[23:54 Tue,26.August 2025 by Thomas Richter] |
For several weeks, an AI image model known only by the code name "Nano Banana" has been causing a sensation with its excellent results. Now it has finally been officially presented: none other than Google has developed the new model, which now operates under the name Gemini 2.5 Flash Image. The model had - quasi anonymously - conquered the top position of AI image editors within a short time on the benchmark page ;)
Character Consistency Across Multiple ImagesThe specialty of Nano Banana, in addition to generating new images, is the editing and recombination of images via prompt. Google&s new model excels particularly in the - previously difficult - consistency of characters, i.e., it can very realistically depict a person or any object given by photo in other scenes defined by prompt - even together with other people or objects. ;) Image Editing via Text InputWhen generating images or image components, the image model draws on the world knowledge of Google&s multimodal Gemini AI to understand and realistically implement prompts. In this way, the model can make precise edits and maintain image consistency. The input consists of images and colloquial prompts that describe exactly what should be edited in an image, thus enabling object-oriented image editing. ;) Users can thus make targeted changes to specific image elements - such as blurring the background, removing objects, changing colors or adjusting details such as the pose of a person. These semantically controlled interventions allow a much more intuitive and flexible editing than conventional, UI-based tools - without any masks. An image can thus be edited step by step without the central motif becoming unrecognizable. It is also possible to use the classic function of transferring a style or texture from one image to another or combining different - even abstract - image elements. In the following example, the model is able to consider the plant species that would thrive on the Dutch coast, and then consider the aesthetics of the plants that would look good. ;) Multi-Image Fusion: Combine Images via PromptWith the new multi-image fusion feature, multiple images can be merged into a coherent scene. For example, objects can be inserted into realistic scenes or interiors can be redesigned with different colors and textures - all controlled via a single text prompt. Integration and AvailabilityGemini 2.5 Flash Image is now available as a preview via the Gemini API, Google AI Studio, and Vertex AI - free for a limited number of trials, after which each generated image costs ./create.pl.039. All images created with the model also contain an invisible SynthID watermark that identifies them as AI-generated.  more infos at bei deepmind.googledeutsche Version dieser Seite: Google Gemini 2.5 Flash Image "Nano Banana" bringt Charakterkonsistenz und Multi-Imag |
 |
|
|