chatterbox-direct-input-splitting

name	chatterbox-direct-input-splitting
description	Prepare direct user-provided text for this Chatterbox project by saving it into `data/` as one or more `.txt` files and enforcing the project's hard `800`-token limit per generated audio file. Use when the user pastes narration, story text, scripts, or other audio content directly into chat and wants input files or audio generation.

Chatterbox Direct Input Splitting

Instructions

When the user provides text directly in chat for this repository:

1. Treat 800 tokens as a hard maximum for each generated audio file.
2. Never put a pasted script that may exceed that limit into a single .txt file.
3. Split long text into multiple sequential .txt files under data/ before running generation.
4. Prefer splits at paragraph, sentence, or dialogue boundaries. If needed, use smaller clause-level breaks to stay under the limit.
5. If exact token counting is unavailable, split conservatively. It is better to create more files than to risk exceeding the limit.

File creation rules

• Put generated text input files under data/.
• Preserve any user intent about naming.
• If the user gives a base name, use ordered suffixes such as story-1.txt, story-2.txt, story-3.txt.
• If matching an existing numbered pattern, continue that pattern consistently.
• Keep chunk order obvious from filenames.

Working rule

Before generating audio from direct chat input, first decide whether the text needs splitting. If there is any reasonable chance a single file would cross the 800-token limit, split it immediately instead of asking for confirmation unless the user explicitly asked for a different structure.

Output expectations

When you create split files from direct input:

• Mention that the text was split to respect the hard 800-token-per-audio limit.
• List the created files briefly.
• Preserve the original text content as closely as possible aside from safe chunk boundaries.

Example

User provides a long horror script directly in chat and asks to generate audio.

Expected behavior:

1. Save the text into multiple files such as data/horror-1.txt, data/horror-2.txt, data/horror-3.txt.
2. Keep each chunk safely below the hard 800-token limit.
3. Only then proceed with the normal generation workflow.