Flow Music 音乐生成与处理（9 合 1）

This is the whole Flow Music API collapsed into one ComfyUI node: generate a track, write lyrics, upload your own audio, extend it, replace a section, make a cover, split stems, download, and render a music video - nine operations behind a single operation dropdown. If you've used the pack's other 31-in-1 Suno node, this is its leaner cousin; the node hides every control that doesn't apply to the action you picked, so the interface stays honest about what it needs.

Like everything in T8mars/ComfyUI_Seedance, it's an API wrapper. No local generation, no model files - a key from api.seedance.nz and an internet connection are the whole setup.

How it works

Each operation maps to a documented Flow Music action. The node runs a field whitelist per action, so a hidden field never leaks into the request - switch operations and the UI collapses to just the relevant controls. Music generation takes either a sound_prompt (style/sound description) or lyrics (or both); lyric generation takes a prompt up to 3000 characters. The genuinely neat bit is chaining: upload a local AUDIO (or paste a public audio_url), and the node hands back a clip_id you can wire straight into the next node's clip_id input for extend/replace/cover/stems/download/video operations. No external storage, no file juggling.

There's a version selector too. lyria-3.5 is only documented for generation, extend, replace and cover - set it on other actions and you're sending a field upstream won't read.

The inputs that matter

• operation - the nine actions: generation, lyrics, upload-audio, extend, replace, cover, stems, download-audio, video-clip. This is the steering wheel; everything else is contextual.
• sound_prompt / lyrics / prompt - the creative core, depending on operation.
• bpm (min 1 - the API refuses 0), length (1–240s).
• clip_id - chain output from a previous Flow Music node.
• extend_from_s / extend_s - where extension starts in the source, and its duration (max 164s).
• start_s / end_s / instruction - the replacement range and what to replace it with.
• strength (0–1) - cover edit intensity. format - downloaded audio format. preset - the video template for video-clip.

Outputs are the full spread: audio1/audio2 (decoded AUDIO), video (for the video-clip action), text, the reusable clip_id, primary_url/result_urls, primary_path/result_paths, task_id and response.

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, then connect a Seedance API Config node (or set SEEDANCE_API_KEY). The pack's only dependency is requests.

Common issues

• Wrong operation, fields ignored - you're filling in controls the selected action doesn't send. Switch operation and watch the node hide them.
• clip_id not connecting - it's a STRING output; you can wire it directly into the next node's clip_id input, which is how every example workflow in the pack does it.
• bpm error - set to at least 1.
• Long generation feels stuck - music jobs genuinely take a while; the node polls with a generous timeout (up to 1800s), so a 240-second track at high fidelity isn't a hang.

One note: audio decode in this pack prefers torchaudio but falls back to FFmpeg (found via SEEDANCE_FFMPEG, PATH, or the portable build). If stems come back as empty paths, the result URL probably expired mid-download - re-run; the pack keeps URL ordering and only fails the whole job when every artifact is gone.