OpenAI has released "the best image generator": What we know about GPT-Image 2

OpenAI has released a new version of its image generator, GPT-Image 2, which has emerged as a worthy rival to Nano Banana.

Prompt: A photorealistic candid travel scene of a person standing at a coastal roadside turnout on an overcast morning, shot on 35mm film. Natural imperfect framing, visible grain, ambient light, muted colors, wind in clothing and hair, cinematic realism, and the feeling of a lived-in documentary photograph.

What’s new in GPT-Image 2?

• One of the main advantages is improved text processing. Now the AI handles fine details and reproduces captions on images in multiple languages flawlessly. While the previous version had issues with this, things are much better now.
• The model generates any format: realistic images, drawings, games, ads, handwritten text, newspaper-style photos, and more.
• Improved generation of realistic photos that are indistinguishable from the real thing. The generator renders lighting much better and captures microtextures.
• You can now edit images: select the desired area and modify only that part using text edits.

Prompt: A photorealistic snapshot portrait of two friends outside a venue at night, shot on a compact point-and-shoot camera with direct flash. Close subject distance, crisp foreground detail, deep shadow falloff, slightly raw spontaneous energy, nightlife atmosphere, and the unmistakable look of an early-2000s flash photograph.