A tutorial for building an n8n text to video workflow using a VideoPrompt agent and WaveSpeed Veo3, including request polling, prompt engineering, and notification delivery.
This walkthrough shows how to build a text-to-video workflow using n8n, a VideoPrompt agent (an LLM-based prompt engineer), and WaveSpeed’s Veo3 text-to-video models. It follows the same high-level pattern as a text-to-image pipeline but includes important differences for video generation:
The chat trigger forwards the user specification to a VideoPrompt agent instead of an ImagePrompt agent.
The model endpoint is a text-to-video model (Veo3) on WaveSpeed.
Video generation takes longer and costs more than image generation, so the workflow includes polling logic and loops to wait for completion.
Overview diagram and prompt guidance
What this workflow does
Accepts a natural-language description from a chat trigger.
Sends that description to a VideoPrompt agent that returns a polished, production-ready prompt.
Posts the prompt to WaveSpeed Veo3 via an HTTP Request node.
Polls the prediction status until the result is completed.
Sends a notification (Gmail/Slack/Telegram) with the final video URL.
VideoPrompt agent: prompt engineering as a service
The VideoPrompt agent is pre-populated with a system prompt instructing it to be an expert text-to-video prompt engineer. Use any LLM to craft that system prompt and paste it into the agent node. The agent transforms user chat input into a concise, vivid prompt suitable for Veo3.Example VideoPrompt UI and sample output
Sample VideoPrompt output (polished prompt)
“A lively, cinematic scene of five gorillas on a wooden fishing boat in the middle of a sunlit lake, laughing and cheering as they reel in big, thrashing fish; splashes of water in the golden sunlight; lush forested shores in the background; playful camaraderie; vibrant colors; dynamic camera movement capturing their joyous expressions and energetic fishing action.”Step 1 — Submit a generation request to WaveSpeed Veo3
Add an HTTP Request (POST) node in n8n to call the WaveSpeed Veo3 endpoint. In WaveSpeed’s model browser, filter by “Text to Video” and choose the Veo3 variant that matches your needs (e.g., veo3-fast for speed; veo3 for higher fidelity).
Veo3 models (especially higher-fidelity variants) can be expensive. While iterating, use short durations and lower resolutions to reduce cost.
Example cURL (POST) to submit a Veo3 Fast request:
curl --location --request POST 'https://api.wavespeed.ai/api/v3/google/veo3-fast' \ --header "Content-Type: application/json" \ --header "Authorization: Bearer ${WAVESPEED_API_KEY}" \ --data-raw '{ "aspect_ratio": "16:9", "duration": 8, "generate_audio": true, "prompt": "A charismatic woman in a dynamic ad set, wearing stylish clothes. She shouts with dramatic excitement: \"Google Veo 3 Fast is now on WaveSpeed AI! It'll blow your mind - go try it now!\"", "resolution": "720p" }'
n8n configuration notes for the POST node
Import the cURL into an HTTP Request node (e.g., name it WaveSpeedPost) or build the POST manually.
Use the same WaveSpeed credentials for authorization (e.g., header auth with Authorization: Bearer ${WAVESPEED_API_KEY}).
Disable automatic header sending if your credential node already supplies the Authorization header.
Use raw JSON body mode if automatic cURL parsing doesn’t map fields correctly.
Replace the prompt field with an expression that pulls the output from the VideoPrompt agent so prompts are dynamic.
Example request body (as JSON) — use expressions to inject the VideoPrompt output for prompt:
{ "aspect_ratio": "16:9", "duration": 8, "generate_audio": true, "prompt": "A lively, cinematic scene of five gorillas on a wooden fishing boat in the middle of a sunlit lake, laughing and cheering as they reel in big, thrashing fish, splashes of water in the golden sunlight, lush forested shores in the background, playful camaraderie, vibrant colors, dynamic camera movement capturing their joyous expressions and energetic fishing action.", "resolution": "720p"}
Step 2 — Handle the POST response
A successful POST returns a JSON response containing an id you will use to poll for results. Example response (abridged):
Step 3 — Poll for the prediction result
Because video generation is asynchronous, add a polling loop:
Add a Wait node (e.g., 15 seconds) after the POST.
Add a GET HTTP Request node that requests:
https://api.wavespeed.ai/api/v3/predictions/<returned-id>/result
Replace <returned-id> with the id from the POST response using an n8n expression.
Evaluate the GET response status field.
If status is completed, proceed to your notification/output node and include the video URL (outputs.urls or data.urls). If not, loop back through a Wait node and poll again.
Construct the GET call in n8n either by importing the cURL below or building it manually.Example cURL (GET) to fetch result:
curl --location --request GET "https://api.wavespeed.ai/api/v3/predictions/${requestId}/result" \ --header "Authorization: Bearer ${WAVESPEED_API_KEY}"
HTTP Request GET configuration example (visual)
Sample GET response when the prediction progresses to completion (abridged):
Polling logic: If node behavior
Place an If node after the GET request to check whether status == "completed". If true, route to your notification/output node (Gmail, Slack, Telegram, etc.) and include the final MP4 URL. If false, route back to a Wait node and then to the GET node to continue polling.Illustrative pseudo-configuration:
if: condition: status == "completed" true: proceed to output false: - wait 15 seconds - goto GET HTTP Request node
Step 4 — Notify and retrieve the final video
When the status becomes completed, read the outputs.urls (or data.urls) field from the GET response. Send this URL to users via your chosen notification method. The MP4 file is usually hosted on WaveSpeed’s CDN; users can click or download the file.Example message body in n8n (use expressions)
Use an expression to read the completed GET response’s URL.
Optionally, add the current timestamp using n8n expressions.
Best practices and parameter guide
For cost vs. quality:
Use veo3-fast for faster results and lower cost.
Use full veo3 for higher fidelity (higher cost).
Development tips:
Iterate with short durations (e.g., 3–8 seconds) and 720p resolution.
Disable audio generation while testing if you don’t need it.
Prompt engineering:
Use the VideoPrompt agent to refine framing, camera movement, color, and action.
Keep prompts descriptive but concise; avoid including metadata or system instructions in the prompt text.
