Generating Images with ChatGPT (DALL-E 3)
ChatGPT with DALL-E 3 is the friendliest place in the world to start generating AI images. You don't pick models, you don't tweak sliders, you don't memorize commands — you just chat, and pictures come out. By the end of this lesson you'll know every trick that matters for getting beautiful images out of ChatGPT, including the conversational refinement workflow that no other tool does as well.
What You'll Learn
- How to generate your first image in ChatGPT (free and Plus tiers)
- The conversational refinement workflow — making changes by asking
- How to control aspect ratio, style, and consistency
- How to put readable text inside images correctly
Generating Your First Image in ChatGPT
Open chat.openai.com and sign in. Free users have access to limited DALL-E 3 generations per day; Plus users have effectively unlimited access. Just type a prompt:
Generate an image: a calm Mediterranean village with whitewashed
houses on a cliffside, blue domed church, deep blue sea below,
golden hour light, oil painting style, 16:9 aspect ratio.
ChatGPT will produce one image. (Older versions used to give you two — DALL-E 3 inside ChatGPT now defaults to one.) That's it. You're already generating AI art.
The trigger word "image" or "picture" or "photo" tells ChatGPT to invoke the image tool. Sometimes you don't even need it — "Show me a watercolor of..." also works.
The Killer Feature: Conversational Refinement
This is what separates ChatGPT from every dedicated image tool. After you generate, you can just talk to make changes:
You: Make the church domes more golden and add a few people walking
on the path below.
You: That's too cluttered, remove the people, but keep the golden domes.
You: Change the aspect ratio to 9:16 vertical, like an Instagram story.
You don't need to rewrite the entire prompt. ChatGPT remembers the previous image and adjusts. This is enormous for fast iteration.
A few things to know:
- It often regenerates from scratch rather than truly "editing." So small tweaks may also change other parts of the image.
- Asking it for the prompt ("show me the prompt you used") sometimes works and sometimes doesn't — OpenAI changes this often.
- In recent versions you can select an area of the image and ask ChatGPT to change only that region (inpainting). On the web app, click the image, then use the highlighter tool.
Controlling Aspect Ratio
Just say what you want in plain English:
Generate an image of a sushi platter, top-down flat lay, on dark
slate, dramatic lighting. Aspect ratio: 1:1 square for Instagram.
DALL-E 3 supports approximately:
- 1:1 square — Instagram posts, profile pictures
- 16:9 widescreen — YouTube thumbnails, slide decks, banners
- 9:16 vertical — Instagram stories, TikTok, phone wallpapers
- 3:2 and 2:3 — closer to traditional photo dimensions
Putting Text Inside Images
DALL-E 3 is one of the best at rendering readable text — much better than older models. Just include the exact text in quotes:
A vintage travel poster that says "VISIT MARS — RED DUNES AND
BLUE SUNSETS" at the top in bold retro lettering, art deco style,
warm orange and turquoise palette, 24x36 poster proportions.
Tips for text:
- Quote the exact text so the model knows it's text, not a description.
- Keep it short — 3-6 words is the sweet spot. Long sentences get garbled.
- Pick a clear style ("bold retro lettering," "elegant serif," "graffiti spray paint").
- Regenerate if letters are wrong. It's the cheapest fix.
Style Consistency Across Multiple Images
Need a series of images in the same style — like illustrations for a school project? ChatGPT handles this beautifully because it remembers context.
Example workflow:
You: Generate an image of a forest fox in flat-vector children's
book illustration style, soft pastel colors, simple shapes.
You: Now generate the same fox sleeping in a den, same exact style.
You: Now the fox looking up at the moon, same style.
The model carries forward the visual vocabulary. For even tighter consistency, refer to the previous image: "in the same style as the previous fox image."
Three Beginner Exercises (Do These Today)
Exercise 1: The Coffee Shop Mood Generate an image of a cozy coffee shop in autumn, watercolor style, warm lighting. Then ask ChatGPT to make three variations — morning rush, late-night studying, snowy winter — all in the same watercolor style. Save all four images.
Exercise 2: The Logo Concept Imagine you're naming a study app called "Ember." Ask ChatGPT to generate a minimalist logo featuring a small flame icon and the word "Ember" in clean modern sans-serif lettering, on a transparent or white background. Generate three variations.
Exercise 3: The LinkedIn Banner Ask ChatGPT for a 16:9 LinkedIn banner that represents your career interest. Example:
Generate a 16:9 LinkedIn banner for a computer science student
interested in AI: abstract neural network pattern in dark blue
and electric teal, minimalist tech aesthetic, leave the right
third of the image clear for text overlay.
Download it. Add a quote or your name in Canva. You now have a banner that's better than the default blue gradient — and it cost you nothing.
Common ChatGPT Image Pitfalls
- It refuses certain requests. Real people, copyrighted characters, certain brands — ChatGPT will decline. Describe a similar look without naming the person/brand.
- Free tier limits hit fast. If you get blocked, switch to Microsoft Designer (also DALL-E 3, free) for the day.
- It changes too much when refining. If a tweak completely altered the image, paste the original prompt again with your new requirement added.
- Hands and crowds still misbehave. Crop or generate close-ups when fingers don't matter.
Key Takeaways
- ChatGPT with DALL-E 3 is the easiest entry point — type a prompt, get an image, refine by chatting
- Aspect ratio, text in images, and style consistency all work via plain-English requests
- Use conversational refinement instead of rewriting prompts from scratch — that's ChatGPT's superpower
- Free tier is enough for the assignments in this course; switch to Microsoft Designer when you hit limits