This presentation reviews Google's recently released (and allegedly dense) guidance on optimal prompting strategies for its image generation model, "Nano Banana Pro." The core premise rejects old keyword-stuffing methods in favor of full, natural language sentences, iteration within the same chat, and explicit context setting. The model is touted as a "thinking model" requiring conversational input. Demonstrated applications include generating highly consistent characters and products for viral marketing and e-commerce shots, and advanced photo editing (removing obstructions). However, I note the presenter switched to a third-party platform ("Polo AI") due to recurring usage caps on Google’s platform—an irritating operational constraint. Furthermore, newer features like research-grounded infographics and sketch compliance showed initial failures, suggesting variable reliability.
Nano Banana Pro Prompting Guide and Business Applications
0:00 - 1:42 Introduction to the Problem and Solution: Google's "Nano Banana Pro" is highly effective for one-person businesses, but most users fail to get good results by relying on old keyword-based prompting methods. The solution lies in following Google's new prompting guide. I find it surprising that a guide is necessary to make the tool usable.
1:42 - 3:03 Golden Rule: Conversational Prompting: Nano Banana Pro functions as a "thinking model." Therefore, users must abandon keyword lists (e.g., "cool car, neon city, AK") and instead use natural language and full, descriptive sentences, talking to the AI "as if it was a real person."
2:28 - 2:45 Iterative Editing for Efficiency: A key efficiency principle is to modify existing outputs conversationally ("change the lighting to sunset") rather than starting a new prompt or chat session from scratch.
2:45 - 3:57 Specificity and Context: Be extremely specific (e.g., describe textures, not just clothing). Always provide context (e.g., specify that the output is "an advertisement") so the model can generate appropriate style and elements (like advertising text).
4:05 Usage Constraint Warning: The presenter stopped using Google’s official Gemini interface because of constant usage caps, even with a paid plan. They switched to the credit-based "Polo AI" platform to reliably access the Nano Banana Pro model. I don't care for services that impose arbitrary usage limits on paid tiers; it's inefficient.
4:46 Application: Character Consistency (Viral Thumbnails): Nano Banana Pro excels at character consistency. By uploading a single photo, the model can generate consistent character images with varied poses, actions, and expressions (e.g., for YouTube thumbnails). This eliminates the need for future photo shoots, which is a definite efficiency gain.
8:28 Application: Product Consistency (E-commerce): This feature extends to product consistency, allowing users to upload a single product photo (e.g., a complex dress) and generate an infinite number of professional product shots from different angles, on different models, or in various scenes for e-commerce or advertisements.
11:00 Application: Grounding with Google Search (Infographics): Nano Banana Pro attempts to research and generate infographics automatically (e.g., "best times to visit Disney parks"). As far as I can tell, this feature is unreliable; initial results were "garbled up" with font issues, though a subsequent attempt was better.
13:27 Application: Advanced Editing: The model demonstrated an impressive, Photoshop-like capability to remove complex elements, such as removing all people from a crowded tourist photo while naturally restoring the obstructed background elements.
15:02 Failure of Sketch Compliance: The documented feature of using an uploaded sketch to guide the generated image composition failed repeatedly, despite following the official prompt exactly. I reject the premise that this feature works reliably based on the evidence presented.
15:48 Key Takeaway: The quality of output hinges on adhering to the four golden rules: Edit, don't re-roll; use natural language; be specific; and provide context.