fishaudio/fish-speech · warning · HTTPException

Text is too long, max length is {app_state.max_text_length}

Error message

Text is too long, max length is {app_state.max_text_length}

What it means

Raised by the TTS endpoint when the submitted text exceeds the server-configured max_text_length limit. It is an HTTP 400 BAD_REQUEST returned before any model inference runs. The limit exists to bound GPU memory usage and generation time.

Source

Thrown at tools/server/views.py:160

            HTTPStatus.INTERNAL_SERVER_ERROR, content="Failed to decode tokens to audio"
        )


@routes.http.post("/v1/tts")
async def tts(req: Annotated[ServeTTSRequest, Body(exclusive=True)]):
    """
    Generate speech from text using TTS model.
    """
    try:
        # Get the model from the app
        app_state = request.app.state
        model_manager: ModelManager = app_state.model_manager
        engine = model_manager.tts_inference_engine
        sample_rate = engine.decoder_model.sample_rate

        # Check if the text is too long
        if app_state.max_text_length > 0 and len(req.text) > app_state.max_text_length:
            raise HTTPException(
                HTTPStatus.BAD_REQUEST,
                content=f"Text is too long, max length is {app_state.max_text_length}",
            )

        # Check if streaming is enabled
        if req.streaming and req.format != "wav":
            raise HTTPException(
                HTTPStatus.BAD_REQUEST,
                content="Streaming only supports WAV format",
            )

        # Perform TTS
        if req.streaming:
            return StreamResponse(
                iterable=inference_async(req, engine),
                headers={
                    "Content-Disposition": f"attachment; filename=audio.{req.format}",
                },

View on GitHub (pinned to befe400174)

Solutions

1. Split the text into chunks under max_text_length and synthesize each separately, concatenating the WAV/PCM output.
2. Increase max_text_length in the server configuration (e.g. config file or KOKORO_TTS_MAX_TEXT_LENGTH env var) if your hardware allows longer generations.
3. Set max_text_length to 0 to disable the limit entirely (only if you accept unbounded generation cost).

Example fix

# before
resp = requests.post(url, json={"text": long_article})

# after
for chunk in chunks(long_article, max_len - 1):
    resp = requests.post(url, json={"text": chunk})
    ...

max_len = get_server_config().max_text_length
text = "..."
assert max_len == 0 or len(text) <= max_len, f"text {len(text)} > {max_len}"

When it happens

Trigger: POST /v1/tts (or OpenAI-compatible /v1/audio/speech) with req.text longer than app_state.max_text_length (when max_text_length > 0). Long documents, unsubtitled full articles, or clients not splitting text trigger it.

Common situations: Default max_text_length too low for batch narration workloads; clients pasting entire articles; changing max_text_length config but not restarting; misunderstanding that the limit counts characters, not tokens.

Related errors

• Streaming only supports WAV format
• Number of prompt text ({len(prompt_text)}) and prompt audio
• Directory {path} does not exist.
• No audio generated, please check the input text.

AI-assisted analysis of fishaudio/fish-speech@befe400174 (2026-08-27). Data as JSON: /api/errors/4c9e02809bb76255. Report an issue: GitHub.