Today Google Formally announcing the launch Gemini 2.5 Flash Image(Internal development codenamenano banana“, which is the one we introduced earlierThe world’s most powerful image model, but at the time Google had not announced that it was its product)Gemini 2.5 Flash Image is currently the most advanced image generation and editing model, not only continuing Gemini 2.0 Flash With its low latency and high cost-performance advantages, it has also made significant improvements in image quality, character consistency, semantic understanding, and multi-image fusion.
Google officially launches its most powerful image model, Gemini 2.5 Flash Image.
Earlier this year, Google Gemini 2.0 Flash The native image generation feature was introduced for the first time. At that time, developers highly praised its low latency, ease of use, and cost-effectiveness, but they also raised explicit requirements:Higher image quality and more powerful creative control.The Gemini team responded to this feedback through Gemini 2.5 Flash Image, not only resolving the aforementioned issues but also marking a significant step forward in the precision of image editing and generation. The model has already been through Gemini API The text is empty. Please provide the Traditional Chinese content you’d like translated. Google AI Studio Provided for developer use, and can be used at Vertex AI The platform serves enterprise customers. In terms of pricing, the price per 1 million output tokens is 30 US dollars, each single image is approximately 1290 tokenapproximately 0.039 USD）。
Gemini 2.5 Flash Image (nano-banana), on the Model Capability Arena lmarena The blind test results previously given to all users show that its performance significantly outperforms all image generation models, including GPT Image 1, FLUX.1 Kontext, Qwen Image Edit, etc., and ranks first:
Feature Highlights and Application Scenarios
Character Consistency: Building a Coherent Story and Brand Image
In the field of image generation,Maintain the consistency of characters or objects across different scenes and angles.It has always been a major challenge. Gemini 2.5 Flash Image provides an answer to this pain point. Whether placing the same character in different environments, presenting the same product from multiple angles, or generating an entire set of brand assets with consistent style, Gemini 2.5 maintains accuracy and coherence, avoiding the problem of “character distortion.”
Google even AI Studio Among them, a customizable template application was launched, demonstrating this feature in Unified visual design for real estate display cards, employee ID badges, and product catalogspotential in scenarios such as these.
Semantic-Driven Image Editing: Precise Modifications Through Natural Language
Traditional image editing requires professional software and complex operations, but Gemini 2.5 Flash Image makes all of this far more intuitive. Users can simply input natural language to make targeted modifications:
-
Blurred background
-
Remove stains from clothing
-
Remove an entire person from the photo.
-
Change the subject’s pose.
-
Convert black and white photos to color.
Google has provided one in AI Studio.Image Editing Template App, combining UI Operations and Text Prompts, allowing developers and designers to easily experience the flexibility of this technology.
Integrating World Knowledge: Semantic Understanding Beyond Aesthetics
Most image generation models excel in aesthetic quality but often lack a deep understanding of the real world. A major breakthrough of Gemini 2.5 Flash Image lies in its combination of Gemini’s World Knowledgeenabling it to handle more complex semantic tasks. For example, Google has built an interactive educational template application that allows the model to read and understand hand-drawn diagrams, and even answer real-world questions and execute complex edits in the same step. This represents generative AI gradually moving toward Knowledge-oriented visual understanding, rather than relying solely on random generation.
Multi-image fusion: creating innovative visual combinations
Another powerful feature is Multi-image FusionGemini 2.5 can understand and combine multiple input images, naturally integrating different elements into a single scene (such as dressing a character in specified clothing).
This ability lies in E-commerce, Interior Design, and the Advertising IndustryParticularly promising.
-
Seamlessly integrating a single product into different scenarios.
-
Restyle the room with new colors and materials.
-
Quickly generate product display or catalog images.
Google has also launched a template app for this feature, allowing users to drag and drop products into scenes and instantly generate realistic images.Experience: Click here）
To help more developers get started right away, Google… Google AI Studio has significantly updated its “Build Mode,” offering template applications, free remixing, and one-click deployment features. Developers can even push generated applications directly to GitHub, or share directly via AI Studio.
More importantly, Google has begun collaborating with industry partners:
-
OpenRouter.ai: has introduced Gemini 2.5 Flash Image to its platform, providing Over 3 million developersUse it; this is also the platform’s first model to support image generation.
-
fal.ai: As a development platform for generative media, fal.ai is also bringing Gemini 2.5 to its community, expanding the technology’s reach.
As AI-generated imagery becomes increasingly indistinguishable from real photographs, identifying whether images are authentic or AI-generated content has become an important challenge. Google has also stated that all images created or edited through Gemini 2.5 Flash Image will carry Invisible SynthID Watermark, ensuring these works can be identified as AI-generated or edited.
Now all users can directly use Gemini 2.5 Flash Image in Gemini for various image creation tasks, or go to Google AI Studio to use the model directly for creation as well. If you’re interested, give it a try.
Source: KOCPC Chinese