Google’s nano banana pro just killed Photoshop...

Key Concepts

• Nano Banana Pro: The new, advanced version of the Nano Banana image generation model.
• Gemini 3 Pro: The underlying AI model powering Nano Banana Pro, providing enhanced capabilities.
• Chain of Thought (CoT): A reasoning process where the model breaks down a complex task into intermediate steps, allowing for iteration and refinement.
• Google Search Grounding: The ability of the model to use Google search results to inform and ground its image generation.
• 4K Resolution: The capability to generate images at a high resolution of up to 4000 pixels.
• Aspect Ratios: The proportional relationship between the width and height of an image, with Nano Banana Pro supporting multiple options.
• Infographics: Visual representations of information or data, which Nano Banana Pro is surprisingly adept at creating.
• Targeted Edits: The ability to make specific modifications to generated images based on textual prompts.
• Virtual Try-ons: The application of generated images to simulate how clothing or accessories would look on a person.
• Context Window: The amount of information an AI model can retain and process within a single conversation or task.

Nano Banana Pro: A Leap in Image Generation

The highly anticipated Nano Banana Pro has been released, promising significant advancements over its predecessor. Built on the powerful Gemini 3 Pro model, it introduces several "neat new capabilities" that aim to "break the internet."

Enhanced Image Generation Capabilities

• Higher Resolution Output: A major upgrade is the ability to generate images up to 4K resolution. This addresses a key complaint with the original Nano Banana, where images were often too low-resolution for professional use.
• Chain of Thought (CoT) for Iteration: Nano Banana Pro utilizes CoT during image generation. This means the model first "thinks about how to approach the problem," generates an "initial design," and then "looks at the rendering to ensure that the user requests are fully included" before producing the final creation. This iterative process is a departure from traditional "single shot" image generation models.
• Blending Multiple Objects: The model can now blend up to 14 different objects into a single image, offering greater control and complexity in creations.
• Professional-Grade Images with Text Prompts: Users can create professional-grade images using text prompts, with improved rendering capabilities, especially for text in multiple languages.
• Infographic Creation: Surprisingly, Nano Banana Pro demonstrates strong proficiency in creating infographics, a task not typically associated with image generation models.
• Multiple Aspect Ratios: The model supports the creation of images in various aspect ratios, with an "auto" setting that allows the model to select the most appropriate ratio.

Grounding with Google Search

A groundbreaking new feature is the ability to ground answers with Google search. This allows the model to access real-time information to inform its image generation.

• Example: When asked to create a "highly detailed infographic" on the "new capabilities of the newly released Gemini 3 model," Nano Banana Pro first performed a Google search. The CoT revealed it "initialized a search for the recently unveiled Gemini 3 model from Google."
• API Key Requirement: This feature is noted to require an API key and is available only on paid accounts.
• Accuracy of Search Results: The generated infographic accurately reflected Gemini 3's capabilities, including "Advanced reasoning and deep think PhD level logic," "Benchmark dominance" (mentioning the ELO score), and even attempted to create a logo for "anti-gravity" (though it got the logo wrong, the attempt itself is highlighted as impressive).

Demonstrations and Examples

The video showcases numerous examples to illustrate Nano Banana Pro's capabilities:

• Orthographic Blueprint: A prompt to create an "orthographic blueprint that describes this building in plan elevation and section" demonstrated the CoT process and the ability to define aspect ratios.
• Scenery Generation (4K): A comparison between Nano Banana Pro and the original Nano Banana for scenery generation highlighted the significant improvement in resolution and detail. The Pro version produced "extremely high resolution images which are extremely detailed and you can potentially use for professional work."
• Targeted Edits: The ability to make targeted edits was demonstrated by changing the weather in a generated scene to winter, with the lake freezing over. The CoT showed the model thinking about "changing the scene to winter" and iterating on the image.
• Instruction Following with Visual Cues: An example involved providing an image with textual instructions and visual cues, asking the model to "generate image following textual requirements in the image and remove instructions." The output successfully placed a bookshelf, a person playing a guitar, and a plant as requested, with minimal alteration to the original person's appearance.
• Style Recreation with Specific Lighting: The model was asked to "recreate the second image in the style of the first one," specifically focusing on replicating lighting conditions. The CoT showed the model creating intermediate images and the final output accurately matched the requested lighting.
• Image Recoloring: The ability to recolor existing images was demonstrated.
• Disassembly of Objects: A toy was "disassembled into components of antenna, head, body, hands, neck, and wheels." This showcased the model's "special understanding of what exactly it's looking at" and its ability to decompose images into constituent parts.
• Gemini Theme Restaurant Dish:
  • Initial Prompt: "Create a picture of a nano banana dish in a fancy restaurant with Gemini theme." The model initially interpreted "Gemini theme" from an astrological perspective.
  • Follow-up Prompt: "Create a picture of a nano banana dish in a fancy restaurant with Gemini model theme from Google." This prompt led to the inclusion of the Gemini logo on a TV in the background, demonstrating the model's ability to access and incorporate specific branding elements.
• Virtual Try-ons: Similar to the original Nano Banana, virtual try-ons are supported, but with "much better quality." The model can also change the dress color and the scene (e.g., from a casual setting to a professional photography studio or a fancy restaurant) through conversational prompts.
• Conversational Editing: The ability to chat with the model for further edits was highlighted, with the example of changing a dress color and the scene. However, it's noted that "if you keep continuing the conversation for too long you're going to see some degradation," due to the limits of the context window.

Comparative Analysis: Nano Banana vs. Nano Banana Pro

The video provides direct side-by-side comparisons to highlight the improvements:

• Lady in Front of Mirror:
  • Prompt: Create an image of a lady standing in front of a large mirror, with a neutral expression in reality and screaming in terror in the reflection.
  • Nano Banana: Followed instructions "pretty closely."
  • Nano Banana Pro: Produced a "much better" quality image, with the "quality of the image definitely is much better for the pro version."
• Comic Strip Story:
  • Prompt: A 2x2 comic strip telling a story of a robot learning to paint from a human.
  • Nano Banana: Followed the theme but had issues with scene consistency (human painting on robot, two humans instead of robot and human).
  • Nano Banana Pro: Consistently produced the correct narrative, with a robot and human smiling at a creation.
• Fisheye Effect in Water:
  • Prompt: Create an image with a fisheye effect in the water.
  • Nano Banana: The effect looked "unrealistic or unnatural."
  • Nano Banana Pro: Created a "lot more natural" effect, demonstrating a "much better understanding of the real world."
• Spatial Understanding:
  • Prompt: A clear glass cube on a leather book, with a floating thunderstorm inside, a peeling red apple on the left, and a burning candle on the right.
  • Nano Banana: Placed objects correctly but showed a "bite from the apple."
  • Nano Banana Pro: Generated "perfect results" at a "much higher resolution and a much higher quality."
• Hand-Drawn Diagram:
  • Prompt: A hand-drawn diagram of a fictional machine called "dream omatic" with labeled parts (input slot, dream processing unit, output labeled as reality).
  • Nano Banana: Did a "pretty decent job."
  • Nano Banana Pro: Took it to "a whole different level," adding an extra element ("turn to dream") and achieving "extremely good" quality.
• Rendering Multiple Items:
  • Prompt: Render multiple different items.
  • Nano Banana: Did a "decent job" but the gun had a "pretty weird shape."
  • Nano Banana Pro: "Nails it," demonstrating superior quality.

Conclusion and Future Outlook

Nano Banana Pro represents a significant advancement in image generation, offering enhanced control, superior quality, and novel capabilities like Google search grounding and iterative refinement through Chain of Thought. The presenter believes that "image editing is going to look very different in the upcoming year," with a shift towards prompting models rather than using traditional editing tools. This evolution suggests a potential reshaping of the AI image generation landscape with new players emerging.