Replace Music Section · V4.5PLUS

Three sliders on this node do roughly the same job from different angles - style_weight, weirdness_constraint and audio_weight, each 0–1, each defaulting to 0.65. Move all three at once and you'll never know which one fixed the bridge. There's also a ten-second minimum on the replacement window, and that one isn't a preference.

What it is

KIE_Next_V4_5PLUS_a1f85c48 regenerates a slice of a Suno track through KIE.ai, pinned to the V4.5PLUS provider version. The KIE Next pack generates a ComfyUI node per KIE Market operation and provider version - nine of them for section replacement alone - so the version is baked into the class, not exposed as a dropdown.

KIE's note for V4_5PLUS is that it "offers richer timbres, new creative methods, and supports durations of up to 8 minutes". It's also listed as deprecated, along with the rest of the V4 family, in favour of V6 / V6_MINI / V6_WILD. So this is the node for revisiting an older project, or for a specific timbre you already know you liked.

How it works

Infill on the provider's side: you identify the source track, give a time window, describe what the section should become, and hand over the complete lyric sheet as it should read afterwards. The model re-renders that stretch in context and returns a full track.

Locally it's the pack's standard path - assemble the documented request, submit, poll until timeout_seconds (1200 default), download the first result URL into a ComfyUI AUDIO, and print the credit receipt on the node. Suno has no open weights, so there's no story here about running it yourself; that's the deal with the whole audio branch of ComfyUI right now.

The inputs that matter

The source: task_id for the parent generation and audio_id for the specific track inside it. KIE's tooltips describe exactly those two roles. Both are optional fields in the schema and both are de facto required - the clean wiring is your music-generation node's task_id socket here, with the audio ID copied from that task's raw_json.

The window: infill_start_s and infill_end_s, seconds to two decimals, start < end, and at least ten seconds between them. A chorus fits; a word doesn't.

The words: full_lyrics is the whole song after replacement (required), lyrics is only the new segment (optional). Required prompt, tags, title are a short description, style tags, and a name. negative_tags defaults to Rock and is a real exclusion, not decoration.

The dials: variety 0–4 sets how far the new section may wander - 0 is strictly the same style, 4 is "unreasonably varied" in the tooltip's phrasing. Then style_weight (adherence to your tags), weirdness_constraint (experimental deviation) and audio_weight (relative weight of the audio elements). Change one, render, listen, change another. vocal_gender (m/f) only shifts the odds. persona_id/persona_model are inert on a V4-era pin.

Outputs

audio is the finished track; Save Audio, or KIE • Assemble Video + Music if you're scoring a clip with it. task_id is a fresh task ID, which is how you chain a second window. url and all_urls_json are the provider links, raw_json the raw response, credits_consumed/credits_left the spend. The node is not an output node, so if nothing downstream consumes the audio, nothing runs.

Installing it

ComfyUI Manager → search "KIE.ai Nodes Next", or:

comfy node install kie-ai-nodes-next

Manual:

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

Restart ComfyUI, then Settings → KIE.ai Nodes Next → Connection and paste your KIE.ai API key. The pack validates the key against KIE before overwriting a saved one, then shows only a masked fingerprint - the full key stays out of your workflow JSON. Environment-variable keys aren't supported.

Dependencies: requests, Pillow, numpy, PyYAML. No model downloads, no VRAM requirement, because the rendering happens on Suno's machines.

Where people get burned

Tuning three dials blind. Each run costs money, so a one-variable-at-a-time habit isn't just good method, it's cheaper.

A window under ten seconds. The pack's local preflight rejects it before upload - free - which is the good failure. Right-click → KIE Next · Check inputs & model controls to see the cached documented limits without submitting.

A full_lyrics field holding only the replacement. You'll get a track that believes the whole song is that one section.

Expecting this to be the current model. It isn't; it's a legacy pin. If what you actually want is "the best Suno section replacement available today", start at the V6 node and only come back here for reproduction.