Skip to content

Using a tool

Variables and batch runs

Type a curly brace in a prompt, describe a list inside it, and one press fires one run per item.

Type an opening curly brace anywhere in a prompt and describe a list inside it.

Prompt with a variable
a ceramic vase in the style of {give me 20 design movements},
studio light, plain grey backdrop

Press Enter to commit it. A drawer opens on the left of the card, filling with the twenty items, and Run becomes Run ×20.

The region you are typing shows amber while it is open, so you can see the tool is waiting for you to close the thought. On Enter, the instruction goes out to a language model that expands it into a list, and the drawer opens immediately in a loading state rather than making you wait on a blank card.

Expanding the list is free. No credits are reserved for it, because it is a short text call rather than a generation.

The drawer is an editor, not a preview. Before you spend anything you can:

  • Edit any item’s wording
  • Delete the ones that were never going to work
  • Add your own
  • Re-roll the whole list if the model misread the instruction

A machine-written list of twenty usually holds a few directions you would never have thought of alongside several that are the same idea twice. Pruning it before you press Run is what turns it into a batch worth paying for.

Binding a variable puts the card into batch mode, which is lock mode driven for you. One press fires N runs, each one substituting its own item into the prompt, and every result lands in its own slot in one collecting frame.

The price on the button is the sum, not the unit. Run ×20 shows what twenty runs cost, before you press it.

The substitution happens per run, so the model receives a ceramic vase in the style of Bauhaus rather than the instruction you typed.

That matters beyond the image. Each result’s record carries its own resolved prompt, so six weeks later the asset can tell you which of the twenty it was, and searching your library for “Bauhaus” finds it.