OpenAI teases GPT-Image 2 with an AI-generated screenshot that looks completely real

The story is built almost entirely on one image OpenAI chose to release, which is about the weakest possible evidence for a photorealism claim. There's no third-party evaluation, no model card, no information about what prompts or generation attempts were discarded before that screenshot was picked.

OpenAI has been on a rapid release cadence across multiple product lines this month. The Codex expansion covered by The Verge on April 16 already bundled image generation as one of several new capabilities, suggesting image output is being positioned as a horizontal feature rather than a standalone product. Meanwhile, Google's Gemini integration with Google Photos (Ars Technica, April 16) shows the competitive pressure is shifting toward personalization and grounded generation, not raw photorealism. GPT-Image 2 appears to be OpenAI's answer on the realism axis, but it's not yet clear whether that axis is where the real competition is heading.

Watch whether independent researchers can reproduce the photorealism claim on diverse prompt sets within the first two weeks of public access. If outputs require significant prompt engineering or show systematic artifacts outside studio-style compositions, the single-screenshot launch will look like careful curation rather than a capability floor.