YuE2 for Video Creators: Soundtracks, Beds and Songs That Fit Your Cut

YuE2 is usually discussed as a songwriting model: give it lyrics, get a structured song back. But the same things that make it good at songs, reliable structure and a melody planned before any audio is rendered, also make it useful for video. Sections land where you put them, and an instrumental still has a real melody instead of an aimless loop.

This guide is for YouTubers, short-form creators, editors and anyone who needs music that fits a cut. It covers when to use a vocal song versus an instrumental bed, how to write style tags that stay out of the way of a voiceover, how to plan sections against your video length, and how to export and split stems. It assumes you're running YuE2 in Song Creator Pro, but the prompting advice applies anywhere you run the model.

If you're new to YuE2 prompting in general, read the YuE2 prompting guide first. For the broader workflow of getting any AI track into an edit, see how to add AI music to a video.

An Honest Note Before You Start

Song Creator Pro is a music generator, not a video tool. It has no video import, no timeline sync, and no "make this exactly 47 seconds" slider. You generate audio, export it, and do the placing, trimming and fading in Premiere, DaVinci Resolve, CapCut, Final Cut or whatever editor you use.

That's a real limitation if you want music generated to hit cut points automatically. What you get instead is control over structure, unlimited local generations to try variations, and files you own outright. The workflow below is built around that reality.

Step 1: Decide Vocal Song or Instrumental Bed

This is the first and most important choice, and it's mostly decided by whether people are talking.

Your video has	Use	Why
Voiceover, narration, interview, talking head	Instrumental bed	Sung words fight spoken words for attention
Montage, b-roll, travel sequence with no dialogue	Vocal song or instrumental	A vocal hook can carry a sequence by itself
Intro or outro sting	Either; often vocal for a branded hook	Short, memorable, no competing speech
Music video, lyric video, trailer for a song	Vocal song	The song is the content
Tutorial or screen recording	Instrumental bed	Viewers need to hear instructions

A useful habit for mixed videos: generate the vocal version first, then use stem separation (Step 6) to get an instrumental from the same song. You can sit the instrumental under the talking sections and bring the vocal back for a montage or the closing beat, and it'll sound like one piece of music because it is.

Step 2: Write Style Tags for a Bed That Sits Under a Voice

Good background music for speech has a specific job: fill space, set mood, and never compete with the voice. YuE2 responds to comma-separated tags, and earlier tags carry more weight, so lead with the qualities that keep the bed out of the way.

Tags that help a track sit under narration:

• Mood and energy: calm, steady, understated, warm, gentle, light
• Sparse arrangement: minimal, sparse, soft pads, light percussion, brushed drums
• Instruments that leave room for the voice: felt piano, muted guitar, soft synth pads, upright bass, marimba
• Tempo: a BPM tag such as "90 BPM" when you want a steady pace (treat it as guidance, not a guarantee)
• Exclusions: negative tags like "no vocals" and "no lead guitar" keep out prominent midrange parts

Things to avoid in a voiceover bed: busy lead lines, prominent saxophone or lead guitar, heavy distortion and big drum fills. They all live in the same frequency range as the human voice.

Example style prompts:

calm lo-fi, soft felt piano, warm pads, light brushed drums, minimal, steady, no vocals

corporate ambient, gentle marimba, soft synth pads, light percussion, optimistic, understated, 100 BPM, no vocals

cinematic, slow strings, low piano, sparse, reflective, building gently, no vocals

For a montage or intro where the music leads, flip the priorities: lead with genre and energy, add vocal tags, and let the arrangement get busy.

More tag ideas by genre and use case are in the prompt library.

Step 3: Set Up the Lyrics Field for an Instrumental

YuE2 plans structure from the lyrics field, so even an instrumental needs section labels. Leave the labels empty:

[Intro]

[Verse]

[Chorus]

[Outro]

Remove every vocal reference from the style prompt (no "female vocal", no "airy vocal"), and keep planning mode on full. Full planning composes a melody and chord progression; with no lyrics, an instrument carries the melody instead of a singer. That's what makes a YuE2 bed sound like a composed piece rather than a texture.

If you'd rather have a quicker, simpler bed, Song Creator Pro's ACE-Step 1.5 model has a dedicated instrumental mode and generates faster. It's often the better pick for straightforward background music, and YuE2 is the better pick when you want a strong, structured melody.

Step 4: Structure Sections to Your Cut Length

Since there's no length-to-cut control, you plan length through structure. Early YuE2 users consistently report each labelled section coming out at roughly 30 seconds of music. That's a community observation, not an official spec, and it shifts with tempo and how much lyric text is in a section, but it's a workable planning unit.

Video length	Sections to plan	Example structure
15 to 30 s (short, reel)	1 to 2	[Intro] [Chorus]
60 s	2 to 3	[Intro] [Verse] [Chorus]
90 s	3 to 4	[Intro] [Verse] [Chorus] [Outro]
2 to 3 min	4 to 6	[Intro] [Verse] [Chorus] [Verse] [Chorus] [Outro]
5 min+	8 to 10	Full song structure with a [Bridge]

Three tips that make the fit easier:

1. Generate slightly long. Trimming the end of a track and adding a fade is easy. Stretching a track that's too short isn't.
2. Put your energy change where your video changes. If your video shifts from setup to payoff around the midpoint, place a [Chorus] there. YuE2 follows section order closely, so the lift lands roughly where you planned it.
3. End on a real ending. An [Outro] section gives you a natural resolution to land your final shot on, which sounds far better than a fade in the middle of a phrase.

For short-form specifics, see AI music for YouTube Shorts and TikTok. For branded openers, see how to make intro and outro music with AI.

Step 5: Generate a Batch and Audition Against Picture

YuE2's best results come from picking among several takes. Its top benchmark score is a best-of-8 number, and the gap between an average and a great take is noticeable. Set the batch size to 2 to 4 and generate.

Then audition the takes against your actual edit, not in isolation. A track that sounds great alone can be too busy under a voice, and a track that seems plain on its own can be exactly right under footage. Since generation is local and unlimited, extra candidates cost only time.

When you find a take you like, note its seed. Reusing the seed while changing one style tag lets you nudge a track (a little sparser, a little warmer) without losing its character.

Step 6: Export and Split Stems

Export in the format your workflow needs:

Format	Use it for
WAV or FLAC	Editing masters, anything you'll process or mix further
MP3	Quick drafts, sharing with a client for approval, small files for web previews

For video work, export WAV or FLAC for your final edit. Lossless files survive your editor's processing and the platform's re-encoding better.

Then use stem separation. Song Creator Pro includes Demucs with four separation models, so you can split any finished track into vocals and instruments, or into drums, bass, vocals and other. For video, that unlocks:

• An instrumental from a vocal song, for talking sections
• Dropping the drums under a quiet emotional beat, then bringing them back
• Ducking just one element (say, the lead melody stem) under dialogue instead of the whole track

The stem extraction guide walks through the process.

Step 7: Place, Trim and Mix in Your Editor

Back in your editor:

1. Drop the track on its own audio lane and line up the first downbeat with your first cut.
2. Trim the end at a section boundary if you can, then add a short fade.
3. Duck the music under speech, by hand or with your editor's auto-ducking. A bed usually needs to sit well below the voice.
4. For a mid-video music change, crossfade between two YuE2 takes generated with the same style prompt. They'll share a sonic character even with different melodies.

Rights and Platforms

Tracks you generate in Song Creator Pro are files on your PC, made on your hardware, and the app includes a commercial licence. YuE2 is included with direct permission from its developers. See the Terms of Service for details.

Platform claim systems are a separate question from licensing. If you're worried about Content ID, read will AI music get a Content ID claim? and can you monetize AI music on YouTube? before you publish. For short-form platforms, see royalty-free music for TikTok and Instagram Reels.

Quick Reference

Goal	Setting
Bed under voiceover	Empty section labels, no vocal tags, sparse instruments, "no vocals"
Montage with a hook	Vocal tags, energetic genre lead, chorus where the edit peaks
Length	Plan about 30 seconds per section, generate slightly long, trim
Best take	Batch of 2 to 4, audition against picture, keep the seed
Flexibility in the edit	Export WAV or FLAC, split stems with Demucs
Fast simple background	Try ACE-Step 1.5 instrumental mode instead

---

Ready to score your next video? Get Song Creator Pro on the Microsoft Store for $49.99 (free trial included) or on itch.io for $44.99. One-time purchase, YuE2 and ACE-Step 1.5, stem separation, unlimited local generations on your Windows PC.