Google's New AI Image Generator: Remixing Images into Unique Creations

Google's Whisk: A New Approach to Image Generation

Google Labs, the division within Google dedicated to experimental projects, is currently evaluating a novel image generator known as Whisk.

This innovative tool distinguishes itself by accepting image prompts rather than traditional text-based inputs.

How Whisk Works

Whisk leverages Google’s advanced image-generation model, Imagen 3, to synthesize images from three distinct sources.

These sources include an image defining the subject, another establishing the scene, and a final image dictating the desired style.

For example, a user could utilize a personal photograph as the subject, a depiction of a futuristic cityscape as the scene, and an anime illustration to define the stylistic elements of the resulting image.

The system automatically crafts a comprehensive caption describing the uploaded images.

This caption then serves as the guiding input for Imagen 3, directing the creation of a remixed image.

Refining Results with Text

Users aren't limited to image prompts alone.

Textual prompts can be incorporated to further refine the desired outcome, allowing for specific details such as “Subject is riding a flying bike.”

Understanding Potential Variations

Google acknowledges that, due to Whisk’s focus on key characteristics, the generated results may not always perfectly align with user expectations.

Variations in attributes like height, weight, hairstyle, or skin tone of the subject may occur.

The underlying prompts used by the system are fully accessible for review and modification by the user.