Upload And Cover Audio · V5.5

Covers are the one music job where an API is still clearly the right answer. You hand over a track, say what style you want it in, and get a new performance back. This node is KIE's "Upload And Cover Audio" pinned to Suno V5.5.

The context matters here. The local music story in ComfyUI is ACE-Step, which is genuinely good at instrumental music and consistently weak at vocals and lyrics - the same complaint people raise about Suno, just with a lower ceiling. This node exists for the case where you need the vocal performance to hold up and you're willing to pay per call and send the source audio off your machine. For a handful of covers that's a fine trade. For a hundred, run the numbers against a local model first.

How it works

The pack generates one node class per model version from KIE's own API documentation at startup, so V5_5 is baked into this class and there's no model selector. The node assembles the documented request body, submits it to KIE's task endpoint - envelope model ai-music-api/upload-and-cover-audio, inner model V5_5 - and then polls KIE's record-info endpoint every couple of seconds (backing off toward 15) until the task completes. On success it downloads the returned mp3 and decodes it into ComfyUI's native AUDIO type; on failure it raises with the task id and the provider's message.

The inputs that matter

Three are required:

• upload_url - the source track to cover. A URL KIE can fetch, not a local path, and the docs ask that you keep source audio under 8 minutes. upload_url alone is enough to get a result.
• prompt - the docs are precise about a trap here: it's used as lyrics only when lyrics is absent. If you set lyrics, prompt stops being lyrics. And if instrumental is true, don't pass it at all.
• instrumental - defaults to true, which is the opposite of what most people expect from a "cover" node. True means no vocals: no lyrics, no prompt-as-lyrics, no vocal_gender. Set it to false when you actually want a sung cover.

After that: style (the genre, e.g. Classical, up to 1000 characters on this tier), title (up to 80 characters), negative_tags, vocal_gender (m or f - the docs admit it only nudges the probability), style_weight, weirdness_constraint, audio_weight, lyrics (takes priority over prompt), variety (0–4), duration (10–360 seconds; the docs bless it for V5.5 and V6, so it does apply here), plus persona_id / persona_model if you've got a Suno style or voice persona, and the usual timeout_seconds and callback_url.

Outputs: audio for downstream audio nodes, url, all_urls_json (covers come back with a source and a cover URL - this is where you keep both), task_id, raw_json, and credits_consumed / credits_left.

Getting a local file in

upload_url is a text widget, so you can't drag an AUDIO onto it. Use KIE • Upload Audio from KIE Next/Utility/Media: it takes an AUDIO and returns a public URL. Then right-click upload_url, pick Convert widget to input, and wire the string in. Same trick works for feeding the cover back around for a second pass.

Installing

ComfyUI Manager, search "KIE Nodes Next". The README still frames registry publication as prepared rather than shipped, so if it isn't in Manager yet:

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 in Settings → KIE.ai Nodes Next → Connection. Only a masked fingerprint is shown, the key is validated before it overwrites a working one, and it never ends up in your workflow JSON. On a headless box, KIE_API_KEY takes priority. Four small Python dependencies, no torch, no checkpoints.

Where people get burned

• Instrumental confusion is the number one here. If you set instrumental true and fill in prompt or lyrics, you're contradicting the endpoint's own rules. Pick a lane.
• No key → an explicit error telling you where to paste it. Start there.
• Mixed-version installs are refused at load: if __init__.py and nodes/generated.py disagree, delete the whole pack folder and reinstall.
• V5.5 is marked discontinued in KIE's docs. It runs; the V6 sibling node is the current tier, and if you're building something new, use that.
• A timeout isn't a cancellation. The node stops polling at timeout_seconds (1200 default); the remote job may still land. Check credits_consumed before firing again.

One editorial note: the pack's README promises native media inputs everywhere, and mostly delivers. upload_url on the cover and extend nodes is the exception - it's still a URL string, which is why the Upload Audio node exists. Use it and you'll never think about it again.