Skip to main content

Video & audio

Make a fictional museum audio guide with AI

Create one fictional exhibit label and 45-second audio stop with a guarded prompt, consistency checks and a manual fallback.

Concept illustration for producing audio and video
AI-generated illustration: Video & audio

This beginner project creates one invented exhibit, a short wall label and a 45-second audio draft. The exhibit must be clearly marked as fictional, and the label and narration must agree. PickByTask did not generate or listener-test the audio.

ElevenLabs currently lists text-to-speech and an editor for voiceovers among its creative tools. Use it for a short narration test only after the fictional label and script pass the same fact checklist. Check the current plan and usage shown in your account before generating. ElevenLabs · Plans

Contents — jump to a section

Invent one exhibit with fixed facts

Write a five-row source card: exhibit name, maker, invented year, material and one unusual detail. Add the sentence “This exhibit and its history are fictional.” Keep the project original; do not imitate a living person's voice or borrow a museum's label.

Allow about 45–75 minutes for writing, checking, one short audio experiment and a manual backup. The official pricing page lists plan-specific features and credits, so use the live page rather than assuming a generation is free.

Copy this starter prompt

Use only the five facts below. Write:
1) a 70-word museum label;
2) a 95-word audio script that takes about 45 seconds;
3) a checklist showing where each fact appears.
Begin both pieces by saying the exhibit is fictional.
Do not add a real museum, expert, quotation, discovery, date or provenance.
If a detail is missing, write NEEDS FACT.
[PASTE THE FIVE-ROW SOURCE CARD]
ElevenLabs official website: product introduction and examplesVisit ElevenLabs ↗
ElevenLabs website · captured 2026-09-13 · Source

Build and check the stop

  1. Freeze the five facts and give each one an ID, F1–F5.
  2. Run the prompt and remove every sentence that lacks an ID.
  3. Compare label and audio line by line; names, year and material must match.
  4. Read the script aloud and shorten it before any voice generation.
  5. Generate only a short test with a permitted voice and record the displayed usage.
  6. Ask a listener to name the exhibit and identify that the story is fictional.

Worked fictional sample

F1 names the exhibit “The Rain Compass.” F2 says it was made by fictional craftsperson Mali Arun in 1912. F3 says it is brass and blue glass. The label and audio may describe those facts, but cannot claim that sailors used it or that a real museum owns it unless those facts were deliberately added to the fictional source card.

The stop passes when both versions state that it is fictional, all five facts agree, no unsupported history appears, and a listener can repeat the exhibit name and material. Keep the approved text beside the audio filename.

Recover without the voice tool

If the narration adds facts, rewrite from one source row at a time. If timing runs long, cut adjectives before facts. If voice generation is unavailable or the plan limit is reached, record yourself reading the approved script or present the label with a play-button mockup. Never present an untested synthetic voice as a real curator.

Product and pricing pages were inspected on 2 October 2026. This is a fictional storytelling exercise, not evidence of audio quality, pronunciation, plan entitlement or museum use.

View ElevenLabs details · Open the related comparison table