Opinions on Generative AI?

Heyhey folks,

I wanted to know what the opinion of folks here was on the use of AI image generators for Playdate development. I'm talking specifically of large-scale models like StableDiffusion, Dall-E (and it's integration into ChatGPT), and Midjourney.

Also if there's an official stance from Panic for use of AI in catalog-submitted games, I'd be interested to hear too.

  • AI use is always unethical
  • AI use needs to be clearly attributed
  • AI generators are invaluable tools to small indie devs
  • Whenever possible, real artists should be preferred over generative AI
  • AI is a tool like a paintbrush and neither ethical nor unethical
  • AI use is fine only for non-commercial games and prototypes
  • AI use is fine only when companies pay royalties to artists whose work was used
0 voters


Are you running this survey with other populations, too? I think your survey results will be very skewed if you only use a specific community like the Playdate dev forum. I get the impression that, for Playdate especially, there are many developers who consider themselves as artists who use games as a medium, and so the responses may disproportionately favor human artists.

Feel free to ignore the rest of this, but if you care to hear them, here are a few thoughts that tend fly through my head when this type of question comes up:

-All of the tools you have listed use any publicly accessible image data on the internet to train. It can be tempting to claim that it isn't unethical because a human can also train on copyright-protected works... But humans are allowed a clear exception because our own enjoyment is kind of the point of this media in the first place; and we are still not allowed to publish works which overly leverage copyright-protected material. A generative tool is not a person who gains harmless enjoyment from reproducing published works. It is a tool which is made to specifically copy image data using various statistical procedures to produce similar images on command. Legally, this would not be ok using screen capture or a digital camera, nor would it be ok using a photo editing application to make existing work look somewhat different. Therefore no exception should be given to these new tools

-Depending on the type of prompt used, it can be impossible to determine how much of any number of copyright-protected images are used in a single generated image, meaning it is impossible to divvy royalties appropriately. An argument could even be made that royalties could be owed to literally all copyright holders whose images were used to train the models, which is also not possible. Therefore an offer to pay royalties is not satisfactory

-As generative AI improves (images, animations, sounds, music, maybe even entire movies and video games) it will become easier and easier to just assume that some or even most aspects of media like games was created with AI. Even statements specifically to the contrary will be under scrutiny... this really sucks for those that prefer not to use generative tools because the creation process is their entire reason for engaging in the industry in the first place. Therefore "just don't use them if you don't like them" isn't a satisfactory argument

-It is such a shame that there is a push to automate art

-The danger of the capability of these tools to produce images which look like photographs is obvious

-I'm of the opinion that AI-generated imagery which is intended to look like human-generated work should NEVER be published in any kind of project, unless it is something meant specifically to review/comment on the ability of the tools to do so. I don't see any problem with using these tools to get ideas, storyboards, prototypes, or vague drafts for a human artist to use as source material

-Image generation tools which do not produce imagery meant to look like it was made by a human, which would be very difficult to reproduce by a human, or which have their own specific visual style inherent to their model are artistic unto themselves in a way that is hard to describe (Google's Deep Dream, for one). It has been interesting seeing these used for specific effects in certain games, especially when leveraged to produce disturbing/distressing vaguely anatomical images

1 Like

Interestingly enough, as a thought experiment I actually used GPT-4 to generate a Playdate game from scratch, solely relying on prompts to generative AI with no manual editing. Here's what I was able to produce:


It's simple but surprisingly performant. Initially the large amount of asteroids on screen progressively slowed down the framerate on-device, so I asked it to self-optimize and it auto-implemented a pooling system for the asteroids to reduce the overhead caused by frequent memory allocations and garbage collection.

I think Playdate is a device that encourages indie game creation. AI when used in a way like this that creates and executes a concept independently goes against the ethos of Playdate. In other words, we would be better off if projects like this didn't flood the marketplace.

However it does appear that image generation could empower programmers to create games despite not having any art skills or the funds to afford expensive commissions. I'm a terrible artist for example, and I used DALL-E to generate this spooky city just now, which could act as a background map in a Playdate adventure game where the developer doesn't have any art talent:


background map

I don't think it's necessarily a good thing to outright shun generative AI like this as development aids. As long as it can be used as a tool and nothing more I think it has merit. And as @lv9_engineer said above, sooner or later we generally won't be able to tell.