🎨 Image Generation with Replicate


This template uses the Replicate API to run image generation models like Stable Diffusion XL (SDXL), allowing users to generate images based on a text prompt.

How it Works

The process involves making API calls to Replicate's /predictions endpoint. Here's a short breakdown:

  1. Submit a Prediction Request

    • The frontend initiates the image generation by sending a POST request to the /api/replicate NextJS API route. This request includes a payload with the user's prompt.
    • The API route passes this request off to Replicate's /predictions API endpoint.
  2. Check Prediction Status

    • After the initial request, the frontend keeps checking the status of the prediction by sending GET requests to /api/replicate/[prediction_id]. While the prediction is ongoing, the app displays a loading spinner.
  3. Handling Success or Failure

    • Once the prediction is completed, the app updates the UI. If the prediction is successful, the generated image is displayed to the user.
  4. Uploading to Supabase

    • If the prediction is successful, its details are stored in Supabase in the sdxl_images table. This includes a URL to the image stored on Replicate's database. (Note that Replicate automatically deletes generated images in a few hours, so you need to store them separately in Supabase storage to make them persistent.)

3-Min Walkthrough


Steps taken on the frontend

1. Making a Prediction Request

To generate an image, users enter a prompt on the /dashboard/generate page. The app then sends this prompt to Replicate for processing.

// Example of submitting a prediction request
const handleSubmit = async (e: any) => {
  e.preventDefault();
  const response = await fetch('/api/replicate', {
    method: 'POST',
    headers: {
      'Content-Type': 'application/json',
    },
    body: JSON.stringify({
      input: {
        prompt: e.target.prompt.value,
      },
    }),
  });
 
  // Handling response...
};

2. Checking Prediction Status

The app continuously checks the status of the prediction until it's either succeeded or failed.

// Example of checking prediction status
while (prediction.status !== 'succeeded' && prediction.status !== 'failed') {
  await sleep(1000);
  const response = await fetch('/api/replicate/' + prediction.id);
  prediction = await response.json();
  // Handling response...
}

3. Uploading to Supabase

If the prediction is successful, the generated image is uploaded to Supabase.

// Example of uploading generated image to Supabase
if (prediction.status === 'succeeded') {
  await uploadGeneratedImage(prediction);
}

Setup

Configure the following in your app to start using image generation:

1. Replicate API Token

Go to your Replicate Account, copy the API Token, and set it in your .env file.

# Update this with your Replicate API Key
REPLICATE_API_TOKEN=r8_cyh*****************************

2. Model and Version

Specify the Replicate model and version you want to run in your environment. By default, the .env.example pins this to a version of Stable Diffusion:

# Update this with the model and version you want to run
REPLICATE_MODEL_VERSION=2b017d9b67edd2ee1401238df49d75da53c523f36e363881e057f5dc3ed3c5b2

You can find the version for any model on the Versions tab of the model card on Replicate's website (https://replicate.com/[company]/[model]/versions).

Replicate Version


Last Updated: June 7