Node types
The seven kinds of node on the Workflow canvas, and when to use each.
| Node | What it does |
|---|---|
| Media | An image, video or audio file from the Library, or a named empty slot filled later |
| Text | A prompt or piece of text passed to other nodes |
| Generate | One model from the catalogue, with inputs from other nodes |
| Scene | A final result, lined up in the Scene results strip |
| User Prompt | The user's request, which can hold variables and images |
| System Prompt | AI that answers in a JSON structure you define |
| Condition | Picks the next path from a value |
Media
A Media node brings a file into the flow. It can hold a fixed file from the Library, or an empty slot with a name, for example "Product photo" or "Logo". Empty slots are filled each time the workflow runs, and that slot is what changes per photo when running a batch.
A slot can also be given a role (product photo, logo, model photo, style reference), so Fill from Brand Kit knows which image goes in which slot.
Generate
A Generate node runs one model, with the same settings as in the console. Its input ports follow that model's fields: prompt, reference images, first frame, and so on. Its price is worked out from the model and the settings you choose in the node.
Scene
A Scene node marks a final result. All scenes are lined up in the Scene results strip under the canvas, so a workflow's output is easy to see and download without hunting among the intermediate steps.
User Prompt and System Prompt
This pair is for a step that needs to "think" before an image or video model runs.
- User Prompt holds the request or brief, and can include variables and images.
- System Prompt holds instructions for the AI and the answer structure you define in JSON. Every field in that structure becomes its own output port, so a field such as
scene_1_promptcan connect straight to a Generate node.
The answer structure can be nested (fields inside fields) and can include lists with an exact number of items, for example exactly three scenes. System Prompt can read up to 10 images. It costs 0.2 credits per run, refunded if the answer fails or does not match the structure.
Condition
A Condition node picks the next path from a value, with layered rules. Paths that are not picked do not run, so they are not charged. An example is the Ad with a claim check template: AI judges the ad's claim first, and the image is only made if the claim is judged safe.