Suno v6 Wild · Text to Audio

Three Suno nodes ship in the PopNow pack - v6, v6 Mini and v6 Wild - all at 10.8 credits a run, all with the identical input schema. The names are your only guide to what you're picking, and Wild's one-line description is "experimental, textured songs with vocals".

So: it's the one you reach for when the brief is strange. Not complex - strange. Shoe-gaze with a broken tape machine. A sea shanty about server infrastructure. Something that sounds like a modular synth falling down a stairwell, but singable.

What's different

Same fields as v6, different model. Wild leans into texture and structure you wouldn't ask for: noisy beds, odd time signatures, instruments that arrive late and stay too long, mixes that sound like a room rather than a mix. On a conventional pop brief that reads as a bug. On the right brief it's the reason to be here.

The practical consequence: write a prompt that gives it something to be weird about. "Indie rock, sad, 90 bpm" gives Wild nothing and it'll split the difference into something mediocre. "Lo-fi shoegaze, hazy cassette hiss, detuned guitars, whispering female vocal, half-remembered 90s chorus, drums recorded down a corridor" gives it a texture to chase.

The inputs

prompt - style description, up to 1,000 characters, and with lyrics or an instrumental it's just the style. With neither, the model writes the words from it.

lyrics - your words, plus [Verse] / [Chorus] / [Bridge] section tags that shape the song and aren't sung. Wild does interesting things with irregular structures; a [Bridge] two lines into the song is a legitimate move here and it'll take you up on it.

generate_lyrics (on by default) - only applies when lyrics is empty and instrumental is off.

instrumental - where Wild is genuinely underrated. Textured, weird, instrumental is a film-score generator, and eleven cents for two cues is cheaper than most sample libraries.

weirdness is the dial you'd think matters most: 0 conventional to 1 unusual, default -1 meaning the field isn't sent and the model decides. Honestly, on Wild I leave it alone - the model is already the weird one, and pushing the dial to 1 on top of it mostly buys incoherence. Use it when a Wild result came back too tame.

style_weight (0 to 1) is the counterweight: how closely the track sticks to the style you described. On a Wild run this is the dial I actually touch, because it's how you keep the texture while stopping the song from wandering off. Higher style weight, tamer result.

vocal_gender only applies with lyrics; negative_prompt is where you put the things you never want to hear - "clean production, autotune, EDM drop" is a reasonable start.

variety (0 off, 1 normal, 2 high, 3 extra, 4 max) sets how different the two versions in a run are from each other. On Wild, leave it high. The whole point is that one of the two takes does something you didn't plan.

persona_id + persona_model - reuse an existing Suno persona, style or voice. audio_weight is how strongly that persona's audio shapes the result, vocals only.

The optional autogrow sockets - up to five image_urls, one reference_video_urls (100 MB, 241 seconds), one reference_audio_urls (6 seconds to 30 minutes) - work only when the model is writing the lyrics. Wild plus five mood images plus an instrumental reference is my favourite way to get somewhere unplanned on purpose.

Outputs: audio (a batch of the results; Save Audio writes one file each) and task id.

Install

Manager → search PopNow → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/popnow-ai/popnow-comfyui

Then a key from popnow.ai/api/console/keys into Settings > PopNow, or POPNOW_API_KEY=sk-popnow-... python main.py. Nothing to install beyond the pack, nothing to download.

Where people get burned

Judging it by pop standards. People run Wild once on a normal brief, get a strange mix, and conclude it's worse than v6. It is - for that brief. It's a different tool wearing the same clothes, and the deterministic-pop job belongs to v6.

No seed. No re-rolling a result you liked. Save the audio the moment it's good, and keep the task id if you might want the run's details later.

The open alternative is a different bet. ACE-Step locally is free and its vocals are its weak suit; the KB's audio essay lays out that trade honestly. For experimental, textured, vocal music, hosted Suno is still the shortest path.