Pick the result you want, copy the prompt, then swap the subject for your own. The exact output can vary, but this gives ChatGPT a much clearer visual direction than “make me an image”.
Want the whole set at a glance?
Download the contact sheet, or keep scrolling for every prompt beside its result.
/xray a retro espresso machine, revealing the boiler, pump, pipes, heating element and water tank through a translucent shell. Photorealistic technical visualisation, dark studio, square format. No text, logos or watermark.
/explodedview a retro espresso machine with its housing, boiler, pump, pipes, portafilter, drip tray, switches and fasteners floating apart in assembly order. Neutral studio background, square format. No text, logos or watermark.
/blueprint a retro espresso machine shown in front, side and top views with dimension lines and construction guides. Crisp white linework on deep-blue grid paper, square format. No logos or watermark.
Slice through the outside so people can see within.
/cutaway a retro espresso machine sliced through one side so the boiler, pump, water circuit and heating element are visible. Photorealistic technical illustration, square format. No text, logos or watermark.
/anatomy of an AI-powered content system with five connected layers labelled SOURCE, THINK, CREATE, REVIEW and PUBLISH. Vintage scientific plate on ivory paper, square format. No other text.
/timeline of a content idea moving through IDEA, RESEARCH, DRAFT, REVIEW, PUBLISH and MEASURE. Clean editorial design, six milestones in order, square format.
/flowchart for lead qualification: NEW LEAD flows to QUALIFIED?; YES flows to BOOK CALL; NO flows to NURTURE. Clean arrows, blue and orange accents, square format.
/isometric a tiny AI automation office where miniature people supervise desks, servers, dashboards and robotic task stations. Clean 3D illustration, soft shadows, square format. No text or logos.
/clay a friendly round-bodied AI assistant robot carrying completed task cards and waving. Hand-modelled plasticine texture, pastel background, square format. No text or logos.
/comic a three-panel story about managing customer leads without a CRM: papers pile up, phones ring beside an endless spreadsheet, then the owner searches for one missing lead. Retro ink and halftones, square page, empty speech bubbles.
/sketch a landing-page layout in a square-grid notebook: header, hero, call-to-action, three features, testimonial and footer. Top-down desk photo, realistic pencil and ink. No readable text.
/handdrawn a four-step automation process connected by marker arrows: SOURCE, SORT, CREATE and SEND. Black marker with one yellow accent, white paper, square format.
/handwritten visual explanation of how an AI agent works: ASK, THINK, USE A TOOL and ANSWER. Blue and red pen on cream notebook paper with arrows and tiny doodles, square format.
/stickynotes a content planning board with four columns labelled IDEAS, MAKE, READY and LIVE. Colourful sticky notes on a white wall, natural daylight, square format. No other readable text.
Layer photography, paper and paint into one visual.
/collage about artificial intelligence, human creativity and modern marketing. Torn magazine paper, analogue photographs, circuit fragments, paint strokes and hands arranging ideas. Bold colours, square format, no readable text or logos.
Starting with /xray or /handwritten tells ChatGPT what kind of visual you want. The subject, composition and constraints that follow make the result reliable.
The reliable prompt formula
Use this every time
/[code] + [subject] + [what must be visible] + [visual style] + [format] + [constraints]
Example: /cutaway a boutique coffee machine + show the boiler and water circuit + photorealistic technical illustration + square format + no text or logos.
1
Choose the visual job
Use /timeline for stages, /cutaway for hidden parts, /clay for style, and so on.
2
Name the subject properly
“A coffee machine” is vague. “A compact 1970s Italian espresso machine” gives ChatGPT something useful to build.
3
Say what must appear
List the parts, stages, labels, people or objects that matter. Do not make the model guess.
4
Lock the finish
Add the medium, background, aspect ratio and anything to avoid: no watermark, no extra text, no logos.
/[CODE] [YOUR SUBJECT], showing [THE IMPORTANT PARTS OR STAGES]. [VISUAL STYLE OR MEDIUM], [BACKGROUND OR LIGHTING], [SQUARE / PORTRAIT / LANDSCAPE] format. No [TEXT / LOGOS / WATERMARK / EXTRA OBJECTS].
Six combinations to steal
Use these when one shortcut is not specific enough. The second word acts as another style or layout instruction.
/blueprint /explodedview a compact podcast recording kit, with microphone, interface, headphones, cables and mounting hardware separated in assembly order. Crisp white linework on deep-blue grid paper, square format. No logos or watermark.
/handwritten /timeline a 30-day launch plan moving through IDEA, BUILD, WARM-UP, OPEN, CLOSE and REVIEW. Blue and red pen on cream notebook paper, arrows and tiny doodles, portrait format.
/isometric /cutaway a tiny online business headquarters with content, sales, client delivery and finance rooms visible inside. Miniature people supervising AI assistants, clean 3D illustration, soft shadows, square format. No text or logos.
/clay /comic a three-panel story about an overwhelmed business owner handing repetitive admin to a friendly AI assistant. Hand-modelled plasticine characters, pastel background, expressive poses, empty speech bubbles, square page.
/collage /mindmap of a PERSONAL BRAND with four branches: IDEAS, STORIES, OFFERS and COMMUNITY. Torn magazine paper, analogue photography, marker arrows and paint strokes, bold colours, square format. No logos.
/sketch /flowchart for turning one video into a newsletter, carousel, short clips and LinkedIn post. Pencil and ink on square-grid notebook paper, clear arrows, top-down desk photo, landscape format. Use only the labels VIDEO, EMAIL, CAROUSEL, CLIPS and LINKEDIN.
If ChatGPT ignores the second code
Write it as normal language instead: “Create an isometric cutaway illustration…” The slash is shorthand, not magic syntax.
Quick fixes for wonky results
✍️
The labels are gibberish
Use fewer labels, type the exact words in capitals, and ask for “no other readable text”.
🧩
Important parts are missing
Name every non-negotiable component. If it matters, put it in the prompt.
📐
The layout is wrong
State square, portrait or landscape, then describe where the main subject should sit.
🎨
The style is too generic
Add a medium, lighting, colour palette and texture instead of relying on the slash word alone.
🪜
The image is too busy
Ask for one focal point, a plain background and fewer elements. More detail is not always better.
🔁
It nearly nailed it
Do not start again. Tell ChatGPT exactly what to keep and change in the current image.
Your first move
Choose one code that looks nothing like your usual content. Copy the prompt, replace the subject, and see what happens. Thirty seconds. That is the test.
Want help turning this into content?
Learn the practical AI systems behind the prompts, without the tech-bro carry-on.
