Sketch Switch

Sketch Switch sends the pictures you draw to a third-party vision-language model while the game is running, and shows you what that model says back. This is the core mechanic of the game rather than a side feature.

LIVE-GENERATED CONTENT

When you finish a drawing, the picture on the canvas and the stage prompt are sent to a generative model (Google Gemini, accessed through OpenRouter). The model returns a short label naming what it believes the drawing shows, plus a one-line remark written in the voice of the in-game character Kakiko. Both are displayed to you as returned. In Puzzle Mode that label is also what decides whether the stage is cleared. In Party Mode the model only answers fixed yes-or-no questions about the canvas; the score itself is calculated by the game code, not by the model.

PRE-GENERATED CONTENT

None. All artwork, music, interface text, stage design and character writing in the game were made by hand.

WHAT IS SENT

Only the drawing on the canvas and the stage prompt. No account details or personal information are sent to the model.

GUARDRAILS

Model responses are length-capped and are used only to name and comment on the drawing currently on screen. Because the model runs as a cloud service, Sketch Switch requires an internet connection to play.