Extend Music · V4.5ALL

Extend Music is the "and then what?" node. You generated a track, liked 90 seconds of it, and the thing just stops. This node asks Suno's V4.5ALL model to carry on from a point you choose, and returns the result as a normal ComfyUI AUDIO object you can save, loop, or feed into a music video.

The one thing that trips everyone

The required input is audio_id, and it is not a URL. It's the provider's internal ID for a track Suno made - the kind of string that shows up in the API response as id. So this node is always the second half of a chain: generate first, then extend. Hand it an MP3 link and it fails, because the endpoint genuinely can't read one. Extending your own existing audio is a different node (Upload And Extend Audio) on a different endpoint.

Where do you find the ID? Generate Music returns raw_json, which contains the provider's full response with an id per take. That's your copy-paste source. It's a small friction the pack doesn't hide, and it's the reason a lot of people think this node is broken when it's actually just mis-fed.

What it does under the hood

A remote call. The node builds the extend request, submits it to KIE.ai under the transport model ai-music-api/extend, and passes V4_5ALL as the version inside the payload. Then it blocks - polling the record-info endpoint roughly every 2.5 seconds until the job resolves or timeout_seconds (default 1200 seconds) expires.

continue_at is the dial that matters: the second at which the extension begins, default 60. Set it at the tail of the original. Set it early and you've asked for a re-write, which Suno will happily attempt.

prompt is used as the extension's lyrics, but only if lyrics is empty - lyrics wins when both are filled. style should mirror the source track, or the seam will be obvious. instrumental defaults to false here, so an extension of an instrumental will try to sing at you; turn it on deliberately, and clear the prompt box when you do, because the docs say neither prompt nor vocal_gender should be sent in instrumental mode.

variety, style_weight, weirdness_constraint and audio_weight are the same optional nudges as the generation node. duration and persona_model are documented as V5.5/V6-family only, so they're inert on this version.

Outputs

audio is the decoded extension, ready for SaveAudio. all_urls_json keeps every result URL from the response (Suno generally returns more than one take, and audio is the first), url is the primary result, task_id is the KIE job handle, raw_json is the full payload - grab the new track's id from there if you want to extend it again, which is exactly how people build three-minute songs out of one good idea. credits_consumed and credits_left close the loop on cost, and both are printed on the node face after a run.

Getting it installed

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

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

Restart afterwards. The pack pulls only requests, Pillow, numpy and PyYAML - no weights, no CUDA, nothing to place in models/. Then paste your key once in Settings → KIE.ai Nodes Next → Connection; it's validated against KIE before saving, never written into your workflow, and shown as a masked fingerprint. KIE_API_KEY overrides the stored key for headless installs.

Caveats worth knowing

Version status. KIE's parameter docs list V4.5ALL in the (Discontinued) bucket with V4, V4.5, V4.5PLUS, V5 and V5.5; V6, V6_MINI and V6_WILD are current. Expect this node to be a legacy path, and switch siblings if the version is retired outright.

Extends are metered, and you don't get to audition first. Each call spends credits whether the join is seamless or not. Iterating on continue_at is the expensive kind of fidgeting - decide the cut point before you press Run.

Keep the node's default title field sane. title shows up in file names and player UIs downstream, and the default "Peaceful Piano Extended" ends up stamped on a lot of unexpected tracks.