Google Launches New AI Tool Whisk, Supporting Image-to-Image Generation

Google has recently unveiled Whisk, an innovative AI tool that enables users to create new images by using other images as prompts, moving away from the conventional lengthy text-based prompts.

With Whisk, users can define the desired image's theme, setting, and style by supplying reference images. For each of these three categories, multiple images can be used as prompts. If users do not have ready-made images, they can click a dice icon, allowing Google to automatically populate some AI-generated images as prompts. Additionally, users have the option to enter text at the end of the process to provide extra details about the desired image, though this step is optional.

Whisk generates images based on these prompts and provides a text description for each generated image. If users are pleased with the outcomes, they can choose to save or download the images. For further refinement, users can adjust the images by adding more text in the textbox or by clicking on an image to edit its text prompt.

In a blog post, Google highlighted that Whisk is designed for "rapid visual exploration" rather than precise, pixel-level editing. Google also noted that Whisk might sometimes "deviate from the intended target," hence providing users the ability to modify the underlying prompts.

During the feature testing phase, image generation takes a few seconds, which may present some inconvenience. Although the generated images can occasionally appear unusual, they offer intriguing material for users' iterative processes.

Furthermore, Google announced the latest iteration of its Imagen 3 image generation model, which serves as the foundation for Whisk. Concurrently, Google introduced Veo 2, the next-generation version of its video generation model. According to Google, Veo 2 has the capability to comprehend "unique cinematic language" and makes fewer errors, such as generating extra fingers, compared to other models (others possibly include OpenAI's Sora). Veo 2 will initially be integrated into Google's VideoFX, accessible to users by joining the Google Labs waitlist. In the future, the model is expected to expand to YouTube Shorts and other products, with a release anticipated sometime next year.