How to generate AI images in ChatGPT with DALL-E?

Updated 2026-07-15Asked across Reddit, Quora & Google· DALL-E
Short answer

In ChatGPT, type a description of the picture you want, like 'a neon city street in the rain, cinematic,' and press enter. ChatGPT generates it in seconds using GPT Image (which replaced DALL-E in 2026). Refine by asking for changes in the same chat. Free plan allows about 2-3 images per day; Plus ($20/month) allows more.

Why — the first-principles explanation

Generating an image in ChatGPT is a describe-and-refine loop. Step one, you describe; the model turns your words into a first draft. Step two, you critique in plain language; it revises. This loop mirrors how a human art director works, and it's powerful because you never touch sliders or menus, you just talk.

The quality of your description does most of the work. The model can only draw what you specify, so naming the subject, style, composition, lighting, and mood dramatically improves results. "A dog" yields something generic; "a golden retriever puppy sitting in tall grass at golden hour, shallow depth of field, photorealistic" yields something intentional. GPT Image, DALL-E's successor, is also good at putting readable text in images, so you can ask for signs, labels, or short words.

The refine half is where beginners gain the most. Instead of rewriting from scratch, keep the good draft and adjust: "same scene, but make it winter," "move the subject to the left," "warmer colors." Because it's a conversation, ChatGPT remembers context. When you like the result, download it. Free accounts hit a small daily cap; paying lifts it.

An example that makes it click

Picture giving directions to a sketch artist over the phone. First you say, 'draw a lighthouse on a cliff.' They sketch it. Then you say, 'add crashing waves and make it stormy,' and they adjust the same drawing instead of starting over. You keep nudging until it matches the image in your head. Generating AI images in ChatGPT is exactly that phone call, only it takes seconds per round.

How to do it

  1. Sign in at chatgpt.com or open the ChatGPT app.
  2. In a new chat, type a detailed image prompt: subject, style, setting, lighting, and mood.
  3. Press enter and wait a few seconds for the generated image.
  4. Refine with follow-up messages ('make it winter', 'warmer colors', 'add text that says OPEN').
  5. Download the final image; if you hit the free cap, wait for reset or upgrade to Plus.

Key facts

Infographic: How to generate AI images in ChatGPT with DALL-E — short answer and key facts
Visual summary — How to generate AI images in ChatGPT with DALL-E?
DE
Try DALL-E by OpenAI

OpenAI's image generator, built into ChatGPT.

Affiliate link — we may earn a commission at no cost to you.
Visit DALL-E ↗
▶ The 60-second explainer (script)

Here's how to generate AI images in ChatGPT. It's a simple describe-and-refine loop. First, describe what you want, and be specific. Instead of 'a dog,' type 'a golden retriever puppy in tall grass at golden hour, photorealistic.' Press enter, and in a few seconds you get your image. Then refine it by chatting: 'make it winter,' 'move the puppy to the left,' 'warmer colors.' ChatGPT keeps the good parts and just adjusts, like directing an artist over the phone. A neat bonus: the current model, GPT Image, which replaced DALL-E in 2026, is great at putting readable text in pictures, so you can ask for signs and labels. When you're happy, hit download. Free accounts get a couple images a day; ChatGPT Plus at twenty dollars a month unlocks a lot more.

What authoritative sources say

How to Generate AI Images in ChatGPT with DALL-E (How-To Geek)media — You generate images in ChatGPT by typing a prompt and refining the result conversationally. source ↗
OpenAI API Deprecationsofficial — ChatGPT's image generation runs on GPT Image, DALL-E's replacement as of 2026. source ↗

People also ask

How long does each image take?

Usually a few seconds to under a minute, depending on server load and your plan. Paid tiers tend to generate faster during busy periods.

Can I put text in the image?

Yes. GPT Image renders readable words well, so you can request signs, labels, or short phrases and they'll usually come out legible.

Can I make multiple variations?

Ask for alternatives, like 'give me two more versions in different styles.' Each generated image counts against your plan's limit.

What if the image is wrong?

Describe the fix in your next message rather than starting over. ChatGPT keeps context and revises the existing concept.

Related questions