KIE Suno Music (Advanced)

Most ComfyUI workflows end at pixels. This node is the one that makes sound. KIE Suno Music (Advanced) calls the Suno music-generation API through Kie.ai, and returns not one but two finished tracks plus two cover images per run. If you're building music videos, training-shot audio beds, or just want B-roll that isn't silent, it's the oddest and most fun node in this pack - and the only one whose output is genuinely hard to produce locally.

Advanced here means one thing over the Basic variant: it exposes Suno's three creative steering weights. Same model lineup, same outputs, just more knobs.

How it works

Different backend from the image/video nodes - Suno on KIE uses the /api/v1/generate endpoint rather than the createTask job flow, then polls /generate/record-info until the track reaches SUCCESS. The pack polls through PENDING, TEXT_SUCCESS, and FIRST_SUCCESS states, so you can watch the lyrics get written before the audio lands. Each run returns two songs and two cover images; they come back as ComfyUI AUDIO and IMAGE outputs.

The inputs that matter

• title, style, prompt - the creative core. In custom mode (the default), title and style are required, and the prompt is where you put lyrics when instrumental is off. Style is where Suno's character lives: "slow-burn 80s synth ballad, analog tape warmth" goes a long way.
• instrumental - defaults to true. Toggle off to get vocals (then the prompt becomes your lyrics).
• model - V4_5 by default, with V4, V4_5PLUS, V4_5ALL, and V5 in the dropdown. Prompt-length limits shift with the model (V4 caps at 3000 chars, the newer ones at 5000).
• The three weights (the whole reason to use Advanced over Basic): style_weight, weirdness_constraint, audio_weight - all 0 to 1, all defaulting to 0.65. Style weight pins the vibe to your style text, weirdness lets the model wander (or not), audio weight trades lyrical control against sonic quality. It's Suno's own balance knobs exposed directly; start at the defaults and nudge one at a time.
• vocal_gender - male/female (custom mode only). negative_tags - tags to steer away from.

Outputs: audio_1, audio_2 (both AUDIO), image_1, image_2 (covers, IMAGE), and data (the full record JSON - handy for metadata or logging).

Installing

One install for the whole pack. ComfyUI Manager: search "ComfyUI Kie API". Or:

cd ComfyUI/custom_nodes
git clone https://github.com/gateway/ComfyUI-Kie-API
# restart ComfyUI

No models to download. API key goes in config/kie_key.txt inside the pack folder (copy from kie_key.example.txt).

Where people get burned

Suno generation is slow and the node polls patiently - don't judge a run dead at two minutes. The bigger gotcha is cost: you get two tracks per generation, and each is a separate credit charge, so a "one song" run is secretly two. Iterating on lyrics burns through a balance fast; nail the title/style first, then spend. Suno also has a sensitive-word failure state (SENSITIVE_WORD_ERROR) that rejects lyrics outright - if a run fails with that, it's the prompt, not the node. And if you're feeding the AUDIO output anywhere unusual, remember it's a real audio object, not a filename - wire it to a node that accepts AUDIO.