🎹 ACE-Step 1.5 高级流派融合

Most music prompt nodes give you one style box. This one assumes you want the song to change personality halfway through - rap verse into an opera chorus, trap beat into a piano ballad - and it writes the caption that tells ACE-Step to do exactly that.

What it is, and why you'd want it

ACE-Step is the open-weights music model everyone actually runs: 1.5 is MIT-licensed and runs under 4GB of VRAM, and the honest community split is that instrumentals are genuinely compelling while vocals and lyrics are the weak suit. You prompt it with a caption (style/tags, in English) plus separate lyrics.

Standard captioning is a list. 🎹 ACE-Step 1.5 高级流派融合 refuses to do lists. It writes a structural caption: an overall base genre, then a verse clause, then a connector, then a chorus clause. The author's own framing is that this is the difference between "整体风格" (one overall style) and per-section style. That's a real capability - section-aware genre contrast is one of the few musical things a text caption can actually steer, and it's how you get a track that reads as a fusion rather than a mush.

Where it fits: it replaces the system-instruction/caption half of your music graph. The pack's sibling nodes (🎵 ACE-Step 1.5 音乐提示词 (手动) and (预设)) build the rest of the prompt; this one is the one you swap in when you don't want a single genre.

How it actually builds the string

The node maps your Chinese dropdown choices to English tags via an internal table, then assembles one sentence:

A experimental C-Pop track. The verse features aggressive, fast-paced Hip-Hop elements,
transitioning into a grand, emotional, soaring Opera chorus. The production creates a unique
fusion of styles.

(Yep, the template opens with "A experimental." ACE-Step does not care. It's a music model, not a copy editor.)

That's the whole mechanism, and it explains the dropdowns: you're not writing prose, you're picking from a fixed vocabulary of genre tokens the model was trained on. The output is English even though the UI is Chinese, which is deliberate.

The inputs that matter

Six inputs, all required, so there's no "leave it blank" escape hatch here:

• 整体基调 - the overall base genre, 20 choices, defaults to C-Pop (华语流行). This is the frame everything else sits in.
• 主歌风格 / 副歌风格 - the verse and chorus genres. Both lists offer the same 20 genres and both exclude "无" and "标准完整版", so you can't opt out of either half. Defaults are 嘻哈/说唱 (Hip-Hop/Rap) and 歌剧 (Opera) - the node ships demonstrating its own trick.
• 主歌形容词 / 副歌形容词 - free-text English descriptors. Defaults are aggressive, fast-paced and grand, emotional, soaring. These are in English because they get dropped into the English sentence verbatim; type Chinese here and you're feeding the caption raw Chinese adjectives.
• 融合连接词 - the six ways the two halves meet: smooth transition, blended, suddenly switching, evolving into, clashing with, alternating. Pick 交替出现 (alternating with) and you get call-and-response rather than a one-way ramp.

One output: 结构化系统指令(Caption), a STRING. Wire it into the caption field of your ACE-Step node instead of typing a style string.

Install

It's a text node in a text-node pack - no model downloads, no checkpoints, nothing to put in models/. The pack's requirements.txt is a single line (torch>=2.0.0), which you already have if ComfyUI runs.

ComfyUI Manager: search prompt-formula and install it. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/a63976659/ComfyUI-prompt-formula.git

Restart ComfyUI. You'll find it under the 📕提示词公式 category (right-click → Add Node, or hit N for the node library). The pack is Chinese-first - the author's stated goal is that you don't need a translation patch to read it - so expect Chinese labels throughout, which is why these articles name the display emoji as well as the class.

Gotchas

Picking a genre is mandatory on both halves. There's no "none" in the verse/chorus lists, unlike most other dropdowns in this pack. If you want a plain single-genre track, use the sibling 音乐提示词 node instead of this one - this node's whole reason to exist is the contrast.

Don't expect it to write a song. It produces roughly one sentence of caption. Lyrics, duration and everything else live in the other ACE-Step nodes. People fire this up, get three lines back and think it's broken; it isn't, it's a caption fragment.

It's a caption, not a guarantee. ACE-Step follows structure loosely and vocals are its weak suit, so a verse/chorus flip lands as a vibe shift more than a clean section boundary. Instrumental fusion - Chinese traditional into cinematic, glitch into orchestral - is where this node punches hardest.

A caveat about ComfyUI-side ACE-Step generally. The community's recurring complaint is that the native ComfyUI nodes lag behind ACE-Step's own Gradio UI in feature coverage. Prompt quality helps, but it can't add features the node doesn't expose.