音频｜Suno 音乐生成

Suno is the music-generator that basically defined AI songwriting for everyone, and the Tikpan Suno node is a full wrapper around it - write a lyric, get back a finished song. It's the audio wing of the Tikpan Pro pack, and it's genuinely different in kind from the image/video nodes: instead of one generation call you get a produce-a-song pipeline with four modes, and every run returns two versions to choose from, plus audio streams you can wire straight into ComfyUI's AUDIO type.

Reach for it when your workflow needs an original soundtrack - background music for a product video, a branded jingle, a loop for a social cut. The pack is aimed at TikTok e-commerce, and short-video background music is exactly the slot this fills.

The unavoidable caveat, same as every node in this pack: it's an API wrapper, metered per call, and your lyrics/prompt leave the machine to Tikpan's servers. Music generation is also slow - you're waiting on a cloud render, not a local sampler (external-api-nodes.md).

How it works

Four 生成模式 (generation modes), chosen by the dropdown:

• 灵感模式 - you write creative direction, Suno writes the song (lyrics auto-generated).
• 自定义模式 - you write the complete lyrics yourself.
• 续写模式 - extends an existing song; needs the 🎤_续写_歌曲ID and optionally the ⏱️_续写起始秒数 (from which second to continue).
• 歌手风格 - mimics a specific artist's persona; needs 🎭_歌手风格_PersonaID or 🎙️_歌手风格_参考音频ID.

Model choice: 模型版本 runs from chirp-v3-0 up through V4, V5, and Fenix/Auk experimental builds. The tooltip is a decent guide - V5/Fenix for latest quality, V4 for stability, V3 to match old works. 风格预设 (16 presets like Pop, Cinematic, EDM, Hip-Hop) sets the genre; 风格标签 adds free-form English keywords; 负面风格标签 excludes things you don't want; 生成纯音乐 (default off) drops the vocals for instrumental-only output.

Under the hood it submits a task to Tikpan's Suno endpoint, polls suno/fetch/{task_id} until both versions render, then downloads them locally and decodes them into AUDIO outputs. Advanced Suno parameters (vocal gender, auto-lyrics, style weight, creativity) are hidden behind 发送高级Suno参数 - off by default to keep the old stable payload, on when you want fine control.

Inputs and outputs that matter

A beginner's first run: pick 自定义模式, write real lyrics in 创作提示词, choose a 风格预设, hit run. Set API_密钥 first. The outputs are generous - two pairs of 📁 audio paths and 🔗 audio links, the 创作提示词, task ID, status, clip IDs, tags, title, and two 🎧音频流 outputs of type AUDIO. Wire an audio stream into a save-audio or a video-track node; or just read the file path.

Installing

Pack install as usual - Manager search "Tikpan", or git clone https://github.com/htrert/ComfyUI-Tikpan-Pro into custom_nodes, restart. Key from tikpan.com. No local model, no GPU.

Common issues

• 续写/歌手风格 modes silently no-op. They need their respective IDs. No ID, no extension - the node returns empty paths with a log line. Fill the required fields for the mode you chose.
• Renders take a while. Two versions of a full song is a real cloud job. Don't kill the run because the progress bar crawls; check the log's task ID and use the Task Fetcher node if you need to collect it later.
• Lyric language mismatch. If you write Chinese lyrics but the model picks an English vocalist, you get mispronounced vocals. 风格预设 with a language-relevant preset (e.g. Chinese Pop) helps; the vocal-gender advanced field helps too.
• Cost. Music generation is among the pricier API calls in this pack. Every run makes two versions - that's part of why it costs what it costs. Check the meter before a batch of tracks.