Extend Music · V4

V4 is the original of the Suno line the pack exposes - "improves vocal quality, max 4 min", in KIE's own description - and this node continues a V4 track that already exists. It's here because the API page still documents the version and the pack builds a node for every documented model version. Whether you should use it is a different question, and the answer is mostly "only if you're already in V4".

How it works

The node composes an extend request and submits it to KIE.ai under the transport model ai-music-api/extend, with V4 as the version inside the payload. It then holds the queue while it polls for the result, waiting up to timeout_seconds (default 1200 seconds) before giving up. The finished audio is downloaded and decoded into a ComfyUI AUDIO object locally, using ComfyUI's own torchaudio or PyAV.

The inputs

audio_id is required and is a Suno track identifier, not a file - the id value from a previous generation's raw_json output. You can't point this at your own MP3; the Upload And Extend Audio node is the one that takes a URL instead. Practically, this node runs downstream of a Generate Music node and nowhere else.

continue_at (default 60 seconds) is where the extension starts. Land it at the end of the original. Land it earlier and you've asked the model to redo the rest, which is a reasonable trick for a weak middle section and a bad idea if you liked it.

prompt is required as a widget and serves as lyrics for the new section, unless lyrics is filled - lyrics wins. style should match the source, and here V4's tighter budget matters: the docs cap V4's style at 200 characters, against 1000 for the later versions. Long, stacked style prompts will get truncated or rejected, so be terse. instrumental defaults to false on this node, so an instrumental source will get vocals unless you turn it on - and if you do, clear prompt and vocal_gender, since the docs say neither should be sent in that mode.

negative_tags, variety and the style_weight / weirdness_constraint / audio_weight dials round it out. duration and persona_model are scoped to V5.5/V6-family versions per the docs, so they do nothing on V4.

Outputs

audio is the extended track. all_urls_json holds every result URL the response contained (more than one take is normal; audio is the first), url is the primary link, task_id is KIE's job handle, and raw_json is the whole payload - including the new track's id, which is how you'd extend again. credits_consumed and credits_left report cost and balance, also rendered on the node after each run.

Installation

Through ComfyUI Manager, search ComfyUI-KIE-Nodes-Next; or:

cd ComfyUI/custom_nodes
git clone https://github.com/felipederosilva/ComfyUI-KIE-Nodes-Next

Restart ComfyUI. The pack installs four small Python packages (requests, Pillow, numpy, PyYAML) and downloads no model files at all. Set your KIE.ai key once in Settings → KIE.ai Nodes Next → Connection → Paste / replace KIE.ai API key. It's validated before being saved, stays out of your workflow JSON, and appears as a masked fingerprint; KIE_API_KEY in the environment overrides it for headless setups.

Read this before you build on it

V4 is the oldest and most constrained rung. KIE's parameter docs list V4, V4.5, V4.5PLUS, V4.5ALL, V5 and V5.5 all as (Discontinued), with V6/V6_MINI/V6_WILD current. V4 additionally has the 4-minute track ceiling and the 200-character style limit. If you're choosing a version today, this isn't the one.

No cross-version extending. A V4 audio_id belongs to a V4 track. Extending a V6 generation from this node is not a thing.

Credit-burning loop. Chaining extensions is the fastest way to spend real money in this pack, and the joins are only audible after you've paid. Decide the extend point, run once, listen, then decide again.