Google rolls out Gemini app feature for editing images with prompts

May 1,TG盗号软件黑产破解技术 2025  17:41

Google's Gemini app is rolling out a powerful new image editing feature that allows users to modify both AI-generated images and personal photos uploaded from devices using simple text prompts, as reported by TechCrunch and Google's official blog.

Native Image Editing Features

Gemini 2.0 Flash's native image editing capabilities represent a significant advancement in AI technology. Unlike previous multimodal systems that relied on separate models working together (such as pairing a language model with Imagen 3), Gemini processes both text and images within a unified system. This native approach eliminates inter-model communication and reduces latency, resulting in more seamless editing experiences.

The editing features are impressively versatile, allowing users to:

• Change backgrounds, replace objects, and add new elements to images
• Maintain context throughout multi-step editing sessions
• Add text directly onto images with improved rendering quality
• Colorize photos with natural results
• Generate consistent series of images for visual storytelling
• Create visual representations of concepts like "what you'd look like with different hair colors"

All images created or edited through Gemini include SynthID's invisible digital watermark, with Google also experimenting with adding visible "AI" watermarks in the corner of generated images. The feature is gradually rolling out to users in over 45 languages across most countries.

SynthID Watermarking Technology

SynthID is Google DeepMind's innovative watermarking technology that embeds imperceptible digital signatures directly into AI-generated content. This invisible watermark is embedded at the pixel level for images and video frames, in spectrograms for audio, and through subtle probability adjustments for text generation. Unlike traditional visible watermarks, SynthID doesn't compromise visual quality while remaining detectable by specialized software even after common modifications like cropping, filtering, or compression.

The technology uses a dual deep learning approach – one model embeds the watermark while another detects it. Already integrated across Google's AI ecosystem, including Imagen and now Gemini's image editing features, SynthID serves as a transparency tool to help users identify AI-generated or heavily edited content. While not foolproof against extreme manipulations, this technology represents a significant step toward addressing misinformation concerns as AI-generated content becomes increasingly realistic and prevalent. Google has also open-sourced SynthID through its Responsible Generative AI Toolkit, extending its application beyond Google's own products.