OpenAI’s ChatGPT Rivals Google’s Nano Banana Pro in Image Generation

OpenAI has rolled out a significant upgrade to its image generation capabilities within ChatGPT, introducing a new model that delivers faster speeds, greater accuracy, and more nuanced control for users. This advancement positions the tool as a more formidable competitor in the AI image space, challenging other leading models. The improvements are not just incremental; they represent a meaningful step forward in how AI interprets creative instructions and assembles visual content.

The newly released model, referred to here as GPT-Image 1.5, boasts several key enhancements. It follows user prompts more accurately, preserves intricate details better, and generates images up to four times faster than its predecessor. A practical new feature allows users to queue additional image requests while others are still processing, streamlining the creative workflow. This model is now accessible to all ChatGPT users and through the developer API.

According to app CEO Fidji Simo, this upgrade is part of a broader evolution for ChatGPT. The vision is to move beyond a reactive text interface toward a “fully generative UI” that intelligently brings together the right tools based on the user’s intent, whether that involves writing, analysis, or visual creation.

A major area of improvement is in image editing. The updated model makes targeted adjustments, like adding or removing elements, without disrupting the overall coherence of the picture. It maintains consistency across lighting, composition, and facial features far more reliably. This opens up practical applications such as sophisticated photo editing, virtual try-ons for fashion and hairstyles, and complete style transformations. Demonstrations from OpenAI show the model skillfully combining subjects from separate photos into a single cohesive scene or applying complex aesthetic filters, like converting a portrait into a vintage Hollywood movie poster.

Perhaps most impressively, the model now demonstrates a much stronger grasp of complex, multi-step instructions. In internal testing with a detailed grid requiring specific objects in specific cells, the new version succeeded where the old one failed. This capability is crucial for users who need precise control over layout and element placement in their generated images.

Text rendering within images has also seen notable progress. The model can produce legible snippets of articles, simple tables, or infographics with numbers, handling denser and smaller text than before. However, OpenAI acknowledges ongoing challenges with longer text passages, uncommon fonts, images containing multiple faces, and generating content in languages other than English.

In comparative testing with a demanding, photorealistic prompt, a scene featuring a horse riding an astronaut, the new model performs on par with other top-tier generators like Google’s Nano Banana Pro, a marked improvement over previous versions. Initial observations suggest a difference in stylistic interpretation; ChatGPT’s images often have a more intense, polished editorial look, while Nano Banana Pro’s outputs can appear more literal and casually photographic, though this may be influenced by prompt engineering.

For developers, the news includes a welcome economic benefit. Despite the enhanced performance, API costs for image generation have dropped by approximately 20 percent. The new pricing structure is set at $8 per million input tokens and $32 per million output tokens for images, with text tokens priced separately. This makes the powerful technology more accessible for integration into various applications.

OpenAI also notes that the model does a superior job at preserving brand logos and specific visual elements, a critical consideration for marketing and e-commerce use cases. The previous version of the image generation model remains available as a custom GPT option for users who may have specific workflows built around it.

(Source: The Decoder)