Amazon listing images
The video claims Google's Nano Banana Pro, used through the Gemini app and fed with ordinary phone photos of a physical product, is the first AI that can produce a full, conversion-ready Amazon listing image set with no photographer or graphic designer. The speaker presents it as an iterative, chat-directed process that takes about 10 minutes and costs little or nothing.
The speaker's starting claim is that Amazon listing images used to need a photographer, a graphic designer, or even a 3D rendering artist. Nano Banana Pro is said to be the first model that replaces all of them. The video says large brands and aggregators have started trusting it over photographers and designers for their top-selling products.
The proof offered is a live walkthrough on two real products: an electric bike pump and a ceramic essential-oil diffuser. The model renders fine product detail (button engravings, display, hose and nozzle, light states) and produces legible, non-gibberish text on graphics. The speaker presents this as the first time a chat with a model yields a graphic ready to upload to an e-commerce listing.
The video does not present the model as reliable on the first try. The speaker shows a lifestyle image with the wrong scale, a features graphic that was corny and not mobile optimized, and a failed correction attempt. The speaker frames the work as directing a photo shoot in a chat: you look at the shots, make adjustments and work around roadblocks.
The quality bar is whether you would accept the result from a paid designer, and the speaker also compares outputs with typical competitor listings. Even the weak first features graphic is described as better than many sellers' images, which the speaker says are poor and not mobile optimized. The final features graphic is called work for which a designer's 'salary was well earned.'
The speaker says a human still has to supply the commercial judgment. The model's wording ('cool mist aromatherapy') missed the product's real selling point of ultra-quiet operation. For your own product you may not trust the AI to pick the most important benefits, and you will make tweaks.
The economic argument is speed and cost. The speaker says the whole process takes about 10 minutes, and that it is a fraction of the time of working with a live photographer and a graphic designer. It is free on the free tier, which hits limits quickly. A Pro plan at roughly $19–20 a month is one the speaker says they have never maxed out.
The speaker says an Amazon listing has about 22 types of images, and the video demonstrates only a few of them. The demonstrated set is the primary and lifestyle images, plus the benefits, features, instructional and competitor-comparison secondary graphics. The rest sit in an SOP linked from the speaker's free guide, which also holds a prompt library and further examples such as products with labels.
Switch Gemini to Thinking with 3 Pro — In the Gemini phone app, changing the mode selector from 'Fast' to 'Thinking with 3 Pro' puts you on Nano Banana Pro, signalled by a 'thinking' circle. Apply: Download Gemini on iPhone or Android, open it, tap the mode selector at the bottom and switch from Fast to Thinking with 3 Pro before generating any product images.
Multi-angle, close-up product photo capture — Giving the model several phone photos at different zoom distances and angles so it learns the full shape and features of the physical product. Apply: Take three to four photos that show the whole product plus close-ups of the button engravings, display and charge ports. Upload them with the plus button in Gemini.
Scale reference in frame — Including an object of obvious size in the product photos so the AI can judge how big the product really is. Apply: Hold the product in your hand or place it next to money, a laptop or a phone, and check later images against objects in the scene such as a coaster or books.
Plain-language prompting — Writing short, human-language prompts instead of long engineered prompts, because the model is meant to understand human language. Apply: Skip AI-written page-long prompts and camera make-and-model specifications, and ask directly, for example 'Create a clean product photo rendering of this product to be used as the primary image on an Amazon listing.'
Corrective prompting — Telling the model what is wrong with a result and asking for a retry in the same chat. Apply: State the flaw and request a new setting, as in 'Try again in a different comfy aesthetic living room. You made the diffuser look a little too big.'
Competitor reference-graphic prompt — Uploading a competitor's graphic as a style reference and asking for a bigger, mobile-optimized version in that style. Apply: Upload the reference image and ask for it in that style with larger text and icons optimized for mobile. The speaker notes it returned basically the same image in that attempt, so be ready to reset.
New-chat reset — Abandoning a degrading chat and starting fresh when the model hits a wall or gets worse. Apply: Tap the square pencil icon in the upper right for a new chat and re-upload the original product photos. Alternatively, download the latest image and upload it into the new chat.
Features graphic prompt with labeled icons — Asking for a clean, aesthetic features graphic that shows the product with labeled icons pointing to each listed feature. Apply: List the features (for example ultrasonic diffusion, seven color mood lighting, totally silent, works with any essential oils) and say 'features' rather than mixing it with 'benefits'. Ask for a clean aesthetic graphic for an Amazon listing with large text and icons.
Targeted text edit by chat — Correcting wording inside a generated graphic by naming the exact section and the replacement text, with no external editor. Apply: In the same chat, say 'replace the text in the lower left section from cool mist aromatherapy to ultra quiet aromatherapy', then check the result.
Momentum / piggyback prompting — Once one graphic has the right style, asking for a similar graphic for another content type so the new image inherits the look. Apply: After a good features graphic, prompt 'create a similar graphic but for benefits instead of features'. The speaker gave no benefits, and the model chose them itself.
Step-by-step instructions graphic — A graphic that shows numbered usage steps with icons to make the product look very simple to use, which the speaker calls a very powerful way to increase conversions. Apply: Prompt 'Now create a clean aesthetic step-by-step instructions graphic to be used on an Amazon listing' and list the steps (fill with water, add essential oils, press power button, select color).
Listing image set sequence — The set of listing images the video builds in order: primary image, lifestyle image, features graphic, benefits graphic, instructions graphic and a competitor comparison. They are drawn from about 22 possible image types. Apply: Produce each type in turn within one chat, and use the speaker's SOP for the full list of 22 options and examples of high-converting images in each category.
Designer-standard and mobile-readiness check — A quality heuristic for judging outputs: would you accept this from a paid designer, is the text large enough for mobile, and is the scale believable?. Apply: Review each output for corny design, tiny text, wrong scale and wrong selling points. Iterate or reset until it would pass as designer work, then upload it.
Scale is a recurring failure point. A product shot alone against a wall or floor makes its size hard for the AI to judge, and the first lifestyle image made the diffuser too big against a coaster and books. The speaker checks scale against the other objects in the scene.
Prompt vocabulary matters. The first features-graphic prompt accidentally mixed 'benefits' and 'features', and the speaker cites this as part of why the result was off.
The recovery that worked was a clean restart, not the correction. After the competitor-reference retry returned basically the same image, a new chat with the original photos re-uploaded produced a very good graphic. The speaker then realized they had never uploaded the competitor image, and it worked anyway.
The model generated four plausible benefits with no benefits or context supplied (long-lasting quality, enhanced atmosphere, improved wellness, effortless relaxation). The speaker notes one of them is really a feature-benefit mix.
For the instructions graphic the model chose fitting icons (water, drip, power, switcher) without ever being shown the inside of the product. The speaker was amazed by this.
Text in the output can be changed by chat alone. A one-line instruction to replace the lower-left text was executed perfectly, with no Canva or Photoshop round-trip.
Quality compounds within a chat. The speaker says 'you kind of gain momentum', so once one graphic has the right vibe, the next ones are easier. Long chats eventually go off the rails, though, and that is the signal to reset.
The speaker uses a phone because the product is physical. You need to keep taking new pictures, angles and close-ups to give the model more context, which a desktop workflow doesn't support.
«this is the first AI ever that lets you make a full Amazon listing image set with no photographer, no graphic designer whatsoever.»
— 00:08
«for the first time in history, large brands and aggregators have started trusting this AI model instead of photographers and designers to create high converting images for their top selling products on Amazon.»
— 00:16
«you need to be able to constantly take new pictures and new angles and close-ups of it to give the model more context.»
— 01:46
«You kind of hit a wall with the model and it's not getting any better. And sometimes it'll just get worse and worse. When you've hit that point, it's time to start a new chat.»
— 11:41
«This is really freaking good. We went from one that was really bad to one that's mobile optimized and really good.»
— 12:41
«This is the first time in human history that you can chat back and forth with the model and get a graphic like this that's actually ready for prime time to upload straight onto your e-commerce listings.»
— 14:00
«It's still going to take you a fraction of the time that it would have working with a live photographer and a graphic designer.»
— 16:00
«this whole process is about 10 minutes.»
— 30:24
«you're directing the model and you're getting different angles and you're looking at the shots and then you're making adjustments.»
— 30:51
«if you're using the free version, you're going to hit limits pretty quick.»
— 31:02
«That's only four. There's actually like 22 different options.»
— 32:09
Reception
Viewers found the content useful and engaging, but many complained that the creator's eating and mouth noises on camera made the video hard to watch, and spam promotions for other tools are mixed in.
A hands-on, demo-driven walkthrough whose main value is the iteration playbook: multi-angle photos, plain prompts, corrective retries, new-chat resets and chat-based text edits. It also serves as a lead-in to the speaker's free guide, prompt library and 22-type image SOP.

32:38