How do you make your first image with DALL·E 2?

DALL·E 2 is OpenAI's image generator: you describe a picture in plain words, and it creates one. It works by a method called diffusion, starting from random dots and gradually shaping them into an image that matches your description. You don't need any drawing skill, only a clear idea of what you want to see and the patience to try a few versions.

The same OpenAI account you use for ChatGPT works for DALL·E. Go to labs.openai.com and sign in. You type a description in the text box and press Generate; your earlier creations are kept so you can come back to them.

How credits work

Each time you press Generate, DALL·E uses one credit and returns four images. An edit or a request for variations of an image also costs one credit and returns three. Keep an eye on how many you have left. If you run out, OpenAI sells extra credits in packs of 115 for $15.

Because every attempt costs a credit, it pays to think about your description for a minute before you click, rather than firing off ten vague tries.

A worked first image

Suppose you want an image for a blog post about spring gardening. A first attempt many people type is:

a garden

That will produce four pleasant but generic pictures, because you have left almost every choice to the model. A stronger first prompt names four things:

A wooden wheelbarrow full of tulip bulbs beside a freshly dug vegetable bed, watercolor illustration, soft morning light, wide shot

• Subject: a wooden wheelbarrow full of tulip bulbs beside a vegetable bed. Concrete objects in a concrete place.
• Style: watercolor illustration. Other options include "35mm photograph," "pencil sketch," "3D render" or "children's book illustration."
• Lighting: soft morning light. Lighting changes the mood more than people expect.
• Composition: wide shot. This tells DALL·E how close the "camera" is.

Now refine, changing one thing at a time so you can see what each change does.

If one of the four images is almost right, open it and choose Variations instead of rewriting the prompt. DALL·E creates new images inspired by that one, keeping its general look while changing details. For small fixes, such as a misshapen glove, the Edit tool lets you erase an area and describe what should fill it.

Keep in mind the content rules. DALL·E rejects requests for violent, adult or political content and attempts to depict public figures, and it won't edit uploaded photos of realistic faces. OpenAI also lets you use the images you create commercially, including selling and printing them.

What makes an image prompt work?

A good image prompt reads like a short brief to an illustrator. Name the main subject first and make it specific: "an orange tabby cat asleep on a stack of library books" gives DALL·E much more to work with than "a cat." Then set the medium or style, the lighting and the framing. Those four parts cover most of what a picture is.

Second, describe what should be visible, not what should be absent. Words in the prompt tend to show up in the picture, so "a beach with no people" can work against you; "an empty beach at dawn" says the same thing positively. Likewise, keep the scene manageable. DALL·E handles one or two main subjects well, but a long list of objects, each with its own color and position, often gets jumbled, with details attached to the wrong item.

Third, expect weak spots. Hands, written words inside the image and exact counts ("seven birds") are often unreliable, so plan to regenerate, use Edit on the problem area, or leave text off and add it later in a design tool. Finally, keep a note of prompts that worked. Reusing a phrase like "watercolor illustration, soft morning light" across a series gives your images a consistent look, which matters more for a blog or a class handout than any single striking picture.