Using Seed Values

A Seed Value is useful for recreating Generations, testing output consistency, and creating tighter variations

Every time you create a generation on Scenario, the output is assigned a random seed value. This value can be used in a variety of ways.

After clicking on an image generated, you can see the seed value in the image information panel:

Screenshot 2023-12-07 at 11.07.50 AM

In this example, we are using the Seed Value: 538251828046456490 and the Model: Psychedelic Bubblegum Pop with the following prompt:

3D render of a sci-fi baroque concept design of anatomically correct brain device, steampunk, intricate details, scientific, hyper detailed, photorealistic

 

Set the number of output Images to 1 and paste your seed value into the Seed field:

Screenshot 2023-12-07 at 11.22.39 AM

Make sure to use the exact prompt and you will get the same output as the original generation:

Screenshot 2023-12-07 at 11.26.06 AM

Maintaining the Seed value while making small changes to the prompt can closely guide explorations and iterations on your ideas.

Creating Variations using a Reference Image and Seed value

You can create tighter iterations of your output by keeping the seed value and using the output as a reference image:

Screenshot 2023-12-08 at 12.05.44 PM

Changing the Seed value by small increments while using the Reference Image will also create variations that are guided closely by the original result:

Screenshot 2023-12-08 at 12.21.23 PM

Using the original output as a Reference Image with it's Seed value, you can restyle the output by simply selecting a different model:

Screenshot 2023-12-08 at 1.10.27 PM

Using ControlNet with a seed value will not produce the exact same result. The seed value is related to the prompt and model used. ControlNet by its nature layers an additional variable on top of the generation that will alter the output. The seed value used with ControlNet still applies some consistency however and can be usefully experimented with in similar fashion to the above explorations on variation.