ChatGPT Images 1.5 launches with 4x speed improvement and new in-app discovery surface
Dec 16, 2025
Key Points
- OpenAI launches ChatGPT Images 1.5 with 4x faster generation and a new in-app discovery surface offering preset styles to lower the creative barrier.
- The model struggles with complex spatial reasoning and narrative coherence, failing internal benchmarks like Where's Waldo that require precise detail at scale.
- OpenAI hires a Google executive to lead mergers and acquisitions as communications chief Hannah Wong departs in January.
Summary
OpenAI has launched ChatGPT Images 1.5, a new flagship image generation model that is 4x faster than its predecessor. The release includes a dedicated images feature within the ChatGPT app that offers preset styles and ideas rather than requiring users to write detailed prompts. This UI addition lowers the barrier to image generation by addressing a known friction point: when users face a blank text box, they tend toward generic requests like "a dog" or "a happy dog." The results are technically impressive but lack novelty because photorealistic dog images already exist at scale online.
Midjourney solved this partly through social proof by hosting generation in shared Discord chat rooms where users saw what others were creating and borrowed ideas. OpenAI is taking a more direct approach by embedding inspiration directly into the product through preset styles and starter ideas, giving users a jumping-off point without requiring them to arrive with a fully formed creative vision.
When tested on a Where's Waldo prompt, ChatGPT Images 1.5 produced results that look convincing from a distance but break down on inspection. The images approximate the isometric perspective of the source material but lack the precise storytelling detail that makes Where's Waldo special. Martin Hanford, the original illustrator, spent up to eight weeks on a single image; the AI version generates details that don't quite cohere at scale and has failed to include a proper Waldo figure in test cases. The gap suggests that while generation speed and broad creative capability have improved significantly, the model still struggles with complex spatial reasoning, consistent character inclusion, and narrative coherence across a single image. One potential path forward is a tile-by-tile generation approach that builds the image piece by piece and inserts a Waldo at the end rather than generating the full scene in a single pass.
OpenAI also hired an executive from Google to lead mergers and acquisitions. Hannah Wong, the company's first chief communications officer and a key figure during the ChatGPT launch, is departing in January.