ChatGPT's New Image Engine Tested: Better Typography, Web Access, and Reasoning Capabilities

ChatGPT’s new image engine, Images 2.0, has launched with a suite of enhancements, including better typography, access to the web, and built-in reasoning capabilities. To assess its practical applications, Axios put the tool to the test.

Why This Update Matters

Previous image engines gained initial popularity among consumers but often fell short in addressing key flaws that limited broader business adoption. The new version aims to bridge that gap.

Key Features of ChatGPT Images 2.0

• Wide Range of Aspect Ratios: Supports diverse formatting needs.
• Standard and 'Thinking' Modes: The standard version is available to all users, while the 'thinking' mode—equipped with reasoning features—is reserved for paid subscribers.

Real-World Testing: Tasks and Results

In a day of hands-on testing, the new model demonstrated versatility across multiple tasks:

• Memorial Image Creation: A friend requested an image memorializing her recently deceased cat alongside two favorite toys. The tool generated a highly personalized sympathy card-style image.
• Photo Album Simulation: It transformed two wedding photos to appear as if they were placed in an old-style photo album with photo corners.
• Fictional Event Poster: Colleagues suggested creating a poster for a fictional event—a Mike Allen look-alike contest in Washington Square Park. The tool delivered a poster design, noting the event’s fictional status unless attendees appeared.
• Infographic Creation: The tool generated an infographic titled “The Case Against Candy Corn,” which was used unsuccessfully to persuade colleagues that candy corn is neither candy nor corn.
• Room Organization Visualization: A photo of a cluttered bedroom was uploaded, and the tool provided a cleaned-up version, revealing additional space. While the user found it inspiring, their partner dismissed it as a tease, stating they’d believe it when ChatGPT could physically clean the room.
• Trading Card Generation: It converted photos of the user playing softball and their 13-year-old playing soccer into trading cards, complete with names, positions, and team logos extracted from their uniforms.

Limitations and Challenges

The testing also uncovered areas where the tool fell short:

• Newspaper Creation Issues: When tasked with creating a faux newspaper titled “Smart Brevity Times” based on the latest Axios headlines, the first attempt used outdated articles. A second attempt pulled current stories but produced a mock-up rather than a polished final product.
• Design Polish: Some creations lacked elegance, such as a mahjong cheat sheet that was accurate but unrefined.
• Processing Time: The added reasoning capabilities can significantly increase the time required to generate images, requiring users to plan ahead.

Final Verdict

ChatGPT’s Images 2.0 shows promise with its improved features and versatility, but it still faces challenges in delivering consistently polished and timely results. The tool’s potential for personal and professional use is evident, though refinements are needed for broader adoption.