Copyright © Nicholas A. Gasper Effective August 2026
The Policy, Plainly
No AI-generated art ships in Fate’s Edge. Not in the Art Bible, not in the web client, not in any adventure module, not on the dev portal or the blog, not in marketing. If it’s an image and it’s official, a human made it — by hand, digitally, traditionally, whatever your tools are, as long as the image itself is your own authorship and not a diffusion model’s.
This extends to submissions. Fan art is welcome (see the Fan Art Policy) — but art generated by an AI tool is not eligible to be featured, credited, or used anywhere official, even as fan art. If you submit something, the honest answer to “did a generative model make this” needs to be no.
This is a policy about images specifically. It is not a blanket “no AI” stance — Fate’s Edge uses language models extensively elsewhere, and the next section explains why that isn’t a contradiction.
Why Art, Specifically
I use LLMs. I’m not going to pretend otherwise — a large amount of the documentation, code maintenance, and content templating across this project’s repositories was done with an AI coding assistant helping me keep a sprawling codebase in sync. I’ll get into exactly what that looks like below. But I draw a hard line at image generation, and the line isn’t arbitrary — it’s because text-generation-as-a-tool and image-generation-as- a-product are not the same thing, ethically or technically.
Code and documentation are structural, and LLMs are good at structure. Keeping a multi-repo codebase internally consistent, updating docs when an API changes, converting content into a templated form — this is exactly the kind of highly-patterned, best-practices-driven work language models are suited for. It’s revision and maintenance, not authorship. The design decisions, the architecture, the game’s actual rules and lore — those are mine, arrived at the slow way. The model helps me hold the sprawl together long enough to keep working on it. I have a hard time tracking and maintaining structure across a large body of text on my own, and this is the accommodation that lets me do it anyway.
Image generation is derivative in a way text generation, used this way, isn’t. An AI image model doesn’t compose an illustration from first-principles the way a human artist does — it interpolates between techniques and, in well-documented cases, wholesale segments of existing artists’ work, without those artists’ consent and usually without their knowledge. There’s no equivalent move available to me with code: there’s no “artist” whose brushwork gets stitched into a merge commit. Code generation from an LLM draws on public patterns and conventions that were never anyone’s proprietary expression to begin with. Art generation, as it currently works, draws on people’s actual output, one image at a time, without asking.
The resource question cuts the same direction. Whether an LLM’s compute cost is “worth it” is genuinely debatable and I’m not going to pretend I’ve settled it for myself. But of the ways to spend that compute, keeping a text-based knowledge base consistent is about as light a lift as generative AI gets. Image generation is comparatively heavy — and I’d rather not spend that cost on something that’s also displacing the working artists I’d want illustrating this setting anyway.
So: least harm on one side, real harm on the other. That’s the line.
What This Means in Practice
- The Art Bible and all shipped illustrations are made by hand — mine or a credited contributor’s. See the Call for Art if you want to be one.
- Fan art submissions are reviewed on the honor system. If you tell me something’s AI-generated, or it’s obviously AI-generated, it doesn’t get featured or credited as fan art. This isn’t a judgment on you for making it for yourself — it’s that “official” and “featured” spaces are specifically reserved for human-made work.
- What you do on your own time is your business. I’m not going to police what you generate for your own private table or your own enjoyment — that’s between you and your conscience. This policy governs what Fate’s Edge itself uses and features, not what you do with the setting in your own home.
- This policy can’t retroactively un-train the models it’s written about. I have no way to verify that the base models any AI coding assistant relies on were trained cleanly, and I’m not going to claim otherwise. What I can control is what I generate images with, and what I ship — and the answer to both is: nothing generated.
The Honest Version of “Where I Use AI”
Since the whole point of this policy is not pretending the line is somewhere it isn’t, here’s the fuller picture, including the part that’s more directly AI than “helped me keep docs in sync.”
Most of the code and documentation work across this project’s repositories was done with agentic AI — Claude, specifically, working semi-autonomously across the codebase with me directing, reviewing, and correcting it. That’s not a small detail I’m eliding; it’s the actual scale of how this got built. I would love, genuinely, to have a healthy developer community doing this work with me instead. Right now, it’s just me, and this is how one person keeps a project this size from collapsing under its own sprawl.
One concrete example, since it’s easy to say “I hold the line” in the abstract and harder to point at the moment it cost something: the AI GM Bot could generate a vivid image of every narrated scene as it happens — the tech to do it is trivial to wire in, and it would look great in a demo. I declined that feature. It would help, and I’m not going to pretend otherwise — imagery does a lot of work a wall of text can’t. But it fails the same test everything else in this policy has to pass, so it doesn’t ship. Not “not yet.” Not ever.
The AI GM Bot is the other place people ask about, so: it’s the outgrowth of a heavily structured dataset, not an improvising storyteller. It describes pre-written scenes from adventure modules and calls for rolls; a deterministic state machine — not the language model — handles the actual mechanics and advances the adventure. I try to keep its token counts tidy, and I’d personally prefer people self-host it with something like Ollama — but hosting costs for that are real, and API costs are cheap by comparison, so the default path is what it is. The Bot does not replace a human GM. I have no illusions about that, and I’d ask you not to have them either. It can run the pre-written adventures capably, and it’s a fine way to explore the world solo. But a solo session with the Bot is a curiosity. It’s the other people at the table who bring you back to it. Nothing replaces that, and nothing here is trying to.
AI is expensive. Disability is expensive. Cars are expensive. Everything costs something, and there are more and less ethical ways to spend that cost — for AI as much as anything else. I’ve drawn my line where text support is a tool I use to keep working, and image generation is a product I won’t ship or feature. You’re free to draw yours somewhere else. This is just where mine is, and why.
Questions
If you’re not sure whether something you want to submit or feature qualifies, ask — support@fates-edge.com, the Discussions board, or the Discord. I’d rather have the conversation than guess wrong in either direction.