What could I make
with a little direction?
This workbook helps you turn an idea into clear instructions for ChatGPT. You’ll answer a few questions, copy a prompt built from your answers, and use it to create or improve an image. Then you’ll learn how to judge what comes back.
You don’t need a finished brand or design jargon. Describe your idea in everyday words—type it, or speak using your AI app’s or keyboard’s microphone if available. You can change your answers as your idea becomes clearer.
A clear brief
Know what you are making, for whom, and what must stay true.
A prompt I understand
See how your answers become instructions, then practise writing your own.
A reviewed result
Create in ChatGPT, make a focused revision, and save the image there.
Author portrait to be added.
Ray Sgambati.
Known as Bruce.
Creator of Bruce’s Bespoke Brands.
I’m a designer, entrepreneur, and creator. Bruce is the creative identity I’ve built alongside my brands and community—the entrepreneurial, imaginative, design-driven side of who I am.
Bruce’s Bespoke Brands is my creative studio, built around branding, design and AI-assisted creative direction. It brings together branding, design, technology, and customer experience.
“This guide isn’t just about how Bruce’s Bespoke Brands looks—it’s about how I think.”
My approach is driven by intention, craftsmanship, curiosity, and creating experiences that feel timeless rather than trendy.
In this workshop, I share that way of thinking so you can make more deliberate decisions for your own work.
What do I want this image to do?
Start with the reason for the image. Maybe you want to introduce a product, explain an idea, or invite someone to try something. A short answer is enough. Together, your answers become your brief—a simple description of what you want to make.
Would I rather talk than type?
My answers, brought together
This summary uses what you type on this page. It helps you check that you have explained your idea before asking ChatGPT to help.
What happens next? Read the prompt, copy it into ChatGPT, and send it. ChatGPT will help you check your plan before making the picture.
A “brief” is simply your plan: what you are making, who it is for, and what you want it to do. The next page helps you add the details.
What does ChatGPT need to know about my idea?
Now that I know what my picture is for, what details should it include?
Attach references here to organize them.
Pick a purpose for each image. These files stay in your saved workshop. You must also attach them in ChatGPT; copying text does not transfer images.
How do I turn my answers into a first draft?
My idea and details are ready. Now I can use them to ask ChatGPT for one picture.
Open ChatGPT using the button below. Sign in, or create an account if needed. Start a new chat.
Use the attachment control beside the message box to add your original reference images. Wait until they appear.
Preview your image prompt here. Copy it, click ChatGPT’s message box, paste, and send.
If ChatGPT asks a necessary question, answer it. If image creation is unavailable, follow the account’s displayed guidance or ask for a layout plan first.
Exact logos, labels, and faces can still change. If exact preservation fails, keep the original asset and use AI for the background or layout direction only.
Official ChatGPT image guide ↗What do I like—and what would I change?
I have a first picture from ChatGPT. What do I want to keep, and what needs changing?
Continue in the same ChatGPT conversation.
Make sure the current image is present. Paste your revision prompt. Compare the new image with the previous version.
Specific: “Keep every word and the palette. Make the main headline larger than the workshop name. Check that I notice the headline first.”
Am I ready to share this?
I have tried a change. Now I will check the picture before sharing it.
A tick records your check. The workbook does not inspect the image for you.
What can I take into my next project?
I can keep my answers and use what I learned next time.
Your creative brief
Your task, audience, facts, style choices, and reference instructions.
Your starting prompt for next time
Copy a personalized starting instruction into your next ChatGPT conversation.
Your reviewed image
The image you made and downloaded in ChatGPT. This workshop does not download it for you.
Create a background in Higgsfield.
Nano Banana is not part of ChatGPT. This optional exercise takes place on the separate Higgsfield website. You do not need it for the ChatGPT practice exercises.
Open Higgsfield. Create an account or sign in. You can skip this exercise and stay in ChatGPT.
Open Image → Nano Banana Pro. Choose a reference-based input mode and upload your lighting reference.
Copy the background prompt from this workbook and paste it into Nano Banana’s prompt box on Higgsfield. Check the displayed size and credit cost, then generate once.
Download your background from Higgsfield. This exercise is complete. Optional: attach it in a separate ChatGPT conversation if you want help planning where to place your real asset.
Plain link to Higgsfield. No referral or affiliate link is used.
What changed when the instructions changed?
Both outputs below were generated from the same practice graphic, once per prompt. These are actual image-tool results from this session, not screenshots of the ChatGPT website.
See the starting graphic
A · Broad request
The headline became clearer, but the tool also added an icon, gold accents and corner decorations.
B · Specific request
The headline became larger while the dark background and simpler arrangement remained. The rendering still changed slightly; this is not an exact pixel-preserving edit.
Choose one image
You will ask ChatGPT to revise the same image twice, then compare the results.
Nothing ready? Use the supplied practice graphic.
Images stay in this workbook. You will attach the original separately in both ChatGPT conversations.
Try a broad instruction
Open a fresh ChatGPT conversation and attach the original image.
- Click the main button below to copy prompt A.
- Paste it into ChatGPT with the original image, then send.
- Save the generated image to your device.
Bring back result A
Add the image you saved from ChatGPT. Keep the original unchanged.
If image generation is unavailable, save your progress and return when your account has access. No Higgsfield purchase is needed.
Give the next revision a clear job
Use another fresh ChatGPT conversation. Attach the same original—not result A.
See the complete prompt before copying
Use the same image tool and size where possible. This uses your ChatGPT allowance.
Bring back result B
Save the second result from ChatGPT and add it here.
Compare what actually changed
Check both images against the same standard. A different result is not automatically a better result.
Two results are a learning exercise, not proof that one prompt always wins. Save mixed results too.
My idea. My words. My prompt.
You have practised describing, making and improving an image. Now write the instructions yourself. Clear, ordinary words are enough.
Can someone understand my request?
How can I say what not to do?
“Do not add discount badges or extra words. Keep my logo and the words I supplied unchanged. Do not invent a price, customer review or product claim. If you need a date or another missing fact, ask me first. If you cannot keep my logo accurate, tell me instead of replacing it.”
Use only the instructions that fit your project. Add them to your prompt in your own words; ticking these boxes does not add text for you.
Why do these parts help?
The task gives direction. The audience gives context. The priority helps focus attention. Your real details reduce guessing. Saying what to leave out and what must not change sets clear limits. Asking about missing facts helps avoid made-up details. Your check gives you a way to request a useful correction.
These ticks record your own review; they do not automatically assess the prompt or guarantee the image quality.
Good-looking can still be the wrong idea.
Imagine I wanted a quiet kitchen ad with one sentence. If I only say “make it special,” I have not told the tool that.
A · I left the choices open
The tool chose mountains at sunrise, a brand name, several slogans, icons and writing on the mug. None of those specifics were supplied.
B · I described the important choices
The requested kitchen setting, soft window light and single headline are visible. This fits the described idea more closely.
What am I picturing?
A kitchen, soft light and one short sentence.
What must I supply?
A real product photo if the exact mug matters. Neither example used one, so both mugs are invented.
What must it leave out?
Extra slogans, brand names and claims I have not approved.
Actual generated outputs: one run per prompt, no output retouching. Fictional products and branding for teaching, not client work. Both prompts were generated independently without product references. This illustrates a brief difference, not a controlled proof that longer prompts always win. Generated with the available image tool; no claim of GPT Image 2.5.
Bruce’s Bespoke Brands · Created by Ray Sgambati (aka Bruce) · Edition 4.17