Name: DiffBIR API: Image Restoration and 4x Upscaling
Brand: magicapi
SKU: api-magicapi-blind-image-restoration-zsxkib-diffbir
Availability: InStock
Author: API Market

Shashank Agarwal

# 🪄 Image Restoration and Upscaling API (XPixelGroup DiffBIR)

## Overview
DiffBIR is a blind image restoration model from XPixelGroup that uses a generative diffusion prior. Blind restoration means it repairs photos without knowing exactly how they were degraded, for example by blur, noise or compression. This listing is packaged by zsxkib.

Send an image URL in `input`. A SwinIR preprocessing step first removes degradations, then the diffusion model restores detail and upscales the image by `super_resolution_factor` (1 to 4, default 4). Choose `faces` or `general_scenes` models for restoration and upscaling. The API returns a list of image URLs; the full restored image is the last item.

## Features
- **Blind restoration**: repairs blur, noise and other unknown degradations in real photos.
- **Upscaling**: `super_resolution_factor` from 1 to 4 (default 4) in both width and height.
- **Faces or general scenes**: `upscaling_model_type` and `restoration_model_type` pick models for portraits (`faces`) or everything else (`general_scenes`).
- **Face mode options**: face detection model, `has_aligned`, `only_center_face`, and a background upsampler (`RealESRGAN` or `DiffBIR`) with its own tile settings.
- **Quality controls**: `steps` (1 to 100, default 50), `repeat_times` (1 to 10), `color_fix_type` (`wavelet`, `adain` or `none`) and a fixed `seed` (default 231).
- **Large images**: `tiled` processes the image in patches of `tile_size` with `tile_stride` overlap.
- **Guidance (general scenes)**: optional latent image guidance with scale, start and stop step, space and repeat settings.

## Use Cases
1. **Old and damaged photos**: restore scanned prints and family photos with blur or noise.
2. **Portrait clean-up**: restore faces in low-quality portraits with the `faces` models.
3. **Low-resolution images**: upscale small or compressed images up to 4x while restoring detail.
4. **Archive digitization**: improve image quality in bulk before cataloging.

## Visual Examples
### Example Input
A low-quality photo with a face in `input`, upscaled 2x (`super_resolution_factor` 2) with the `faces` upscaling model, the `general_scenes` restoration model and the `DiffBIR` background upsampler. All other settings are the defaults:

![DiffBIR example input: a low-quality photo with a face](https://blog.api.market/wp-content/uploads/2026/10/diffbir-input-api-market.jpg)

```json
{
  "seed": 231,
  "input": "https://blog.api.market/wp-content/uploads/2026/10/diffbir-input-api-market.jpg",
  "steps": 50,
  "tiled": false,
  "tile_size": 512,
  "has_aligned": false,
  "tile_stride": 256,
  "repeat_times": 1,
  "use_guidance": false,
  "color_fix_type": "wavelet",
  "guidance_scale": 0,
  "guidance_space": "latent",
  "guidance_repeat": 5,
  "only_center_face": false,
  "guidance_time_stop": -1,
  "guidance_time_start": 1001,
  "background_upsampler": "DiffBIR",
  "face_detection_model": "retinaface_resnet50",
  "upscaling_model_type": "faces",
  "restoration_model_type": "general_scenes",
  "super_resolution_factor": 2,
  "disable_preprocess_model": false,
  "reload_restoration_model": false,
  "background_upsampler_tile": 400,
  "background_upsampler_tile_stride": 400
}
```

### Example Output
The request returned three images. The first two are images of the detected face; the last is the full restored image:

![DiffBIR example output 1: detected face](https://blog.api.market/wp-content/uploads/2026/10/diffbir-output-1-a5wv6csf-api-market.png)

![DiffBIR example output 2: detected face](https://blog.api.market/wp-content/uploads/2026/10/diffbir-output-2-a5wv6csf-api-market.png)

![DiffBIR example output 3: full restored image, upscaled 2x](https://blog.api.market/wp-content/uploads/2026/10/diffbir-output-3-a5wv6csf-api-market.jpg)

## Input Parameters
| Field | Type | Required | Default | Allowed values | Description |
| --- | --- | --- | --- | --- | --- |
| `input` | string (URL) | Yes | none |  | Input image URL to restore |
| `upscaling_model_type` | string | No | `general_scenes` | `faces`, `general_scenes` | Model type for upscaling: `faces` for portraits, `general_scenes` for everything else |
| `restoration_model_type` | string | No | `general_scenes` | `faces`, `general_scenes` | Restoration model that removes degradations; pick the one that matches the image content |
| `reload_restoration_model` | boolean | No | `false` |  | Reload the SwinIR restoration model before running. Normally leave this `false`. |
| `steps` | integer | No | `50` | `1` to `100` | Number of diffusion steps. More steps can give a clearer image but can also introduce artifacts. |
| `super_resolution_factor` | integer | No | `4` | `1` to `4` | Factor by which width and height are increased, for example 4 makes the image 4 times wider and taller |
| `repeat_times` | integer | No | `1` | `1` to `10` | Number of times the enhancement is repeated, feeding the output back as input. Can refine the result but can also over-enhance it. |
| `disable_preprocess_model` | boolean | No | `false` |  | Skip the initial SwinIR preprocessing (restoration) step. Set to `true` if the input image is already of high quality. |
| `tiled` | boolean | No | `false` |  | Use patch-based sampling, useful for very large images |
| `tile_size` | integer | No | `512` |  | Size of each tile when `tiled` is on |
| `tile_stride` | integer | No | `256` |  | Distance between the starts of neighbouring tiles when `tiled` is on. A smaller stride means more overlap. |
| `use_guidance` | boolean | No | `false` |  | Use latent image guidance during enhancement |
| `guidance_scale` | number | No | `0` |  | For `general_scenes`: scale of the guidance |
| `guidance_time_start` | integer | No | `1001` |  | For `general_scenes`: step at which guidance starts |
| `guidance_time_stop` | integer | No | `-1` |  | For `general_scenes`: step at which guidance stops |
| `guidance_space` | string | No | `latent` | `rgb`, `latent` | For `general_scenes`: space in which guidance operates (`rgb` or `latent`) |
| `guidance_repeat` | integer | No | `5` |  | For `general_scenes`: number of times the guidance process is repeated |
| `color_fix_type` | string | No | `wavelet` | `wavelet`, `adain`, `none` | Color correction applied after enhancement: `wavelet`, `adain` or `none` to skip it |
| `seed` | integer | No | `231` |  | Random seed. The same seed and input give the same output. |
| `has_aligned` | boolean | No | `false` |  | For `faces` mode: whether the input is already a cropped and aligned face. If not, the model detects and aligns faces. |
| `only_center_face` | boolean | No | `false` |  | For `faces` mode: when several faces are found, only enhance the one closest to the center |
| `face_detection_model` | string | No | `retinaface_resnet50` | `retinaface_resnet50`, `retinaface_mobile0.25`, `YOLOv5l`, `YOLOv5n`, `dlib` | For `faces` mode: model used to detect faces |
| `background_upsampler` | string | No | `RealESRGAN` | `DiffBIR`, `RealESRGAN` | For `faces` mode: model used to upscale the background around faces |
| `background_upsampler_tile` | integer | No | `400` |  | For `faces` mode: tile size used by the background upsampler |
| `background_upsampler_tile_stride` | integer | No | `400` |  | For `faces` mode: distance between the starts of background tiles. A smaller stride means more overlap. |

## Pricing
- **Pay As You Go (PAYG) pricing**: you only pay for what you use.
- **Price**: $0.105 per API unit.
- **1 API unit = 1 successful request with up to 96 seconds of processing.** A request that takes longer is billed one more unit for each started block of 96 seconds (for example, 193 seconds of processing is 3 units, $0.315).
- More `steps`, a higher `super_resolution_factor`, more `repeat_times` or larger images take longer and can use more units.
- **Example cost**: the example request above took 55.4 seconds of processing, so it used 1 API unit and cost $0.105.
- **Failed or cancelled predictions are not billed.**
- **GET requests to check prediction status** are free and don't count toward API unit usage.
- **No monthly subscription fee.**

## API Documentation

### Base URL
All API requests go to: `https://prod.api.market/api/v1/magicapi/blind-image-restoration-zsxkib-diffbir`

### Authentication
Send your api.market API key in the `x-api-market-key` header:
```
x-api-market-key: YOUR_API_MARKET_KEY
```

### Asynchronous Processing
**IMPORTANT**: This API is asynchronous. The flow is:
1. Send a POST request to `/predictions`. A successful request returns **201 Created** with a prediction `id` and a `status` of `starting` or `processing`.
2. Poll GET `/predictions/{id}` with that ID.
3. Keep polling while `status` is `starting` or `processing`.
4. Stop when `status` is `succeeded` (read `output`), `failed` or `canceled` (read `error`).

A polling interval of 1 second and an overall timeout of 600 seconds (10 minutes) work well.

### Endpoints

#### Create a Prediction
```bash
curl -X POST "https://prod.api.market/api/v1/magicapi/blind-image-restoration-zsxkib-diffbir/predictions" \
  -H "Content-Type: application/json" \
  -H "x-api-market-key: YOUR_API_MARKET_KEY" \
  -d '{
    "version": "51ed1464d8bbbaca811153b051d3b09ab42f0bdeb85804ae26ba323d7a66a4ac",
    "input": {
      "seed": 231,
      "input": "https://blog.api.market/wp-content/uploads/2026/10/diffbir-input-api-market.jpg",
      "steps": 50,
      "tiled": false,
      "tile_size": 512,
      "has_aligned": false,
      "tile_stride": 256,
      "repeat_times": 1,
      "use_guidance": false,
      "color_fix_type": "wavelet",
      "guidance_scale": 0,
      "guidance_space": "latent",
      "guidance_repeat": 5,
      "only_center_face": false,
      "guidance_time_stop": -1,
      "guidance_time_start": 1001,
      "background_upsampler": "DiffBIR",
      "face_detection_model": "retinaface_resnet50",
      "upscaling_model_type": "faces",
      "restoration_model_type": "general_scenes",
      "super_resolution_factor": 2,
      "disable_preprocess_model": false,
      "reload_restoration_model": false,
      "background_upsampler_tile": 400,
      "background_upsampler_tile_stride": 400
    }
  }'
```

Response (201 Created):
```json
{
  "id": "a5wv6csfp9rmt0d1d1d82s10hm",
  "version": "51ed1464d8bbbaca811153b051d3b09ab42f0bdeb85804ae26ba323d7a66a4ac",
  "status": "starting",
  "created_at": "2026-10-23T10:41:43.526Z"
}
```

#### Check Prediction Status
```bash
curl -X GET "https://prod.api.market/api/v1/magicapi/blind-image-restoration-zsxkib-diffbir/predictions/{id}" \
  -H "x-api-market-key: YOUR_API_MARKET_KEY"
```

Response when the prediction has finished (200 OK):
```json
{
  "id": "a5wv6csfp9rmt0d1d1d82s10hm",
  "status": "succeeded",
  "output": [
    "https://blog.api.market/wp-content/uploads/2026/10/diffbir-output-1-a5wv6csf-api-market.png",
    "https://blog.api.market/wp-content/uploads/2026/10/diffbir-output-2-a5wv6csf-api-market.png",
    "https://blog.api.market/wp-content/uploads/2026/10/diffbir-output-3-a5wv6csf-api-market.jpg"
  ],
  "error": null,
  "created_at": "2026-10-23T10:41:43.526Z",
  "started_at": "2026-10-23T10:44:42.877Z",
  "completed_at": "2026-10-23T10:45:38.274Z"
}
```

`output` is a list of image URLs. The last item is the full restored image. In face mode the list also contains images of the detected faces, which come before the full image.

### Status Codes
- **201 Created**: prediction created (POST `/predictions`). This is the normal success response.
- **200 OK**: prediction details returned (GET `/predictions/{id}`).
- **400 Bad Request**: invalid input, for example a missing `input`, a `super_resolution_factor` above 4 or `steps` above 100.
- **401 Unauthorized**: missing or invalid API key.
- **404 Not Found**: no prediction with that ID.
- **429 Too Many Requests**: rate limit exceeded.
- **500 Internal Server Error** or **503 Service Unavailable**: temporary server problem; retry later.

## Rate Limits
All plans allow 10 requests per second. Requests above that return `429 Too Many Requests`; back off before retrying and honour any `Retry-After` header. Polling a prediction with GET is free but still counts toward the per-second limit, so poll every 1 to 5 seconds rather than in a tight loop.

## Error Handling
- Check the HTTP status code of every request and read the response body for the error message.
- A prediction can also end with `status` set to `failed`; the `error` field then explains why. Failed predictions are not billed.
- Retry 429, 500 and 503 responses with exponential backoff. Fix the input before retrying a 400.

## Implementation Guide
### Example Code

#### JavaScript
```javascript
const apiKey = "YOUR_API_MARKET_KEY";
const baseUrl = "https://prod.api.market/api/v1/magicapi/blind-image-restoration-zsxkib-diffbir";

async function createAndWaitForPrediction(input) {
  const response = await fetch(`${baseUrl}/predictions`, {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      "x-api-market-key": apiKey
    },
    body: JSON.stringify({ version: "51ed1464d8bbbaca811153b051d3b09ab42f0bdeb85804ae26ba323d7a66a4ac", input })
  });

  if (response.status !== 201) {
    throw new Error(`API error: ${response.status} ${await response.text()}`);
  }

  const prediction = await response.json();
  return pollForResults(prediction.id);
}

const delay = (ms) => new Promise((resolve) => setTimeout(resolve, ms));

async function pollForResults(predictionId) {
  const maxTimeout = 600000;
  const pollInterval = 1000;
  const startTime = Date.now();

  while (Date.now() - startTime < maxTimeout) {
    const response = await fetch(`${baseUrl}/predictions/${predictionId}`, {
      headers: { "x-api-market-key": apiKey }
    });

    if (!response.ok) {
      throw new Error(`API error: ${response.status} ${await response.text()}`);
    }

    const prediction = await response.json();

    if (prediction.status === "succeeded") {
      return prediction.output;
    }

    if (prediction.status === "failed" || prediction.status === "canceled") {
      throw new Error(`Prediction ${prediction.status}: ${prediction.error || "Unknown error"}`);
    }

    await delay(pollInterval);
  }

  throw new Error("Prediction timed out after 10 minutes");
}

createAndWaitForPrediction({
  "seed": 231,
  "input": "https://blog.api.market/wp-content/uploads/2026/10/diffbir-input-api-market.jpg",
  "steps": 50,
  "tiled": false,
  "tile_size": 512,
  "has_aligned": false,
  "tile_stride": 256,
  "repeat_times": 1,
  "use_guidance": false,
  "color_fix_type": "wavelet",
  "guidance_scale": 0,
  "guidance_space": "latent",
  "guidance_repeat": 5,
  "only_center_face": false,
  "guidance_time_stop": -1,
  "guidance_time_start": 1001,
  "background_upsampler": "DiffBIR",
  "face_detection_model": "retinaface_resnet50",
  "upscaling_model_type": "faces",
  "restoration_model_type": "general_scenes",
  "super_resolution_factor": 2,
  "disable_preprocess_model": false,
  "reload_restoration_model": false,
  "background_upsampler_tile": 400,
  "background_upsampler_tile_stride": 400
}).then((output) => console.log(output));
```

#### Python
```python
import time
import requests

API_KEY = "YOUR_API_MARKET_KEY"
BASE_URL = "https://prod.api.market/api/v1/magicapi/blind-image-restoration-zsxkib-diffbir"


def create_and_wait_for_prediction(input_data):
    headers = {"Content-Type": "application/json", "x-api-market-key": API_KEY}
    response = requests.post(
        f"{BASE_URL}/predictions",
        headers=headers,
        json={"version": "51ed1464d8bbbaca811153b051d3b09ab42f0bdeb85804ae26ba323d7a66a4ac", "input": input_data},
    )

    if response.status_code != 201:
        raise Exception(f"API error: {response.status_code} {response.text}")

    return poll_for_results(response.json()["id"])


def poll_for_results(prediction_id):
    max_timeout = 600
    poll_interval = 1
    headers = {"x-api-market-key": API_KEY}
    start_time = time.time()

    while time.time() - start_time < max_timeout:
        response = requests.get(f"{BASE_URL}/predictions/{prediction_id}", headers=headers)

        if response.status_code != 200:
            raise Exception(f"API error: {response.status_code} {response.text}")

        prediction = response.json()

        if prediction["status"] == "succeeded":
            return prediction["output"]

        if prediction["status"] in ("failed", "canceled"):
            raise Exception(f"Prediction {prediction['status']}: {prediction.get('error') or 'Unknown error'}")

        time.sleep(poll_interval)

    raise Exception("Prediction timed out after 10 minutes")


print(create_and_wait_for_prediction({
    "seed": 231,
    "input": "https://blog.api.market/wp-content/uploads/2026/10/diffbir-input-api-market.jpg",
    "steps": 50,
    "tiled": False,
    "tile_size": 512,
    "has_aligned": False,
    "tile_stride": 256,
    "repeat_times": 1,
    "use_guidance": False,
    "color_fix_type": "wavelet",
    "guidance_scale": 0,
    "guidance_space": "latent",
    "guidance_repeat": 5,
    "only_center_face": False,
    "guidance_time_stop": -1,
    "guidance_time_start": 1001,
    "background_upsampler": "DiffBIR",
    "face_detection_model": "retinaface_resnet50",
    "upscaling_model_type": "faces",
    "restoration_model_type": "general_scenes",
    "super_resolution_factor": 2,
    "disable_preprocess_model": False,
    "reload_restoration_model": False,
    "background_upsampler_tile": 400,
    "background_upsampler_tile_stride": 400
}))
```

## FAQ
- **Should I use `faces` or `general_scenes`?**
  - Use `faces` for portraits and photos where faces matter most, and `general_scenes` for landscapes, objects and everything else. You can set the upscaling and restoration models separately.
- **Which output image is the final result?**
  - The last URL in `output` is the full restored image. In face mode the earlier items are images of the detected faces.
- **How much can it upscale?**
  - Up to 4x in width and height with `super_resolution_factor` (1 to 4, default 4). Use 1 to restore without enlarging.
- **How do I get the same result twice?**
  - Keep the same `seed` (default 231) and the same input and settings.
- **How is a request billed?**
  - $0.105 per request for up to 96 seconds of processing, plus one unit for each further started block of 96 seconds. The example took 55.4 seconds and cost $0.105.
- **Is the 201 response from POST an error?**
  - No. 201 Created means the prediction was accepted. Use the returned `id` to poll GET `/predictions/{id}` until `status` is `succeeded`, `failed` or `canceled`.
- **What if my prediction fails?**
  - Failed predictions are not billed. Read the `error` field, adjust the input and try again.

## Contact Us
If you have any questions, reach out to us through the chat on the bottom right of the page. You can also email our support team at hello@magicapi.com.


Restore blurry, noisy or compressed photos and faces and upscale them up to 4x with DiffBIR, a diffusion-based blind image restoration model.