DALL-E 3 review

What it is

DALL-E 3 is OpenAI's image model, used almost entirely through ChatGPT rather than as a standalone product. You describe an image in conversation, it generates, and you refine by describing changes in plain language.

Where it shines

Prompt adherence. Give it a paragraph specifying five objects, their positions, the lighting and a mood, and it will attempt every constraint rather than producing a beautiful image that ignores half of them. For diagrams, explanatory illustrations and any image where the content matters more than the styling, that literalness is the right trade.

The conversational loop is the other advantage. Because the generator sits inside the assistant, you can say make the background darker and remove the second figure without rewriting the prompt, and the model handles the translation. For non-designers this is far more approachable than parameter-based tools.

It also has no marginal cost for anyone already paying for ChatGPT, which quietly makes it the most-used image generator in the world despite not being the best one.

Where it falls short

Aesthetics. Side by side with Midjourney or Flux, DALL-E output looks like stock illustration — competent, slightly plastic, rarely striking. Skin, hands and complex lighting are noticeably weaker.

The content filters are the other frequent complaint. They block public figures and trademarked characters as expected, but they also reject a surprising number of innocuous requests, and the refusal message rarely explains which part was the problem. Power users also miss the controls other tools offer: no seed, no weighting, no style reference, no negative prompt in the consumer interface.

Pricing in plain English

A small free allowance inside ChatGPT; realistic use requires Plus at $20 per month. API pricing starts around four cents per image depending on size and quality.

Who should use it

ChatGPT subscribers who need images occasionally, educators and writers making explanatory visuals, and anyone who would rather describe an edit than learn a parameter syntax. Serious visual work belongs in Midjourney or Flux.