site banner

Tinker Tuesday for May 26th, 2026

This thread is for anyone working on personal projects to share their progress, and hold themselves somewhat accountable to a group of peers.

Post your project, your progress from last week, and what you hope to accomplish this week.

If you want to be pinged with a reminder asking about your project, let me know, and I'll harass you each week until you cancel the service.

2
Jump in the discussion.

No email address required.

On my tank game, I got a simple grenade-throwing enemy working using unity's Behavior system, my collaborator got his messaging system working, and we got a bunch of old systems cleaned up and working properly again. He's working on engine and transmission mechanics, while I've had to hop over to another project where I'm trying to build a pixel-art environment tile set for another collaborator's RPG. After much experimentation for painting techniques for high-detail pixel art, I caved and got myself a midjourney subscription. The plan is currently to use midjourney to generate chunks of terrain detail, which I will then cut out into raw sprites. I will then clean up and normalize these to remove generation artifacts and inconsistencies, then use them to assemble the final terrain sprites. So far, it's working pretty well, and is the first time I've used AI for serious art production.

@Southkraut, @ArjinFerman, how go the projects?

All of my tinkering time went into that trans post, so didn't get anywhere this week.

I do have a question though - you have any recommendations for AI tools? I'm currently using an open source / creative commons sprite sheet. I tried creating something fancier recently, but when I tried asking for a spritesheet it tended to spit out the same kind of retro style that I was trying to upgrade from, and when I asked for an animation that I could later turn into a spritesheet, it would do goofy stuff like get out of the sprite bounds... Any hints?

damn opsec....

I'm currently using Midjourney, and I'm only just dipping my toes in the shallow end. Here's what I've learned so far:

Midjourney does not seem capable of generating a clean sprite sheet, nor tiling textures. If there's a way to do it, I haven't found it yet. What I have been able to do is to generate good-looking pixel-art images of terrain, then chop chunks out of that terrain and compile them into a sheet of raw chunks. My plan is to process these chunks using photoshop magic (desaturate to greyscale, clamp number of colors to a small value, gradient map to recolor with a consistent pallete) and possibly some cleanup by hand to eliminate rendering artifacts and inconsistencies in the chunks, and then use the chunks like lego pieces to assemble production-quality sprites. Creating tiling textures is fairly trivial in photoshop using filter > other > offset and fill > content-aware or minor editing.

I've started experimenting with re-rendering of images I've painted myself, but so far the results have been unusable. I think this is probably just lack of knowledge on my part, so I'm gonna keep digging on it.

I've also started testing generation of monster sprites. So far the results for that have also been unusable, but did have some interesting bits. I'm hoping the re-rendering might help a lot here if I can get it working properly.

You got some examples of the kind of pixel art you're going for? any details on the sorts of sprites you're trying to make?

Well, this is the spritesheet I'm using right now (or rather, the 3rd and 5th rows, as it's all from the top-down perspective, and I just rotate the sprites). Here's one I found for the marine / player character, and this is the style I'd like to go for. Not necessarily realistic, but not the retro-RPG cartoonish style either. Something "serious" / played straight, someting fitting for a sci-fi / horror. Honestly, the typical AI-slop style would even work for me, I thought, but the issue is it doesn't understand the assignment. I ask for top-down view, and it gives me a perspective view. If I specifically ask for a sprite, I get something closer to what I need, but then it reverts to the retro/cartoon style I want to avoid... I ask it for a transparent background, and it just draws checkers in the background... and that's without bringing up the animations.

On the other hand it did make me a pretty kick-ass logo.

It's not so important right now, either way. Perfectly happy to use the retro-RPG one as a placeholder, and when I'm done with the thing I can comission someone, but since you mentioned actually using AI for your project, I thought I'd ask if there's a better way to use it, than what I was doing.

DM me the prompt you're using to try for enemies?

Partially provoked by the fact that Github Copilot is going to become almost prohibitively more expensive to use (I paid 10€ for what github tells me will cost 200€ from June onwards), and partially because the Copilot was unable to cope with the growing complexity of the project, I decided a few weeks ago to restructure the codebase and my workflow to mostly go back to the pre-copilot days of me doing the coding, but especially the structural design of the code, myself. Vibe coding was great to make things happen without knowing in advance how to get Unreal Engine to do what I wanted, but it turned out to be really damn bad for maintainability. I've always had a sort of cartographic understanding of code, carefully shaping it to be visually intuitive to grasp, and taking great pains to make sure naming and hierarchies made it easy to traverse. But of course the LLM agent does not, does not need to, and has a hard time replicating that even if I try to tell it to (which might be poor prompting on my part). So right now I'm doing the utterly unpleasant work of going through the tens of thousands of lines written by the LLM to roughly hammer them into a shape that corporeal old me can make sense of. Long overdue I suppose.

As so often, this presents me with the same dilemma I've mentioned several times before - is it better to do all this review, refactoring, cleanup and correction work, or should I just start over? Sunk costs on the one hand, along with a big dose of "It's time to buckle down and get something done; if you keep starting over you'll never see results!". But on the other hand, the codebase is a mess and I do now have a much better understanding of the Engine and of how to use (and especially how not to use) LLM agents, and a fresh start might actually be a lot faster if I can salvage some of the more hard-won functionalities like the procedural meshes or my physical-entity-framework.

Another issue I have right now is that I apparently botched my project's upgrade from UE 5.4 to 5.7. I followed the documentation on how to upgrade, but now I can't open the project in the Unreal Editor anymore because apparently some modules haven't been fully upgraded, and attempting to upgrade them fails. Hm. On the one hand, this might be fate pointing me towards a fresh start. On the other hand, if this happens again in the future, I can't start over every time!

Issues like these make me think of going back to Unity, that simpler world of yesteryear, where I had everything under control. But then I think of the strange directions the Unity Engine has been going, and I recall Zorba's words on the topic of comparing game engines.