01 · Reality check
What ChatGPT actually does well here
Good at
- Turning a plain-language description into a usable image in seconds
- Iterating conversationally: describe a change instead of rewriting the prompt
- Any aspect ratio, so you can generate for a specific placement
- Concepts, illustrations, mockups and social visuals
- Rendering readable text inside an image, which older tools handled badly
Not the right tool for
- Exact reproduction of a specific real person, product or brand asset
- Pixel-precise control that a designer needs; you steer with words, not tools
- Guaranteed consistency of a character or style across many separate images
- Anything requiring true photographic accuracy for documentation or evidence
- Replacing a designer when brand precision genuinely matters
02 · The method
Step by step
- 1
Describe the subject, then the treatment
Lead with what the image is of, then how it should look: style, mood, lighting, colour. 'A ceramic coffee cup on a wooden desk, soft morning light, shallow depth of field' beats 'a nice coffee photo'.
- 2
Set the aspect ratio before you refine
Use the aspect-ratio picker to match where the image will actually be used. Changing ratio later regenerates the image, so pick it early to avoid losing a version you liked.
- 3
Refine by describing the difference
Do not rewrite the whole prompt. Say 'make the lighting warmer and remove the plant on the left'. Conversational iteration is the main advantage over one-shot generators.
- 4
Turn on thinking for harder briefs
On a paid plan, Images with thinking lets the model plan and search before generating. It helps most when the image needs accuracy, several elements arranged deliberately, or reference to something real.
- 5
Be specific about text in the image
If the image needs words, state the exact text in quotes and where it sits. This model handles in-image text far better than older ones, but only if you tell it precisely what to write.
- 6
Check the details before you use it
Look at hands, faces, small text and repeated patterns. These are still where generated images break, and they are easy to miss at a glance.
03 · Use this now
Copy-paste prompt structure for a better image
Create an image of [subject, described concretely]. Style: [photorealistic / flat illustration / 3D render / watercolour]. Composition: [close-up / wide shot / overhead], with [subject] positioned [where]. Lighting and mood: [soft morning light / dramatic side light / bright and clean]. Colour palette: [describe or name colours]. Aspect ratio: [16:9 / 1:1 / 9:16]. If the image needs text, include exactly this text: "[your text]", placed [where]. Do not include [things you do not want]. After you generate it, ask me what to adjust rather than regenerating from scratch.
04 · Avoid these
Common mistakes
- Writing a vague prompt and blaming the model. Specificity about style, lighting and composition does most of the work.
- Rewriting the whole prompt to make one change, instead of describing just the difference.
- Expecting a named real person or an exact brand logo, which it will not reliably reproduce.
- Trusting third-party claims about daily image limits. OpenAI publishes no numbers; the app tells you when you hit yours.
- Publishing without checking hands, faces and small text, which is where generated images give themselves away.
05 · Questions
Frequently asked questions
Can ChatGPT create images for free?
Yes. OpenAI states that ChatGPT Images 2.0 is available on all tiers, including Free, though the free plan is described as limited and slower for image creation. Paid plans add faster generation and Images with thinking, where the model plans and can search the web before generating.
How many images can I generate per day?
OpenAI does not publish a numeric limit for any tier. The product tells you when you have hit the current limit and when it resets, and image creation has its own limit separate from the main chat limit. Treat any specific number you see quoted online as unverified, because it is not in OpenAI's documentation.
Does ChatGPT still use DALL-E?
Not by default. Image generation now runs on ChatGPT Images 2.0, released in April 2026. DALL-E still exists as a legacy path through the DALL-E GPT for Free, Plus, Business and Pro users, but it is no longer the engine behind normal image requests.
What is Images with thinking?
It is a mode where the model plans and refines before generating, and can search the web as part of that. OpenAI lists it as available on Plus, Pro and Business, and not on Free. It helps most when the brief is complex, needs several elements arranged deliberately, or references something real.
Can ChatGPT make an image of a specific person?
It will not reliably reproduce a specific real individual, and there are safety restrictions around generating images of real people. For likeness-accurate work you need a different approach entirely. ChatGPT is best for original concepts, illustrations and generic subjects rather than specific identifiable people.