Text → Image → Video
Follow the real editor controls from a saved project to one short test clip.
Create and name the project
Sign in and create a project from the Workspace dashboard. The opened account project has its own workflow URL and automatic remote save; a local draft in a development environment is not the same as an account project.
- Click New Canvas and wait for the editor.
- Give the canvas a meaningful name, such as “Night market first shot”.
Write the scene in Text
Text is the actual node; Prompt is a mode within Text rather than a separate add-menu node. Keep the scene small enough to diagnose one result at a time.
- Right-click blank canvas, choose Text, move the placement ghost and left-click.
- Write “Rainy night market, one vendor under warm lanterns, medium shot, 16:9”; keep a copy for revisions.
Give the scene a visual anchor
Add Image Generation and connect the Text output to its Text input. Image model options include GPT Image 2 and Nano Banana routes, subject to the deployment configuration.
- Right-click empty space, choose Image Generation and place the card to the right of Text.
- Connect Text → Image Text, select a model and check its ratio before pressing the Image node Run control.
Animate what changes
Add Video Generation, connect the completed Image result to its Image input, choose a preset that accepts images and explicitly select that image in the @ reference picker. The video prompt should describe action rather than restating the still.
- Place Video to the right and connect the finished Image output into its compatible image handle.
- Write “Vendor turns toward camera; lanterns sway once; slow dolly forward” and select the image thumbnail in @ Reference Media.
Run one shot, not the entire graph
Before a paid video run, verify input count, output ratio, duration and provider. Asynchronous video tasks can remain queued; repeated clicks may create duplicate tasks rather than make one finish sooner.
- Press Run on the Video node once and wait for a finished playable result.
- If the node fails, read its error and check the image URL, @ selection, current model ports and server availability before retrying.
Turn a test into a repeatable process
Keep the source Text and Image nodes rather than flattening everything into the final video. To extend the story, extract the last frame and feed the new Reference image into another Video node.
- Compare the first video against the original brief; revise one variable at a time.
- Save the useful image or clip in the asset library and continue with a second shot or Video Edit.