# Face blurring

> Automatically detect faces in an image and blur or pixelate them.

CoreIX automatically detects every face in an image and blurs or pixelates the face regions. This is useful for privacy protection, content compliance, and similar scenarios.

## Parameters

| Parameter    | Description                                                                           |                  Default                  | Values             |
| :----------- | ------------------------------------------------------------------------------------- | :---------------------------------------: | ------------------ |
| `face-blur`  | **Face blur**. The value sets the strength; 30 or higher is recommended               |                     –                     | 1 \~ 100           |
| `face-pixel` | **Face pixelation**. The value sets the strength; 40 \~ 60 is recommended             |                     –                     | 1 \~ 100           |
| `face-pad`   | **Coverage expansion**. Expands the covered area outward, as a ratio of the face size |                    0.3                    | 0 \~ 2             |
| `face-shape` | **Coverage shape**                                                                    | `ellipse` for blur, `rect` for pixelation | `ellipse` / `rect` |

- Use either `face-blur` or `face-pixel`, not both.
- `face-pad` and `face-shape` must be used together with one of the two above.

  <https://demo.bitiful.com/girl.jpeg?face-blur=50>

  <https://demo.bitiful.com/girl.jpeg?face-pixel=50&face-pad=0.5&w=800>

## How it works

- **Strength is independent of size**: strength is calculated relative to face width, so the same strength looks the same on the original image and on a thumbnail.
- **Err on the side of coverage**: detection prioritizes not missing any face. Profile views, small faces, and partially obscured faces are all covered; faces on posters and mannequins may be covered too.
- **Fail closed**: if face detection fails, an error is returned; an image with uncovered faces is never output.
- **Cover first, then resize**: faces are covered on the original image before resizing, so no recognizable faces remain in thumbnails.
- **HDR images**: HDR information in the same regions is processed in sync, and the output is still HDR.

> **Warning**
>
> Face blurring supports static image output only. An error is returned if the source is an animated image or the output is video.

> **Tip**
>
> If you only need face positions without blurring, use `fmt=faces`. See [Image analysis](/en/coreix/features/analysis).
