Generate sounds · V6_MINI

It isn't a music node

Everything else in the Suno corner of this pack makes songs. KIE_Next_V6_MINI_beaa4b19 makes sounds - a whoosh, a sci-fi door, a riser, a rain loop, ambience - from a short text prompt, using Suno's sounds endpoint on the V6_MINI model. It sits in the KIE Next/Audio/Suno category rather than Music Generation for exactly that reason.

That makes it quietly one of the more useful nodes here if you're doing video. Sound design is the step where a decent generated clip stops feeling like a generated clip, and local options for it are thin - you're usually digging through sample packs or a subscription library. Being able to describe a sound and have it land in the same graph as your video generation, already an AUDIO object you can mux, skips a whole detour.

It's still an API call: the prompt goes to KIE, the sound is built on their hardware, and it costs credits.

How it behaves

The node builds the request from four small widgets, submits it to KIE's task queue, and then waits for it - polling at 2.5 seconds and backing off toward 15 until the job resolves or timeout_seconds runs out (1200 by default, up to 7200). The result is downloaded and decoded into ComfyUI's AUDIO format. The V6_MINI model version is sent inside the request body; KIE's market model name carries the outer envelope.

The inputs, all of them

The schema here is refreshingly small:

• prompt - required, and short: the tooltip states a hard limit of 500 characters. This is a sound description, not a song idea. "Metal shutter slamming twice, reverb tail, dirty recording" does more than a paragraph.
• sound_loop - true by default. On, you're asking for a cyclic sound, which is what you want for ambience or a bed that will be repeated in an edit. Off, you get a one-shot: a hit, a riser, a sting.
• sound_tempo - integer, 1 to 300, default 166. Set it to your edit's tempo and the sound sits in time; leave it and you're at the default's mood.
• sound_key - 24 options covering both majors and minors (D#m, Cm, C#m, Dm, Em, Fm, F#m, Gm, G#m, Am, A#m, Bm and the major equivalents), defaulting to D#m. Use it when the sound has pitched content - a riser or a tonal ambience - so it doesn't clash with the music underneath.
• grab_lyrics - true by default. The author's tooltip explains it as asking for lyric subtitles after completion; on a sound effect it's mostly inert. It exists because the endpoint shares plumbing with the music ones.

Outputs

audio - the generated sound as a native ComfyUI audio object, ready to save or mux. url - the first result URL. all_urls_json - every URL in the response, worth checking for alternates. task_id - the job handle. raw_json - the verbatim provider response, which is where a failure reason appears. credits_consumed and credits_left - the cost of the run, also painted on the node.

Install

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

Restart ComfyUI afterwards. Dependencies are requests, Pillow, numpy and PyYAML; there are no model files, no downloads and nothing touching your GPU. Windows users can run the one-click installer from the repo's dist/ folder. ComfyUI Manager support is where the project is headed (registry metadata is already in the repo under publisher felipeosilva) - until then, the clone is the dependable install.

Then set the key once in Settings → KIE.ai Nodes Next → Connection. The pack verifies it against KIE before replacing a saved key and shows a masked fingerprint instead of the key itself; KIE_API_KEY in the environment takes priority on servers.

Practical notes

Descriptions that read like music requests tend to come back as music, which is the one thing this node can't give you - if you want a track, use the Generate Music nodes. Prompts over the 500-character ceiling are a provider-side problem, so keep them tight. And if a run times out, don't instantly re-queue: the job is often still finishing, and KIE • Task Status or KIE • Wait for Task on the task_id will tell you before you spend credits on a duplicate.