Add Instrumental to Music · V6

Hand it a song, describe the style, get an instrumental back. This is Suno's "Add Instrumental to Music" endpoint on V6, the current tier - the sibling of the V5 and V5.5 versions, and the one to use for new work now that KIE's docs mark the older tiers discontinued.

Be clear about what it does, because the name misleads. This is not stem separation. It doesn't split your file into vocal and instrumental halves. It takes your source audio as a reference and generates an instrumental performance guided by it, steered by your tags, negative_tags and style_weight. Sometimes that beats separation - a real mix, no watery artifacts. Sometimes it's worse - the arrangement drifts from what you wrote. If you want the actual instrumental of a track you own, use a separation tool first, it's local and free. Come here when you want a re-performed backing in a style you choose.

Worth noting the local economics: ACE-Step makes instrumental music for free on modest hardware. Instrumental is the part of the job where local already wins. Reach for this node when the backing needs to match an existing song's character, or when you want the result to sit in the same family as other Suno output.

How it works

The pack generates its node classes from KIE's official API documentation at startup, one per documented model version, so V6 is baked in - no model dropdown. The node assembles the documented request body, submits it to KIE's task endpoint (envelope model ai-music-api/add-instrumental, inner model V6), polls KIE's record-info endpoint every couple of seconds with backoff until the task succeeds, then downloads the resulting mp3 and returns a native ComfyUI AUDIO. Failed tasks raise with the task id and the provider's message.

The inputs that matter

Four are required, and the naming is a trap for anyone coming from its siblings - this endpoint calls the style field tags:

• upload_url - the source audio, as a publicly reachable URL. The docs: source audio plus title, negative_tags and tags is enough to generate. Upload-family limit: under 8 minutes.
• title - required; appears in the player and the filename.
• tags - what to include, e.g. relaxing, piano, soothing.
• negative_tags - what to avoid, e.g. heavy metal, fast drums.

Optional: vocal_gender (schema-complete, practically pointless when the target is instrumental), style_weight (0–1, default 0.61 - how hard the tags pull), weirdness_constraint (default 0.72), audio_weight (how much the source audio counts), lyrics (optional; the source is enough on its own), variety (0–4, per the docs' own ladder from "strictly the same" to "unreasonably varied"), timeout_seconds (1200) and callback_url (leave empty).

Outputs: audio - wire it into Save Audio or anything downstream - plus url, all_urls_json, task_id, raw_json, credits_consumed and credits_left.

Getting your song in

upload_url is a string widget, not an AUDIO socket, so a native audio cable won't attach. Add KIE • Upload Audio from KIE Next/Utility/Media, connect your AUDIO to it, then right-click upload_url on this node, choose Convert widget to input, and wire the URL string across. Two nodes, once, and the chain runs unattended after that.

Installing

ComfyUI Manager, search "KIE Nodes Next". The README describes the registry listing as prepared rather than published, so if Manager comes up empty:

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

Restart, then set the key once under Settings → KIE.ai Nodes Next → Connection. It's validated before overwriting a working key, only a masked fingerprint is shown, and KIE_API_KEY takes priority for headless installs. That's the whole setup - four small Python packages, no torch, no checkpoints, no VRAM.

Where people get burned

• Hunting for a style field. It's tags on this endpoint. Same job; the docs are just inconsistent across the family.
• Expecting stem separation. It generates; it doesn't isolate.
• Local paths as upload_url. KIE needs to fetch it. Upload first.
• No key configured throws an explicit error naming the settings page - always the first check.
• Mixed-release installs are refused at import. If the pack's version and its generated-node build disagree, delete ComfyUI-KIE-Nodes-Next and reinstall.
• Credits, every attempt. credits_consumed and credits_left print on the node so iterating is visible rather than surprising.

Spend your tuning effort on tags before style_weight. Concrete tags - lead instrument, genre, mood, a tempo word - move the result far more than nudging the adherence dial with a vague list.