OpenAI upgrades gpt-image-1: accurate text, photorealistic portraits, and inpainting API

Until a few months ago, asking ChatGPT to generate an image with text written inside was a gamble: sometimes it came out well, often the words were mangled or unreadable. With gpt-image-1, OpenAI solves this problem reliably. You can generate a flyer, a poster, a social media post with legible text directly from the AI.

Photorealistic portraits are another leap: people look photographed, not generated. The most useful thing for content producers is character consistency: if you create a character in one image, you can use it in other images maintaining the same appearance, without having to start from scratch each time.

There's also inpainting via API: you can upload an image, indicate a zone to modify with a mask, and the model fills that zone coherently with the rest. Useful for removing objects, changing backgrounds, adding elements. With this update DALL-E 3 is effectively retired as the primary backend: gpt-image-1 becomes the standard model for image generation across the entire OpenAI platform.