ChatGPT Image AI

AI Image Prompts: 6 That Fix What Usually Goes Wrong

With the official DALL-E GPT retired on 30 August, ChatGPT Images is now the default for a lot of people who never chose it. These six are built around the failures that waste the most attempts: text that comes out garbled, styles that drift between images, compositions that ignore what you asked for, and the specific problem of getting the same character twice. They work in ChatGPT Images, Midjourney and Gemini with minor wording changes.

⌨️ 6 prompts 🕐 Updated Aug 30, 2026

💡 How to use these prompts: Replace everything in [BRACKETS] with your specific details before sending. Click Copy to copy any prompt to your clipboard instantly.

Get text in the image to come out readable

Text rendering fails when the model treats it as decoration. Making it the priority is what fixes it.

Create an image containing the exact text: [YOUR TEXT]

Requirements:
- The text must be spelled exactly as written above, no substitutions
- Place it on a plain area with high contrast, not over busy detail
- Use one typeface for the whole thing
- Keep it large enough to read at thumbnail size

Everything else in the image is secondary to the text being correct.

Keep a style consistent across a set

Style drift happens because each generation reinterprets vague words. A fixed specification you paste every time removes the ambiguity.

I need several images in a matching style. Here is the style specification I will reuse:

Medium: [e.g. flat vector illustration]
Palette: [3 to 4 named colours]
Line treatment: [e.g. no outlines, uniform weight]
Lighting: [e.g. flat, no shadows]
Background: [e.g. single solid colour]

Subject for this image: [SUBJECT]

Apply the specification exactly. Do not add stylistic flourishes that are not listed.

Describe composition, not just contents

Listing objects gives you objects arranged arbitrarily. Describing the frame gives you a usable image.

Create an image with this composition:

Foreground, lower third: [WHAT]
Middle, centre: [WHAT]
Background: [WHAT]
Camera angle: [e.g. eye level, slight low angle]
Focal length feel: [e.g. wide, or compressed telephoto]
Where the eye should land first: [WHAT]

Leave the upper right quadrant relatively empty for text overlay.

Get the same character twice

Character consistency is the most requested thing image models do badly. A fixed block plus a do-not-alter instruction gets closer than describing afresh each time.

I need this character to appear in multiple images. Here is the reference description I will reuse verbatim:

[AGE, BUILD, HAIR: colour, length, style, FACE: three specific features, CLOTHING: exact items and colours, DISTINGUISHING DETAIL: one memorable specific thing]

For this image, place that exact character: [SCENE]

Do not alter any listed feature. If something is not specified, keep it neutral rather than inventing.

Fix one thing without regenerating everything

Editing prompts usually regenerate the whole image. Naming what must not change is what preserves the parts you liked.

Keep this image exactly as it is, and change only: [THE ONE THING]

Do not adjust: composition, lighting, colour grading, style, or any element other than the one named.

If the change cannot be made without affecting something else, tell me what would be affected before doing it.

Write a prompt for a different tool

Different tools want different phrasing. This gets a translated prompt plus enough explanation to edit it.

I want this image: [DESCRIBE WHAT YOU WANT]

Write me a prompt optimised for [TOOL NAME], following the conventions of that tool for structure, parameters and phrasing.

Then explain in one line per item which parts of the prompt do what, so I can adjust it myself rather than starting over.