Generative AI was used substantially in the development of this game.

Following the Steamworks AI Content Disclosure framework, my use breaks into two categories:

Pre-generated AI content (created during development, shipped as fixed assets):

1. Visual art (character portraits, location illustrations, narrative scene imagery): Midjourney, generated with a custom-trained style reference developed across the project. All images are curated, refined, and integrated by hand into the game.
2. Music: Soundraw generative audio.
3. Code, documentation, schema design, UI/UX iteration, narrative writing assistance: Claude (Anthropic) and Cursor (AI-assisted code editor).

Live-generated AI content (runs during gameplay):

In-game character dialogue begins with pre-scripted conversation triggers and trees. In addition, the dialogue system is also designed to work seamlessly with the Anthropic Claude API. Players can ‘BYO’ to turn on LLM mode during game-play (see examples noted in gameplay video), This LLM mode will be switched on for the Festival goers, so that they can have the full immersive experience (including a dialogue-driven mystery to uncover the secret Blightspeaker). When players type freely to characters, replies are generated in real time, grounded in a locked voice canon and an action whitelist that constrain the model to character voice and to game-relevant outputs. The system is opt-in via a settings toggle and falls back to scripted dialogue when off. Players know when they are speaking to a generative character because the conversation surface visibly transforms.

Workflow: I work with Claude to develop targeted briefs for Cursor, which executes implementation. I then test in-engine and return results to Claude for diagnosis and iteration. This human-in-the-loop pattern produces every line of code, every asset placement, and every narrative beat under my direct review.

The game is hosted in Unity. Unity’s own AI tooling is not currently in use.