my own format and tooling is less than a day of work, thanks to AI.

**Does programmed art look good?** Yes and no. It's hit and miss. But producing art that looks good isn't a question of how good it looks at the first attempt. If I don't like the result, I fix the parameters and workflow and try again. The traditional approach would be throwing away days or weeks of hard work. And for placeholder art (thay may even be demo-ready), it's good enough any day. Oh, and you get consistency and precision for free.

**Is it cost effective?** Hell, yeah. More cost and time effective than AI art. Token cost is once for the tooling plus dirt cheap for every iteration of making and improving the artwork.

**Is it fun?** What do you think it is? I'm a programmer doing what I love doing the most: Building tools, automating things, deconstructing and solving problems. You better believe it that I'm having fun. What I'm not doing is learning a skill I won't have the time to use (pixel art and modeling), but I now have the resources to gradually learn all the essential skills of art direction. What I'm also not doing is staring at a code editor writing and fixing C++ code for a component or tool where the code isn't the interesting part.

**Does it still look like programmer art?** Absolutely. But I made it in hours, not months, including the workflow, and I can remake it in a better style a dozen times a day.

**Does it look like AI sl.op?** I don't know. But being able to control the parameters at several levels, mechanically and precisely, avoids AI producing default AI style.

**Does it have soul?** I don't know. What is a soul anyway? But if art has soul, then I don't think you'll find it in a pixel grid while pixeling the shadow of pose 56 of character 195 on day 160 of making the artwork for your game pixel by pixel.

document fetched 2026-10-25 00:25 · canonical https://reddit.com/r/gamedev/comments/1xe3sth/programmed_art_can_be_the_new_and_better