The rise of AI in creative work

Businesses are rapidly adopting AI tools to enhance performance and streamline creative processes. The impact is already visible across industries. Fashion companies, for example, can now generate hyper-realistic images of models and settings in hours rather than weeks, cutting production costs by more than two-thirds. Marketing is seeing a similar shift, with more teams integrating generative AI into their creative workflows.

These tools may look simple on the surface, but understanding the core principles behind them is what separates average results from exceptional ones. Fortunately, those principles aren’t difficult to learn.

Understanding how to work with AI

Before generating any image, you need a strong prompt, and AI language tools like ChatGPT can help you build one. To get the most out of them, follow three key rules:

• Talk to the AI like a human. Use natural language and a conversational tone.
• Be specific. Include only relevant details, but make them precise.
• Push for better. After receiving an answer, ask the AI to go deeper and provide more detail.

A good image prompt doesn’t need to be long, it needs to be structured. Prompts are built using “locks”: reusable building blocks that each control a specific element of the image.

Core prompt elements (locks):

• [STYLE LOCK] — overall visual tone and medium
• [CHARACTER LOCK] — person/model appearance
• [ENVIRONMENT LOCK] — location and setting
• [CAMERA LOCK] — lens, depth of field, framing
• [LIGHTING LOCK] — light source and behaviour
• SCENE — what’s actually happening
• COMPOSITION (optional) — framing and angle
• MICRO DETAILS (optional) — textures, imperfections
• EMOTION (optional) — mood and expression
• NEGATIVE LOCK — what to avoid (e.g. no plastic skin, no AI artefacts)

The style lock

A style lock is the foundation of visual consistency across multiple images. It sets the tone, look, and feel so you can reuse it across every prompt without starting from scratch. Think of it as your creative brief translated into AI language.

A well-built style lock can sometimes be enough on its own, just add a scene description and you’re ready to generate. However, since it’s a summarised version of all prompt elements, some details may still be missing.

Key structural elements of a style lock:

1. Visual medium:
   Overall style (e.g. editorial photography, cinematic still)
2. Realism level:
   How real vs. stylised the image feels
3. Subject treatment:
   How people are rendered (natural, unretouched, imperfect)
4. Lighting philosophy:
   Light source, direction, and behaviour
5. Colour system:
   Palette, contrast, and grading style
6. Environment design:
   Location type and materials
7. Camera & optics:
   Lens type, depth of field
8. Composition language:
   Framing, perspective, use of space
9. Mood & emotional tone:
   The overall brand feeling
10. Negative lock:
    What to avoid at all times

You don’t need to be an expert in cinematography, ChatGPT can help fill in each element with the right terminology. The important thing is understanding what each element does, so you can adjust it to match your creative vision.

The examples below showcase two versions of the same style lock. The first is a shortened version of the second. Depending on how much detail you want visible in your images and the model you use, you can chose which one to apply. Both give similar images but the images generated with the second style lock are slightly more detailed and realistic.

Example style lock (condensed):

Hyper-realistic editorial photography. Natural human subjects, visible skin texture, no retouching. Soft natural daylight through large windows, high dynamic range, no artificial lighting. Colour palette: dark green, walnut brown, muted beige, soft gold — restrained saturation. Modern luxury interior with dark wood, velvet, stone. Full-frame camera, 35mm lens, shallow depth of field. Calm, refined mood — never staged. No AI artefacts, no plastic skin, no over-stylisation.

Example style lock (detailed):

Hyper-realistic editorial photography, natural human subjects (non-model look), visible skin texture (pores, fine lines, subtle imperfections), no retouching.

Lighting: bright natural daylight entering through large architectural windows, softly diffused but directional. High dynamic range with luminous highlights and gentle shadow gradients. No artificial studio lighting.

Colour grading: balanced contrast between light and dark — warm highlights with deeper, rich shadows. Dominant palette includes dark green, deep walnut brown, muted beige, and soft gold accents. No oversaturation, cinematic but restrained.

Environment: modern luxury hotel or high-end lounge setting. Interior features dark wood finishes (walnut, mahogany), deep green upholstered furniture (velvet or textured fabric), refined materials like stone, brass, and glass. Clean, architectural lines with a premium, calm atmosphere. Subtle reflections on polished surfaces.

Material realism: ultra-detailed textures — wood grain, fabric depth, slight wear on surfaces, realistic reflections and light absorption.

Camera: full-frame sensor, 35mm or 50mm lens, shallow depth of field with natural falloff, subject sharply defined while environment remains legible.

Composition: editorial framing with intentional negative space, layered depth (foreground, subject, background), natural perspective.

Mood: calm, refined, intelligent, understated luxury — never flashy, never staged.

No AI artefacts, no plastic skin, no symmetry correction, no artificial lighting, no over-stylisation.

The character lock

A character lock works like a style lock but focuses entirely on a specific person. It ensures your character looks consistent across every image, regardless of scene, setting, or composition.

Two types of character locks:

1. Appearance-based — You describe physical traits (skin, gender, age, clothing, expression). The AI generates different images of similarly looking people each time.

2. Named character — Same physical details, but you give the character a name. The AI will now generate the same exact person in any scene. Ideal for building campaigns or visual stories with a consistent fictional model.

How to build a character lock (step-by-step)

Step 1: Ask ChatGPT for the right prompt structure.

Tell it: “I want you to build a model prompt that I can use to ask you to build a character lock. I want hyper-realistic looking people -non-model look -extremely realistic skin details -natural imperfections -micro details -relatable friendly look -creative/professional -fashionable yet casual chic -behavioural details -expression details -additional realism details -keep the prompt as short as possible.”

ChatGPT will return a detailed structured prompt template covering all necessary fields.

Step 2: Review and clean up the output. Remove any fields that aren’t relevant to your needs. Keep it focused.

Step 3: Send the refined prompt back to ChatGPT. It will generate a full character lock. Check the result, and if anything doesn’t match your vision, simply ask ChatGPT to adjust specific details while keeping the rest intact.

Key character lock elements:

• Identity — gender, age, ethnicity, overall appearance
• Facial structure — face shape, jawline, asymmetry
• Skin — tone, texture, pores, imperfections, micro-details (critical for realism)
• Eyes, lips, hair — shape, colour, texture, natural irregularities
• Body & posture — build, stance, natural physical traits
• Wardrobe — style, materials, fit, wear and imperfections
• Expression & behaviour — micro-expressions, eye behaviour, candid vs. posed
• Realism enhancers — preserve asymmetry, no retouching, natural skin contrast

Example character lock (condensed):

Woman, 31 years old, European appearance. Slightly asymmetrical face — left cheek fuller than right, soft jawline. Dark blonde hair, fine texture, shoulder-length, natural flyaways and subtle frizz. Fair skin, neutral undertone — visible pores on T-zone, faint acne scarring on cheeks, slight redness near nostrils. Faint under-eye darkness with creasing, small broken capillaries near nose, barely visible peach fuzz. Grey-blue eyes, slightly hooded, minimal asymmetry. Lips: natural tone, slightly dry with fine vertical lines. Expression: relaxed, unposed, subtle tension around eyes.

The skin lock

If you’re working with only a style lock and scene description without a character lock, you may need a separate skin lock to ensure realistic skin rendering.

Build it the same way as a character lock:

Step 1: Tell ChatGPT:

“I want you to give me a prompt to build the most detailed skin lock for image generation. I want hyper-realistic skin details -natural imperfections -facial details -no plastic skin -high resolution -keep the prompt as short as possible.”

Step 2: Review, remove redundant fields, and refine.

Step 3: Send it back to ChatGPT and let it generate your skin lock.

A good skin lock includes: pore-level texture, micro-contrast, natural tone variation across the face, visible imperfections (redness, pigmentation, fine lines), and realistic light interaction, with no smoothing or artificial gloss.

The negative lock

When crafting a detailed image prompt, one element you should never overlook is the negative lock: a targeted list of everything you don’t want to appear in your final image.

By explicitly naming unwanted elements, you signal to the AI’s training system to automatically filter out any training images tagged with those characteristics. It’s a simple but powerful way to take back creative control.

Start small. Begin with just a handful of exclusions and evaluate the results. Often, a few well-chosen terms are enough to steer the generation in the right direction.

If the AI still slips in unwanted details, revisit your negative lock and add those specifics to the list, fine-tuning as you go.

Here’s the catch, though: more isn’t always better. Overloading your negative lock with too many restrictions can actually backfire. When the list grows too long, the model struggles to track what should and shouldn’t be included. And the very elements you’re trying to eliminate can end up creeping back in anyway.

The sweet spot? Precise, intentional exclusions, just enough to guide the AI without overwhelming it.

Outcome & results

With a solid foundation in prompt engineering and lock implementation in place, the real work begins: putting them to the test. We ran our prompts across multiple GenAI tools and model variations. Not just to validate, but to understand where they hold, where they bend, and where they need sharpening.

Here’s what we found:

Key takeaways

• Locks = consistency. Style, character, and skin locks are reusable building blocks that save time and keep your visuals coherent across campaigns.
• You don’t need to be a technical expert. ChatGPT can help you construct every element — you just need to understand what each one does.
• Iterate and refine. Always review AI output, remove what you don’t need, and ask for adjustments. The best results come from a back-and-forth process.
• Start simple. A style lock + scene description is often enough to get started. Add character and skin locks when you need more control.