Comet 音乐

The CometAPI pack does images, video, text, and - a little out of left field - music. "Comet 音乐" is a Suno-style generation node: give it a song idea or lyrics and it returns an audio clip you can listen to and save. It's the odd node in the family because it's the only one wired to a single channel (runninghub is the only channel choice), and the models are all Suno versions (v5.5, v5, v4.5 in the dropdown).

If you already run the pack for images, this is a free extra creative tool - but it's not a "compose the whole track" button. Suno-class models generate songs, complete with vocals, from a prompt; they don't give you stems or a mix you can rearrange. Know what you're getting into.

The inputs that matter

The node has two mode switches that shape everything else:

• music_mode - 生成歌曲 (generate a song) vs 生成歌词 (generate lyrics only). The second one is text-only output, which is a genuinely useful way to iterate on words before spending a generation on audio.
• music_submode - 灵感模式 (inspiration mode: describe the vibe, the model writes the lyrics), 自定义歌词 (you provide lyrics), 纯音乐 (instrumental only), or 生成歌词. The submode you pick is what makes "I have lyrics already" vs "I just have a mood" both workable.

Then the actual creative fields:

• prompt - lyrics, a lyric theme, or a purely instrumental description. This is the main thing you'll write.
• title - the song title.
• tags - style tags like pop, electronic. These matter more than most people think; Suno-class models lean heavily on them for genre.
• gpt_description_prompt - a longer description of inspiration, style, mood, or scene. Fill this when you're in inspiration mode.
• vocal_gender - 自动/男声/女声 (auto/male/female). Only takes effect for song generation.

Outputs are audio (the AUDIO asset, which the 音频卡片 node can save and display) and summary (a STRING with the run's task ID, mode, model, title, output path, and - if you generated lyrics - the lyrics themselves). The summary is your audit trail, so read it.

How it works

The execute() path normalizes your mode/submode pair (some combinations are collapsed - e.g. selecting 生成歌词 as the main mode forces the lyrics submode), sends the job to the runninghub channel, waits for it, downloads the resulting audio to your output folder, and returns the asset. If you only asked for lyrics, the lyrics come back in the summary instead of as an audio file. The node handles the "generate lyrics, then generate song from those lyrics" hand-off as two separate runs rather than one magic call.

Install and caveats

Install with the whole pack - ComfyUI Manager (search "ComfyUI-CometAPI") or git clone https://github.com/jieg9341-lab/ComfyUI-CometAPI into custom_nodes, restart - and set a runninghub key in the 设置中心. Dependencies are just requests and aiohttp; there are no local models.

Where people get burned:

• One channel, one key. This node can't fall back to another reseller if runninghub has a bad day. If it errors, check that channel's status first.
• Generation is billed and non-refundable. A song that comes back "close but not right" still cost a generation. Iterate cheaply in 生成歌词 mode first, then spend generations on audio.
• The audio lands in ComfyUI/output (the pack's cometapi_assets area) - make sure your output folder is where you expect, or a day of song runs becomes a scavenger hunt.
• Copyright reality check: you're generating with a model that was trained on other people's music. For personal fun, fine. For release, read the platform's terms - this pack's non-commercial license applies to the software, but the generated song's terms are the provider's, and they're not nothing.