Projecting Images To A Generative Model Based On Gradient-free Latent Vector Determination

In a content creation digital medium environment, a computing device comprising:

• a processor;
• and
• computer-readable storage media having stored thereon multiple instructions of an application that, responsive to execution by the processor, cause the processor to perform operations including:
• receiving a target image;
• identifying a first latent vector and a first class vector that result in a first image generated by a generative model by:
• determining a second latent vector by applying a previously determined loss between the target image and a previously generated image to a gradient-free technique;
• determining a second class vector by applying the previously determined loss to a gradient-based technique;
• generating, from the second latent vector and the second class vector, a second image;
• determining a loss between the target image and the second image;
• repeating the determining the second latent vector, the determining the second class vector, the generating, and the determining the loss until a loss condition is satisfied;
• identifying the second latent vector that resulted in the loss condition being satisfied as the first latent vector;
• identifying the second class vector that resulted in the loss condition being satisfied as the first class vector;
• providing the first latent vector and the first class vector to the generative model, the generative model including multiple weights;
• and
• adjusting the multiple weights to reduce the loss between the target image and a first image.

A system comprising:

• a generative model having multiple weights and configured to generate an image;
• means for identifying a first latent vector by applying a gradient-free technique and identifying a first class vector by applying a gradient-based technique that result in a first image generated by the generative model;
• and
• means for adjusting the multiple weights to reduce a loss between a target image and the image generated by the generative model from the first latent vector and the first class vector.