You can use Runchat to design content templates and then create infinite variations on these templates in parallel. This can be really useful for concept exploration.

You can define a schema for any object using the Output Format parameter on the Prompt Node. This allows you to specify key-value pairs that will always be present in the generated content. Formats are useful for defining templates for generated objects. Let’s say for instance we are creating a series of design concepts for a client. Each design concept needs to consider each of the following:

To ensure that the Prompt Node always responds with content for each theme, we could define an Output Format as an object with properties of materiality, colour, style and precedent. If we wanted to generate multiple concepts that match our template, we have two options:

  1. Define our schema as an array of objects with the same properties - the Prompt Node will then output a list
  2. Provide a list input to one of the parameters of the Prompt Node. The node will then generate our template once for each input item.

Parallel Prompting

Let’s say you want to:

  1. Discover some things to do in a city
  2. For each of these things, estimate the price, distance from where you’re staying and duration

If you only wanted to do this for a few things, a single prompt might work fine. But if you wanted to run this for more than a handful of options then a single prompt would be a) slow and b) suffer from hallucinations.

Instead, you could create one prompt and use the array format to generate a list of text responses.

Then, connect this list as context to the second prompt, and use the object format to describe your properties of price, distance and duration. This second prompt would be executed as a parallel request (all of the requests made at the same time), which means that the response time will only be as long as the longest single request, rather than the sum of all requests. You’ll also get the response as a nice table!

image.png

Generating Templates