Split stems before the UGC music bed

Split the generated song into stems before you mix it under UGC. Split a song into stems (separate_music_vocals) returns vocal and instrumental from a track you generated in-app. Lay the instrumental under the talking-head as the bed. Keep the vocal stem off the timeline if the ad already has a person speaking. If you mix the full song under a founder read, the caption engine will highlight the snare, the hook, and the wrong instrument. The mute viewer thinks the edit is drunk.

This page is the stem gate before a UGC bed. It is not karaoke isolation, and it is not “generate a lo-fi loop and hope.”

The bed cannot include a second talker

UGC ads are usually a face plus a pack plus a line. UGC lives or dies on that line being readable on mute and audible on speaker. A generated track that still has a singer is a second mouth. Mix mode will keep both. The recognizer will hear both.

Dirty beds that fail this clock:

• A Suno generate with lyrics, mixed at “background” volume that is still a chorus.
• A native-audio clip that already baked a score, then a full song on top — two beds, one dialogue.
• A karaoke-the-product job where you wanted an instrumental and you attached the master.
• Word-timed captions on a smash of speech plus vocal stem. Phrase-level BASIC looks forgive a little drift. Karaoke does not.

Suno Sounds V5.5 is 2 credits, text-to-audio, with looping, tempo, key, and lyrics subtitle capture on the card. Lyrics in the generate are a feature when the song is the product. They are a defect when the song is a bed under a person. Write the prompt as instrumental if you can. Split anyway if the take came back with a singer you did not want to throw away.

separate_music_vocals is specific to Versely-generated tracks. You need taskId and audioId from a prior generate_music or extend_music result. It is not for an arbitrary uploaded MP3. Uploaded commercial records are a licence problem first. For a recording that is not an in-app generate, isolate vocals from a track (isolate_audio) is the general-purpose strip — a different door, and the 09-28 karaoke page already covers that order. This page is the in-app split before a UGC mix.

Billed per split, on top of the original track’s generation cost. Confirm with estimate_cost. Removing a vocal you already paid to generate does not refund the song.

Procedure: split, then mix, then caption

1. Generate the track you actually like — genre, mood, tempo — or reuse one you already have in history. AI music generator is the write step.
2. Pull taskId and audioId from that result (get_generation_history / fetch_user_generations).
3. Run split a song into stems. Select the stem type you need. You want the instrumental for a bed under speech.
4. Listen to the instrumental on headphones. If a ghost vocal remains, it will still confuse captions. Do not mix a leaky stem at full song volume.
5. Attach with add music to video in mode mix. music_volume low relative to original_volume so the founder stays intelligible. Mode replace strips the video’s original audio. That is the wrong setting if the UGC take already has the line.
6. Caption the speech, not the bed. Add captions to video: 165 language codes, 21 BASIC looks at 1×, 9 DYNAMIC looks at 2×. If you needed a transcript first to correct SKUs, that is a document job, then a burn.

If the UGC plate is silent on purpose and the only voice is a TTS read you will add, you still split if the music file has lyrics. A singer under TTS is the same two-mouth problem. Mix the instrumental. Keep the vocal stem for a different cut — a lyric video, a karaoke variant — not this ad.

If the track is too short for the spot, extend the music, not the picture. extend_music continues a Suno generate from an audioId. Veo extend is 7 seconds of picture at 720p, 40 credits. That is not a bed.

What you are not doing

You are not mastering a record. Volume controls on mix are not a compressor. Duck by ear: if the line recedes on a phone speaker, pull music_volume. Splitting again will not fix a bed that is simply too loud.

You are not clearing a chart song. In-app generates are the tracks this split tool accepts. A ripped instrumental from a hit is still a licence.

You are not captioning the drummer. If the brief was karaoke-on-the-song, isolate the vocal and caption that stem — different job, different first file. For UGC, the speech is the founder, the creator, the TTS. The bed should have no words.

You are not using the vocal stem as “energy” under a talking head. Energy is arrangement, not a second lyric. If the instrumental feels thin, generate a new instrumental prompt. Do not sneak the singer back in at −12 dB and hope the word clock ignores it.

Kapwing’s 2026 UGC round-up still cites the usual lift numbers — 4× CTR, about 50% lower CPC, 92% trust in peers — without a Versely rate card. None of those stats require a chorus fighting the offer. The bed is furniture. The line is the product.

The test before you burn type

Mute the timeline. Can you still read the offer in captions that were timed to the face, not to the kick? Unmute. Can you still hear the SKU? If the chorus sits on the price, the split did not make it into the mix, or the mix never used the instrumental.

Seed Audio 1.0 is 4 credits text-to-audio with preset voices — a speech row, not a bed. Do not mix a TTS take into the music slot and then stem it. Stem is for a song you generated as music. The UGC line is the video’s original audio, ducked under the instrumental.

If you need a seat to split the last generate and mix the instrumental under the take, sign up. Bring the audioId. Do not bring a YouTube URL.

FAQ

Can I split any MP3 into stems on this tool?

No. separate_music_vocals only works on tracks you generated in-app via generate_music or extend_music, referenced by taskId and audioId. For an arbitrary upload, use isolate_audio.

Should I caption before or after the split?

After the mix you will ship. Caption the speech on the timeline that no longer contains a singer. A burn timed to a dirty bed will highlight the wrong instrument.

Mix or replace when I add the instrumental?

Mix, if the UGC take already has dialogue you want to keep. Replace strips the original audio. Replace is for a silent plate or a track you are throwing away on purpose.