Earlier this year, Google introduced native image editing capabilities to its Gemini AI app. Now, the company has upgraded the feature with a new model from Google DeepMind, which serves to provide the user with more creative control.
The main focus of this update is consistency. Typically, AI models struggle with maintaining the likeness of a character from one image to another. According to Google, this update makes it so that Gemini can keep the subjects in the image looking consistent throughout the edits. The company suggests prompting the app to change a person or pet’s location or outfit to see this improvement in action.
Naturally, enhancements to this feature allows for better multi-turn editing as the model can now preserve parts of the image the user wishes to keep unchanged. This means that the app can continue to make tweaks to specific areas without affecting the rest of the picture.
In addition to this improvement, the user can now upload multiple photos and combine them into a single new image. As an example, the user can use their picture and another of their dog to create a portrait of the two at a basketball court. Aside from that, the blending capabilities also allows for applying the attributes of one image to another. For instance, the user can tell Gemini to design a clothing item with the colour or pattern of something else.
As usual, the images edited with Gemini will feature a visible watermark, as well as an invisible SynthID digital watermark. These serve to mark the pictures as AI-generated.
(Source: Google blog)
