OpenAI has integrated image generation directly into ChatGPT. This is not just "image support"; it’s a full upgrade. The generation process is not through diffusion like DALL-E, but sequentially, "letter by letter" — from left to right, top to bottom. Thanks to this, Sora has finally achieved what seemed impossible: correctly writing text on images without getting confused about objects, colors, and shapes.
You can ask for a comic, a Newton diagram, or a poster with text — and the model will produce a coherent image where everything is in its place. Right inside ChatGPT.
Live Demo: https://www.youtube.com/watch?v=2f3K43FHRKo
