Voice 1 + Voice 2
Authorized duet · Local conversion · WAV
V Mode · consent-first AI duet studioOne song.
One song.
Two voices.
Upload the song you want to re-sing, then add a second authorized singer reference. V Mode keeps Song 1's melody, words, timing and music while creating a second performance in Voice 2 and arranging both singers as a duet.
Beta 8 introduced SoulX-Singer Studio, phrase protection, verified resumable model setup and automatic legacy fallback. Beta 9 keeps every improvement.8Fast Studio steps
16Balanced Studio steps
32Maximum Studio steps
4 WAVlocal exports
THE NEW V WORKFLOW
A duet—not a third song.
Only two recordings are needed. Song 1 is the complete musical foundation. Song 2 is used only to learn the second authorized voice for this render.
Keep the song people recognize
Song 1 supplies the words, melody, timing, original singer and instrumental. V Mode does not rewrite its performance.
Use only the authorized second voice
The separator finds the clearest active vocal section in Song 2 and prepares it as a short Voice 2 reference. Song 2's music and lyrics are not carried into the duet.
Re-sing without changing the tune
SoulX-Singer Studio follows Song 1's measured pitch contour while transferring the authorized Voice 2 character and protecting duration, words and phrasing.
Choose how the duet shares the song
Alternate lines with shared hooks, build a call-and-response exchange, or let both voices perform together.
Tune Voice 2 without rebuilding the song
Adjust Voice 2 pitch and level while preservation locks protect melody, lyrics, timing and the original instrumental.
Receive every useful stem
Download the complete duet, Voice 1, AI-transformed Voice 2 and Song 1's instrumental as local WAV files.
BETA 8 STUDIO ENGINE
Cleaner identity. Protected performance.
Every profile converts the full Song 1 vocal. SoulX-Singer Studio uses the measured source pitch instead of asking a speech model to guess the performance, then handles long recordings in protected segments.
8 Studio steps
The fastest full-song Studio render for testing a voice match.
16 Studio steps
The default balance of timbre detail, stability and waiting time.
32 Studio steps
The highest-refinement profile for final exports on capable GPUs.
LOCAL PROCESSING PIPELINE
Separate. Convert. Arrange. Master.
The progress display identifies every active step. Files stay on the computer, but first setup and conversion speed depend on available GPU, CPU, memory and storage.
Confirm recording rights and voice consent
V Mode will not start until the user confirms rights to both recordings and the second singer's consent for AI transformation.
Separate both recordings locally
Song 1 becomes Voice 1 plus its instrumental. Song 2 is reduced to a clean vocal reference and its old music is discarded.
Convert Song 1's vocal to Voice 2
SoulX-Singer Studio performs pitch-conditioned singing conversion, protects phrase boundaries and preserves the original duration.
Build and export the duet
BLENDLINE applies the chosen vocal arrangement, gentle separation, ducking and normalization, then writes four WAV files.
WHAT EACH INPUT CONTRIBUTES
Nothing is ambiguous.
V Mode is designed to avoid accidentally mixing three unrelated songs or carrying Song 2's old backing into the result.
SourceUsed from itNot used from it
Song 1Words, melody, timing, Voice 1 and instrumentalNothing is rewritten
Song 2Authorized Voice 2 characterOld lyrics and music
Final duetBoth voices over Song 1's instrumentalFour local WAV exports
HONEST LOCAL REQUIREMENTS
Free to run. Demanding to compute.
V Mode has no BLENDLINE per-song meter because conversion runs locally. That does not make hardware, electricity, storage or waiting time free.
About 6–8 GB total
The verified SoulX-Singer Studio checkpoint is about 2.8 GB; its runtime, support files and cache use additional space.
6 GB VRAM minimum recommended
Studio is selected on compatible CUDA GPUs. BLENDLINE can automatically use an existing Seed-VC installation as a legacy fallback.
The reference still matters
A clean, mostly solo vocal reference produces a better identity transfer than noisy, heavily processed or harmony-filled material.
CONSENT & DISCLOSURE
A voice is not just another instrument preset.
V Mode has no celebrity presets. Use only a singer who has agreed to the transformation, label the result as AI-transformed, and review the finished audio before publishing. Recording rights and singer consent are separate requirements.
Authorize both recordings
Uploading a file does not prove ownership or grant a publishing licence.
Obtain explicit singer consent
Do not clone or impersonate a person without their knowledge and permission.
Disclose the transformation
Tell listeners and platforms when a performance contains an AI-transformed voice.
OPTIONAL OPEN LOCAL ENGINE
View engine project ↗Download Beta 9 →Powered by SoulX-Singer Studio.
The Apache-2.0 Studio engine installs only when you choose it. Its model download resumes after interruption and is verified before use. Seed-VC remains available only as an automatic legacy fallback.