# Image Processing

<Note>
  Since parallel steps are not yet available in
  [workflow-py](https://github.com/upstash/workflow-py), Python example is implemented in a sequential manner. See our
  [Roadmap](/workflow/roadmap) for feature parity plans and
  [Changelog](/workflow/changelog) for updates.
</Note>

## Introduction

This example demonstrates how to process images using Upstash Workflow. The following workflow will upload an image, resize it into multiple resolutions, apply filters, and store the processed versions for later retrieval.

## Use Case

Our workflow will:

1. Receive an image upload request
2. Resize the image into different resolutions
3. Apply various filters to the resized images
4. Store the processed images in a cloud storage

## Code Example

<CodeGroup>

```typescript api/workflow/route.ts
import { serve } from "@upstash/workflow/nextjs"
import {
  resizeImage,
  applyFilters,
  storeImage,
  getImageUrl,
} from "./utils"

type ImageResult = {
  imageUrl: string
}

export const { POST } = serve<{ imageId: string; userId: string }>(
  async (context) => {
    const { imageId, userId } = context.requestPayload

    // Step 1: Retrieve the uploaded image
    const imageUrl = await context.run("get-image-url", async () => {
      return await getImageUrl(imageId)
    })

    // Step 2: Resize the image to multiple resolutions
    const resolutions = [640, 1280, 1920]

    const resizedImages: { body: ImageResult }[] = await Promise.all(resolutions.map(
      resolution => context.call<ImageResult>(
        `resize-image-${resolution}`,
        {
          // endpoint which returns ImageResult type in response
          url: "https://image-processing-service.com/resize",
          method: "POST",
          body: JSON.stringify({
            imageUrl,
            width: resolution,
          })
        }
      )
    ))

    // Step 3: Apply filters to each resized image
    const filters = ["grayscale", "sepia", "contrast"]
    const processedImagePromises: Promise<string>[] = []

    for (const resizedImage of resizedImages) {
      for (const filter of filters) {
        const processedImagePromise = context.call<ImageResult>(
          `apply-filter-${filter}`,
          {
            // endpoint which returns ImageResult type in response
            url: "https://image-processing-service.com/filter",
            method: "POST",
            body: JSON.stringify({
              imageUrl: resizedImage.body.imageUrl,
              filter,
            })
          }
        )
        processedImagePromises.push(processedImagePromise)
      }
    }
    const processedImages: { body: ImageResult }[] = await Promise.all(processedImagePromises)

    // Step 4: Store processed images in cloud storage
    const storedImageUrls: string[] = await Promise.all(
      processedImages.map(
        processedImage => context.run(`store-image`, async () => {
          return await storeImage(processedImage.body.imageUrl)
        })
      )
    )
  }
)
```

```python main.py
from fastapi import FastAPI
from typing import List, TypedDict
from upstash_workflow.fastapi import Serve
from upstash_workflow import AsyncWorkflowContext, CallResponse
from utils import store_image, get_image_url

app = FastAPI()
serve = Serve(app)


class ImageResult(TypedDict):
    image_url: str


class ImageProcessingPayload(TypedDict):
    image_id: str
    user_id: str


@serve.post("/process-image")
async def process_image(context: AsyncWorkflowContext[ImageProcessingPayload]) -> None:
    payload = context.request_payload
    image_id = payload["image_id"]
    user_id = payload["user_id"]

    # Step 1: Retrieve the uploaded image
    async def _get_image_url() -> str:
        return await get_image_url(image_id)

    image_url: str = await context.run("get-image-url", _get_image_url)

    # Step 2: Resize the image to multiple resolutions
    resolutions = [640, 1280, 1920]
    resize_responses = []

    for resolution in resolutions:
        response: CallResponse[ImageResult] = await context.call(
            f"resize-image-{resolution}",
            # endpoint which returns ImageResult type in response
            url="https://image-processing-service.com/resize",
            method="POST",
            body={"imageUrl": image_url, "width": resolution},
        )
        resize_responses.append(response)

    resized_images = [response.body for response in resize_responses]

    # Step 3: Apply filters to each resized image
    filters = ["grayscale", "sepia", "contrast"]
    filter_responses = []

    for resized_image in resized_images:
        for filter in filters:
            response: CallResponse[ImageResult] = await context.call(
                f"apply-filter-{filter}",
                # endpoint which returns ImageResult type in response
                url="https://image-processing-service.com/filter",
                method="POST",
                body={"imageUrl": resized_image["imageUrl"], "filter": filter},
            )
            filter_responses.append(response)

    processed_images = [response.body for response in filter_responses]

    # Step 4: Store processed images in cloud storage
    async def _store_image() -> str:
        return await store_image(processed_image["imageUrl"])

    stored_image_urls: List[str] = []
    for processed_image in processed_images:
        stored_image_url = await context.run("store-image", _store_image)
        stored_image_urls.append(stored_image_url)

```

</CodeGroup>

## Code Breakdown

### 1. Retrieving the Image

We begin by getting the URL of the uploaded image based on its ID:

<CodeGroup>

```typescript TypeScript
const imageUrl = await context.run("get-image-url", async () => {
  return await getImageUrl(imageId)
})
```

```python Python
async def _get_image_url() -> str:
    return await get_image_url(image_id)

image_url: str = await context.run("get-image-url", _get_image_url)

```

</CodeGroup>

### 2. Resizing the Image

We resize the image into three different resolutions (640, 1280, 1920) using an external image processing service.

We call `context.call` for each resolution and use `Promise.all` to run them parallel:

<CodeGroup>

```typescript TypeScript
const resolutions = [640, 1280, 1920]

const resizedImages: { body: ImageResult }[] = await Promise.all(resolutions.map(
  resolution => context.call<ImageResult>(
    `resize-image-${resolution}`,
    {
      // endpoint which returns ImageResult type in response
      url: "https://image-processing-service.com/resize",
      method: "POST",
      body: JSON.stringify({
        imageUrl,
        width: resolution,
      })
    }
  )
))
```

```python Python
resolutions = [640, 1280, 1920]
resize_responses = []

for resolution in resolutions:
    response: CallResponse[ImageResult] = await context.call(
        f"resize-image-{resolution}",
        # endpoint which returns ImageResult type in response
        url="https://image-processing-service.com/resize",
        method="POST",
        body={"imageUrl": image_url, "width": resolution},
    )
    resize_responses.append(response)

resized_images = [response.body for response in resize_responses]

```

</CodeGroup>

### 3. Applying Filters

After resizing, we apply filters such as grayscale, sepia, and contrast to the resized images.

Again, we call `context.call` for each filter & image pair. We collect the promises of these requests in an array `processedImagePromise`. Then, we call `Promise.all` again to run them all parallel.

<CodeGroup>

```typescript TypeScript
const filters = ["grayscale", "sepia", "contrast"]
const processedImagePromises: Promise<string>[] = []

for (const resizedImage of resizedImages) {
  for (const filter of filters) {
    const processedImagePromise = context.call<ImageResult>(
      `apply-filter-${filter}`,
      {
        // endpoint which returns ImageResult type in response
        url: "https://image-processing-service.com/filter",
        method: "POST",
        body: {
          imageUrl: resizedImage.body.imageUrl,
          filter,
        }
      }
    )
    processedImagePromises.push(processedImagePromise)
  }
}
const processedImages: { body: ImageResult }[] = await Promise.all(processedImagePromises)
```

```python Python
filters = ["grayscale", "sepia", "contrast"]
filter_responses = []

for resized_image in resized_images:
    for filter in filters:
        response: CallResponse[ImageResult] = await context.call(
            f"apply-filter-{filter}",
            # endpoint which returns ImageResult type in response
            url="https://image-processing-service.com/filter",
            method="POST",
            body={"imageUrl": resized_image["imageUrl"], "filter": filter},
        )
        filter_responses.append(response)

processed_images = [response.body for response in filter_responses]

```

</CodeGroup>

### 4. Storing the Processed Images

We store each processed image in cloud storage and return the URLs of the stored images:

<CodeGroup>

```typescript TypeScript
const storedImageUrls: string[] = await Promise.all(
  processedImages.map(
    processedImage => context.run(`store-image`, async () => {
      return await storeImage(processedImage.body.imageUrl)
    })
  )
)
```

```python Python
async def _store_image() -> str:
    return await store_image(processed_image["imageUrl"])

stored_image_urls: List[str] = []
for processed_image in processed_images:
    stored_image_url = await context.run("store-image", _store_image)
    stored_image_urls.append(stored_image_url)

```

</CodeGroup>

## Key Features

1. **Batch Processing**: This workflow handles batch resizing and filtering of images in parallel to minimize processing time.
2. **Scalability**: Image resizing and filtering are handled through external services, making it easy to scale. Also, the requests to these services are handled by QStash, not by the compute where the workflow is hosted.
3. **Storage Integration**: The workflow integrates with cloud storage to persist processed images for future access.