Extend Music · V5.5

Songs generated by an API come out short. Wanting more of one is the single most common follow-up, and this node is the answer: Suno's extend endpoint at the V5.5 tier, which continues a track from a point you choose instead of regenerating it from the prompt.

The key thing to get right is what you're extending. audio_id is a Suno track identifier, not a file and not a URL. It's the id that comes back in a generation's raw_json - which means the natural workflow is generate-then-extend inside the same graph, rather than pointing this at an arbitrary mp3. If your source is a file you own, you want the Upload And Extend node instead; it takes a URL. Mixing those two up is the classic first mistake with this pair of nodes.

How it works

Like every model node in this pack, this class was generated at startup from KIE's published API docs, one node per documented model version - hence no model dropdown, just V5_5 baked in. The node builds the documented request body, submits it to KIE's task endpoint (envelope model ai-music-api/extend, inner model V5_5), polls KIE's record-info endpoint every couple of seconds with backoff until the task succeeds, then downloads the resulting mp3 and presents it as a ComfyUI AUDIO. Failures raise with the task id and provider message rather than returning silence.

The inputs that matter

Two required fields, and one of them is the interesting one:

• audio_id - the track to extend. Source audio alone is enough to generate, per the docs; everything else is steering.
• prompt - used as lyrics only when lyrics is empty. Set both and the prompt stops being lyrics. Don't set it at all when instrumental is true.
• continue_at - the seconds mark where the extension should begin, default 60. The docs' own range: greater than 0 and less than the total duration of the source. This is the field that decides whether the result sounds like a continuation or like a different song wearing the same hat.

Then the steering: style (keep it aligned with the original - the docs say so, and they're right), title, negative_tags, vocal_gender, style_weight, weirdness_constraint, audio_weight, persona_id / persona_model, lyrics, variety (0–4), and instrumental (false by default here, unlike the cover node). timeout_seconds defaults to 1200; callback_url stays empty.

Outputs: audio, url, all_urls_json, task_id, raw_json, credits_consumed, credits_left. Note that extending produces an additional track rather than editing the original - all_urls_json is where you get both, and credits_left is printed on the node so you can watch the meter while you iterate on continue_at.

Installing

ComfyUI Manager, search "KIE Nodes Next" - and if it isn't listed yet, that's expected; the README describes registry publication as prepared rather than complete. Manual route:

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

Restart and set the key once at Settings → KIE.ai Nodes Next → Connection. The key is validated before it replaces a working one, the frontend only ever sees a masked fingerprint, and KIE_API_KEY beats the saved key for headless installs. Four small dependencies, no torch, no model files - the install cost here is near zero, which is the whole point of an API node pack.

Where people get burned

• Extending a URL. audio_id is not upload_url. If you have a file, use the Upload And Extend node.
• No key configured gives you an explicit error naming the settings path - always the first thing to check.
• Version mismatch on startup is refused rather than run: if __init__.py and nodes/generated.py disagree, delete the folder and reinstall cleanly.
• V5.5 is discontinued in KIE's own docs. Fine for reproducing old work; for new work use the V6 sibling.
• The 8-minute frame of reference. Extension requests are limited by the source's total duration, and continue_at must fall inside it. A value past the end just fails.

If you're chaining these, the pattern that holds up is generate → extend → extend again, keeping the same style and persona_id across calls. Music is the one audio job where the API still clearly beats local, and this node is a good illustration of why: the model isn't just making sound, it's remembering the song it already made.