According to news on August 27, on Tuesday, Google released a new generation of AI image generation and editing model Gemini 2.5 Flash Image (codenamed "Nano Banana").The model is now fully integrated into the Gemini app’s image editing capabilities.
Gemini 2.5 Flash Image is now available for preview through Gemini API, Google AI Studio and Vertex AI platform, and the stable version is expected to be officially released within a few weeks. This model adopts Token billing model.Pricing is $30 per 1 million output tokens, a single image is fixedly counted as 1290 output tokens (approximately US$0.039 per image), and the billing standards for other input and output modes are consistent with Gemini 2.5 Flash.
Google said in a technical blog post that this upgrade brings four core functions:It supports intelligent fusion of multiple images, maintains the consistency of characters across different screens to enhance narrative performance, enables precise directional editing through natural language, and uses Gemini's world knowledge base for image generation and processing.
Earlier this year, when Google first launched the native image generation function in Gemini 2.0 Flash, its low latency, high cost performance and ease of use were well received by users, but there was also feedback about improving image output quality and enhancing creative control.
In order to lower the development threshold, Google has simultaneously upgraded the “build mode” of Google AI Studio (more features will be launched in the future). Developers can now not only quickly test the AI application capabilities of the model, but also realize creative reorganization and concept implementation through a single command. After completing the application building, users can deploy directly from Google AI Studio or save the code to the GitHub repository with one click.
Developers can try to enter commands such as "Build an image editing application that supports users to upload images and apply different filters", or select preset templates for secondary creation. The above functions are currently free and open to use.
Maintain role consistency
A core challenge that has long faced the field of image generation is how to maintain the consistency of the appearance of a character or object across multiple command input and editing processes. Now, users can flexibly place the same character in different environmental contexts, present a single product in a new scene from multiple angles, or generate brand assets with a highly unified visual style, all while ensuring that the main characteristics remain unchanged.
Google provides a customizable template application (supporting visual programming) in Google AI Studio to visually demonstrate the model's powerful capabilities in role consistency.
In addition to character consistency, the model also has excellent visual template adaptation capabilities. Currently, developers have applied it to a variety of practical scenarios such as real estate listing cards, uniform employee badges, and dynamic renderings of product catalogs. All of this can be achieved with just a basic design template.
Command-based image editing
Gemini 2.5 Flash Image allows users to achieve directional transformation and precise local editing through natural language instructions. Specific functions include blurring the image background, removing T-shirt stains, completely removing a person from the photo, adjusting the subject's posture, colorizing black and white photos, and any other editing needs that users can achieve through natural language description.
In order to intuitively present these functions, Google developed a dual-mode photo editing template application in Google AI Studio, which provides both a visual interface and pure command control interaction methods.
Integrate native world knowledge
Traditional image generation models tend to focus on aesthetic representation, but have obvious shortcomings in deep semantic understanding of the real world.Gemini 2.5 Flash Image relies on Gemini's huge world knowledge base to break through this limitation and open up a new application dimension for image generation.
For example, Google built a template application in Google AI Studio that transformed a simple canvas into an interactive educational coaching tool, demonstrating the model's comprehensive ability to parse hand-drawn diagrams, answer real-world questions, and execute complex editing instructions in one step.
Intelligent fusion of multiple images
Gemini 2.5 Flash Image can understand and intelligently fuse multiple input images simultaneously. Users can naturally embed objects into new scenes, redefine room styles through color schemes or material textures, and complete multi-image fusion with just one command.
To demonstrate this feature, Google has launched a dedicated template application in Google AI Studio. Users can place products into new scenes simply by dragging and dropping, and quickly generate highly realistic fusion images.
The new leader in the AI imaging competition
The release of Gemini 2.5 Flash Image marks that Google is narrowing the gap with its competitors and intends to directly compete with companies such as OpenAI and Adobe.
The model was previewed in a low-key but successful way before its official debut.After its anonymous appearance on the crowdsourcing blind testing platform LMArena under the pseudonym "nano-banana", it quickly sparked heated discussions in the AI community and became the world's highest-rated image editing model.David Sharon, head of Gemini applications, spoke highly of its market response: "Early user feedback has been very enthusiastic, and it has become the highest-rated image editing model in the world."
This release is undoubtedly an important response from Google to the increasingly competitive AI market. In March this year, OpenAI integrated its GPT-4o image generator directly into ChatGPT, which not only brought about a surge in user participation, but also set a new benchmark for the integration of AI tools through a series of viral content demonstrating its power and ease of use.
At the same time, creative software giant Adobe is also actively strengthening its flagship product Photoshop. The three recently launched features powered by Firefly, namely "Harmonize" that automatically matches color and lighting, "Generative Upscale" that improves resolution, and the improved "Remove Tool", further consolidate its position in the professional field.
Competitive pressure has even forced industry giants to adjust their strategies. After Meta encountered obstacles in internal development, it instead chose to license Midjourney's technology. Wang Tao, head of Meta AI, regards this as a necessary move and emphasizes that the company needs to adopt a "comprehensive strategy" to provide optimal products.
This market dynamic reflects the polarized development of the AI industry: on the one hand, intensified consolidation among large companies, and on the other hand, rapid professional segmentation. More and more emerging companies are focusing on solving specific long-term problems. For example, the FLUX.1 Krea model launched by Black Forest Labs is designed to break through the common "AI style" and achieve more realistic image effects; Alibaba's open source Qwen-Image model has outstanding performance in solving the key technical problem of readable text rendering in the generation system.
Balancing creativity and safety
Google's announcement, and subsequent introduction of new security measures, can be seen as an important response to the setbacks in its early AI image generation capabilities. Previously, Gemini caused widespread controversy due to accuracy issues with the historical images it generated, which once prompted the company to suspend the feature and conduct comprehensive optimization. Today's resurfaced version comes with tighter security protections.
In response to the increasingly rampant threats of deepfakes and disinformation, Google announced that it will implement watermarking on all AI-generated content. Each image not only contains a visible logo, but also embeds an encrypted SynthID invisible digital watermark, ensuring that AI-generated content can be clearly identified.
This approach is in sharp contrast to the legal difficulties faced by competitors. Midjourney is currently facing copyright lawsuits from Disney and Universal Pictures over alleged infringement of training data. Disney's general counsel Horacio Gutierrez has taken a strong stance, saying: "Piracy is piracy. The fact that an AI company engages in this kind of conduct does not lessen its infringing nature."
At the same time, the entire industry is also facing questions about the reliability of AI benchmarks. As AI strategist Nate Jones points out: “When we pursue leaderboard rankings too much, we risk developing models that perform well on specific tests but fail to solve real-world problems.”
all in all,By integrating powerful and easy-to-use editing tools into its flagship AI product, Google is trying to attract mainstream users with convenience and creative control. This strategy not only positions Gemini as a chatbot, but also builds it into a comprehensive creative engine to occupy a place in the rapidly evolving field of content generation AI.
Related articles:
Google releases its most advanced image generation model Gemini 2.5 Flash Image