Replace Music Section · V6_WILD

V6_WILD exists to "push creative boundaries for bolder, more distinctive musical expression", in KIE's phrasing. Translated: it's the version that will give you a section you didn't ask for, which is either the point or a disaster depending on which section you're replacing. Use it on a middle eight, not on the chorus that the whole track hangs on.

What it is

KIE_Next_V6_WILD_e0986c4f regenerates a time window inside an existing Suno track via KIE.ai, pinned to the V6_WILD provider version. The KIE Next pack generates one ComfyUI node per KIE Market operation and provider version - which is why there are nine Replace Music Section nodes and why this one's class name is a hash rather than something human.

It's current, not legacy: KIE recommends the V6 family for this operation, and personas work here. That's the difference between this node and the five V4/V5-era ones stacked above it in the menu.

How it works

Infill on the provider side. You identify the source track and a time window, describe what should now be there, and supply the complete corrected lyric sheet. The provider re-renders that stretch conditioned on its surroundings and hands back a whole track - not a stem, not a patch - so the untouched parts stay untouched.

Locally it's the pack's standard generated-node path: assemble the documented request, submit, poll until timeout_seconds (1200 by default), download the first result URL into a ComfyUI AUDIO, and show credits spent alongside the live balance. Nothing renders on your hardware; Suno has no open weights.

The inputs that matter

Source: task_id (parent generation) plus audio_id (the specific track). Both optional in the schema, both what the request actually needs - wire the generator's task_id output and lift the audio ID from its raw_json.

Window: infill_start_s and infill_end_s, seconds with two decimals, start < end, at least ten seconds apart. Non-negotiable, and locally checked before upload.

Lyrics: full_lyrics (required) is the entire song as it should read after the replacement; lyrics (optional) is the replaced segment only, capped at 5000 characters on V6, V6_MINI and V6_WILD. Required prompt, tags, title are description, style tags, name - and the prompt tooltip's "the lyrics after replacement" wording is a KIE doc slip, not a field alias.

Now the wildness control stack, which on this node is the whole argument. Start with variety 0–4: 0 is strictly the same style, 1 balanced, 2 clearly different, 3 bold, 4 explicitly "may differ in style very significantly". V6_WILD's own disposition already leans creative, so pairing it with variety 4 is how you get a section from a different album entirely. weirdness_constraint (0–1, default 0.65) is the second lever in the same direction - raising it encourages experimental deviation. style_weight and audio_weight pull the other way, toward your tags and the surrounding audio. negative_tags (default Rock) excludes. persona_id/persona_model work here as they do on V6 and V6_MINI.

My read: on V6_WILD, drop variety to 1 or 2 and let the model's natural boldness do the work. Stacking both is how you end up with a render you can't use and can't explain.

Outputs

audio → Save Audio, Preview Audio, or KIE • Assemble Video + Music for the video-scoring case. A fresh task_id comes out for chaining another window. url, all_urls_json, raw_json, credits_consumed, credits_left complete the sockets. Not an output node: connect something downstream or it won't run.

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 saved one and shows a masked fingerprint afterwards; the key never enters the workflow JSON, and environment variables are deliberately not read.

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

Where people get burned

Replacing the hook with a completely different song. The tool's job is to surprise you; your job is to point it at the expendable part. Try the intro, the outro, the second verse.

Ten-second minimum. Same as every node in this family. Widen the window.

Paying repeatedly to explore. Each render is billed, so change one dial per run and keep notes - or use KIE • Task Board to see what the last twenty submissions cost. Right-click → KIE Next · Check inputs & model controls reads the cached documented limits locally; KIE • Reuse Completed Image/Video/Audio fetches a finished result again without a second charge.