ChatGPT Image Prompts That Actually Work (With Examples)

ChatGPT’s image model reads prompts like a person does. That makes it easy to use, but it also means clear sentences beat keyword lists. Here is how to get the most out of it.

The basic ChatGPT image prompt

Describe the picture in three to six sentences: subject, style, framing, light, mood, size. Start with “Create” so ChatGPT knows you want an image and not a description.

Create a photorealistic photograph of a steaming cup of masala chai on a rainy windowsill. Close-up from a slightly low angle. Soft grey daylight from the window with raindrops on the glass. The mood is cozy and quiet. Warm browns against cool blue light. Vertical 4:5 format. No text or watermarks.

You can build prompts like this in seconds with our AI image prompt generator. Choose “ChatGPT” as the model.

Editing your own photo

This is where ChatGPT shines, and it is what most people search for. Upload a photo, then describe the new scene. The key is one clear instruction about identity, placed first.

Use the uploaded photo as the reference and keep the face and identity exactly the same. Create a cinematic portrait of this person in a black hoodie leaning on a car at night, city lights blurred behind. Moody low-key light with a soft rim light on the face. Vertical 2:3 format.

Tips that help:

• Use a sharp photo with the face clearly visible and no sunglasses.
• Keep the pose simple if the face matters most.
• If the result changes your face, reply “keep my face exactly like the photo, change only the background” and regenerate.

Many prompts in our portrait library and couple library are written for this.

Text inside images

ChatGPT is good at short text. Put the exact words in quotes and say where they go:

Create a minimal poster with the words "CHAI O'CLOCK" in bold rounded letters at the top. Below, a flat illustration of a clay kulhad with steam. Cream background, burnt orange and brown palette.

Keep text short (a few words), and spell out the font style you want: bold, handwritten, serif, rounded.

Keeping a character consistent

To reuse a character across images, describe them once in detail and repeat that description every time, or keep working in the same chat and say “same character as before”.

A cheerful 3D cartoon girl, about 8 years old, with a curly black ponytail, round glasses, a yellow raincoat and red boots.

What to do when the result is not right

1. Ask for a specific change rather than starting over: “make the lighting warmer”, “zoom out to show the full body”.
2. Remove conflicting words. “Bright” and “moody” pull in opposite directions.
3. Regenerate two or three times. Each run is different.
4. Simplify. If a prompt has ten ideas, the model will drop some. Keep the three that matter most.

Ready-to-use ChatGPT prompts

Every prompt in our library was tested in ChatGPT and shows the image it created. Open any prompt and press Open in ChatGPT to run it with one click, or Remix to change it in the generator.