Generate Music · V5.5

This is the node that makes songs. Text in, track out, at Suno's V5.5 tier. It's the busiest node in the Suno family - sixteen-ish widgets - and most of the confusion around it comes from one field: custom_mode.

If you've looked at local music generation, you know the trade already. ACE-Step runs on a 4GB card and is genuinely good at instrumental music; its vocals and lyrics are its weak spot. Suno's vocal quality is the reason people still pay per call. So the decision is roughly: local for instrumental beds and experimentation, this node when the song needs a singer that doesn't sound like a compromise. Every call spends KIE credits and sends your prompt off the machine.

How it works

The pack builds its node library from KIE's official API docs on startup, splitting one documentation page into one node per published model version. That's why there's no model dropdown here - the class is fixed to V5_5. The node assembles the documented request body, submits it to KIE's task endpoint (envelope model ai-music-api/generate, inner model V5_5), polls the record-info endpoint every couple of seconds with backoff until the task reports success, then downloads the first audio result and returns a native AUDIO. Failures raise with the task id and the provider's message.

The inputs that matter

Three are required, and together they define the mode you're in:

• custom_mode - default true (detailed control). Set it false for simplified mode and the endpoint mostly ignores the detailed controls.
• instrumental - default true (no lyrics). In custom mode with instrumental true, only style and title are needed; with it false, style, title and prompt are required, and prompt becomes the exact lyrics. Confusing? Yes. Read the tooltip on the node; it spells out the whole matrix.
• prompt - in custom mode it's sung strictly as lyrics; in non-custom mode it's the core idea and the lyrics get written for you.

Then the controls that only bite in custom mode: style (up to 1000 characters on this tier), title (up to 80 characters), negative_tags, vocal_gender (m/f, and the docs admit it only nudges probability), style_weight, weirdness_constraint, audio_weight (docs: not supported when there are no vocals), persona_id / persona_model, duration (10–360s - valid here, since it's blessed for V5.5 and V6), variety (0–4), and lyrics (takes priority over prompt; 5000-character cap on the current tiers).

And the reference sockets, which are the best part of this node: image_urls takes an IMAGE (a style or mood reference), and audio_1/2/3 and video_1/2/3 take native AUDIO and VIDEO inputs - the pack uploads them for you and passes the URLs along. No URL pasting.

Outputs: audio, url, all_urls_json (the endpoint's own example response contains two tracks - this is where you keep both), task_id, raw_json, credits_consumed, credits_left.

The trap nobody reads about

Connect an IMAGE and leave custom_mode at its default and KIE rejects the request after uploading your file. The pack defends you: when it sees image references on Generate Music, it forces non-custom mode before submitting and drops the now-invalid custom-only fields (duration, style, title, negative_tags, vocal/weight fields, persona_id, instrumental), printing what it changed. Good behavior - but it means a graph that looks like it's set up one way submits another. Read the console line, don't be surprised by the result.

Installing

ComfyUI Manager, search "KIE Nodes Next"; the README describes registry publication as still-in-preparation, so the clone route is the reliable one:

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

Restart, then paste your key at Settings → KIE.ai Nodes Next → Connection - validated before it replaces a working key, masked fingerprint only, with KIE_API_KEY taking priority on headless machines. Four small dependencies, no torch, no checkpoints.

Where people get burned

• Mode mismatch. Setting custom_mode true, leaving instrumental true, and filling in lyrics gets you no lyrics. The mode matrix is not decoration.
• Two tracks per request. audio gives you the first one. To keep the other, read all_urls_json and pull the second URL through a download node.
• No key configured produces a clear error naming the settings path.
• Mixed-release installs are refused on startup - if the version and generated-node build disagree, delete the folder and reinstall.
• V5.5 is discontinued in KIE's own docs. The V6 node is the current tier; this one is for reproducing what already exists. The one thing V5.5 still buys you over V5 is duration, which the V5 tier's docs don't support.

Start with the defaults, run one call, and listen. Then change exactly one knob per run. Music is the job where the API still wins - but it wins on iteration, and iteration is what costs credits.