This guide will walk you through all of the steps to make your own latent space walk video. Commonly referred to as "morphing" videos, these exports walk us through the latent vector space of image generative models. These videos can only be produced using synthetic output images, and not with original content. If you'd like to learn how to produce one with your own aesthetic or style, this guide can help you get there.
First, choose your model - it must be image generating (rather than a stylization model for example). This can be a model you trained, or one from our directory. After adding it to your workspace, click "Export." A screen like this should appear:
From here, you can select your images you'd like to use in your final export. These can be random selections, or images you've previously starred using the vector space (save up to 50 images). We also provide some features for further customization, such as easing options, and transition times. The truncation slider on the right can be used to generate new random images: greater truncation = greater variety, but may sacrifice quality. Have fun experimenting and playing around!
The export resolution of your latent walk video will depend on the model. For trained models, it will be dependent on the pre-trained option that was selected; more on that here. The maximum resolution supported for exports at the moment is 1024x1024, and the aspect ratio is always 1:1. Videos can be upscaled using our Super Resolution model.
Please note, at this time we are unable to support saving/reopening latent walk projects.
Above: 2 images, 5s, ease in/out