Add a bed after native-audio generate

Mix the bed under the take. Native audio already spoke. Add music to a video in mode mix keeps the original soundtrack and layers a track with separate volumes. Mode replace strips the dialogue you paid a native-audio sticker for. Do not double the talker with a song that also sings the offer. Duck the instrumental. Leave the line.

This page is the bed after a sounded generate. It is not silent-then-TTS, and it is not replace-audio-after-TTS.

Native audio is already a mix

A native-audio generate is picture plus a baked performance: speech, sometimes Foley, sometimes a score the model invented. Published examples: Veo 3.1 is 40 credits, 4/6/8s, 4K, audio true. Seedance 2.0 is 69 credits, 4–15s, 1080p, audio true. Grok Imagine Video is 7 credits, 720p, 6/10/15/20/30s, audio true. Flux 3 text-to-video is 9 credits, 5–20s, audio true.

Listen to the keeper before you attach anything.

• If the take already has a loud score, you may not need a bed. A second bed is mud.
• If the take is speech-plus-room and you want furniture, you need an instrumental at low volume.
• If the take’s invented score fights the brand, you still mix a new bed only after you accept you cannot surgically mute the baked music without also touching speech. Native audio is not stems. If the baked score is the problem, a silent generate plus TTS plus a bed was the other path — too late for this file unless you re-generate.

Add music to video: attach_audio_to_video copies the video stream losslessly. Only audio is touched. generate_music writes a track from genre, mood, tempo if you do not already have one. Suno Sounds V5.5 is 2 credits on that family. Ask for estimate_cost. Attach is a flat fee regardless of video length; the song is a separate bill.

Mode is the whole decision:

• mix — original stays, new track layers. music_volume versus original_volume. This is the bed.
• replace — original gone. This is a new VO on a silent plate, or a music video that should not keep the generate’s speech. Wrong after Veo if you wanted those words.

If the generated track has lyrics, split it first. A singer under a native-audio talker is two mouths. The caption engine will hear both.

Duck so the line still reads on a phone

Short-form plays on speakers that crush beds and on mute that ignores them. You are mixing for both.

1. Generate or choose an instrumental. Prompt without a vocal if the bed is furniture.
2. Attach in mix. Start music_volume clearly below original_volume. There is no full mastering chain here. If the line recedes, pull the bed. Do not ride a compressor you do not have.
3. Listen on headphones, then on a phone speaker. If you cannot transcribe the offer by ear, the bed is too loud or the native take was never intelligible.
4. Caption from the dialogue, not the chorus. Add captions to video after the mix you will ship. 21 BASIC looks at 1×; 9 DYNAMIC at 2×. Karaoke on a smash of speech-plus-singer highlights the wrong syllable.

Do not write “add upbeat music” into a second generate prompt and pay another 40 credits to recast the room. The picture is locked. The bed is an attach.

Do not stack a whoosh pack, a native score, and a Suno chorus. Pick one furniture layer. Native Foley can stay; it is part of the take. A second drum loop is not.

If the native line is wrong — legal, SKU, a cousin of the offer — a bed will not fix it. Replace the audio with a TTS or a founder read, or lipsync, or re-generate. Mixing a cheerful bed under the wrong sentence makes the wrong sentence travel farther.

What this is not

This is not replace the take’s audio after TTS. That page assumes a silent plate. Native audio is the opposite assumption: the plate already talks. Replace would mute it.

This is not picking Turbo versus a sounded row. You already picked sounded. You are finishing.

This is not extending the picture so the song can finish. If the track is short, extend the music (extend_music on a Suno audioId). If the picture is short, that is a video extend row — Veo extend is 7s at 720p, 40 credits — and a different decision. Looping a 6-second Veo under a 30-second song is an edit, not a generate.

YouTube labels (YouTube Blog, 27 May 2026): photoreal / meaningfully AI-altered Shorts get an overlay on the video. A bed does not change whether the picture is photoreal. Disclose the file you actually made. Labels alone do not impact recommendations or monetization eligibility on that announcement.

Mix, then caption, then stop

The order after a native-audio keeper:

1. Picture lock.
2. Optional stem split if the song sings.
3. Mix the instrumental under the take.
4. Burn captions on that mix.
5. Do not generate again for “more energy.” Energy is volume and arrangement, not a 40-credit cousin.

Volumes are not a mastering chain. If the native take already ducks itself — a quiet line after a loud Foley hit — a static music_volume will sit too high on the whisper. Pull the bed until the quietest word of the offer is still intelligible on a phone speaker. If that makes the chorus inaudible, the chorus does not belong under this take. Generate a thinner instrumental. Do not compensate by replacing the soundtrack.

Seedance V1.5 Pro text-to-video is 6 credits with native audio at 1080p, 4–12s — a cheaper sounded row than Veo. The mix rule does not change because the sticker is 6 instead of 40. Mix under. Do not replace. Do not add a singing bed. Caption the dialogue you kept.

If you need a seat to mix a bed under the Veo take without muting it, sign up. Bring the keeper. Use mix.

FAQ

Will add-music delete the native dialogue?

Not if you use mode mix. Mix keeps the original and layers the track. Replace is the mode that strips the original. Do not use replace after a native-audio generate unless you intend to throw the speech away.

The generate already has music. Do I still add a bed?

Only if you still need furniture after listening. A second score on a baked score is mud. If the baked score is unusable and you cannot isolate it, you are in a re-generate or a silent-then-TTS path, not a mix.

Should I caption before adding the bed?

Caption the mix you will ship. A burn timed to a pre-bed file is fine if the picture and speech did not move; attach copies video losslessly. If you change replace/mix later, recaption.