Upload And Extend Audio · V4.5

Most track-extending in this pack wants a Suno-generated audio_id - annoying when the song you want longer is one you already have. This node skips that requirement entirely: give it a URL to an audio file, tell it where to start extending, and Suno's V4.5 model picks up the thread.

What it is

A remote API call wrapped in a ComfyUI node. The request goes to KIE.ai under the transport model ai-music-api/upload-and-extend-audio, with V4_5 as the model version; then the node blocks, polling the record-info endpoint every few seconds until the job finishes or timeout_seconds (default 1200 - twenty minutes) gives out. The finished audio comes back decoded, as a ComfyUI AUDIO you can save, preview, or route into video work.

Inputs, in order of how much they matter

upload_url is required and takes a public URL. Not a path, not a file:// reference, not a Google Drive preview link - KIE's servers have to fetch the bytes themselves, and anything they can't reach fails the task. Keep the source under 8 minutes; that's a documented hard limit, not a suggestion. The current build exposes this as a plain text field rather than an audio socket, so tracks generated in the same graph need pushing somewhere public first - the pack's upload helpers under KIE Next / Utility exist for that.

continue_at is the dial you'll be fiddling with: the second at which the extension starts, default 60. Aim it at the end of the original. Aim it mid-track and you're asking for the remaining section to be re-performed, which is sometimes exactly what you want when a bridge is wrong and sometimes a surprise.

prompt is required as a widget and acts as lyrics for the extension, but only when lyrics is left empty - lyrics wins when both are present. style should match the source or you'll hear the handover. instrumental defaults to true, and when it's on the docs say don't send prompt, lyrics or vocal_gender; clear the text box rather than leaving the placeholder sentence in it.

title and negative_tags are administrative. style_weight, weirdness_constraint, audio_weight and variety are optional nudges. duration and persona_model are documented as V5.5/V6-only, so they don't do anything here - leave them where they are.

What you get back

audio is the result, decoded and ready to wire into SaveAudio. all_urls_json contains every URL the provider returned (there's usually more than one take, and audio is just the first), url is the primary result, task_id is KIE's job handle if you want to dig into the task later, and raw_json is the full response - the place to check what the API actually saw. credits_consumed and credits_left come back as floats and are printed on the node after each run.

Installing it

ComfyUI Manager → search ComfyUI-KIE-Nodes-Next. Manual:

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

Restart. Dependencies are requests, Pillow, numpy, PyYAML - nothing heavyweight, no checkpoints. Then one-time setup: Settings → KIE.ai Nodes Next → Connection → Paste / replace KIE.ai API key. The pack validates the key against KIE before replacing a working one, stores it outside your workflow JSON, and shows a masked fingerprint plus connection status. Server installs can skip that and export KIE_API_KEY instead; the env var wins.

Things that catch people

V4.5 vs V4.5ALL vs V4.5PLUS. Three nodes, near-identical schemas, different model versions - and per KIE's parameter docs, V4.5 caps style at 1000 characters (V4 caps it at 200). All three are tagged (Discontinued) next to V5 and V5.5, with V6/V6_MINI/V6_WILD current, so treat the whole V4.5 row as legacy-but-still-documented.

Extensions are paid and unauditionable. Every call costs credits, and you only find out whether the join works after the fact. Decide continue_at first.

One extension is not a whole arrangement. If the source is a 90-second loop, you'll get more of that loop - not a second verse with new instrumentation. For that, the honest local alternative is ACE-Step, which is fine on instrumentals and weak on vocals, but free to retry.