Gemini gets powerful update with generative image editing
Google has announced a major update to its image editing tools in the Gemini app. Developed by the DeepMind team, the new model has already received the highest scores in early tests and is available to all Gemini users. The model allows you to create and modify photos so that the people or animals depicted remain recognisable and natural.
| Place | Model | Score | 95% CI (±) | Votes | Organisation | Licence |
|---|---|---|---|---|---|---|
| 1 | gemini-2.5-flash-image-preview (nano-banana) | 1362 | ±2 | 2 521 035 | Google. | Proprietary |
| 2 | flux-1-kontext-max | 1191 | ±3 | 357 196 | Black Forest Labs | Proprietary |
| 3 | flux-1-kontext-pro | 1174 | ±2 | 2 015 530 | Black Forest Labs | Proprietary |
| 3 | gpt-image-1 | 1170 | ±3 | 1 026 399 | OpenAI | Proprietary |
| 5 | flux-1-kontext-dev | 1152 | ±3 | 1 584 400 | Black Forest Labs | Proprietary |
| 6 | qwen-image-edit | 1145 | ±2 | 1 585 904 | Alibaba | Apache 2.0 |
| 6 | seededit-3.0 | 1142 | ±4 | 1 285 080 | Bytedance | Proprietary |
| 8 | gemini-2.0-flash-preview-image-generation | 1093 | ±3 | 1 700 785 | Proprietary | |
| 9 | bagel | 1044 | ±5 | 12 774 | Bytedance | Apache 2.0 |
| 10 | step1x-edit | 1017 | ±4 | 138 399 | StepFun | Apache 2.0 |
Source: lmarena.ai
What has changed
The new version of Gemini works with a focus on authenticity. Previously, artificial intelligence could "reshape" facial features or change the appearance during editing, but now the system minimises such deviations. This means that friends in photos will look like friends, and pets will look like pets, even after changing their hairstyle, clothes, or environment.
Gemini has learned to
- transfer people or objects to a different environment without losing recognition;
- add or replace things and furniture in a photo while leaving other elements unchanged;
- apply the style of one part of the image to another (for example, the colour of flowers on clothes);
- combine photos to create entire new scenes.
An example of changing the roles of a character. Illustration: Google
Examples of use
- Family photos: if someone blinks their eyes, the algorithm will adjust only the facial expression.
- Pets: a dog with a new collar will remain the same pet, not a "new breed".
- Fashion experiments: you can try on different clothes or shoes right in the photo.
- Renovation and design: it's easy to imagine what a room will look like with new furniture or wall colour.
- Fantasy scenes: friends from the backyard can find themselves in Paris or on Mars - and still be recognisable.
An example of mixing two images. Illustration: Google
Why it is important
Google positions the updated Gemini as a tool for a new level of personalisation and visual storytelling. The model opens up opportunities for creativity, online communication, and professional work with visual content - from creating family albums to advertising campaigns.
An example of a design combination. Illustration: Google
According to the developers, all calculations are performed while maintaining user privacy, and the update is only the first step towards even more flexible editing functions.
Source: Google