image-redraw

Image Redraw — FastAPI + R2 + Queues + Workflows + Workers AI

Upload a picture and get it back redrawn as a wobbly MS Paint doodle.

This example demonstrates how to leverage multiple Cloudflare services to create a full-featured image processing pipeline using Python Workers.

What it uses

• Python Workers — the runtime
• FastAPI — the HTTP API, served over ASGI
• R2 — stores uploaded originals and redrawn outputs
• Queues — hands work off the request path
• Workflows — durable, retrying background execution
• Workers AI — image generation guided by the uploaded reference
• Pillow — image normalization inside the Workflow
• Static Assets — the plain HTML/CSS/JS frontend

flowchart LR
    Browser["Browser"]
    Worker["Python Worker<br/>FastAPI"]
    Queue[["Queue"]]
    Workflow["Workflow"]
    R2[("R2")]
    AI["Workers AI"]

    Browser --> Worker
    Worker --> R2
    Worker --> Queue
    Queue --> Workflow
    Workflow --> R2
    Workflow --> AI

Loading

The flow

1. The browser POSTs the image bytes to the Worker.
2. FastAPI validates the image and stores the original in R2, and enqueues the job.
3. The queue consumer turns each batch of IDs into Workflow instances.
4. The Workflow reads the original from R2, normalizes it with Pillow, calls Workers AI, and stores the result back in R2.

Setup

First ensure that uv is installed: https://docs.astral.sh/uv/getting-started/installation/#standalone-installer

Workers AI is a remote binding, even during local development. wrangler.jsonc declares "ai": { "binding": "AI", "remote": true }, so inference always runs on Cloudflare's network and bills against your account. Log in before starting the dev server:

uv run pwrangler login

How to Run

uv run pywrangler dev

Then open http://localhost:8787/ in your browser.

How to deploy

uv run pywrangler deploy