Replace Music Section · V5.5

KIE's one-line description of this provider version is "a dedicated model tailored to your unique taste", which is marketing for: it does personas well. The catch is the two words next to it in the same note - deprecated. So read this as a compatibility node, not a destination. If you want taste-matching section replacement today, it happens on the V6 family.

What it is

KIE_Next_V5_5_746e60fb rewrites a time window inside a track you already generated, using Suno V5.5 through KIE.ai. Like everything in this folder it's a generated node: the KIE Next pack reads KIE's catalog and provider docs and creates one ComfyUI node per operation/version, so the version lives in the class name rather than a widget.

Its practical value is continuity. A track generated on V5.5 has a specific vocal character and mix, and replacing a section on a different version can sound like an edit. Staying on the same version is how you keep a repair invisible.

How it works

Infill, on the provider's side. You name the source track, give it a window, describe the new section and hand over the corrected lyric sheet; the model re-renders that stretch in the context of the audio surrounding it and returns a whole track. Not a stem, not a patch.

The pack wraps the transport: build the documented request, submit, poll until timeout_seconds (default 1200), download the first result URL into a ComfyUI AUDIO, and show credits spent plus the live balance on the node. Suno's models aren't open weights, so this is a remote render by definition.

The inputs that matter

Source identity: task_id for the parent generation and audio_id for the individual track. Both are optional in the schema and both are what the provider actually needs; the tooltips spell out those roles. Wire task_id from your generator node and copy audio_id out of its raw_json.

Window: infill_start_s and infill_end_s in seconds, two decimals, start < end, at least ten seconds wide. Non-negotiable, and the pack checks it locally before uploading.

Lyrics: full_lyrics is required and must be the complete song as it should read after the replacement. lyrics is optional and holds only the replaced segment. Required prompt, tags, title are the description, style tags and title - with the usual warning that KIE's prompt tooltip says "the lyrics after replacement", which is a slip in their docs rather than an instruction.

Personas: persona_id takes an ID from KIE's Generate Persona / Generate Voice endpoints, and persona_model says which kind it is - style_persona for style personas, voice_persona for Suno Voice IDs. KIE's own tooltip lists persona support for V5, V5_5, V6, V6_MINI and V6_WILD (with V5 and V5.5 flagged discontinued), which is why this node is the interesting one in the deprecated stack.

Shaping dials: variety 0–4 (0 = strictly the same style, 4 = may differ very significantly), style_weight, weirdness_constraint, audio_weight each 0–1 at 0.65 default, negative_tags defaulting to Rock, and vocal_gender (m/f) which nudges rather than guarantees.

Outputs

audio goes to Save Audio / Preview Audio, or to KIE • Assemble Video + Music if the song is scoring picture. The node returns a fresh task_id, so a further window can chain from it. url is the remote file, all_urls_json the list, raw_json the raw response, credits_consumed and credits_left the receipt. It isn't an output node, so nothing executes until the audio is consumed.

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, then Settings → KIE.ai Nodes Next → Connection and paste your KIE.ai key. The pack validates it before overwriting a working key and shows a masked fingerprint afterwards; nothing is written into the workflow. Env-var keys are unsupported by design.

Dependencies: requests, Pillow, numpy, PyYAML. No weights, no VRAM floor.

Where people get burned

Reading "discontinued" as "broken". It still runs. It's just not where KIE points new work, so don't build a 2026 pipeline around it.

Persona model mismatch. A voice ID with style_persona selected tends to produce a result that politely ignores the persona.

Window under ten seconds, or a full_lyrics field that only holds the new part. Both are self-inflicted and both are cheap to avoid. Right-click → KIE Next · Check inputs & model controls reads the cached documented limits locally, so you can confirm a window is legal before spending a credit.