MiniMax 音乐/语音/声音克隆（4 合 1）

MiniMax is one of the bigger Chinese AI labs, and its audio side covers three genuinely different jobs - music generation, speech synthesis (HD and Turbo tiers), and voice cloning - plus this node's fourth trick: it hands you all three behind a single model dropdown, and the UI hides whatever doesn't apply. One node, four jobs, no audio-engineering degree required. It's the most flexible audio node in the T8mars/ComfyUI_Seedance pack, and the one you'll want if your workflow touches both music and voice.

As ever with this pack: API wrapper, key from api.seedance.nz, requests only, nothing runs locally.

How it works

The model choice selects the path, and the required inputs change meaning accordingly:

• Music - prompt describes the style; lyrics takes lyrics with optional structure tags. is_instrumental makes it a pure instrumental (lyrics are then omitted), and lyrics_optimizer tells the model to write lyrics from the prompt when lyrics is empty.
• Speech (HD / Turbo) - prompt is the text to read, voice_id picks a system or cloned voice, and you get speed, volume, pitch, language_boost and channel (mono/stereo).
• Voice Clone - custom_voice_id names your new voice (8–256 chars, must start with a letter), clone_target_model picks the target speech model, and the cleanup toggles (need_noise_reduction, need_volume_normalization) are here. This is the only path that reads the optional reference_audio input - one clip, 10 seconds to 5 minutes.

Shared technicals: output_format, sample_rate, bitrate. The node uploads, submits, polls, downloads and decodes to ComfyUI AUDIO (1 second of silence on skip_error, so batch pipelines keep running).

The inputs that matter

For a beginner, the practical trio is model (which job are you doing), prompt (style, text, or preview text), and - for cloning - reference_audio + custom_voice_id. Everything else is either per-mode polish or technical output format.

Outputs: audio (AUDIO), audio_url, audio_path, result_text, task_id, response. The result_text output is handy for music - it can carry lyrics or generation notes.

Installing it

ComfyUI Manager → search "ComfyUI Seedance", or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/ComfyUI_Seedance.git
cd ../..
python -m pip install -r custom_nodes/ComfyUI_Seedance/requirements.txt

Restart, key from api.seedance.nz/console, Seedance API Config into api_config.

Common issues

• Music rejects your lyrics - if is_instrumental is on, lyrics are deliberately omitted. Turn it off to use lyrics.
• Clone fails validation - custom_voice_id must be 8–256 chars starting with a letter, and reference_audio must be 10s–5min. Short clips get refused.
• Tweaking speed does nothing on music - speech-only control; the node shows/hides fields by mode, but if you switched models mid-workflow the leftover value can confuse. Re-check which model you're on.
• Output format surprises - output_format/sample_rate/bitrate apply across modes; if your downstream audio tool chokes, you usually picked an odd combo. mp3 with a sane sample rate is the safest default.

Voice cloning is the feature people actually come for, and it's genuinely good - but remember the safety frame: a cloned voice leaves your machine and lands on the gateway's servers. Don't run someone's voice without permission; this is exactly the tool that makes that obvious and effortless.