DALL-E 3: images that actually follow instructions
In one sentence OpenAI launches DALL-E 3 integrated into ChatGPT: dramatically improved prompt adherence over DALL-E 2, automatic caption synthesis for training, more readable text in images.
With DALL-E 2, asking for "a cat wearing a red hat reading a newspaper" produced chaotic results: the hat vanished, the newspaper became unreadable, and the cat did whatever it wanted. DALL-E 3 fixes exactly this — the generated image follows the text almost literally.
The trick is that OpenAI rewrote all training data captions using GPT-4: instead of short, vague descriptions, every image gets a detailed caption. The model then learns to match precise words to precise visual details.
Direct integration into ChatGPT lowers the barrier: no API, no specialized prompt engineering. You ask in natural language and get a professional-quality image.
Companies
OpenAI
Tools
DALL-E 3, ChatGPT
Tags
Sources