The video compares OpenAI’s GPT Image 1.5 with Nano Banana Pro, concluding that while GPT Image 1.5 shows notable improvements and unique features like transparent image generation, Nano Banana Pro remains superior in accuracy, detail, and handling complex tasks. Despite GPT Image 1.5’s promise and free accessibility, it has not yet surpassed Nano Banana Pro as the leading AI image generator.
The video presents a detailed comparison between OpenAI’s newly released image generator, GPT Image 1.5, and the current leading AI image generator, Nano Banana Pro. Both models offer similar functionalities such as editing images of people, changing hairstyles, clothing, backgrounds, and creating deepfakes. However, the creator focuses on testing the limits of GPT Image 1.5 with more challenging prompts that typically trip up other image editors. Initial tests, such as generating Pokémon images and rendering complex emotions, show that while GPT Image 1.5 performs well, Nano Banana Pro generally produces more accurate and visually appealing results, especially in detailed renderings.
The video also explores GPT Image 1.5’s ability to solve math homework by generating handwritten solutions, which it does impressively well, even outperforming Nano Banana Pro in preserving the original background. However, both models struggle with biology homework, failing to correctly label organelles. When tasked with generating complex image transformations like infrared thermal maps, segmentation maps, depth maps, and color inversions, Nano Banana Pro again outperforms GPT Image 1.5, producing more accurate and visually coherent results. In tricky prompts involving clocks, rare frogs, manga creation, and translating manga to Chinese, both models show strengths and weaknesses, but Nano Banana Pro often edges out GPT Image 1.5 in accuracy and detail.
Further tests include generating images of existing celebrities and anime characters, where Nano Banana Pro demonstrates superior world knowledge and character consistency. GPT Image 1.5 is limited by guardrails that restrict certain content generation, which Nano Banana Pro handles more freely. When it comes to spatial understanding, such as creating floor plans from photos, and generating gameplay scenes or technical diagrams, Nano Banana Pro consistently delivers more accurate and coherent outputs. GPT Image 1.5 shows promise but falls short in these complex tasks, often missing key details or producing less precise visuals.
The video also highlights some unique advantages of GPT Image 1.5, such as its ability to generate transparent images, which Nano Banana Pro currently cannot do. Both models perform well in remastering classic game screenshots and generating sprite sheets, with GPT Image 1.5 offering some novel features. However, in data visualization tasks like converting tables into charts, Nano Banana Pro excels by accurately capturing data and performing complex calculations to create meaningful graphs. Both models struggle with highly complex scenes like a “Where’s Waldo” image, producing distorted faces and obvious placements.
In conclusion, while GPT Image 1.5 is a significant improvement over its predecessor and offers free access to users, Nano Banana Pro remains the superior AI image generator in terms of accuracy, detail, and world knowledge. The video acknowledges that GPT Image 1.5 has removed previous issues like the yellow tinge and improved text rendering, but Nano Banana Pro still sets a very high bar that is difficult to surpass. The creator encourages viewers to try GPT Image 1.5 themselves and shares information about Chat LLM by Abacus AI, a platform that integrates multiple AI models for enhanced productivity. Overall, GPT Image 1.5 is promising but has not yet dethroned Nano Banana Pro as the king of AI image generation.