程序员的那些事·2025-12-20 02:10Core Viewpoint - OpenAI has launched its new flagship image model, GPT Image 1.5, which claims to outperform Google's Nano Banana Pro in various benchmarks, but initial user feedback has been largely negative, suggesting it may be a case of "high scores but low capability" [1][12][26]. Group 1: Model Features and Performance - GPT Image 1.5 has significantly improved capabilities, including enhanced instruction comprehension, detailed image retention, and a generation speed that is four times faster than its predecessor [4][6]. - In benchmark tests, GPT Image 1.5 achieved a score of 1264, ranking first in Text-to-Image generation, while Google's Nano Banana Pro scored 1235, placing second [10][15]. - The model's editing capabilities allow for precise modifications while maintaining consistency in lighting, composition, and character appearance across multiple edits [28][30][67]. Group 2: User Feedback and Comparisons - Despite its high ranking, user tests revealed that GPT Image 1.5 struggles with certain tasks, such as interpreting handwritten notes, leading to criticism that it does not live up to expectations [12][14][21]. - Users have expressed disappointment, with some stating that Google's Nano Banana Pro remains superior, and others labeling OpenAI's release as "meaningless" [21][26]. - Comparisons between images generated by GPT Image 1.5 and those from Nano Banana Pro show that the latter often produces more realistic results, raising concerns about OpenAI's competitive edge [183][189]. Group 3: Market Context and Competitive Landscape - OpenAI's rapid response to Google's advancements, including the recent release of GPT-5.2 and now GPT Image 1.5, indicates a competitive urgency in the AI image generation market [163][164]. - The AI image model competition is intensifying, with other players like Qwen-Image and Black Forest Labs also making strides in the field [167]. - OpenAI aims to capture the enterprise market by enhancing its image generation capabilities, which are now 20% cheaper for API users compared to the previous version [152][153].