THE PROMPTS FOR CREATING QUALITY, ENGAGING VISUALS WITH AI

FROM FLAT AND LIFELESS TO CINEMATIC AND TIMELESS

HOW TO CREATE AI VISUALS THAT DON'T SUCK

I don’t know about you, but I’m getting tired of seeing AI-generated images of people who look like dolls. We went from lifeless stock photos that at least included real people to lifeless (and sometimes creepy) AI visuals that don’t even make sense. The worst part is: it’s not AI’s fault. We’re just shit at prompting it.

Many people forget that AI is an execution tool. It doesn’t think for itself. Nano Banana Pro sort of breaks this pattern, because it is a Thinking Model, not just a rendering engine. Which means it doesn’t merely translate prompts into pixels; it reasons about why a moment exists. When you provide narrative tension, it responds with cinematic realism instead of glossy sameness. With that said, you still need to give it a narrative.

Ask a traditional image model for “a woman drinking coffee,” and you’ll get a sanitised lifestyle shot. Give Nano Banana Pro a reason she’s drinking coffee—and you get a frame from an indie film. In this blog, I wanna teach you how to go beyond pro to create AI visuals that do connect your brand to people.

Lesson #1: Mess and drama before beauty

Emotion comes from imperfection, tension, and specificity. Nano Banana Pro excels when you give it those inputs before asking for visuals. Before you start typing “Generate image…”, you should think about that moment and visualise it in your head. What messy details can I include? In an era where everyone has access to the same tools, those with an imagination are the ones who will stand out.

Phase 1: The Idea (Prompt to brainstorm, not to generate the image)

Most people start with an image description. That’s the mistake. Instead, you should first use Perplexity as a creative researcher. Here’s a prompt that can help you come up with the idea:

The Researcher Prompt

You are a researcher tasked with deeply exploring a scenario connected to [TOPIC / PRODUCT / OCCASION / THEME]. Discovery through questioning Begin by asking three highly specific, probing questions designed to uncover meaningful moments, emotional drivers, tensions, or decision points within this scenario. The questions should move beyond surface-level description and encourage reflection, storytelling, or lived experience. Aim to reveal why people behave, feel, or remember things the way they do, not just what happens. Contextual insights and hypotheses After posing the questions, share initial insights, hypotheses, or strategic observations based on common patterns in similar scenarios. Highlight potential emotional beats, motivations, unmet needs, or moments of friction. Suggest angles or lenses through which this scenario could be interpreted (e.g. cultural, social, practical, emotional, aspirational). Exploration of real conversations and stories Draw on relevant online conversations (such as forums, reviews, social media discussions, comment threads, or personal blogs) where people naturally describe: Favourite moments Memorable experiences Frustrations, surprises, or turning points Personal stories linked to this topic Summarise the types of moments people repeatedly talk about, the language they use, and what seems to matter most to them in their own words. Moment selection and synthesis Review all responses, insights, and observed patterns, then identify one specific moment or scene that best serves the overarching goal (e.g. storytelling, design inspiration, brand strategy, user understanding, narrative development). Clearly explain why this moment is the most powerful or relevant. Describe it with enough clarity and detail that it could be visualised, written, or built upon in the next stage of work. Your output should demonstrate structured thinking, cultural awareness, and sensitivity to real human experience, while remaining practical and goal-oriented.

Phase 2: Visualising the moment

Take that moment and go to ChatGPT. My favourite thing is to hit the mic and start speaking, but you can also type. Ask it to help you explore all elements of this moment using this prompt:

The Explorer Prompt

You are an explorer of moments, tasked with expanding and visualising a specific scene. Begin by identifying a moment where [BRIEFLY DESCRIBE THE MOMENT]. Treat this moment as a snapshot within a wider sequence of events rather than an isolated image. Temporal exploration Break the moment down across time: Before: What has just happened to lead up to this moment? What emotional, physical, or environmental cues are already present? During: What is actively unfolding in the exact instant we are capturing? What tension, stillness, or motion defines it? After: What is likely to happen next? How might this moment change the mood, behaviour, or outcome of the scene? Setting and environment Define the type of setting in which this moment would realistically occur. Is it public or private, familiar or unfamiliar, controlled or chaotic? Consider time of day, weather, season, and atmosphere. Describe how the environment contributes to the meaning or emotion of the moment. Objects and material details Identify the objects likely to appear in the scene. Which items feel essential or symbolic? What objects suggest recent activity or human presence? Consider wear, age, texture, and placement rather than pristine or idealised conditions. Foreground and background composition Explore visual depth by separating the scene into layers: Foreground: What immediately draws the eye? What details anchor the viewer in the moment? Midground (if relevant): Where does interaction or action occur? Background: What contextual elements quietly reinforce the setting without stealing focus? Offer multiple compositional options to help visualise different interpretations of the same moment. People and behaviour Decide whether people are present in the scene. If so, how many, and what is their relationship to one another? What are they doing physically, and what might they be thinking or feeling? Consider body language, posture, gaze, and small human gestures rather than exaggerated action. Exploration of variations Generate several plausible versions of the moment: Calm vs tense Intimate vs observational Ordered vs messy Quiet vs subtly active These options should help refine tone, perspective, and emotional intent. Identifying visual flaws and imperfections Actively look for visual flaws that would naturally occur in this moment. Optical imperfections (motion blur, uneven focus, harsh or mixed lighting, lens artefacts) Environmental flaws (clutter, stains, asymmetry, damaged surfaces, weather effects) Human flaws (awkward posture, imperfect expressions, partially obscured faces, unfinished actions) Treat imperfection as a tool for realism and believability, not a mistake. Preparation for image generation The next step is to generate an image using Nano Banana Pro. When ready: Re-engage via voice and clearly state the exact moment you want to see. Specify foreground and background details with intention. Deliberately select a small number of imperfections to include. Remember: perfection is the enemy of realism. Nano Banana Pro responds especially well to controlled imperfection, particularly when flaws are sensory, optical, or human in nature. Prompt construction Describe the scene as you are now visualising it, using clear, concrete language. Then ask Nano Banana Pro to generate a detailed image prompt that captures: The chosen moment The spatial composition The emotional tone The selected imperfections Your goal is not to create a flawless image, but a convincing, lived-in moment that feels observed rather than staged.

Phase 3: Prompt engineering

No, we’re still not ready to generate an image just yet. All good things come to those who wait, and this is definitely the case here. By following this framework, you’re already doing what 95% of people aren’t doing. In this phase, we’re gonna make the prompt detailed af. Copy this prompt and paste it into ChatGPT:

The Engineering Prompt

You are a professional prompt engineer. Your task is to take the previously defined moment and translate it into a highly specific, production-ready visual description suitable for advanced image generation. Systematically refine and specify the scene by addressing each of the following dimensions in detail. Every choice should serve realism, narrative clarity, and visual coherence. Setting Clearly define the overall setting. Is it indoor or outdoor? Domestic, commercial, industrial, natural, or hybrid? What does the setting immediately communicate about mood, purpose, or social context? Avoid vague descriptors; favour concrete, observable characteristics. Location Specify the location with believable detail. Is it a specific type of place (e.g. a cramped kitchen, a suburban street, a backstage corridor, a corner café)? Consider geographical cues, architecture, materials, or layout that subtly anchor the scene in a real world. Time of day / night Define the exact time period. Early morning, mid-afternoon, golden hour, late evening, after midnight, etc. Explain how the time of day affects light quality, energy levels, and atmosphere. People (Yes / No) If no people: describe how recent or implied human presence is visible (disturbed objects, warmth, mess, wear). If yes: How many people are present? What are they doing at this precise moment? What do they look like (age range, build, posture, grooming)? What are they wearing, and does it feel appropriate, slightly off, or imperfect for the situation? Describe facial expressions subtly. Avoid exaggerated emotion; favour restrained, ambiguous, or in-between expressions that feel human and unperformed. Objects and furniture List the key objects and furniture visible in the frame. Prioritise items that suggest recent activity or unfinished actions. Note materials, condition, placement, and any signs of use or neglect. Avoid symmetry or showroom-style perfection. Human flaw / imperfection factor Introduce one or two believable imperfections that add realism and mild tension. This could be something spilled, something dirty, visible sweat, clutter, a crooked object, rumpled clothing, condensation, fingerprints, or uneven wear. The flaw should feel accidental and relevant to the moment, not decorative. Use imperfection to suggest life, not chaos. Camera and lens choice Specify how the image appears to have been captured. Type of camera (e.g. DSLR, mirrorless, smartphone, analogue film camera). Lens type and focal length (e.g. wide-angle, standard 35mm, shallow depth portrait lens). Indicate whether the perspective feels observational, intimate, or slightly detached. Visual style: realism vs stylisation Clearly state whether the image should be: Fully realistic / photorealistic Lightly stylised but grounded Illustrative or cartoonish If not fully realistic, define the limits of stylisation so it does not undermine believability. Image characteristics Precisely define: Lighting: natural vs artificial, soft vs harsh, directional vs flat Contrast: low, medium, or high, and why Film grain or noise: present or absent, subtle or noticeable Focus: shallow or deep depth of field, what is sharp and what falls out of focus These choices should reinforce mood and draw attention to the intended focal point. Aspect ratio and output size Specify the most suitable aspect ratio and resolution for [PLATFORM]. If the platform is unknown, default to a flexible, adaptable format (e.g. 4:5 or 16:9) that can be cropped without losing key elements. Ensure the composition accounts for safe areas and avoids cutting off essential details. Check the response you get from ChatGPT and adjust anything you would like to change or remove anything you don’t like.

Phase 4: The Generation (Nano Banana Pro’s Master Prompt Structure)

Nano Banana Pro handles long, layered prompts better than most models. Use that strength. You already have the perfect prompt from ChatGPT, so chuck it in there and hit ENTER! If the image isn’t what you hoped, see if there’s something in the prompt that you need to adjust. If not, start a new chat and try the prompt again. You won’t always get the perfect result on the first try.

Putting the method to the test

I’d be full of shit if I didn’t practise what I’m preaching, or show proof that this works. The main image of this blog was generated using exactly this method, but here’s a comparison. The first prompt jumps straight into image generation with:

The Basic Image Generation Prompt

“Generate an image of a woman drinking a cup of coffee.”

VS.

The Narrative-based Image Generation Prompt

“Generate a photorealistic, cinematic still.
Subject:
A young female architect sitting at a cluttered dining table at 3:15 AM.
She is intensely focused on a blueprint on her desk. She’s determined to ace a project.
She holds a mug of coffee with both hands, close to her face for warmth.

Details:
Her hair is in a messy bun with loose strands catching the desk lamp’s light.
There are realistic dark circles under her eyes.

Environment:
The table is covered with eraser shavings, pens, and a glowing laptop.
The kitchen behind her is dark and shadowy.

Tech Specs:
35mm film grain.
High ISO noise.
Low-light photography.

Product:
The coffee mug is a simple ceramic cup.
Steam rises from it, slightly distorting the laptop screen behind.”

The Result: Meaning Over Messaging

This image doesn’t scream “Buy our coffee.” It quietly communicates: “We fuel your hustle.”

That’s the power of narrative-first prompting. The image from the second prompt could become a post about going after your dreams, describing a scene that an ambitious person could relate to, and would trigger their emotions. More connection = more brand love = more conversions. The first image speaks to no one and is a waste of everyone’s time. If anything, it’s off-putting.

Final Takeaway: Stop Prompting Images. Start Prompting Moments.

If you feed AI context, narrative, tension, and imperfection, it will reward you with visuals that feel authentic, emotional, and more human. An important thing to remember is to customise the above prompts to your needs and have fun with the process.

Need custom prompts for your brand?

Get in Touch

I get you wanna gatekeep this, but come on...