Boost Music Style

Writing music style prompts is the same craft as writing image prompts, minus every visual intuition you've built. "Pop, Mysterious" is what most people manage, and it's why their Suno tracks all sound like the same generic pop with a reverb pedal on it. This node wraps KIE's Suno "boost style" endpoint: give it a short style description, get an expanded one back that you can reuse, edit, or hand to the actual music generation node.

It's the smallest, most single-purpose node in the pack, and it's the one people skip straight past - which is a mistake if you care about a recurring sound, because consistency in music lives in the style string much more than in the seed.

Inputs and outputs

One required input. content is a STRING with a default of Pop, Mysterious and the author's tooltip spells out the intent: describe the style you want in concise, clear language. Not a paragraph, not a mood board - a few tags. Comma-separated descriptors behave best because that's what the style field downstream expects.

Then the pack's standard optional plumbing: timeout_seconds (30–7200, default 1200) and callback_url, which you leave empty for normal ComfyUI use. It's there for servers that can receive KIE's webhook instead of polling.

Outputs are the audio-family set: url, all_urls_json, task_id, raw_json, plus credits_consumed and credits_left, and an audio socket. One honest note on that audio socket - a style-boosting call returns text, not a track, so treat the returned style string as living in raw_json and don't wire the audio output into a player expecting a song. It exists because the node is filed under the audio family and generated nodes share a family-level output signature.

In practice: run it, copy the boosted style out of raw_json, paste that into the prompt/style widget of a Suno Generate Music node in the same pack. Keep it in a text file if the track is part of a series - that's your house style, and it beats re-inventing descriptors every session.

Why this is worth the credits

Local music generation closed most of the gap in 2026 - ACE-Step is genuinely the open Suno, and on a decent card it does minutes of audio in seconds with LoRAs and an XL variant on top. What it still doesn't match is the hosted model's vocal quality, which is the same complaint people level at Suno itself. So the honest position is: if you're doing instrumental beds for video, local is the sane default and this node is irrelevant. If you're chasing a vocal-forward track, you're already on the API, and this node is the cheap step that makes the expensive one come back right the first time.

And that's the actual economics. Style boosting is a text call; generation is the metered one. Spending a tiny amount to revise the prompt is a much better deal than running four full generations to discover your style description was thin.

Install

Manager → search KIE.ai Nodes Next (publisher id felipeosilva) → install. If Manager finds nothing - the README's registry publishing is still described as planned rather than live - clone it and install requirements yourself:

cd ComfyUI/custom_nodes
git clone https://github.com/felipederosilva/ComfyUI-KIE-Nodes-Next
pip install -r ComfyUI-KIE-Nodes-Next/requirements.txt

Restart ComfyUI. Requirements are requests, Pillow, numpy, PyYAML; there are no model files and no audio libraries to install, because all the inference is on KIE's side. Then set the key once: Settings → KIE.ai Nodes Next → Connection → Paste / replace KIE.ai API key. It's validated before it replaces a saved key, and only a masked fingerprint ever reaches the frontend. KIE_API_KEY in the environment wins on headless installs.

Gotchas

Node not in your menu? The Suno nodes come from the live catalog, so run KIE • Refresh Model Catalog - and if this specific endpoint is missing while other Suno operations are present, check it with KIE • Inspect API Definition (Advanced), since the pack only builds widgets from what KIE's docs publish.

Long-winded input, vaguer output. Tag soup with contradictions in it ("ambient, aggressive, lo-fi, orchestral") gets you a boosted version of the same contradiction. Two or three concrete descriptors beat eight aspirational ones.

This is a hosted call, so your prompt leaves your machine and the meter is running - small, but real, and it applies to every node in this pack. The corollary: it also can't be run offline, and if KIE's API is having a bad day you get an error rather than degraded output. The rest of the pack is built to survive transient trouble (generation results still return when the balance service is down), but the request itself needs the network.