Complete Workflow Tutorial

Learn how to combine multiple models to create complex AI generation workflows.

Overview

This tutorial shows how to chain multiple WaveSpeedAI models together to create a complete content generation workflow.

Example: Text → Image → Video → Digital Human

Create a talking head video from just a text description. Each submission returns a prediction ID, not the final media. Poll data.urls.get starting around every 2 seconds (and less often for long-running tasks) until the step reaches a terminal status before starting the dependent step.

Generate a face image (Text-to-Image)
Generate speech audio (Text-to-Speech)
Animate with lip sync (Digital Human)

Step 1: Generate a Face Image

curl --fail-with-body --connect-timeout 10 --max-time 60 --request POST 'https://api.wavespeed.ai/api/v3/wavespeed-ai/z-image/turbo' \
--header "Content-Type: application/json" \
--header "Authorization: Bearer ${WAVESPEED_API_KEY}" \
--data-raw '{
  "prompt": "Professional headshot of a young woman, neutral expression, studio lighting, white background, high quality",
  "size": "1024*1024"
}'

Poll the returned result URL. After the prediction is completed, save data.outputs[0] for step 3.

Step 2: Generate Speech Audio

curl --fail-with-body --connect-timeout 10 --max-time 60 --request POST 'https://api.wavespeed.ai/api/v3/minimax/speech-2.6-hd' \
--header "Content-Type: application/json" \
--header "Authorization: Bearer ${WAVESPEED_API_KEY}" \
--data-raw '{
  "text": "Hello! Welcome to WaveSpeedAI. Let me show you how easy it is to create AI-generated content.",
  "voice_id": "Friendly_Person",
  "emotion": "happy",
  "speed": 1
}'

Poll the returned result URL. After the prediction is completed, save data.outputs[0] for step 3.

Step 3: Create Digital Human Video

Use the image and audio URLs from the previous steps:

curl --fail-with-body --connect-timeout 10 --max-time 60 --request POST 'https://api.wavespeed.ai/api/v3/wavespeed-ai/infinitetalk' \
--header "Content-Type: application/json" \
--header "Authorization: Bearer ${WAVESPEED_API_KEY}" \
--data-raw '{
  "image": "https://...your-generated-face-image...",
  "audio": "https://...your-generated-audio...",
  "resolution": "480p",
  "seed": -1
}'

Poll the returned result URL. When the prediction is completed, the final video is in data.outputs[0].

Example: Character Animation

Make a person in an image follow movements from a reference video:

Upload your image (the person to animate)
Upload a reference video (the movements to follow)
Generate the animated video

Step 1: Upload Files

Upload your image and reference video to get URLs.

Step 2: Generate Animation

curl --fail-with-body --connect-timeout 10 --max-time 60 --request POST 'https://api.wavespeed.ai/api/v3/wavespeed-ai/wan-2.2/animate' \
--header "Content-Type: application/json" \
--header "Authorization: Bearer ${WAVESPEED_API_KEY}" \
--data-raw '{
  "image": "https://your-uploaded-image-url",
  "video": "https://your-reference-video-url",
  "mode": "replace",
  "resolution": "480p",
  "seed": -1
}'

Poll the returned result URL before reading data.outputs[0]. The person in your image will replicate the movements and expressions from the reference video.

Example: Concurrent Image Generation

There is no dedicated batch-submission endpoint. To generate multiple images, make individual API calls concurrently within your account limits and track each returned prediction ID.

# Image 1
curl --fail-with-body --connect-timeout 10 --max-time 60 --request POST 'https://api.wavespeed.ai/api/v3/wavespeed-ai/z-image/turbo' \
--header "Content-Type: application/json" \
--header "Authorization: Bearer ${WAVESPEED_API_KEY}" \
--data-raw '{
  "prompt": "A cinematic portrait on a city street",
  "size": "1024*1024"
}'
 
# Image 2
curl --fail-with-body --connect-timeout 10 --max-time 60 --request POST 'https://api.wavespeed.ai/api/v3/wavespeed-ai/z-image/turbo' \
--header "Content-Type: application/json" \
--header "Authorization: Bearer ${WAVESPEED_API_KEY}" \
--data-raw '{
  "prompt": "A cinematic portrait in a coffee shop",
  "size": "1024*1024"
}'

For character or style consistency with a compatible LoRA model, see LoRA Training & Usage and use that model’s documented schema.

Best Practices

Error Handling

Check both the HTTP status and the response envelope’s code, message, and data.status
Retry idempotent result-query GET requests only for network errors, 429, and 5xx; honor Retry-After
Do not blindly retry submission POST requests because an ambiguous disconnect can create a duplicate prediction
Start polling around 2 seconds and increase toward 5–10 seconds for long-running tasks
Stop on completed, failed, cancelled, timeout, or your own deadline
Use webhooks for long-running tasks

Cost Management

Test with lower resolution first
Use webhooks instead of polling for long tasks
Monitor credit usage via API

Rate Limiting

Add small delays between requests when making many calls
Use webhooks to avoid polling overhead
Batch similar operations when possible

Next Steps

How to Create a Digital Human Overview