Nano Banana 2 · The Complete Guide to Creating Images, Ads, and Automated Systems.

Think about how you used to produce a product image until today · taking a photo, opening Photoshop, removing the background, adding a shadow, fixing the lighting, leaving Photoshop, uploading to the website. An hour of work for every single image. If you wanted a video ad · the budget started at 3,000 shekels. Nano Banana 2 changes all of that. You feed it a raw product image · it removes the background, creates a studio background, fixes the lighting, and returns an image that looks like it was shot on a professional shoot day. All in a single prompt. And if you take it a step further · you connect it to Seedance or Veo, and it returns a complete video ad · with editing, sound, and all the scenes. This guide is the essence of what we learned in session 18 of the club. We will cover the 3 levels of working with the tool · beginners (image creation and background changes), professionals (Gems that generate packshots and automated ads), and masters (a system in Claude Code that does the entire process on its own). At every level · ready-to-copy prompts.

What you will have by the end of the guide

• **Ability to produce a professional product image** · a packshot with a studio background, from a raw photo you took on your phone
• **A method to make variations at scale** · taking a single product and generating 10, 20, 50 different versions for a campaign
• **Your own personal Gem in Gemini** · an agent that receives an image and returns a professional image · without needing to write a prompt every time
• **A 9-frame video storyboard** · one image with 9 scenes + directing instructions, ready to feed into a video tool
• **Connecting Nano Banana to Seedance / Veo 3.1** · turning a product image into a 9:16 video ad for Instagram
• **A complete system in Claude Code** · input an image · receive a video ad with editing and music. Without touching other tools

What is Nano Banana 2 anyway

💡 The model is available for free with a limit of 20 images per day (Gemini Free). On the paid subscription · up to 100 per day. If you use the tool professionally or at scale · it is better to work in Google AI Studio with API · it is more flexible and allows integration with Claude Code.

Level 1 · Beginners · Image creation and background changes

Prompt 1 · Creating a product image with a studio background (for copying)

I want a high-quality professional product photo on a clean white studio background. The product should look like it was shot in a Nike-style commercial - sharp lighting from above, soft shadow underneath, no logo or text anywhere in the frame, no overlay graphics, no models. Vertical 1:1 format. The product should be centered, well-lit, with realistic reflections and material textures. [Drag your raw product image here · can be a photo from your phone]

Prompt 2 · Combining 3 images into a campaign scene (for copying)

I'm giving you 3 separate images: 1. A product (shoe insole) 2. A pair of running shoes 3. A tennis court background Combine them into a single vertical 9:16 advertisement composition. The insole should be displayed elegantly next to the shoes, placed on the tennis court ground. Cinematic lighting, no text overlay, no logos, no captions. The composition should feel like a premium sports brand campaign shot.

Field tips for working at Level 1

• **The more product reference images you have · the more accurate the result.** If you are upgrading jewelry or an item with very precise details · input 3-4 different angles of the same item
• **Tell it what you do not want as well.** 'Make sure there is no logo', 'Do not add text', 'Do not change facial features' · the tool sometimes oversteps and adds elements without you asking
• **Jumping to Studio · it is better to break it down into steps.** Instead of one prompt with 5 actions · request one action at a time. Change background · then change lighting · then add text. This way it does not get confused
• **If the model refuses or starts acting up after 6-7 iterations** · open a new chat, upload the latest image, continue from there. This resets the model's memory on this topic
• **Source image quality matters.** Shoot next to a window with natural light, use a tripod, not in the dark. The tool cannot compensate for a poorly shot photo

⚠️ Watermark · Nano Banana in Gemini adds a small watermark in the corner. If you are using the image for commercial purposes · it is better to work from Google AI Studio which does not add a watermark, or use an editing plugin to remove it. And before using the image for paid ads · make sure there are no copyright issues with elements the tool incorporated.

Level 2 · Professionals · Building a Gem that works automatically

How to build a Gem in Gemini · A 5-minute process

• Go to gemini.google.com → Gems (left menu) → New Gem
• Give it a name · for example 'Packshot Studio'
• In Instructions · paste a system prompt that describes what the Gem needs to do (you can ask ChatGPT to write this prompt for you · 'Write me a system prompt for a Gem that takes a product image and returns a packshot on a white background')
• Default tool · select Create Image
• Knowledge · upload a reference image that shows it what the result should look like (optional · but very helpful)
• Save. Now · every time you upload an image to this Gem · it automatically performs the action. Without a prompt

System prompt for Gem · Automated packshot (for copying and use in Instructions)

You are a professional product photographer AI. Your only job: take any product image the user uploads, and return a clean studio packshot on a pure white background. For every image uploaded: 1. Remove the original background completely. 2. Place the product centered on a seamless white studio background. 3. Apply soft, even lighting from above with a subtle shadow underneath the product to ground it. 4. Preserve all material textures, reflections, and product details exactly. 5. Do NOT add any text, logos, captions, watermarks, or overlay graphics. 6. Do NOT modify the product itself (no color changes, no material swaps). 7. Output should be square 1:1 format, suitable for e-commerce or social media. If the user uploads multiple images, treat each as a separate packshot request. If the user asks you to do something other than a packshot, politely redirect: "I only handle product packshots. For other tasks, please use a different Gem."

Advanced Gem · 9-frame storyboard for an ad

The AI Master Club opens the door for you: a live meetup every week, recordings, and a community. 47 ILS per month, cancel anytime.

System prompt for Gem · 9-frame ad storyboard (for copying)

You are a creative director AI specializing in short-form vertical video ads for social media (Instagram Reels, TikTok). When the user uploads a product image, you do this: 1. Create ONE vertical 9:16 image divided into a 3x3 grid of 9 cinematic storyboard frames. 2. Each frame should show a different scene of a 10-second product commercial: hook (frame 1), product reveal (frame 2-3), benefit demonstration (frame 4-6), social proof or context (frame 7-8), final CTA shot (frame 9). 3. Number each frame clearly (1-9) in its top-left corner. 4. Below or beside each frame, include a short director's note in clear text: what's happening, camera movement, lighting, mood. Keep notes concise - one line per frame. 5. Use cinematic lighting and a consistent visual style across all frames. 6. Do NOT add any branding, captions for end-users, or logos - these are internal storyboards only. The output image will be fed into a video generation model (Seedance / Veo) which will read the frames and director's notes to produce a final commercial. Make sure the notes are legible and actionable for an AI video model.

🎬 Once you have the storyboard from Gemini · feed it into Seedance (via fal.ai or Heyxfield platform) or Veo 3.1 (on paid Gemini subscription). The tool reads the storyboard, understands the instructions, and returns a ready video ad with editing, scene transitions, and sound. If you want custom sound · you can add it in CapCut afterwards.

Top 8 business use cases for Level 2

• **Store product packshots** · even if your store is on Shopify / WooCommerce / Cardcom · updating 20 products = 20 drags into a Gem. An hour of work turns into 10 minutes
• **Presentation banners** · any presentation you give · 5-10 professional banners with your typography. Instead of searching in Canva · create them in the tool
• **Instagram post gallery** · a single product image · 9 variations with different backgrounds · 9 ready posts for a whole month
• **Landing page images** · the landing page hero is what counts. Instead of a stock photo · a packshot of the real product on a careful background
• **Interior design / fashion mockups** · photograph an empty room · ask the tool to add a sofa in a specific style, furniture, a painting. The client sees before spending a dollar
• **A/B campaign variations** · a campaign with 5 visual variations · assets from a single Gem that generates all of them
• **Upgrading old images** · black-and-white grandparent photos, low-quality images · upgrade to 4K. An asset for blogs, books, company website
• **Video ad storyboard** · the critical transition to Level 3 · turning an image into a full ad via a video model

Level 3 · Masters · Full system in Claude Code

💡 You can watch the full Claude Code workshop (4 sessions · from installation to building a WhatsApp agent) on Skool. In this guide, I provide the prompt that describes to Claude what to build · but you need to have Claude Code installed and Google AI Studio API + Seedance API connected to it.

Prompt for Claude · Building a full automated ad system (for copying to Claude Code)

I want to build a system that automatically generates a video ad from a product image. Process: 1. I place a product image in a folder named `input/` on my computer. 2. The system takes the image, sends it to Google AI Studio (Gemini API · Nano Banana 2 model) with a prompt that creates a 9-frame storyboard in 9:16 format including text directing instructions. 3. Receives back the storyboard image · saves it in `storyboards/`. 4. Sends the storyboard to the Seedance API (via fal.ai) with an instruction to create a 10-second 9:16 video ad based on the scenes in the storyboard. 5. Downloads the video to the `output/` folder. 6. Sends me a notification (Slack / WhatsApp / email) that the ad is ready with the link. Work in Plan Mode. First show me the plan · which APIs are needed, what environment variables (API keys), what the folder structure is. After I approve · proceed to building. Work with Python. Write clean code with error handling so that if something fails midway · the system recovers and sends me a clear error message.

What needs to be installed before starting Level 3

• **Claude Code on your computer** · at claude.ai · installation via the workshop (4 recorded sessions on Skool)
• **Google AI Studio API key** · go to aistudio.google.com → Get API key → create a free key
• **fal.ai account with credits** · fal.ai hosts Seedance · paid credits are required (starting from $10 gives plenty of ads)
• **Working folder on your computer** · create a folder named 'ad-pipeline' with input/storyboards/output subfolders
• **Slack / WhatsApp / Resend connection** (optional) · if you want the system to notify you when the ad is ready · connect the integration via Claude Code Connectors

✅ Once the system is built · operation takes one second. Drag an image into the folder · go grab a coffee · come back · you have a ready ad. If you are a store owner who uploads a new product every week · 52 ads a year with zero effort.

Solutions to common issues when working with Nano Banana 2

8 solutions to common issues

• **It changes my facial features in a profile picture** · in the prompt explicitly add: 'Do not modify the face, hair, or any facial features in any way. Preserve the person's identity exactly as in the source image.' Also input 2-3 photos of the same person from different angles
• **It sends me text instead of an image / refuses to execute** · the model switched to a cheaper version (you ran out of free credits). Check · is the new image significantly worse? · you switched to an older model. You need to wait until the next day or subscribe to paid
• **After 6-7 modifications on the same image · quality degrades** · open a new chat, upload the latest image, continue from there · this resets the memory
• **It adds logos / text without me asking** · in the prompt add: 'No logos, no text, no captions, no watermarks, no brand names anywhere in the frame.' Be explicit about what you do not want
• **A 2D architectural plan turns into 3D with changes for me** · request only a single action · 'Convert this 2D floor plan to a 3D isometric view. Do not change room layout, dimensions, or wall positions. Only add depth and material textures.' Work in steps · after this comes out · next request
• **It mixed a reference image into the final product** · if you provided a reference image in Gem · it sometimes copies elements from it into the output. Remove the reference image from Knowledge, or input it only as a visual example in the prompt ('Style reference only, do not copy elements')
• **Want to input more than 10 images at once** · create a single document in Canva / Figma with all the images (up to 20 items), save as PNG, and input this PNG into the tool as a single image
• **Want to work in RTL in the prompt** · the prompt itself works well in English · spoken Hebrew is also understood, but it is more precise in English. If you want Hebrew text in the image · explicitly write it in Hebrew inside the prompt

Frequently asked questions

Nano Banana 2 is completely free up to 20 images per day in Gemini. For a private user who produces one or two images a day · the free tier is enough. If you are a business owner producing images for a campaign · the paid Gemini subscription is $20/month and gives 100 images per day. For those working at commercial scale (advertising agency, designers) · switch to the Google AI Studio API · pay per call (a few cents per image).

Both are excellent. The small differences · Nano Banana 2 is great at manipulations on existing images (background replacement, combining multiple images, changing a single item without touching the rest). ChatGPT's image model (GPT-4o image gen) is especially strong in Hebrew text and complex scenes with many elements. If you are already paying for one · the other is not mandatory. If you want the complete package · start with Nano Banana (cheaper and more flexible) and add ChatGPT as needed.

Yes, you can. In Veo 3.1 (on paid Gemini subscription) · you work with faces without problems. In Seedance · there are limitations in various tools that host it · especially after copyright lawsuits in the industry. The most convenient way to work with faces today is via fal.ai (a tool that hosts Seedance and allows face uploads). If you generate content with other people's faces · ensure you have explicit usage rights.

Yes · this is one of the most popular uses. The prompt that works best: 'Change the clothing to [clothing description] but preserve the face, hair, skin tone, and body proportions exactly as in the original image. The new clothing should fit naturally on the body.' Input 2-3 photos of the same person from different angles · this way it will learn the face better and not change it.

This is precisely the case for Level 3 · a system in Claude Code. You build a pipeline that takes all images from an input folder, loops them through the Google AI Studio API, and returns each packshot to an output folder. One run · 200 packshots ready. API cost for Gemini with 200 images · a few dollars. In Session 18 we showed an example with 5 products · the exact same principle.

It understands Hebrew well · but the model was trained mostly on English, so an English prompt comes out more precise. If you want Hebrew · write the prompt in simple and precise Hebrew, without clichés. If it does not work out · translate the prompt to English (ChatGPT/Claude will do it in a second) and feed the English · the result will be better. If you want Hebrew text inside the image · write the text in Hebrew in the prompt explicitly: 'Add the Hebrew text "מבצע השבוע" at the top of the image in a bold sans-serif font.'

Yes · via a Gem or via a system in Claude Code. In Gem · in Knowledge, upload a document with your Brand Guidelines · color palette, font, image style. In Instructions, define · 'Always use these brand colors: #FF5733, #1A1A1A, #F5F5F5. Always use a clean modern sans-serif font. Style: minimalist, premium, lots of negative space.' Every image this Gem generates · will appear in the same brand language. If you want this at an automated system level · build a skill in Claude Code that holds the Brand Guidelines and adds them to every API call.

The main risk · copyright on elements that the model incorporated. If you uploaded a photo of Nike shoes (a photo by a specific photographer) · there is a risk you might infringe on copyrights. To avoid this · 1) Use only images you photographed yourselves or have a license for. 2) If you want an image inspired by a specific brand · ask the tool to generate a new image in a certain style · without uploading the original image. 3) For large commercial ads · keep all prompts and versions. If there is a lawsuit · you have documentation.

💪 Want guidance along the way?