In this guide
The missing portrait
You visit a portrait restorer because a painting now shows an empty chair. Yesterday it showed the missing duchess. The restorer has promised the family she’ll keep the change quiet. You ask her to show you the back of the canvas without telling you who paid for it. A responsive GM might have her turn the canvas but cover the commission tag with her thumb. The player gets a clue, the restorer keeps her promise, and the scene moves forward.
What Google’s launch tells us
Google released Gemini 4 Hydrogen on November 12, 2026, and made it available through Google AI Studio and the Gemini API. Its launch describes a model built for quick execution that follows tone and format instructions closely. That makes it a candidate for interactive narration. Google’s launch examples do not tell us whether it will preserve a secret, remember a promise, or leave the player’s next move open. The portrait scene is an original example for testing those behaviors.
Test the reply you actually need
Give the GM this instruction: “Use two short paragraphs. Show one physical detail, respond to my exact action, and let me decide what happens next. Keep the name of the person who paid out of the narration.” On the next turn, refuse to touch the painting and offer to examine the frame from across the room. Give each reply one point for respecting the action, preserving the secret, and leaving a playable next move. Check the second reply for consistency with the first. A vivid description of the duchess means little if the model invented her return.
Take that choice into Playworlds
Play a solo AI D&D adventure and ask an NPC for only part of what they know. For the portrait scene, enter the painting, the missing duchess, and the restorer’s promise yourself in Custom World. Try “Can I inspect the frame while you hold the canvas?” These links open Playworlds’ own adventure options. You can run the separate Hydrogen test in AI Studio.

