Summary
Nate Herk runs a structured 30-prompt head-to-head between GPT Image 2 and Google’s Imagen 3 (accessed as Nano Banana 2), covering artistic styles, photorealism, product packaging, UI screenshots, portraits, physics diagrams, and 3D modeling. To add a layer of objectivity, Herk uses Claude Opus 4.7 as an automated judge, which scores GPT Image 2 as the overall winner — with advantages particularly pronounced in photorealism and complex scene composition. Pricing is also compared directly: Imagen 3 ranges from $0.04 to $0.09 per image by resolution quality tier, while GPT Image 2 is a flat $0.06 per image through key.ai, an OpenRouter-style multi-model API.
The second half of the video catalogs practical use cases for GPT Image 2, starting with pitch-ready product packaging: cereal boxes, pill bottles, and branded bags all with accurate nutrition labels, readable text, and scannable barcodes. Herk then demonstrates handwriting cleanup — uploading a crumpled handwritten page and generating a clean, crease-free scan that matches the original handwriting with high fidelity.
Additional use cases include spatial layout control for thumbnail and product mock-up composition, web search integration that pulls real brand logos into UI screenshots, cartoon portrait generation from reference photos, and multi-image iPhone mockups for app store listings. The video functions as both a model selection guide for teams choosing between the two leading image generators and a practical use-case library for content creators and product designers.
📺 Source: Nate Herk | AI Automation · Published April 22, 2026
🏷️ Format: Comparison
