Got stems from a vocal remover? Here's how to get your video back

You ran your video through a vocal remover. It worked beautifully — clean vocal track, clean instrumental track, exactly the split you wanted. Then you looked at what you actually received: two audio files. WAVs, maybe. And the video — the one with your footage, your grading, your cuts — is nowhere in the download folder.

You now own world-class stems and a problem: the video is gone, and the audio inside it is still the original mix.

This is the standard output of most vocal separation tools, and it's not a bug — it's an architecture designed for musicians. But if your source was a video, your job is only half done. This article covers the two ways to finish it: the manual way (any editor) and the lossless way (remuxing), plus the shortcut that skips both.

Why this happens

Separation tools evolved from music production, where the deliverable has always been audio. The pipeline looks the same everywhere: extract the audio from your container, run the AI model on it, hand back the stems. The video stream is treated as packaging — discarded once the audio is out.

For a producer working in a DAW, that's fine. For a videographer, a vlogger, an editor working on client footage, or anyone making karaoke or fan-edit style content, the video is the deliverable. The stems are just ingredients.

Option 1: The manual way (any video editor)

The traditional finish:

1. Import your original video and the separated stems into an editor (Premiere, DaVinci, CapCut, anything).
2. Mute or detach the original audio track.
3. Lay the stem you want — the instrumental for a music-only version, the vocal for a speech-only version — underneath the picture.
4. Check sync. This is the step that bites. If the extraction tool trimmed silence, changed sample rate, or added padding, your stem drifts against the picture and you'll be nudging frame by frame.
5. Export. The whole timeline re-encodes, which costs quality and render time.

Total cost: fifteen to thirty minutes per clip, plus one generation of encoding loss, plus the mental tax of sync-checking. It works. It's also the reason many "separated" projects die in a folder — the last mile never gets done.

Option 2: The lossless way (remuxing)

Here's what editors often don't realize: you don't need to re-encode the video to replace its audio. Modern containers (MP4, MOV, MKV, WEBM) store video and audio as separate streams inside one file. You can copy the video stream bit-for-bit and swap only the audio track — a process called remuxing, done with tools like FFmpeg on the command line:

ffmpeg -i original.mp4 -i instrumental.wav -map 0:v -map 1:a -c:v copy -c:a aac output.mp4

The -c:v copy flag is the important part: the video stream is copied, not recomputed. Every frame is identical to your original — no re-encode softness, no color shift, no generation loss. The render is fast because almost nothing is actually being processed; it's container surgery.

The cost is that you're now maintaining a command-line workflow: correct stream mapping, audio codec flags, faststart for web playback. Powerful, free, and not for everyone.

Option 3: The shortcut — a tool that returns the video

The reason we built BGM Remover is that the last mile shouldn't exist. Drop your video once and the pipeline:

1. extracts the audio track in your browser (the video file itself never leaves your device — only the few-MB audio track goes to the AI, deleted within 24 hours),
2. separates voice from music,
3. remuxes each stem back onto your original video stream — voice-only video, music-only video, plus both raw stems for your editor.

The remux in step 3 is the same -c:v copy surgery described above, automated. One-time 5-minute free trial after sign-in, no watermark, and the picture comes back pixel-identical because it was never touched.

Which path fits your situation

| Your situation | Best path | | --- | --- | | One-off job, you live in an editor anyway | Manual: stems + timeline, accept the re-encode | | Batch of clips, quality matters | Remux: scriptable, lossless, fast once set up | | You want the result, not a workflow | Tool that returns video directly | | Karaoke / sing-along content | Music-only video (remuxed) — picture and lyrics overlay intact | | Client delivery with a deadline | One-pass tool; every manual step is a delay risk |

Frequently asked questions

Will putting the audio back change the picture quality?

Not if you remux. Video stream copy is lossless — the picture is bit-identical to the original. Only a full re-encode (the manual timeline export) degrades anything.

The vocal remover trimmed the start of my audio and now it's out of sync. Fix?

This happens when tools strip leading silence or resample without padding. In an editor, nudge the stem to re-align; with FFmpeg, add the offset back with an audio delay filter before remuxing. Or avoid it: tools that process the full track without trimming keep sync by construction.

Can I get a video with the voice kept AND a separate one with the music kept?

Yes — that's the useful version of this job. Separation produces both stems in one pass, so a tool that remuxes can hand you a speech-only video and a music-only video together. BGM Remover does exactly that, plus the raw stems.

My vocal remover is web-based and uploads the whole video. Is that normal?

It's common, but not necessary. The separation only needs the audio track — a few MB per minute. Tools that extract the audio in your browser and upload only that keep your footage on your device, which matters for unreleased or client material.

The bottom line

Stems without the video is half a deliverable. Finish the job the way that fits your skills — editor, FFmpeg, or a tool that returns the remuxed videos directly — but don't let the last mile convince you the separation was the whole task. Get your video back with the audio already swapped.