Recently, the tech media The Decoder reported that Google DeepMind has launched a new Gemini2.5Flash image editing model. This upgraded model provides users with a more accurate image editing experience in the Gemini app, allowing users to make significant modifications to photos through text instructions without affecting the appearance of people and animals.
Compared to previous image generation tools, Gemini2.5Flash demonstrates higher accuracy when processing complex text instructions, even surpassing GPT-4o used by ChatGPT in several tasks. This advancement makes it easier for users to realize their creativity when editing images.
A highlight of Gemini2.5Flash is its "character consistency" feature. Even when generating multiple images, the appearance of the characters, animals, or objects specified by the user can remain consistent, regardless of changes in posture, background, or lighting. This feature is particularly valuable for brand series photos and product multi-angle displays, greatly improving the efficiency of material and product catalog production.
In addition, Gemini2.5Flash supports precise local text editing. Users can easily achieve background blur, defect removal, color addition, or object removal without manually selecting areas. It can even merge up to three images at once, such as combining a product photo with an interior photo into a realistic scene. Furthermore, it has a "style transfer" feature that can apply a texture, color, or pattern to another object while maintaining the integrity of shape and details.
The "realistic reasoning" feature of Gemini2.5Flash breaks traditional image editing limitations, simulating simple causal relationships, such as generating a scene where a balloon flies toward a cactus and the subsequent results. These innovative features make Gemini2.5Flash not only a powerful photo editing tool but also a creative platform that allows users to unleash their imagination.
Currently, users can experience this new feature by switching the model to "Flash" within the Gemini app. Notably, the generated images will have visible watermarks and invisible SynthID digital watermarks to ensure copyright protection. Developers can also try it through the Gemini API, Google AI Studio, and Vertex AI, with a cost of $30 per million output tokens, and the cost per image is approximately $0.039.
