What are Predictions

Understanding predictions (tasks) in WaveSpeedAI.

Overview

A prediction is a single request to generate content using an AI model. Each prediction has:

• A unique ID for tracking
• A status indicating progress
• Outputs containing URLs to generated content

Prediction Lifecycle

Created → Processing → Completed (or Failed)

Status Values

Creating a Prediction

Send a POST request to a model endpoint:

curl --fail-with-body --connect-timeout 10 --max-time 60 -X POST "https://api.wavespeed.ai/api/v3/wavespeed-ai/z-image/turbo" \
  -H "Authorization: Bearer $WAVESPEED_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "A cat in space", "size": "1024*1024"}'

Response:

{
  "code": 200,
  "data": {
    "id": "pred_abc123",
    "status": "created",
    "urls": {
      "get": "https://api.wavespeed.ai/api/v3/predictions/pred_abc123/result"
    }
  }
}

Checking Prediction Status

Poll the prediction endpoint:

curl --fail-with-body --connect-timeout 10 --max-time 30 \
  "https://api.wavespeed.ai/api/v3/predictions/pred_abc123/result" \
  -H "Authorization: Bearer $WAVESPEED_API_KEY"

Use a controlled polling interval for each prediction. Start around 2 seconds and increase toward 5–10 seconds for long-running tasks. Avoid querying the same prediction more often than every 2 seconds.

Response when completed:

{
  "code": 200,
  "data": {
    "id": "pred_abc123",
    "status": "completed",
    "outputs": ["https://...generated-image.png"],
    "timings": {
      "inference": 2500
    }
  }
}

Viewing Predictions

Web Interface

View all your predictions at wavespeed.ai/history.

API

List recent predictions:

curl --fail-with-body --connect-timeout 10 --max-time 60 \
  -X POST "https://api.wavespeed.ai/api/v3/predictions" \
  -H "Authorization: Bearer $WAVESPEED_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"page": 1, "page_size": 20}'

Both fields are required. Increment page for the next page, or use the cursor-pagination option described in the Predictions API for long-running synchronization.

Data Retention

• Generated media URLs are temporary and generally expire after 7 days
• Download outputs you need to retain
• Prediction metadata and history may remain available longer than the output files

Alternatives to Polling

Instead of repeatedly checking status, you can use:

• Webhooks — Receive notifications when tasks complete
• Streaming — Get real-time progress updates (supported by some models)

See How to Use Webhooks for details.

Next Steps

• How to Submit Task
• How to Get Result
• How to Use Webhooks