AI image generation
Nano Banana Pro, an AI image-generation model, can construct a complete brand identity from scratch — logo, packaging, product photography, lifestyle imagery, mascot with lore, Instagram grid, landing page, and merch — in under an hour starting from zero assets, photos, or inventory, driven almost entirely by short text prompts.
The video frames itself as a stress test: build an entire matcha brand named 'Verdant' from absolute zero (no assets, no photos, no inventory) using only Nano Banana Pro.
Nano Banana Pro is offered in three image resolutions: 1K, 2K, and 4K.
Text-in-image rendering (a common failure point for other models, which warp or break text) is treated as a critical early test, applied first to the logo's sans-serif font.
The logo is generated first as a standalone vector mark, then extended into a full brand logo while keeping it minimalistic and elegant.
Product packaging (a matcha can) is tested for realistic materials — a transparent can body and an aluminum top — from a short prompt, which the presenter frames as evidence of strong prompt following.
The liquid color inside the can is changed (to orange-yellow, then to a complex mix of blue matcha, deep purple, and a creamy texture) to test how naturally the model blends multiple textures/colors.
The product is placed in different settings (concrete counter in a modern cafe) to test composition-driven product photography.
Multiple cans are laid out in an open cooler with ice to test whether the logo (down to a small dot detail) stays consistent across differently angled/positioned instances of the same product.
The brand name is placed on a hanging t-shirt near the sea to test the model's ability to render text and objects in unusual orientations, including upside down.
A lifestyle shot places the product in the hands of the target audience to test naturalness of pose and hand rendering.
The product is shown on a cafe table and in a Pinterest-style media-kit shot, testing ambient lighting/shadow integration and liquid physics inside the can.
A full matcha shop scene (also named 'Verdant') is generated, testing embossed branding on tile, crowd/street realism, and undistorted patron faces.
A brand mascot (a character in a scarf) is generated, followed by a prompt asking the model to generate backstory/lore for the mascot given only the brand name and product description.
Previously generated images are combined into a mocked-up nine-post Instagram grid shown inside an iPhone screen, testing the model's ability to composite and reproduce prior generations consistently.
A landing page is generated for the brand, evaluated on clarity, layout cohesion, a visible call-to-action, and added brand/product information.
The brand world is expanded further with a matcha recipe, branded merch (scarf, tote bag, hoodie), a pop-up festival stand design (with matcha powder, greenery, textile banners), and a meme using the mascot for social engagement.
The video concludes that a full brand identity — logo, packaging, photos, website layout, and mascot — was built in under an hour from zero starting materials.
The model preserves fine brand-identity details (including a small dot on the logo) consistently across multiple differently-angled product renders in the same scene, which the presenter contrasts with other models' tendency to let objects 'blend or morph.'
The model correctly renders text and objects in atypical orientations (e.g., a brand name upside down on a hanging t-shirt), which the presenter says most models fail at, especially with faces and text.
The model adds plausible, unprompted environmental details — e.g., generating a real MacBook on a cafe table and correct window-light shadows without those elements being requested — suggesting it infers scene context beyond the literal prompt.
The model can generate an entire fictional backstory/lore for a mascot from only the brand name and product category, functioning as a lightweight copywriting layer on top of image generation.
The model can reconstruct earlier-generated images into a new composite (a mocked-up Instagram grid on an iPhone screen) while keeping each individual image intact, effectively doubling as a mockup/compositing tool.
The presenter claims to have independently fact-checked an AI-generated matcha recipe and confirmed it reflects how matcha 'should be made,' suggesting the model's outputs can carry accurate domain knowledge, not just plausible-looking visuals.
«I have zero assets, zero photos, and zero inventory.»
— 00:13
«This is always a critical test for AI because other models have a hard time handling the text.»
— 00:33
«Most models struggle here because realistic textures are difficult. Things often look too shiny or artificial.»
— 01:37
«Most of the models don't know how to place objects, especially faces and text upside down, but Banana Pro, as we can see, can do it perfectly well.»
— 03:28
«And I also like that Nanabanana generated a real MacBook page without us specifically pointing this out.»
— 04:09
«This instantly looks like a curated and intentional brand feed.»
— 05:57
«Okay, guys. Just like that, we built a complete brand identity in less than an hour with zero assets, zero inventory, and zero starting materials.»
— 07:40
The video is a single-project capability demo for one model (Nano Banana Pro) rather than a teaching resource on prompting technique; it's useful as a reference for the range of branding outputs the model can produce (logo, packaging, photography, mascot lore, social/landing-page mockups, merch) but offers little generalizable method beyond the specific prompts shown.