Skip to main content

Build Your First Pipeline

This guide walks through building a narrated product video pipeline in the Pipeline editor. Each run takes one product description and produces:

  1. A refined, detailed prompt
  2. A vertical product image
  3. A short video animating that image
  4. The same video narrated by an AI voice

You'll build it once, save it, and then reuse it for any product by changing only the prompt.

What you need​

  • A computer or tablet with a reasonably wide screen (the editor isn't available on phones)
  • A connected TikTok account (see Connect TikTok)
  • A short line for the narrator to say, for example Built for the long game. Grip greatness.

The finished pipeline​

Refine prompt ──► Generate image ──► Image to video ──► Add voiceover
                                          ▲
Text (fixed prompt) ──────────────────────┘

The Text node gives the video its own motion prompt, so the camera movement is described separately from the still image.

Step 1: Start a new pipeline​

  1. Open Editor from the sidebar
  2. Click New in the toolbar
  3. Click the name Untitled pipeline and rename it, for example Narrated product video

Step 2: Add the nodes​

In the Nodes panel on the left, click each of these once:

  1. + Refine prompt
  2. + Generate image
  3. + Text (fixed prompt)
  4. + Image to video
  5. + Add voiceover (narrate a video)

The counter under the buttons now shows 5/12 nodes. Drag the nodes by their title bars into a left-to-right line so the connections are easy to follow.

Step 3: Connect the nodes​

Drag from each output port (right edge) to the matching input port (left edge):

FromTo
Refine prompt → textGenerate image → prompt
Generate image → imageImage to video → image
Text (fixed prompt) → textImage to video → prompt
Image to video → videoAdd voiceover → video

Leave the prompt input of Refine prompt unconnected. That's where your run prompt goes in.

tip

If you connect the wrong ports, click the connection to remove it, or press Ctrl+Z (Cmd+Z on Mac) to undo.

Step 4: Configure each node​

Click each node and set its options in the Settings panel on the right.

Refine prompt​

In Creative direction, describe the look you want every image to have, for example:

Rewrite as a clean studio product photo on a seamless background, soft directional light, no people, no text.

Generate image​

  • Model: grok-imagine-image-2.0 (Grok Image 2.0)
  • Aspect ratio: 9:16 for TikTok, Reels and Stories

Text (fixed prompt)​

Describe only the motion, since the image already defines the scene:

Slow push-in on the product, gentle light sweep across the surface, subtle parallax in the background.

Image to video​

  • Model: grok-imagine-video-1.5 (Grok Video 1.5)
  • Duration (seconds): 8
  • Resolution: 720p
  • Aspect ratio: Auto (matches the 9:16 image)

Add voiceover​

  • Script: Built for the long game. Grip greatness.
  • Voice: pick one of the preset voices, for example leo
  • Original audio: mix keeps the video's own sound quietly under the narration. Choose replace for narration only

Keep the script short enough to say comfortably within the video's 8 seconds. A longer script is cut off when the video ends.

Step 5: Validate and save​

  1. Click Validate. You should see Graph is valid, plus an Estimated cost per run in the Settings panel
  2. If errors are listed instead, fix them using the validation errors table. Each error names the node by the id shown at the top of its Settings panel
  3. Click Save. The button changes from Save* to Save once your changes are stored

Step 6: Run it​

  1. Click Run…

  2. In the Run pipeline dialog, describe your product and its setting, for example:

    A tan cabretta leather golf glove on a weathered timber bench beside the first tee at golden hour

  3. Click Run pipeline

You're taken to the Pipelines page, where your run appears at the top with its stages open.

Step 7: Follow the run and collect the result​

The stages complete in order: Refine prompt (a few seconds), Generate image, Image to video (usually a few minutes) and Add voiceover. Each finished stage shows a preview, and the Refine prompt stage shows the rewritten prompt.

When the run shows ready, click View job on the Add voiceover stage. That's your narrated video. From the job page you can download it, upload it to your TikTok Ad Library or post it.

The still image and the un-narrated video are saved as their own jobs too, on the Jobs page.

If a stage fails, read its error message, then click Retry on the run. Stages that already finished are not run again. See Retry a failed run.

Variations to try​

  • Start from your own product photo: replace Refine prompt and Generate image with a Your image node, and connect its image output straight to Image to video
  • Blend your brand photos into the image: add up to five Your image nodes and connect each one to the refs input of Generate image
  • Several videos from one image: add a second Image to video node with a different Text prompt, and connect the same image to both. Each extra video node adds to the cost and counts toward your daily video quota
  • Silent cut for your own music: add a Remove audio from a video node after Image to video