Replicate

Create Account and Obtain API Token

Register for a Replicate account and obtain an authentication token to call the API.

Register at replicate.com, go to Account Settings > API Tokens, copy the token and save it to the environment variable `REPLICATE_API_TOKEN`.

Do not commit the token to git, use the .env file and add .env to .gitignore.

Select the Appropriate Model on Replicate Explore

Find a suitable image generation model that meets your needs and review the specific input/output documentation for that model.

Go to replicate.com/explore, search for a model like 'stability-ai/sdxl', view the API tab to know the required parameter format.

Test the model directly on Replicate's web UI first to understand the output before writing integration code.

Install SDK and Test API Call

Install the official client library and make a test call to the model from backend code.

Run `npm install replicate`, then call `replicate.run("stability-ai/sdxl:version-id", { input: { prompt: "..." } })`.

Always pin the correct version ID of the model in code to avoid unexpected output changes when the model is updated.

Handle Asynchronous Processing and Status Checks

Manage situations where the model does not return results immediately, needing to monitor the prediction status.

Use a webhook or polling via `replicate.predictions.get(id)` to check the status 'starting', 'processing', 'succeeded'.

Prefer using webhooks instead of continuous polling to reduce load and get results faster once processing is complete.

Save Results and Limit Costs

Save the resulting images to a separate storage and set limits to control usage costs.

Download the image from the result URL to save it in S3/Supabase Storage, while setting limits on the number of calls per user per day in the application code.

Monitor the usage dashboard on Replicate weekly to detect any unusual cost increases early.