Upload And Cover Audio · V4

Cover generation is the sleeper feature of the Suno API: hand it a track you already have, and it rebuilds it in Suno's production - same melodic idea, new everything else. Vocals, arrangement, mix, all regenerated. This node is that endpoint, pinned to the V4 model version.

Worth saying plainly up front: this is one of a family of nodes. The pack splits every documented Suno version into its own node, so you'll also see V4.5, V5 and V6 variants of the same cover endpoint in your menu. V4 is what you're looking at here, and KIE's own docs mark V4 as discontinued - the tooltips in this node literally say "V4 (Discontinued): Maximum 200 characters" for style. It still works, and it's the cheapest of the family, but if you're starting fresh, the V6 nodes are where Suno's current quality lives.

Why anyone uses the cloud for this

Local music generation got real - ACE-Step runs on a 4GB card and instrumental output is genuinely close - but vocals are still the weak spot, and covers are mostly vocals. That's the honest argument for paying per generation here.

How it works

It submits a Market task to KIE, waits, and downloads the result as a native AUDIO object. Under the hood this pack does something important for Suno specifically: KIE's endpoint takes a provider version (V4) nested in the request body while the outer Market model is ai-music-api/upload-and-cover-audio. The pack handles that envelope mapping for you, which is the difference between a working node and a 400 you'd spend an evening on.

Inputs

upload_url is required and is a URL string to the source track, not an audio socket. KIE's tooltip caps it at eight minutes - trim longer material first. In practice you wire this from KIE • Upload Audio, which takes a native AUDIO and gives you a hosted URL; that node is the missing half of this one.

prompt is required in the widget and is the lyrics for the cover - but read the tooltip before filling it in: it's only used as lyrics when you don't provide lyrics, and per KIE you should not pass it at all when instrumental is on.

instrumental (default on) is the mode switch. On, you get no vocals and the prompt and lyrics are meant to be absent. Off, the text you supplied gets sung. One practical trap: this pack sends every widget it can see, with its default value, so if you leave the example lyric text sitting in prompt while instrumental is on, that text goes to the provider anyway. Clear the box rather than trusting the mode flag.

Optional fields worth knowing: style (the genre steer, with that 200-character V4 ceiling), title (80 characters), negative_tags to push things out, vocal_gender (m or f - the tooltip is honest that this only nudges the odds), the three fine-tune weights style_weight, weirdness_constraint and audio_weight, lyrics, variety (0–4, default 1, where 0 is "stay exactly on style" and 4 is "surprise me"), and persona_id with persona_model for singing in a saved voice. duration also appears, but KIE documents it as valid only on V5.5 and later, so on V4 leave it alone.

Outputs and install

Outputs: audio (wire into a core Save Audio or preview), url, all_urls_json, task_id, raw_json, and the credit receipt pair credits_consumed / credits_left.

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

Or install via ComfyUI Manager (search "KIE.ai Nodes Next") and restart. Then set your key once under Settings → KIE.ai Nodes Next → Connection; KIE_API_KEY covers headless installs. Every generation is metered, so use credits_left rather than discovering the balance mid-batch.