**Cost**: 5 credits per successful call.

Image Object Erase supports two input modes:

1. **JSON URL mode**: send `application/json` with the existing URL field.
2. **Direct upload mode**: send `multipart/form-data` with a local image file in the `image` field. The compatible field name `file` is also accepted for the primary image. Codia checks credits before reading and uploading the file, so insufficient-credit requests return `402` without uploading anything.

Direct upload example:

```bash
curl 'https://openapi.codia.ai/v2/open/image/object_erase' \
  -H 'Authorization: Bearer {codia_api_key}' \
  -F 'image=@./photo.png' \
  -F 'mask=@./mask.png'
```

Remove specific regions of an image as identified by a **mask** image, and inpaint the gap so the result looks like the object was never there. Synchronous.

## What this API is for

Deleting people from background plates, removing unwanted objects from product shots, cleaning up screenshots before redistribution. Unlike `watermark_remove` (which auto-detects), this endpoint requires you to **specify exactly what to erase** via a binary mask.

## Mask format

`mask_url` must point to an image with the **same dimensions** as `image_url`:

- **White (255)** pixels = area to erase + inpaint.
- **Black (0)** pixels = keep as-is.
- Grayscale values in between produce partial blending; for crisp removal, threshold your mask to pure black/white.

The mask can be a separate PNG you generated client-side (e.g. from a brush UI or a segmentation model) or an alpha-channel cutout. Soft edges in the mask produce softer transitions in the result.

## Request

| Field | Required | Notes |
|---|---|---|
| `image_url` | yes | Public URL of the source image. |
| `mask_url` | yes | Public URL of the mask image (same dimensions as source). |
| `model` | no | Model ID. Defaults to `codia_image_v2`. Also: `nano_banana_2`, `nano_banana_pro` (higher fidelity at higher latency). |

## Response

Shared `ImageUrlListSuccess` — `data.image_urls` carries one CDN URL of the cleaned image.

## Billing & limits

- 5 credits per request from the unified Codia Open API balance.
- Synchronous.

## Errors

Standard set; in particular `400` on size mismatch between image and mask.

> **Sync vs. async**
> This synchronous endpoint returns the result inline — the simplest path. For production pipelines that need webhook delivery, automatic retry with idempotency, or explicit task cancellation, call `POST /v2/open/tasks` with `operation=image_object_erase` instead. Inputs and credit cost are identical.

可选，由用户自定义的幂等键（最多 255 个字符）。设置后，成功（2xx）的响应会缓存 7 天。使用相同的键 + 相同的请求体重试，会返回缓存的响应（不额外扣除积分）；使用相同的键但不同的请求体重试，则返回 409。用于确保 POST 请求在网络故障时可以安全重试。对于 multipart/form-data 上传，请求会按表单字段和文件内容进行比较，因此即使客户端生成了新的 multipart boundary，重试的上传也会被重放。

Idempotency-Key

image_url

遮罩图片的公开 URL（白色 = 要擦除的区域，黑色 = 保留）

mask_url

执行局部重绘（inpainting）的模型。默认为 `codia_image_v2`。目前所有支持的模型每次请求均消耗 5 积分。

| 模型 | 适用场景 | 积分 |
| --- | --- | ---: |
| `codia_image_v2` | 干净、快速的局部重绘，适用于照片和 UI。周围内容简单时效果最佳。 | 5 |
| `nano_banana_2` | 具备 AI 理解能力的擦除，可重建复杂的周边内容（Gemini 3.1 Flash）。 | 5 |
| `nano_banana_pro` | 高保真擦除，可重建精细复杂的背景（Gemini 3 Pro）。 | 5 |

model

image

兼容用的源图片文件字段（已弃用：请改用 `image`）。

file

mask

不上传遮罩时使用的遮罩图片 URL

code

message

image_urls

data

请求体格式错误或缺少必填参数

Idempotency-Key 已被用于不同的请求体。

请求频率超限。请等待 `Retry-After` 响应头中指定的秒数后再重试。

移除遮罩标记的区域并自动填充，看起来就像该物体从未存在过。

物体擦除

输入

输出

相关 API