What causes "ElevenLabs returned audio that FFmpeg cannot decode" in harry0703/MoneyPrinterTurbo?

Manually inspect the temp audio bytes (the finally block deletes it, so reproduce with a raw curl to /v1/music/video-to-music) to see whether it is JSON/HTML instead of audio

harry0703/MoneyPrinterTurbo · error · ElevenLabsMusicError

ElevenLabs returned audio that FFmpeg cannot decode

Error message

ElevenLabs returned audio that FFmpeg cannot decode

What it means

Raised after a successful download when the audio ElevenLabs returned fails the local FFmpeg decode check (bgm_service.validate_audio_file with a 120s timeout). It means the HTTP call succeeded but the bytes on disk are not a decodable audio file — e.g. a JSON error body saved with 200, a truncated stream, or an unsupported codec.

Source

Thrown at app/services/elevenlabs_music.py:349

                with response:
                    if not response.ok:
                        raise ElevenLabsMusicError(
                            "ElevenLabs generation failed "
                            f"({response.status_code}): "
                            f"{_safe_response_error(response)}"
                        )
                    total_bytes = _stream_audio(response, temp_audio_path)
        except requests.RequestException as exc:
            # 下载阶段断线也属于请求失败，必须进入任务降级逻辑，不能留下半条
            # 音频或让已经生成的视频因为第三方网络波动整体失败。
            raise ElevenLabsMusicError(
                f"failed to request ElevenLabs music: {exc}"
            ) from exc

        try:
            bgm_service.validate_audio_file(temp_audio_path, timeout_seconds=120)
        except (bgm_service.BgmUploadError, bgm_service.BgmServiceError) as exc:
            raise ElevenLabsMusicError(
                "ElevenLabs returned audio that FFmpeg cannot decode"
            ) from exc
        os.replace(temp_audio_path, output_path)
        temp_audio_path = ""
        logger.info(
            "ElevenLabs background music generated: "
            f"output={output_path}, size={total_bytes} bytes"
        )
        return output_path
    finally:
        _remove_file(temp_audio_path)


def generate_bgm(
    video_path: str,
    output_path: str,
    video_duration: float,
    prompt: str = "",

View on GitHub (pinned to 1f9f19c202)

Solutions

Manually inspect the temp audio bytes (the finally block deletes it, so reproduce with a raw curl to /v1/music/video-to-music) to see whether it is JSON/HTML instead of audio
Confirm ffmpeg is installed and on PATH — a missing FFmpeg makes every validation fail regardless of the audio
Retry the request; truncated streams are usually transient server-side or network issues
If ElevenLabs changed its output format, pin/switch music_model_id between music_v1 and music_v2 and retest

Example fix

# before
bgm_service.validate_audio_file(temp_audio_path, timeout_seconds=120)  # opaque failure

# after
# capture format info for diagnosis before failing the task
probe = subprocess.run(
    ["ffprobe", "-v", "error", "-show_format", temp_audio_path],
    capture_output=True, text=True, timeout=30,
)
logger.warning(f"elevenlabs audio probe: rc={probe.returncode} err={probe.stderr[:200]}")
bgm_service.validate_audio_file(temp_audio_path, timeout_seconds=120)

Defensive patterns

Strategy: validation

Validate before calling

import shutil
assert shutil.which("ffmpeg") and shutil.which("ffprobe"), "ffmpeg/ffprobe required"

Try / catch

try:
    elm.generate_bgm(video, out, duration, prompt)
except elm.ElevenLabsMusicError as e:
    if "cannot decode" in str(e):
        log_probe_details(video)  # ffprobe the raw response once, then give up BGM
    raise_task_degradation()

Prevention

Verify ffmpeg/ffprobe are installed and on PATH in the deployment image
Retry once — truncated audio streams are often transient
Pin music_model_id to a known-good value when ElevenLabs ships format changes

When it happens

Trigger: _stream_audio wrote fewer bytes than expected (connection cut but requests did not raise), ElevenLabs returned an error payload with a 200 status, or the returned audio codec/container is not decodable by the bundled FFmpeg build. validate_audio_file then exits non-zero and raises BgmUploadError/BgmServiceError.

Common situations: MITM proxies that rewrite responses, partial writes when the process is SIGSTOPed mid-download, ElevenLabs serving a new audio format after an API update, or an FFmpeg binary missing from PATH so validation fails for every file.

Related errors

AI-assisted analysis of harry0703/MoneyPrinterTurbo@1f9f19c202 (2026-08-14). Data as JSON: /api/errors/b73a092125f84e75. Report an issue: GitHub.