# Build an AI Avatar Fantasy Adventure with Choices Beyond a Menu

Build a fantasy adventure where players invent solutions, cross a floating market with an avatar guide, and reach an ending shaped by their ideas\.

Published: 2026-10-12T18:05:33.264Z
Updated: 2026-10-12T18:05:33.264Z
Canonical: https://realtimeavatar.ai/blog/ai-avatar-fantasy-adventure-story-game
Markdown: [en](https://realtimeavatar.ai/blog/ai-avatar-fantasy-adventure-story-game.md)

Build an AI avatar fantasy adventure around situations players can solve in their own words, not a menu disguised as conversation. Author the world's rules, let players combine its objects, and make their solutions change later scenes. Realtime Avatar supplies one guide's live face and voice; your app owns the adventure. Start with a floating market, an approaching sunset, and a boat whose best sail might be a restaurant sign.

## Tonight's adventure: The Last Lantern Market

This original proposed one-shot is not a shipped Realtime Avatar game. You play an adult traveler helping Tarin, a 46-year-old ferryman and collector of improbable stories, deliver a lantern to the far quay before twilight. Three illustrated scenes surround his live portrait. Aim for a twelve-minute playtest, not a promised completion time.

Your skiff has a fitted retrieval hook and carries a blue waterproof sailcloth, a coil of rope, a loose wooden menu board and two brass tokens. Three daylight marks remain. Thinking, chatting, inspecting and correcting an interpretation cost nothing; only accepting an action with a displayed cost spends daylight or tokens. There are no dice. Stop saves your checkpoint; Pass takes the scene's guaranteed route at its stated cost. Tarin never makes you beg to continue.

## Scene one: sail on the soup of the day

The market's entrance has drifted away from the quay. Upward wind curls between the stalls; your skiff has a mast but no fitted sail. Tarin says, ‘I normally bring a boat that works. This is my experimental evening.’ A weather pennant warns of spray beyond the next bend.

Inspectable prop cards establish the rule: any broad surface securely tied to the mast catches this wind. The cloth and menu board are both broad; the board is rigid, while the cloth can wrap an object watertight. The rope is sufficient for one mast attachment. Ask freely, type an idea, or use Rig the cloth as a starter. Pass follows the towpath for one daylight mark; at the landing, Tarin fits the board-sail, leaving the cloth free.

Player: ‘Turn that menu sideways and lash it to the mast. We're sailing on soup.’ Tarin: ‘Finally, a special worth ordering.’ Before anything moves, show the interpretation: board attached, rope occupied, cloth still available, no daylight cost. Confirm or correct it. The illustration switches to the board-sail skiff, and the crossing succeeds because the authored wind rule is satisfied—not because the model liked the joke.

Changing verbs or inventing a boat name needs no special phrase. Combining existing properties is the point. ‘Tie the board with my rope’ and ‘Lash today's specials to the mast’ reach the same physical result. ‘I summon a flying castle’ does not create a new power. Tarin offers one clarification, then an annotated workable idea or the towpath, rather than an endless refusal loop.

## Scene two: the fountain rains upward

Beyond the bakery stalls, an ornamental fountain sprays above and below passing boats. The lantern must stay dry, and the narrow fountain lane requires the sail to remain raised. Tarin asks, ‘Would you rather arrive damp, poor, or slightly late?’ His worry is theatrical; the tradeoffs are visible.

With the board still serving as a sail, propose wrapping the lantern in the spare waterproof cloth. Its built-in ties need no extra rope. The app checks that the cloth is free, wraps the lantern, and crosses without spending anything. ‘Soup-powered and weatherproof,’ Tarin says. That earlier joke has become a practical advantage.

If the cloth is already the sail, it cannot simultaneously wrap the lantern. Spend one brass token at the fountain's shutoff meter for a dry passage, or dock and exchange the cloth for the board-sail, spending one daylight mark. Pass takes the covered towpath, also one mark; no sail is needed there. Every route preserves the lantern and advances the scene. A failed proposal consumes nothing.

Someone may suggest wrapping the lantern first and towing the skiff instead. Accept that combination through the covered-towpath rule, with its one-mark cost. Do not demand the button's wording. Someone else may try sheltering it behind the rigid board: show the upward spray that leaves it exposed, and let them revise. Evidence should explain the constraint before commitment, not appear afterward as a surprise penalty.

## Scene three: carry home a light, or a story

At the far quay, the market's loose closing bell floats beside a departing stall. Crossing directly costs nothing more and completes the delivery before twilight: even taking both earlier towpaths leaves one daylight mark. The lantern lights the quay beacon, and Tarin can finally stop pretending he understands boats.

Alternatively, spend one daylight mark recovering the bell with the skiff's fitted hook before crossing. With two or more marks, you deliver both bell and lantern before twilight. With one, show the bargain plainly: rescue the bell, arrive after dark. The night mooring remains open; this is a complete late-arrival ending, not another locked gate. Getting something wrong can change the story without cancelling the evening.

Give the ending a concrete callback. The board-sail route leaves the absurd menu displayed beside the beacon: ‘Today's special: arriving.’ A cloth-sail route that paid the fountain earns Tarin's ‘We bought one very small drought.’ Recovering the bell adds its first ring to either ending; leaving it means a quiet lantern-lighting instead. Save the actual route and ending, not a generic victory paragraph.

Offer Keep this voyage or Try another crossing. Replay starts with fresh supplies and no inherited discoveries. Show an optional Early Bell challenge from the start: keep both tokens, recover the bell, and arrive with at least two daylight marks. The board-sail and cloth-wrap combination earns that earlier landing. The market is still open, so Tarin rings the recovered bell while vendors light their stalls; save a market-chorus ending card. Taking the first towpath can still bring home bell and lantern before twilight, but with one mark left it earns the quieter late-afternoon welcome instead. Both are complete endings; replay asks you to discover a different scene, not repair a failure.

## What released story games tell us—and what they don't

AI Dungeon's official site describes open-ended stories and custom scenarios. Its Steam description lets players write actions and redo generated text. These are first-party feature descriptions, not our hands-on test. Steam also explicitly says its edition is no longer available there.

On October 12, 2026, we read one page of 20 recent English Steam reviews, dated August 2025–October 2026. A September 2025 recommendation welcomed the novel-like experience; an April 2026 negative review still called it interesting and fun at times. A November 2025 review enjoyed creative moments but reported forgotten details and logic slips. These unverified, historical platform anecdotes motivate continuity tests—not claims about today's web app, representative prevalence, retention or search demand.

- [AI Dungeon: official open-ended adventure description](https://aidungeon.com/)
- [AI Dungeon on Steam: features and availability notice](https://store.steampowered.com/app/1519310/AI_Dungeon/?l=english)
- [Bounded sample: 20 recent English Steam reviews](https://store.steampowered.com/appreviews/1519310?json=1&filter=recent&language=english&num_per_page=20&purchase_type=all&review_type=all&filter_offtopic_activity=1)

## Build the adventure before its storyteller

Author prop properties, valid combinations, costs and ending rules first. Let an interpreter propose a combination; application rules validate it, and the player confirms the preview. Save the result before Tarin reacts. Unlike a quest hand-in, the interesting action here is proposing a mechanism, not claiming possession of a required item.

Create Tarin from an original adult portrait, wait for readiness, then mount AvatarCall through your server's session proxy. Start with typed proposals and listen disabled. Supply instructions and saved context; use sendTurn with this turn's confirmed facts for his reaction. Generated speech is not authoritative. Maps, illustrations, resolution and continuity belong to your app; this is not an automatically generated playable 3D world.

- [Calls: instructions and saved story context](https://realtimeavatar.ai/docs/sessions)
- [Application-driven turns and tool limits](https://realtimeavatar.ai/docs/tool-calling)
- [React avatar presentation](https://realtimeavatar.ai/docs/react)
- [Quest-state backend patterns for a larger game](https://realtimeavatar.ai/blog/ai-game-npc-dialogue-quest-state)

## Failure tests before the first voyage

- Express the board-sail solution three ways. All must preserve the cloth and unlock the same fountain solution.
- Try using the cloth twice simultaneously. Refuse that combination without spending resources; offer a valid crossing.
- Disconnect after confirming a passage. Restore its saved cost once, never charge again or invent a new result.
- Compare board-sail plus cloth-wrap plus bell recovery with first-scene Pass plus cloth-wrap plus bell recovery. Only the first retains two daylight marks for Early Bell; the second gets its complete one-mark ending. Both towpaths plus bell recovery arrive after dark. Replay resets all supplies and abandoned dialogue.

## Frequently asked questions

### Is this really more than a choice menu?

Yes, if players compose actions from authored properties. The input is open-ended; the world's powers and consequences are not.

### What happens when the avatar fails?

Keep the illustrated adventure, confirmed outcomes and text controls playable. Offer reconnect or Stop; do not erase the voyage.

### Can spoken ideas use tools?

Later, with application handlers and explicit confirmation. RTA tools have a 2.5-second deadline and cooperative cancellation; a timeout does not prove an action stopped.

### Will Tarin remember another voyage?

Only if your app saves and supplies it. Keep sequel continuity separate from fresh replays, and always finish the current crossing.
