``Last update: March 24, 2026``

***
:::content-center
## Introduction
:::

- This is a [cloud-based](https://docs.aihub.gg/extra/glossary/#cloud-based) tool to run RVC Inference directly in your browser without any local installation, maintained by **r3gm**.

!!!warning Disclaimer
This demo is provided for educational and research purposes only. The authors and contributors do not endorse or encourage any misuse or unethical use of this software. Any use for purposes other than those intended is solely at the user's own risk.
!!!

***
:::content-center
## Choose your Cloud Service
:::
Set up the program using your preferred cloud service guide. Once configured, use the Public Tunnel URL to open the interface and continue with the next steps

<style>
  /* Local styling that only affects this page */
  .cloud-grid {
    display: flex;
    flex-wrap: wrap;
    gap: 1.5rem;
    justify-content: center;
    margin: 2rem 0;
  }
  .cloud-card {
    flex: 1 1 220px;       /* Allows cards to shrink/grow, base width 220px */
    max-width: 280px;      /* Prevents cards from getting comically wide */
    padding: 1.5rem;
    border: 1px solid rgba(150, 150, 150, 0.2); /* Subtle border for light/dark mode */
    border-radius: 12px;
    text-align: center;
    text-decoration: none !important;
    color: inherit !important;
    background: rgba(128, 128, 128, 0.03); /* Barely visible background */
    transition: all 0.2s ease;
  }
  .cloud-card:hover {
    transform: translateY(-4px);
    border-color: rgba(150, 150, 150, 0.6);
    background: rgba(128, 128, 128, 0.08);
    box-shadow: 0 10px 20px rgba(0,0,0,0.1);
  }
  .cloud-card img {
    width: 54px;
    height: 54px;
    margin-bottom: 1rem;
    object-fit: contain;
  }
  .cloud-card h3 {
    margin: 0 0 0.5rem 0 !important;
    font-size: 1.25rem;
    border-bottom: none !important; /* Overrides Retype's default header lines */
  }
  .cloud-card p {
    font-size: 0.9rem;
    opacity: 0.7; /* Makes description text slightly muted */
    margin: 0;
    line-height: 1.4;
  }
</style>

<div class="cloud-grid">
    
    <a href="#google-colab-web-ui" class="cloud-card">
        <img src="/img/google-colaboratory-logo.svg" alt="Colab">
        <h3>Google Colab Web UI</h3>
        <p>Decent performance and easiest setup for standard use.</p>
    </a>
    
    <a href="#huggingface-space-appliox" class="cloud-card">
        <img src="/img/huggingface-icon.svg" alt="HuggingFace Space">
        <h3>HuggingFace Space</h3>
        <p>Best for quick inference. Limited free hardware.</p>
    </a>
</div>

<br><br>

***
:::content-center
### HuggingFace Space
:::

!!!danger HuggingFace Space Service
**Check the [HuggingFace Space Glossary](https://docs.aihub.gg/extra/glossary/#huggingface-space) for more info on Free Tier, Limits, Quotas and other things.**
!!!

!!!danger RVC Zero HuggingFace Space Limits
- The RVC Zero HuggingFace Space is made only for **Inference**. Training is not supported.
!!!

!!!danger RVC HuggingFace Spaces Pausing Issue
**Please be aware that HuggingFace has recently been pausing RVC-related Spaces without providing any reason.**
!!!

#### <u>Accessing the Space</u>

Open the space at [https://huggingface.co/spaces/r3gm/rvc_zero](https://huggingface.co/spaces/r3gm/rvc_zero).

Once you're in the space, the rest of the process is **identical** to what's described below in the [Usage](#usage) section.

***
:::content-center
### Google Colab
:::

!!!danger Google Colab Service
**Check the [Google Colab Glossary](https://docs.aihub.gg/extra/glossary/#google-colab) for more info on Free Tier, Limits, Verification, Pricing and other things.**

**Disallowed Activities:** **Running Web UIs on the Google Colab Free Tier is a violation of Google's Terms of Service**. Please be aware that Google’s detection systems operate mid-run, not just at startup; even if you bypass the initial "Disallowed Code" check through encryption, your session remains at risk. Persistent violations can lead to a progressive penalty: starting with reduced GPU availability and potentially escalating to a permanent restriction of your Google account's ability to execute Colab notebooks. See [Google's official policy](https://research.google.com/colaboratory/faq.html#disallowed-activities) for more information.
!!!

#### <u>Setting Up</u>

- Access the Colab notebook [here](https://colab.research.google.com/github/r3gm/rvc_zero/blob/main/colab_notebook.ipynb) and log in to your Google account.

- Run the **"Run the RVC ⚡ ZERO App"** cell by clicking the play button on its left.

   <img src="../rvc-zero-cloud-img/colab-cells.png" alt="Google Colab run cell" width="600" height="auto">

- The installation will begin automatically. It will take approximately **7 minutes**, wait for it to complete.

   <img src="../rvc-zero-cloud-img/colab-output.png" alt="Colab installation output" width="600" height="auto">

- Once finished, a **public URL** will appear in the cell output. Click it to open the RVC Zero interface.

!!!warning
Don't close Google Colab until you're done using it, or the session will stop working.
!!!

Once you're in the interface, the rest of the process is **identical** to what's described below in the [Usage](#usage) section.

***
:::content-center
## Usage
:::

Both the HuggingFace Space and the Google Colab version share the exact same RVC Zero web interface. The steps below apply to both.

***
## <u>Interface Overview</u>
***

The space features a really clean, single-page layout:

   <img src="../rvc-zero-cloud-img/interface.png" alt="RVC Zero main interface" width="1000" height="auto">

- `TTS (toggle)`: Enable Text-to-Speech mode so you can generate audio directly from text instead of uploading a file.
- `Audio files`: Upload the vocal audio you want to convert.
- `URL-to-Model (toggle)`: Load a model straight from a URL instead of uploading the `.pth` file yourself.
- `Model file`: Upload your RVC model (`.pth` file).
- `Index file`: Upload the matching `.index` file (this is optional, but we highly recommend it).
- `Advanced settings`: Expand this section to tweak things like pitch, index influence, and other fine-tuning parameters.
- `Inference button`: Click this to run the voice conversion!
- `Result`: Your final output audio will show up here once the conversion is all done.

***
## <u>Inference</u>
***

### Standard (File Upload) Method

**a.** **Upload your audio:** In the "Audio files" drop zone, simply drag and drop (or click to upload) the vocal audio you want to convert.

**b.** **Upload your model:** In the "Model file" drop zone, upload your `.pth` RVC model file. For the best accent accuracy, you can also upload the matching `.index` file in the "Index file" drop zone.

   <img src="../rvc-zero-cloud-img/interface.png" alt="Model and Index file upload, Inference button and Result area" width="1000" height="auto">

**c.** **Run Inference:** Click that pink **Inference** button. Once it finishes, your converted audio will appear right in the **Result** section. Just click the download icon to save it!

---

### URL-to-Model Method

If you'd rather not upload a `.pth` model file manually, you can load a model directly from a URL. Just flip the **URL-to-Model** toggle on, and an input field will pop up.

   <img src="../rvc-zero-cloud-img/url-to-model.png" alt="URL-to-Model toggle expanded" width="1000" height="auto">

The field accepts two URL formats:

- **A `.zip` file** containing the model: paste the direct download link. For example:
  `https://huggingface.co/MrDawg/ToothBrushing/resolve/main/ToothBrushing.zip?download=true`

- **Separate `.pth` and `.index` links** separated by a comma. For example:
  `https://huggingface.co/.../model.pth?download=true, https://huggingface.co/.../model.index?download=true`

Once you've pasted the URL, click **Process** to download the model. Then upload your audio and click **Inference** as normal.

!!!
To get a direct HuggingFace download link, go to the model's **Files and versions** tab, right-click the download icon next to the file, and copy the link address.
!!!

---

### TTS (Text-to-Speech) Method

If you don't have an audio file, you can generate one from text directly inside the space. Toggle **TTS** on at the top of the page and the TTS panel will expand.

   <img src="../rvc-zero-cloud-img/tts.png" alt="TTS toggle expanded showing Text field, tts voice dropdown and Process TTS button" width="1000" height="auto">

**a.** In the **Text** field, write the text you want spoken.

**b.** In the **tts voice** dropdown, choose a voice. It's recommended to pick a voice that matches the gender and language of your RVC model for the best results.

**c.** (Optional) Enable **Play** to preview the raw TTS audio before it goes through RVC.

**d.** Click **Process TTS**. The generated audio will automatically be used as the input for RVC inference. No separate upload needed.

Then upload your model and click **Inference** as normal.

***
## <u>Advanced Settings</u>
***

You can expand the **Advanced settings** menu in the interface to fine-tune your conversion. Here is a breakdown of what each setting does:

   <img src="../rvc-zero-cloud-img/advanced-settings.png" alt="Advanced settings and Examples table" width="1000" height="auto">

- `Pitch algorithm`: The algorithm used to detect and shift the pitch. `rmvpe+` is our recommended default and usually gives the best results.

- `Pitch level`: Shifts the output pitch up or down in semitones:
     - `Lower value`: deeper or more masculine voice.
     - `Higher value`: higher or more feminine voice.
     - `Use 0`: used if your model already matches the target pitch.

- `Index influence`: Controls how much the `.index` file's accent is applied:
     - `Lower value`: less accent and fewer artifacts.
     - `Higher value`: brings out a stronger accent or character from the model, but it might introduce noise depending on your dataset's quality.

- `Respiration median filtering`: This is a handy setting to help smooth out breathing and background noise artifacts in your output.

- `Envelope ratio`: Controls how much of the original audio's volume envelope blends with the converted voice. Higher values will preserve more of the original audio's natural dynamics.

- `Consonant breath protection`: Protects consonant sounds from getting over-processed during the conversion, which really helps keep the speech clear and understandable.

***
:::content-center
## Troubleshooting
:::

==- *GPU task aborted or Inference timeout*
ZeroGPU Spaces have a maximum inference time limit for each request. If your audio is too long, the task might get aborted before it can finish. Try splitting your audio into shorter segments and converting them one by one.
===

==- *You have exceeded your GPU quota*
ZeroGPU Spaces give out a specific usage quota per account. If you aren't logged in, your available quota is much lower. To fix this:
- **Log in** to your HuggingFace account to get a higher quota.
- **Wait** for your quota to reset. The error message will actually tell you how much time is remaining.
- Become a **HuggingFace PRO Member** to get 5× more quota!

You can check on your ZeroGPU quota [right here](https://huggingface.co/settings/billing).
===

==- *No GPU is currently available for you after 60 seconds*
Because ZeroGPU is shared across a lot of different Spaces, it might be temporarily unavailable during peak hours. Just give it a moment and try again!
===

==- *The file is too large*
There's an upload size limit for your model and audio files. If you run into this, try using a smaller model or compress your audio file before uploading it.
===

==- I can't use the Cloud GPU after some time.
- You have exhausted the Free Cloud GPU. You will need to wait for it to reset, pay or switch to a different account.
===

==- :question: I couldn't find my answer.
- Report your issue [here](https://docs.aihub.gg/contributions).
===

###### ‎   ‎
:::content-center
#### `You have reached the end.`

[!badge variant="info" size="xl" corners="pill" icon="paper-airplane" iconAlign="right" text="Report Issues"](https://docs.aihub.gg/contributions/)
:::