The short answer
Seed-VC and RVC both turn one voice into another, but they get there differently. Seed-VC is zero-shot: you give it a reference clip of 1 to 30 seconds and it converts audio into that voice with no training step. RVC, Retrieval-based Voice Conversion, trains a model on recordings of a voice first, then converts anything you feed it through that model.
Pick Seed-VC when you need a quick result from a short clip. Pick RVC when you want a voice you can reuse, tune, and share, and you have clean recordings to train on. One more fact matters before you build anything on Seed-VC: its repository was archived by its owner on November 21, 2025, so it is read-only and no fixes are coming.
How Seed-VC works
Seed-VC is an open-source project by Plachtaa, released under the GPL-3.0 license. Its README describes zero-shot voice conversion, zero-shot singing voice conversion, and a real-time mode. It runs on Windows, Linux, and Apple Silicon Macs with Python 3.10 suggested.
It also supports fine-tuning if you want a speaker closer to the target, and the README says each speaker needs at least one utterance. For real-time use it lists about 300 ms of algorithm delay plus about 100 ms of device delay, and it strongly recommends a GPU. Its published benchmark numbers come from an NVIDIA RTX 3060 laptop GPU.
How RVC works
RVC is the technology behind the RVC WebUI and community apps such as Applio. You train a model on recordings of one voice, and the project's README recommends at least 10 minutes of low-noise audio. Training produces a .pth model file and an .index file, and conversion then retrieves matching features from that index. The project is MIT licensed, and its README reports 170 ms end-to-end latency for real-time use, or 90 ms with ASIO audio devices, depending on your hardware.
The training step is the cost, and it is also the point. A trained model is a portable file you can keep in a library, merge with another voice, or hand to a collaborator, and the RVC community shares a great many models. Our guides to training an RVC model and the best RVC settings cover that workflow.
Side by side
Training: Seed-VC needs none for the base zero-shot mode, while RVC needs about 10 minutes of clean audio. Input voice sample: Seed-VC takes a 1 to 30 second reference, RVC takes a trained model. License: Seed-VC is GPL-3.0, RVC is MIT. Project status: Seed-VC is archived and read-only. Reuse: an RVC model is a file you can move between tools, while a Seed-VC conversion starts from a reference clip each time.
We have not run a head-to-head listening test of the two, so this page makes no claim about which sounds better. Quality depends heavily on the source recording, the reference or training audio, and the settings, and anyone who gives you a flat winner without showing the test is guessing.
Does the archived status matter?
For a one-off experiment, not much. The code is still there to download and run. For anything you plan to depend on, it matters: an archived repository cannot receive bug fixes, security updates, or support for new Python and PyTorch versions, so an install that works today can break after a system update and nobody will patch it.
The GPL-3.0 license also matters if you are building software rather than just making audio. It requires derivative works you distribute to be released under the same terms, which is a different bargain from RVC's MIT license.
Where Xport Studio fits
Xport Studio runs RVC-style voice conversion and training locally, with no Python install and no upload. You add 5 to 30 minutes of clean vocals, train a voice on your own machine, and it lands in your Library ready to convert with. It also loads RVC models you already own, and you can pull a clean vocal out of a full song first with the stem splitter.
It needs macOS 15 or later on Apple Silicon, or Windows 10 or later. Voice conversion is a 7 day free trial, then part of Xport Pro, $79 one time, and the other tools are free. See the AI vocal converter and offline voice cloning pages for details.
Which one should you use?
If you want to hear a voice change from a short clip today, are comfortable with Python, and accept an unmaintained project, Seed-VC is a reasonable experiment. If you want a reusable voice, a large community of shared models, and models you can share or merge, RVC is the safer base, and Applio vs RVC WebUI explains the main ways to run it. If you need a place to find ready-made models, see where to find RVC models.