August 27, 2012 - Technology media outlet The Decoder published a blog post yesterday (August 26) reporting thatGoogle DeepMind launches new Gemini 2.5 Flash image editing model.In the Gemini application it is possible to improve the accuracy of the modified images while maintaining a consistent appearance of the characters and animals based on textual commands.
1AI cites a blog post that describes how it is more accurate at making image modifications based on text than previous native image generation tools, and even outperforms the GPT-4o used by ChatGPT in a number of tasks, allowing it to better edit images based on complex text.
The highlight of the new model is the "Character Consistency" feature. After generating multiple images, the same person, animal, or object will look the same even if the pose, background, or lighting changes. This is especially valuable for creating photo series and product displays from multiple angles, and is suitable for batch production of branded materials and catalogs.
Gemini 2.5 Flash supports precise localized text editing, allowing users to defocus backgrounds, remove blemishes, add colors or remove objects without manually circling.
The model can fuse up to three images at a time, such as combining product and interior photos into a realistic scene. It also supports "style migration," which applies a texture, color, or pattern to another object while leaving shapes and details intact. The "Reality Reasoning"-based functionality simulates simple cause and effect, such as generating an image of a balloon flying towards a cactus and the subsequent outcome.
Gemini 2.5 Flash is now available in the Gemini application. Users need to switch the model to "Flash" in order to use the image editing function, which produces an image with a visible watermark and an invisible SynthID digital watermark.
Developers can try it out with Vertex AI via the Gemini API, Google AI Studio, at a cost of $30 per million output tokens and about $0.039 per image.