How I use AI
I am a solo developer and use image models to create art assets in a pipeline for the start of my workflow. I then manually process them to get them to a level I am happy with.
Selecting a concept
I prompt for art assets in a pixel-art style and generate based on a feeling what I am going for. Most of these are wrong: the proportions fight a game camera, the joints do not read, quality is blown out or muddy. This is a messy process and takes a while. It does not deliver ready to use assets, the candidate now requires manual processing.
The candidate with the most potential
This Gemini image was the one I used as a base. It was most like what I was going for and has relatively fake pixel art. It then gets scaled with a nearest neighbor algorithm in Krita and then I start manually fixing it.
Then the work is manual
AI workflow ends here. I convert the candidate to true pixel art, usually I trace the entire assets outer border again manually as it's always janky. Cleanup is pixel-level. Animation is frame by frame. Layering is done manually as well as there are no layer so nothing to work with.
AI cannot do real pixel art (yet) and so the output is actually not directly usable. This asset's base is 7MB and there's no way I can manually manipulate this as it's not real pixel art. So I have to scale it down and clean up all the artifacts that come with that.
- Convert to pixel art Scale the candidate in Krita with nearest neighbor, then fix it by hand on a real pixel grid.
- Cleanup Fix edges, palette banding, and transparency until it holds at game scale.
- Layering Split dome, housing, internals, and laser so each part can animate on its own.
- Draw remaining assets Add the beam, lights, housing, and effects that the prompt never resolved correctly.
- Animating Toggle and offset those layers into idle, charge, breakthrough, and blast sequences for example for this single asset.
Layering in Krita
The drill is single static image. I have to usually manually create all the layers and draw a bunch of stuff in that I want to animate. The gif is that file with layers toggled so you can see what I do manually. For example: charge, fire, and idle states before I commit frames.
Final animation frames
Each state is its own frame sequence. Idle, prepare, breakthrough, and blast sit side by side so the difference in motion is obvious.
Total time spent on this asset: 12 hours.
I would never be able to do this without AI. But it's still not an easy process. I just want to make fun games that people enjoy. I would not be able to make that at the level people are interested in without this workflow. Please feel free to not support my games if you are against AI usage, but please don't hate on the people who use it. Just note that all big studios are using it, they do have money to hire people, the just choose not to.
All assets
All assets are animation heavy work like this. Feel free to slide my game into the "AI slop" category if you want, but note, not all games using AI use the same workflow. I am only writing this because the negativity surrounding my games and the usage of ai. I have spent over a year on Core Splitter already and lots of that time is spent on art. It's disheartening to see people blanketing all games using ai as "AI slop" hopefully we can have some nuance.
Declaring AI use
Despite the fact that some people will behave disproportionately badly towards devs when "admitting" that ai was used, I still choose to put all the necessary disclaimers everywhere where asked. This is a decision I am making out of fairness. I hope you understand that given all I have learned over the last years working on Core Splitter. I could very easily tone down the art resolution a bunch, claim it as my own art and have undetectible, real pixel art that has no traces of AI usage. I am choosing not to do that, because I want to make a fun product that is the best I can make using the tools currently available.
However there are many games that follow the same workflow I describe above and with a few more steps will not be detectable as AI pixel art. I could do the same, there's nothing stopping me doing that and not disclosing it. If you choose to attack actively disclosed AI usage in game dev, you are choosing to support gamedevs who hide their AI usage instead.