YinChao Generate Music

This is the node the whole pack is named for. Type a vibe, optionally paste lyrics, and get a full song back as native ComfyUI AUDIO - no local model, no GPU, no stitching together submit/wait/download/convert nodes. If you've been wrestling with open music generators and their dependency stacks, this is the "just works" alternative, and the tradeoff is exactly what you'd expect from an API call.

What it actually is

A cloud song generator behind a ComfyUI face. The YinChao (音潮) platform runs the actual model on its own servers; this node submits your prompt/lyrics, polls the async task, downloads the MP3, and decodes it into the AUDIO contract ComfyUI expects. It's billed per song - the README quotes roughly ¥0.22 (about $0.03), which is genuinely cheap next to Suno credits. That's also the honest catch: every Run costs money, your text leaves the machine, and you get whatever the platform's model gives you. There are no weights to tune and no community bypasses when the output is mid.

The inputs that matter

• model - a combo dropdown, default v4.0, with v3.5 as the alternative. Start with v4.0; drop to v3.5 if you want the older, presumably cheaper-sounding generation for quick tests.
• prompt - optional multiline. Style and description (under 1000 characters). Think "lo-fi beat with vinyl crackle, 90 BPM."
• lyric - optional multiline. Paste in lyrics from YinChaoGenerateLyrics (the official combo workflow does exactly this) or write your own (up to 3000 characters).

Here's the one rule that trips people: you need at least one of prompt or lyric. Both empty is a hard error. Give it one and the platform fills in the other.

How it works

Internally it's the same pipeline as the pack's other song nodes: submit a normal task with the song count hard-coded to n=1 - a deliberate choice so one run can't trigger the platform's default of two generations and bill you twice - then poll every 3 seconds for up to 10 minutes, download, and decode. The model choice is the only thing you control; the billing and async handling are handled for you.

What comes out

Five outputs: audio (native AUDIO - wire it into any save/audio/video/subtitle node), title, lyric (the lyrics the platform settled on), audio_url (a direct hosted link to the MP3, useful for grabbing the file without ComfyUI), and task_id (the API receipt).

Install

Same pack as the other three YinChao nodes. Search "YinChao Music" in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/yinhcao/ComfyUI-YinChao
python -m pip install -r requirements.txt   # requests, av>=12, numpy

Restart ComfyUI. The av dependency is PyAV, which bundles codecs - no system ffmpeg required. The pack uses the V3 schema, so keep ComfyUI updated.

You need an API key from platform.yinchaoyongxian.com, set under Settings → YinChao → API Key (never as a node input, so it won't end up in your workflow JSON). Headless runs use the env var:

export YINCHAO_API_KEY="your-api-key"

Common issues

• "At least one of prompt or lyrics" - the most common beginner stumble; the node errors clearly.
• Missing key / no balance / moderation rejection all fail loudly with explicit messages. The pack never returns a fake-empty song.
• Task timeouts after 10 minutes happen on heavy jobs; the node surfaces them rather than hanging forever.
• Costs add up faster than you expect - that's the universal API-node lesson. A $0.03 song is nothing until you've generated forty of them iterating on a prompt. Use the free quota to test, then budget.
• Your prompt and lyrics go to a server you don't control. For your own material that's fine; it's why you shouldn't reach for this when the lyric content is sensitive.