Create Images With ChatGPT

Direct AI image generation to build a 3-piece visual campaign.

Introduction

30 Second Summary

You have probably asked AI to create an image and gotten something that looks fine but feels completely generic. That is what happens when AI makes every creative decision for you.

In this project, you will take control of AI image generation by building a 3-piece visual campaign with ChatGPT. You will go from a vague prompt that hands all creative decisions to AI, to intentionally directed images that match exactly what you imagined.

What You'll Build

Imagine showing someone three related images you created that look like they belong to a professional campaign, right next to the generic first attempt you started with.

By the end of this project, you'll have:

A clear eye for the visual decisions AI makes when you give it no direction, and the skill to take those decisions back by adding intentional direction across subject, environment, mood, lighting, and more.
A proven workflow for diagnosing what went wrong in an AI-generated image and making targeted revisions through the chat or the image editor, without starting over.
A 3-piece visual campaign with a consistent look and feel across all three images, built for a purpose you choose.
Secret Mission: Adapt a campaign image for a completely different format and add intentional, readable text.

Are there any prerequisites?

This project is part of the ChatGPT Power User curriculum. You need a ChatGPT account (Free works, though ChatGPT Plus at $20/month is recommended for higher image generation limits) and should have completed Chapter 1 before starting.

Before We Start

Before you start generating images, take a moment to lock in what you're building and why it matters to you. This step is where you commit to a direction for your visual campaign before any hands-on work begins.

Direct the Image With Intention

Direct Your First Image

When you ask ChatGPT for an image without specifying details, it fills in every gap on its own: the lighting, the angle, the style, the mood, the composition. The result looks fine, but it reflects AI's defaults, not your creative intent.

In this step, you will take control of those decisions. You will choose your campaign concept, learn the key dimensions of visual direction, and create your first campaign image with intention.

In this step, get ready to:

Choose what your three campaign images will be.
Direct your first campaign image with intentional visual choices.

Choose your campaign concept

Your campaign is a set of three related images that share a consistent look and feel. Before you start creating, decide what those three images will be.

Career: 3 personal brand visuals (e.g., a headshot background, a portfolio banner, and a social media post).
Creator: 3 content campaign images (e.g., a YouTube thumbnail, an Instagram post, and a story graphic).
Business: 3 product or service visuals (e.g., a hero image, a feature highlight, and a customer scenario).
Student: 3 images for a project or presentation (e.g., a cover image, a concept illustration, and a diagram).
Personal: 3 images for an event, hobby, trip, room, or creative idea.

Pick a campaign type and decide what your three specific images will be.
Write down a one-line description of each image so you have a clear direction before you start generating.

Direct your first campaign image

Visual direction means telling ChatGPT what you actually want across a few key dimensions. Think of these as a mental checklist, not a formula to memorize.

Subject: What specifically is in the scene?
Environment: Where is it? Interior, exterior, city, countryside?
Composition: What is prominent? What is in the background?
Mood: What feeling should the viewer get?
Lighting: What kind of light? Natural, warm, dramatic, flat?
Perspective: Eye-level, overhead, close-up, wide shot?
Important details: Any specific elements that matter?
Intended use: What is this image for?

Do I need to fill in all eight?

Not at all. Some images only need three or four of these dimensions to land well. The checklist is there so you do not accidentally leave out a dimension that matters to your vision.

A good image prompt does not need to be long. In most cases, 1-3 clear sentences are enough.

In ChatGPT, select New chat to start a fresh conversation.
Tell ChatGPT your campaign concept, then write a directed prompt for your first image using the dimensions above.

Why a new conversation?

ChatGPT uses the images and text in your conversation to maintain consistency. Starting a dedicated conversation gives your campaign its own clean context.

✔️ I'll give it a go!

Use the checklist above to write your own directed prompt. Cover at least subject, environment, mood, lighting, and perspective. Type it into the message box when you are ready.

ⓧ Maybe a little help...

Here is an example prompt that covers all eight dimensions. Adapt it for your own campaign concept:

Create an image of a small Japanese-style pour-over coffee bar at golden hour. Warm natural light streaming through a floor-to-ceiling window on the left. Minimal wooden interior with a single barista mid-pour. Shot from a low angle at the counter, looking up slightly. Calm, focused mood. This image is for a specialty coffee brand's Instagram post.

Before you send, picture what you want the result to look like. Which dimension do you think will shape the image the most?

Send your prompt and wait for ChatGPT to generate the image.
Check whether the mood, lighting, composition, and details match what you asked for.

Image not appearing?

If you are on the Free plan and see a wait time, ChatGPT has temporarily rate-limited image generation. Wait for the displayed time to pass, then try again.

Make sure you are in a standard conversation, not inside a custom GPT. Image generation works in regular ChatGPT conversations.

Why isn't my image generating in ChatGPT?

💡 Your first image sets the tone

Spend extra time getting image 1 right. Its color palette, lighting style, and mood become the reference point for everything that follows. Fixing consistency later is harder than nailing it here.

Your first campaign image is ready. Next up, you will learn to diagnose what is not quite right and fix it without starting over.

Diagnose and Fix Without Starting Over

Your directed image from Step 2 looks dramatically different from the vague one. You took control of the creative decisions and the result changed.

That image is still not perfect though. In this step, you will learn to diagnose what went wrong and fix only that part, without losing everything that already works.

In this step, get ready to:

Diagnose a flaw in your directed image and make a targeted chat-based revision.
Use the image editor's selection tool to fix a specific area of your image.
Learn the core image workflow: Describe, Generate, Observe, Diagnose, Change, Repeat.

Diagnose a flaw and revise through the chat

Before you ask ChatGPT to change anything, you need to know exactly what is wrong. Vague feedback like "make it better" gives AI the same unchecked freedom as a vague prompt.

Look at your directed image from Step 2 carefully.
Find one specific thing that does not match your intention. The lighting might be too warm. A person might be facing the wrong direction. There might be an object you did not ask for.

Before you write your revision, answer these three questions:

What specifically is wrong?
What should stay exactly as it is?
What one thing needs to change?

Type a targeted revision into the chat. Be explicit about what to keep and what to change. Your prompt might look something like this:

Change only the lighting to cooler, blue-toned afternoon light. Keep the same composition, camera angle, barista position, and interior exactly as they are.

Why be explicit about what to keep?

When you ask for a revision, ChatGPT generates a fresh image based on your instructions. Without clear direction about what should stay the same, it may change elements you were happy with.

Naming what to keep protects the parts that already work. A prompt like "Change only X. Keep everything else exactly the same" is often the clearest way to guide a precise edit.

Before you send your revision, predict: do you think ChatGPT will keep everything you asked it to keep, or will some details drift?
Send your revision and wait for the new image.

Compare the revised image with the version from Step 2. The element you asked to change should look different. The elements you asked to keep should look mostly the same.

Did ChatGPT change more than you asked?

This is common. Try sending a follow-up revision that is even more specific about what should stay the same. Name the exact objects, positions, and colors that should not move.

If a single element keeps drifting, try the image editor's selection tool in the next substep to target just that area.

ChatGPT keeps changing parts of my image I want to keep. How do I make more precise revisions?

Use the image editor for a region-specific edit

Chat-based revisions affect the whole image. When you need to change something in one specific area, the image editor lets you highlight exactly where the change should happen.

Click or tap on your most recent image to open the editor.
Select the Select tool to activate the selection brush.
Use the brush to highlight a specific area you want to change. For example, an unwanted object in the background or the color of a wall.
In the text field, describe your edit. For example:

Remove the plant in the background corner and replace it with an empty wooden shelf.

Send the edit and wait for the result.

What if the edit changes more than the highlighted area?

This is expected. Highlights are not always precise, and edits may extend beyond the area you selected. If the edit drifts, try again with a smaller selection and a clearer description of what the area should look like.

💡 Using the editor on mobile

On the ChatGPT mobile app, tap your image to open the editor. Tap Select to activate the selection tool. Use the slider to adjust the brush size. After highlighting, tap Next to type and send your edit description.

Compare the result with the previous version. The highlighted area should reflect your change. The rest of the image should remain intact.

Edit not showing up in the right area?

Make sure you highlighted the correct area before describing the edit. If your selection was too small or missed the target, use Undo and try again with a larger brush.

Keep your description focused on the specific area. A description that implies a full scene change may cause ChatGPT to regenerate beyond the selection.

The image editor is not applying my edit to the right area. How do I use the selection tool more precisely?

Remember the core image workflow

You have now used both tools for revising images: the chat for whole-image adjustments and the editor for region-specific fixes. These two tools form the backbone of a repeatable workflow.

The Describe-Generate-Observe-Diagnose-Change-Repeat cycle

This is the process you will use from now on. You describe what you want. ChatGPT generates it. You observe the result. You diagnose what is off. You make one targeted change. Then you repeat.

You do not need to write better prompts. You need to get better at seeing what is wrong and telling ChatGPT exactly what to fix.

Look at your image now compared to the original directed version from Step 2. You should see that your targeted changes were applied while the rest of the image stayed mostly intact.

You can now diagnose and fix any AI image without starting over. Next up, you will build a full visual campaign using everything you have learned.

Build Your Visual Campaign

You can direct a single image with intention, and you can diagnose what went wrong and fix it without starting over. Individual images are practice.

A campaign is the real test. Can you create multiple images that look like they belong together and serve a shared purpose? In this step, you will build a 3-piece visual campaign where intentional direction becomes a transferable skill.

In this step, get ready to:

Choose what your three campaign images will be and generate the first one with full intentional direction.
Generate two more images that maintain visual consistency with the first.
Evaluate all three images as a set and fix anything that breaks the visual identity.

Choose your campaign and create your first image

You already chose a campaign type back in Step 0. Now it is time to decide what your three specific images will be. Here are some ideas depending on the type you picked:

Career: 3 personal brand visuals (e.g., a headshot background, a portfolio banner, and a social media post).
Creator: 3 content campaign images (e.g., a YouTube thumbnail, an Instagram post, and a story graphic).
Business: 3 product or service visuals (e.g., a hero image, a feature highlight, and a customer scenario).
Student: 3 images for a project or presentation (e.g., a cover image, a concept illustration, and a diagram).
Personal: 3 images for an event, hobby, trip, room, or creative idea.

Once you know what your three images will be, it is time to generate the first one.

In ChatGPT, start a new conversation.
Tell ChatGPT your campaign concept and ask it to generate the first image.
Apply everything you have learned: subject, environment, composition, mood, lighting, perspective, important details, and intended use.

Why a new conversation?

ChatGPT uses the images and text in your conversation to maintain consistency. Starting fresh gives your campaign its own clean context so previous coffee shop images do not influence the style.

ChatGPT generates your first campaign image. Study it closely.

Check whether the mood, lighting, composition, and details match what you asked for.
If something is off, diagnose what specifically needs to change and make a targeted revision in the chat or with the image editor.
Be explicit about what to keep and what to change, just like you practiced in the previous step.

Your first image sets the tone

Spend extra time getting image 1 right. Its color palette, lighting style, and mood become the reference point for everything that follows. Fixing consistency later is harder than nailing it here.

Generate images 2 and 3 with visual consistency

Creating one good image is one challenge. Creating three that look like they belong together is a different one entirely. The key is to reference your previous images each time you ask for a new one.

In the same conversation, ask ChatGPT to generate your second campaign image. Reference the first image to maintain consistency. Here is an example prompt:

For the second image in this campaign, keep the same color palette, lighting style, and overall mood as the first image. The subject this time is [[SUBJECT_DESCRIPTION="describe your second image's subject"]].

What does this prompt do?

By explicitly naming the visual elements to carry over (color palette, lighting style, mood), you tell ChatGPT which parts of the first image define your campaign's look. Without this reference, each image would drift into its own random style.

Once image 2 appears, compare it against image 1. Look at the color palette, lighting, and mood. If something drifted, make a targeted revision.

For example, if the lighting shifted, you could tell ChatGPT:

The lighting in this one is too harsh compared to the first. Soften it to match the warm natural light from image 1.

Use the chat or the image editor to fix anything that breaks the visual consistency. You can use the Select tool in the editor to highlight a specific region if only part of the image needs adjusting.

Once you are happy with image 2, repeat the same process for image 3.

Ask ChatGPT to generate the third image, again referencing previous images to maintain the same visual identity.
After each generation, diagnose and revise. If the style drifts, explicitly call out what needs to match from your earlier images.

Images looking inconsistent?

If your images keep drifting in style, try being more specific about what to carry over. Instead of just saying "same style," name the exact elements: "same warm golden-hour lighting, same muted earth-tone palette, same low camera angle."

If the drift is too large to fix with a revision, regenerate the image. Reference the first image again with more precise consistency instructions.

Help me get consistent visuals across my campaign images.

Evaluate your campaign as a set

You have three images. Now step back and look at them as a set, not as individual pieces.

Before you evaluate, predict: if someone who had never seen your campaign scrolled past all three images, would they be able to tell these belong together?

Line up all three images in your conversation and ask yourself these questions:

Do they look like they belong together?
Could someone tell these are part of the same campaign?
What visual elements tie them together (color, lighting, style, composition)?

If something feels off, make one targeted revision to bring it in line. Do not start over.

What makes a campaign feel cohesive?

Visual consistency does not mean identical images. It means shared decisions. A consistent color palette, similar lighting, the same mood, and a matching level of detail are what make three different subjects feel like they belong to one campaign.

You should now have three related images that share a consistent visual identity. You can point to specific visual elements that tie them together.

Your campaign is complete. Next up, you will compare your very first vague image against your strongest campaign piece to see how far your visual direction skills have come.

Secret mission

Adapt for a New Format and Add Text

Your campaign images work as a set. But what happens when you need one of them in a completely different shape? Reformat a campaign image for a new platform, add intentional text to it, and see if it still holds together.

Clean Up Your Resources

This project runs entirely inside ChatGPT, so there are no ongoing costs or paid cloud resources to worry about. Decide whether to keep your images, download any you missed, or delete everything.

Resources you used:

ChatGPT conversations containing your generated images (vague coffee shop image, directed images, campaign images, adapted image).
Downloaded campaign images saved to your device.

Keep everything running

No action needed. Choose this if you plan to keep iterating on your images or want them available for future projects.

Your generated images are automatically saved under Images in your ChatGPT account. You can browse, revisit, and reuse them anytime.
Your downloaded campaign images remain on your device wherever you saved them.
To download any image you missed, select the image in ChatGPT, then select Save.

Pause - I'll come back to this later

This project has no running processes or paid resources. There is nothing to shut down or pause.

Your images remain in your ChatGPT account at no cost. They will be there when you come back.
If you want to free up device storage, move your downloaded campaign images to a cloud drive or external storage before deleting them locally.

Delete - I don't want to use this again

Remove your project images from ChatGPT and your device entirely.

Delete your ChatGPT conversations that contain the generated images. On the web, hover the conversation title, select More options, then select Delete.
On iOS or Android, long-press the conversation title and select Delete.
Delete the downloaded campaign images from your device. Find them in your Downloads folder (or wherever you saved them) and move them to the Trash.

Nice Work!

Nice work! You just built a 3-piece visual campaign using ChatGPT Images, starting from a vague prompt and ending with intentionally directed, visually consistent images.

You've learned how to:

Add intentional visual direction across key dimensions like subject, environment, composition, mood, lighting, and perspective to transform AI-generated images from generic defaults into exactly what you imagined.
Diagnose what ChatGPT got wrong in an image and make targeted revisions using both the chat and the image editor without starting over.
Build a 3-piece visual campaign with consistent visual identity across multiple images, then compare your first undirected image to your strongest directed work to see how far your process has come.
Secret Mission: Adapted a campaign image for a completely different format and added intentional, readable text to the image.

You just learned ChatGPT can create more than text. Next, you will give it something even more powerful: your own information.

Ready to quiz yourself?