Generate sounds · V6_WILD

Sound design, not songwriting

KIE_Next_V6_WILD_33453af5 generates individual sounds from a short text description - hits, risers, ambience, drones, machine noises - using Suno's sounds endpoint on the V6_WILD model. KIE bills that model as "pushing creative boundaries," and for sound design that's the interesting half: WILD is more likely to give you a texture that doesn't sound like a stock library entry, which is precisely what you want when the video is strange and a clean whoosh would ruin it.

If your workflow ends in a clip that needs a soundtrack, this is the node that fills the gaps the music didn't: the impact when a cut lands, the hum under a night scene, the metallic scrape before a reveal. It returns ComfyUI audio, so it muxes straight into whatever you're building.

And, like every node in this pack, it's a paid API call rather than a local model - KIE does the synthesis, and the prompt leaves your machine.

Mechanism, in one paragraph

The node assembles the request from four widgets, submits it to KIE's task API, then polls for the result itself: 2.5 seconds at first, backing off toward 15, bounded by timeout_seconds (1200 by default, up to 7200). When the task succeeds, the audio is downloaded and decoded into ComfyUI's standard AUDIO dict. V6_WILD rides in the request payload while KIE's market model name carries the outer envelope - a split earlier builds had backwards, which is the thing to suspect if a submission is rejected on model grounds.

The whole input surface

• prompt - required, maximum 500 characters per the tooltip. Write it like a sound, with material and motion: "granular metallic drone, slow rising pitch, cold room." Vague poetry gets vague output.
• sound_loop - default true, meaning a cyclic sound. Keep it on for beds and ambience you'll repeat; turn it off for one-shots that need a beginning and an end.
• sound_tempo - 1 to 300, default 166. Matching your edit's BPM is the difference between a sound that sits in the groove and one that fights it.
• sound_key - 24 choices spanning majors and minors (default D#m), and worth setting for anything tonal. A riser in the wrong key is a riser your audience will hear as a mistake.
• grab_lyrics - default true, described in the tooltip as fetching lyric subtitles after completion. On an SFX request it does nothing you'll notice; it's inherited from the shared endpoint.

Outputs

audio is the sound as native ComfyUI audio - Save Audio, or into a mux. url is the first result URL. all_urls_json holds every URL returned, so check for alternates before re-running. task_id identifies the job, raw_json is the full provider response, and credits_consumed / credits_left report the spend on the node itself.

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. The requirements are requests, Pillow, numpy and PyYAML - nothing heavy, no checkpoints, no VRAM. A one-click Windows installer is in the repo's dist/ folder if you'd rather avoid a terminal. The repo ships Comfy registry metadata under publisher felipeosilva, so Manager support is where this is going; the git clone is what works today.

Then the key, pasted once at Settings → KIE.ai Nodes Next → Connection. It's checked against KIE before it overwrites a working key and never displayed in full; headless installs can set KIE_API_KEY, which takes priority.

Where it goes wrong

The classic miss is describing music instead of a sound - this node will happily hand you a few seconds of song, which is rarely what you wanted. Prompts past the 500-character limit are rejected at the provider. Very short prompts come back generic, and since it's WILD, "generic" has a wider definition: expect to run it twice and pick.

Provider-side failures print a reason into raw_json. Timed-out runs are frequently still alive upstream, so check the task_id with KIE • Task Status or KIE • Wait for Task rather than resubmitting, which charges you a second time for the same sound.