How to Create Marketing Images With AI: Prompts and Edits
A repeatable three-step routine for turning a plain idea into a usable marketing image, then keeping every image after it consistent with the first.
Using Images as Prompts: Nine Things to Do With a Photo
Most people only send images out of an AI. Sending one in is where the genuinely useful answers are, because it is a question you could not have typed.
ChatUp GuidesProductivity
Image generation gets the attention. Image input is the capability that quietly changes how you use an assistant, because it lets you ask questions you have no words for — what is this component, why does this look wrong, what does this label say.
Here are nine uses that come up constantly, with what to ask.
1. What am I actually looking at?
Prompt to try
What is this part, what is it for, and what would I search to buy a replacement? If you're not sure, say what it's most likely to be and what detail in the photo would confirm or rule it out.
The uncertainty clause matters. Identification is exactly where a confident wrong answer sends you to a store for the wrong thing.
2. Screenshot to code
Prompt to try
Here's a screenshot of an interface. Build this as semantic HTML with Tailwind classes. Match the spacing and hierarchy rather than guessing exact values, use real elements — buttons as buttons, headings as headings — and make it responsive. Tell me what you had to guess at.
Good for scaffolding, not for a pixel match. Treat the output as a starting structure and expect to do the fine work yourself.
3. Reading documents you photographed
Prompt to try
Transcribe the text in this photo exactly as it appears, preserving structure. Mark anything you can't read clearly with [unclear] rather than guessing at it.
The [unclear] instruction is important. Without it, ambiguous characters get silently resolved into something plausible — which is fine for a shopping list and not fine for a serial number.
4. Charts, diagrams, and dashboards
Prompt to try
Read this chart. What's the trend, what's the most misleading thing about how it's presented, and what would I need to see to know whether the conclusion in the title is justified? Don't read exact values off the axes unless they're labeled.
5. Captions and hashtags from the actual image
Prompt to try
Here's the photo I'm posting. Write five caption options that reference what's actually in the frame rather than generic sentiment. Range from dry to warm. Then suggest hashtags that describe the content, not aspirational ones with millions of posts.
6. Naming and describing a product
Prompt to try
Here's a photo of my product. Describe what a buyer sees — materials, finish, scale, what it looks like it would feel like. Then give me eight name options and tell me what each one signals. Don't claim any feature you can't see in the photo.
7. What can I cook with this?
Prompt to try
Here's what's in my fridge. Give me three things I could make tonight using mostly these, in order of how little else I'd need to buy. Assume basic store-cupboard items. Tell me if anything in the photo looks like it should be used first.
8. Rooms, spaces, and things that look wrong
Prompt to try
Here's a photo of my living room. What's not working, in order of impact? Focus on things I could change without buying furniture — layout, lighting, what's on the walls, what's creating visual noise. Be specific about what to move where.
Diagnosis from a photograph is genuinely good, and it is the step people skip before redesigning something.
9. Solving the practical problem in front of you
Prompt to try
This is the back of my router and the cable I've been given. Which port does it go in, what should the lights look like when it's working, and what should I check first if they don't?
Prompt to try
This is the error screen on my dishwasher and the label with the model number. What does this code mean for this model, what are the common causes, and which ones can I safely check myself?
That last constraint — what can I safely check myself — belongs in every appliance, electrical, or vehicle question.
Getting better answers from images
- Good light, and get close. Most bad answers are bad photographs.
- Include the context. The label, the surrounding area, the scale reference.
- Ask one question about one thing. Multiple subjects in a frame produce muddled answers.
- Say what you already know. “This is a 2019 model and the fault started after a power cut” narrows enormously.
- Ask for the confidence level on anything you will act on.
Where it is unreliable
Dense tables photographed at an angle, small text, handwriting in unusual hands, exact color matching, precise measurements, and reading values off unlabelled axes. Anything that needs to be exactly right off an image should be verified against the source.
Where this fits in ChatUp
Use a currently available model whose catalog capabilities include vision for image questions. File Assistant is the better route for supported multi-page documents rather than photographing each page. Image Gen covers generation and can accept reference images when the selected image model supports them.
Frequently asked questions
Can AI read my handwriting?
Neat handwriting, usually. Difficult handwriting, partially. Always ask it to mark unclear passages rather than guess.
Is it accurate at identifying objects?
Good on common objects, less reliable on specific models, parts, and variants. Ask for confidence and for the detail that would confirm it.
Can I use it for medical images?
No. Photographs of symptoms, scans, and test results need a clinician. It can help you prepare questions for one.
Does it work on screenshots of text?
Yes, and well — screenshots are cleaner than photographs. For long documents, upload the file instead of screenshotting it.
The question you could not type
The value here is not that the model can see. It is that a photograph carries information you would need three paragraphs to describe badly. When you cannot name the thing you are asking about, show it.
Try it in ChatUp
Turn this guide into a workflow.
Run the prompts above against the model that suits the task, keep the useful context across chats, and pick it back up on any device.
Try for Free