← All tools

Plain Captions

Clear, plain-language captions for everyone.

Make any audio or video easier to follow: on-device speech-to-text plus a gentle rewrite into short, plain captions. Nothing is uploaded.

What it runs on

Base model

gemma-class Apache‑2.0

Inference

pipeline

Input

Audio (or an existing transcript). (audio)

Output

Short, plain-language caption lines. (captions)

Author

downlow community

On the map

Offer captioning help to community venues and events listed on the map.

The structure

1. On-device speech-to-text (Whisper) → raw transcript
2. Gemma rewrites the transcript into short plain-language caption lines

The proven prompt

This is the exact prompt the tool uses. It's open — read it, fork it, improve it. Honesty is built in: the prompt is told never to invent facts.

You are Plain Captions. You receive a raw speech transcript.
Rewrite it into short caption lines that are easy to read:
- keep the speaker's meaning exactly; never add facts or opinions;
- prefer short, common words and short lines (max ~7 words per line);
- keep names, numbers and quotes verbatim;
- if a passage is unclear in the transcript, keep it as-is and mark it [unclear] rather than guessing.

Proof it works

A tool earns trust by passing its own golden examples — inputs with a checkable expectation. These are how we move a tool from community to verified.

Subsequently, the committee deliberated extensively regarding the proposed budgetary allocation of twelve thousand dollars.

And then he said [inaudible] about the contract.

Don't have a model yet? →