Prompt writing for designers: a method, not a list of prompts

A good prompt for an image or video model is a brief, not a magic spell. It says what is in the frame, where the camera stands, how the light falls, what things are made of, which colours lead and what the image is for. Designers write briefs like this all the time for photographers and illustrators. Prompt writing is the same skill with a new reader.

I direct AI tools for stills, motion and voice in the video series I make at SkillUp MENA, and I use image models for moodboards, packaging concepts and backgrounds. AI is a tool in my hand: it gives me options fast, and the decisions stay mine. This article is the method I follow, in the order I write a prompt, followed by iteration, seeds, reference images, brand consistency, video briefs and a few example prompts to adapt.

Why a method beats a list of prompts

Lists of "best prompts" are copied from someone else's image, made for someone else's brand. They teach you to collect, not to direct, and they age with every new model version. A method travels: the same thinking works in Midjourney, Gemini, ChatGPT or Adobe Firefly, because it is based on how images are made, not on one tool's quirks. Google's own prompting guide for its Gemini image model makes the same point in one line: "Describe the scene, don't just list keywords." A short paragraph that reads like a brief usually beats a string of disconnected words.

The same guide adds a second habit worth stealing: explain the purpose of the image. "A product shot for a premium coffee brand's online shop" tells the model far more than "coffee bag". If you are new to AI in design work, I wrote a wider overview in AI in graphic design.

The parts of a prompt, in the order I write them

Subject: say exactly what it is

"A coffee bag" gets you the average coffee bag. Describe the shape (a stand-up pouch with a valve), the material, the finish, where the label sits and how big the object is next to anything else in the frame. Keep one main subject per image where you can.

Composition: leave room for the design

Think about the layout before you generate. Where will the headline and the logo go? Ask for that space directly: "the upper third stays empty", "the product sits in the lower right". Use the words a photographer would use: close-up, medium shot, top-down flat lay, three-quarter view, centred symmetry.

Lens and camera language

Photographers' words give you precise control. Google's guide suggests terms such as wide-angle shot, macro shot, low-angle perspective and 85mm portrait lens. In practice:

• 85mm or longer: calm, compressed perspective. Products feel premium and still.
• Wide-angle: drama, space and some distortion. Good for interiors and scale.
• Macro: texture and detail. Coffee grounds, paper fibres, condensation.
• Shallow depth of field: a soft background that pushes attention to the subject.

Choose one lens and one angle per prompt.

Light: the mood lives here

Describe the direction, the quality (hard or soft), the temperature (warm or cool) and the time of day. "Soft light from a large window on the left" and "hard midday sun with sharp shadows" give the same product two different personalities.

Material and texture

This is what makes a packaging shot believable. Use the words a printer would use: uncoated kraft, soft-touch matte, gloss varnish, embossing, hot foil, frosted glass, condensation drops, linen. Describe the finish you can actually print, or the render will promise the client something the budget cannot deliver.

Colour

Name colours in words, and give them a reference: "deep olive green, like a ripe olive", "warm cream, like unbleached paper". Keep two or three leading colours plus neutrals. Some tools accept HEX codes, but do not count on an exact match: check the colours against the brand palette afterwards and correct them in Photoshop.

Style references, and why I don't name living artists

A style phrase tells the model the whole look in a few words: editorial still life, risograph print, flat vector illustration, soft clay render, 1970s travel poster.

Naming a living artist is different, and I avoid it for three reasons. It borrows someone's life work without their consent and competes with them using their own name. Some tools are built to refuse it: OpenAI said DALL·E 3 was designed to decline requests for an image in the style of a living artist. And for client work it is a real risk: a campaign that looks like a well-known illustrator's work can draw complaints, and the client cannot defend it as original.

Describe the qualities you admire instead: the medium, the technique, the era, the line, the palette. You end up closer to the brand and further from someone else's portfolio.

Aspect ratio: decide by destination

Choose the format before you generate, because a different ratio changes the whole composition. Common choices are 1:1 and 4:5 for feed posts, 9:16 for stories and reels, and 16:9 for web banners and video. In Midjourney the format is a parameter at the end of the prompt, such as --ar 4:5. In chat-based tools, write it as a sentence or use the format setting.

Negative instructions: describe what you want instead

Telling a model what not to include can bring that very thing into the picture. Google's guide recommends describing the scene positively: instead of "no cars", write "an empty, deserted street with no signs of traffic". Midjourney has a --no parameter for removing specific elements, such as --no text. Keep negatives short and specific, and reach for them only after a positive description has failed.

Iterate like a designer, not a gambler

Generating twenty images and hoping one works is not a method. Treat each round like a design review:

1. Change one thing at a time. Fix the light, then the lens, then the colour. If you change everything, you learn nothing.
2. Keep a prompt log. For each keeper, save the prompt, the tool, the settings and a note on why it worked.
3. Use the conversation. In chat-based tools, refine with short follow-ups such as "make the light a little warmer". If a character drifts after many edits, Google suggests starting a new conversation with a detailed description.
4. Know when to stop. When the image is close, finish it by hand in Photoshop and Illustrator, not in another fifty generations.

Seeds and variations

A seed is the number that sets the random starting point of an image. In Midjourney you set it with --seed. Its documentation is honest about the limits: a seed only shapes the initial noise, cannot hold a style or a character across different prompts, and may not give the same result in a later session. Use seeds for controlled tests (same seed, one word changed), not to keep a brand consistent.

Variations are more useful day to day. When a result is close, ask for subtle variations of that image instead of a fresh generation, so you keep the composition and explore the details.

Reference images

Often an image explains a style better than any paragraph. In Midjourney, the --sref parameter takes a style reference from an image. In Gemini and other chat-based tools, you can upload images and say what to take from each: "use the light and palette of the first image and the pouch shape of the second". Use your own photos, sketches and approved renders, or images you have the rights to use, and never upload client material you are not allowed to share.

Keeping a brand consistent across many images

One good image is easy. Thirty images that look like one brand is the real job. The video work I direct, such as the AI Video Studio series at SkillUp MENA and the short film for The Coffee Address in Saudi Arabia, depends on exactly this: many generated stills and shots that have to feel like one piece. What keeps a set together:

• A fixed style block. Write one paragraph that describes the look: light, palette, lens, materials, mood, background. Paste it into every prompt, word for word. Only the subject and composition change.
• The same references. Reuse the same style reference images for the whole set.
• One finishing pipeline. The same colour correction, the same type, the logo placed by hand. I set Arabic type myself in the design software, so the letterforms and the brand font are exactly right.
• Review as a set. Put every image side by side on one board before anything is published, and reject the one that drifts.

Writing briefs for video models

A video prompt is a shot description. Google's prompting guide for its Veo video model uses a simple formula: cinematography, plus subject, plus action, plus context, plus style and ambiance. Some practical rules:

• Name the camera move. Dolly in, tracking shot, crane shot, slow pan, aerial view, point of view. One move per shot.
• One shot per prompt. Keep each shot short and simple, then cut them together in the edit.
• For image-to-video, describe the motion, not the image. Runway's guide warns that repeating details already in the image can reduce motion or cause unexpected results. Its example is simple: "The man extends his arm to shake hands, then nods politely."
• Sound, if the model makes it. Google's guide suggests putting dialogue in quotation marks and describing sound effects and ambient noise separately.

In my SkillUp MENA series the work moves from the script to stills, motion, voice and the edit. Each brief for a model should trace back to a line in that script, so nothing is generated without a reason to exist.

Example prompts to adapt

These are starting points written for this article, not results I am showing. Swap in your own brand details, and adjust for the tool you use.

Example 1: packaging shot

A studio product photograph of a matte kraft stand-up coffee pouch with a small round gold foil seal and a plain cream label, standing on a warm grey stone surface. Soft window light from the left, a gentle shadow to the right, a few roasted coffee beans in front, slightly out of focus. 85mm lens at eye level, shallow depth of field, calm premium mood, muted olive and cream palette. The label area stays blank for the design to be added later. Vertical 4:5 format.

Example 2: moodboard

A moodboard for a modern café brand, laid out as a clean grid of nine square tiles on an off-white background: palm-leaf weaving, a sand-coloured plaster wall, a brass coffee pot in close-up, dates on a handmade ceramic plate, morning light through a carved wooden screen, deep green and sand fabric swatches, a small cup of coffee seen from above. Natural light, honest textures, warm and quiet. Photographs only, no lettering. 16:9 format.

Example 3: social post background

A background for a social media post: a top-down flat lay on a deep olive green linen cloth, with a ceramic coffee cup and a small plate of dates in the lower right corner. The upper two thirds stay calm and empty, with soft even light and very gentle shadows, leaving room for a headline. Warm, quiet evening mood. Square 1:1 format.

Example 4: video shot from a still

Slow dolly in towards the coffee pouch. Steam rises gently from the cup beside it. A hand enters from the right, places three coffee beans next to the pouch, and leaves the frame. The light stays soft and constant. Quiet café ambience.

If you are still choosing a tool, I compare the main options in the best AI image generators for designers.

Conclusion

Prompt writing is briefing. Say what the subject is, where the camera stands, how the light falls, what things are made of, which colours lead and where the image will live. Change one thing at a time, and protect the brand with a fixed style block and a human finishing pass. The model gives you speed. The direction is still yours.

If you need a campaign, packaging visuals or an AI video series directed with that kind of care, tell me about your project.