AI Image Prompt Examples: The 8-Part Formula for Realistic Photos

Your Nano Banana generations look too polished. Too smooth. Too fake.

Not because the model is bad — Nano Banana is one of the most capable image generation models available right now. The problem is almost always the same: your prompt has no structure. You are describing a vibe instead of building a blueprint.

Generative AI does not understand feelings. It understands hierarchy. This guide gives you the exact 8-part framework that forces Nano Banana to output raw, candid, smartphone-quality photography every single time.

Quick reference — the 8 layers every realistic AI prompt needs:

1. Subject definition (specific physical descriptors, not "a woman")
2. Action and context (what they are doing and where)
3. Outfit and styling (every garment, fabric, and fit detail)
4. Environment (background with same precision as subject)
5. Camera and lens (focal length, aperture, body)
6. Lighting (source, direction, quality, color temperature)
7. Texture and realism modifiers (skin pores, grain, imperfections)
8. Negative constraints (explicit exclusions)

Why Structured Prompts Produce Better Results

AI image models read prompts sequentially and probabilistically. When you write "a woman at a coffee shop, natural light, realistic," the model fills every undefined gap with the most statistically average answer from its training data. The result is a stock-photo face, a generic background, and that unmistakable AI skin texture that fools nobody.

When you define every layer explicitly, you leave no room for the model to hallucinate details you did not want. Structure is the difference between a first-generation draft and a final image.

---

The 8-Part Prompt Formula

1. Subject Definition

Who or what is the main focus? Avoid vague nouns. Replace "a woman" with "a 28-year-old woman with dark circles under her eyes, slightly asymmetrical face, shoulder-length black hair pulled back loosely."

Real imperfections are what make AI images pass as photographs. Symmetry is a CGI tell. For a deep dive on how to eliminate the plastic look specifically, see our guide on fixing fake skin texture in AI portraits.

2. Action & Context

What is the subject doing, and where? The action anchors the narrative. "Pouring oat milk into a ceramic cup, morning, small independent coffee shop" gives the model a physical and temporal context that constrains every other decision it makes.

3. Outfit & Styling

This is the most skipped layer — and the one that causes the most inconsistency. Describe exact garments, fabrics, and fit. "Oversized beige linen shirt, untucked, top two buttons open, worn-in dark jeans, white low-top sneakers with slight dirt on the sole."

Every detail you leave out is a detail the model invents. If you need to match a specific reference — a jacket from a lookbook, a shoe from a brand shoot — this is where the CookedBanana Outfit Ref system locks those assets in directly, bypassing Nano Banana's guesswork entirely.

4. Environment

Define the background with the same precision as the subject. "Inside a brutalist coffee shop: exposed concrete walls, warm Edison bulb pendant lights, a single window to the left with diffused morning light entering." Vague settings produce generic stock backdrops.

5. Camera & Lens

This single layer separates photorealistic results from CGI renders. Real cameras introduce optical characteristics — depth of field, barrel distortion, focal compression — that our eyes associate with genuine photography.

Use these as a starting point:

| Intent | Lens Reference | |---|---| | Candid street feel | 35mm f/1.8, slight vignette | | Intimate portrait | 50mm f/1.4, shallow depth of field | | Fashion editorial | 85mm f/1.2, background separation | | Raw smartphone candid | iPhone rear camera, slight lens flare | | High-detail editorial | Hasselblad X2D, 135mm, medium format |

6. Lighting

Lighting is the single most impactful variable for realism. A perfectly described subject photographed under flat lighting looks like a headshot from 2009.

Define the source, direction, and quality:

• "Soft natural window light from the left, slight warm color cast, no fill light, natural shadows falling across the right side of the face"
• "Overcast outdoor light, even diffusion, no harsh shadows, slight desaturation in highlights"
• "Late afternoon raking sidelight, strong shadow definition, warm 5400K"

Avoid words like "good lighting" or "bright." They mean nothing to the model.

7. Texture & Realism Modifiers

This layer adds the micro-details that make an image feel lived-in. Without it, skin looks filtered, surfaces look CGI, and the image loses the tactile quality of real photography.

Keywords that consistently push Nano Banana toward authenticity:

• visible skin pores, natural skin texture, no retouching
• steam rising, condensation on glass, fabric wrinkles
• slight film grain, digital noise, jpeg compression artifact
• catchlight in eyes, slightly wet lips, natural eye redness

8. Negative Constraints

Tell the model explicitly what to avoid. Nano Banana responds well to direct exclusions.

Always include: no plastic skin texture, no extra fingers, no symmetrical face, no text overlays, no watermark, no HDR effect, no oversaturated colors, no AI-generated aesthetic

---

Putting It All Together: A Complete Example

Here is what all 8 layers look like assembled into a single, structured Nano Banana prompt:

28-year-old woman, dark circles, asymmetrical face, hair loosely pulled back — pouring oat milk into a ceramic cup, morning — oversized beige linen shirt untucked, worn-in dark jeans — brutalist coffee shop, concrete walls, Edison bulbs, single window left — 35mm f/1.8, slight vignette — soft window light from left, warm cast, natural shadow right side — visible skin pores, steam rising from cup, slight film grain — no plastic skin, no extra fingers, no HDR, no AI aesthetic.

Compare this to the prompt most people write: "realistic barista making coffee, natural light." The difference in output quality is not marginal — it is categorical.

If you want to go further, you can also feed a reference photo through CookedBanana's image-to-prompt engine to extract all 8 layers automatically from any photo you already love.

---

The Real Cost of Writing This Manually

At 10 prompts per session, writing each 8-layer prompt from scratch takes 90 minutes. Multiply that across a week of content production and you have a serious bottleneck.

CookedBanana is built around this exact framework. Upload a reference photo or describe a scene in plain text — the engine outputs a fully structured 8-layer prompt, pre-formatted for Nano Banana's architecture, in under 10 seconds.

The Lock Reference feature handles Part 3 (Outfit) and the identity anchoring in Part 1 automatically — so your subject stays consistent across every generation without manual re-entry.

Start your free trial — 3 generations, no credit card required.

---

Frequently Asked Questions

Why do my Nano Banana images still look fake even with detailed prompts?

The most common cause is missing texture modifiers in Layer 7. Without explicit keywords like visible skin pores or natural skin texture, Nano Banana defaults to its training average — which skews toward over-retouched, idealized photography. Add the realism modifiers from this guide and the difference is immediate.

What is the difference between a photorealistic and a hyperrealistic AI prompt?

Photorealistic means the image resembles a real photograph taken with a real camera. Hyperrealistic pushes beyond — it adds micro-details like skin imperfections, fabric wear, environmental clutter, and optical lens characteristics that make the result indistinguishable from a high-quality documentary photograph. The 8-part formula targets hyperrealism by design.

Do negative prompts actually work in Nano Banana?

Yes — and they are often more impactful than positive descriptors. Nano Banana responds strongly to explicit exclusions. Adding no plastic skin texture, no symmetrical face, no AI aesthetic directly suppresses the model's tendency to default to idealized outputs. Always include at least 5 negative constraints.

How many words should a Nano Banana prompt be?

There is no fixed length, but structured prompts between 80 and 150 words consistently outperform short prompts and unstructured long ones. The key is layer hierarchy, not word count. A 200-word prompt that mixes lighting with clothing details will underperform a clean 90-word prompt that separates each layer correctly.

Can I use this formula for outfit-specific generations without re-describing the entire look?

Yes. This is exactly what the CookedBanana Outfit Ref system solves — you select which outfit elements to reference directly from uploaded images, and the engine injects them as locked references in Layer 3. No manual re-description required across generations.

Can I apply this formula to product photography prompts?

Yes. For product photography, Layer 1 becomes the product Identity Layer (exact color, material, dimensions, logo placement), while Layers 4–6 define the environment and lighting. This two-layer architecture is how e-commerce teams prevent product identity drift across generations. See the complete e-commerce AI photography guide for the full workflow.

What camera reference produces the most realistic skin texture?

Medium-format camera references — Hasselblad X2D, Fujifilm GFX 100S — consistently produce the most textured, pore-accurate skin in Nano Banana. These cameras are associated with unretouched editorial photography in the model's training data. For candid, smartphone-authentic output, use iPhone rear camera, candid snapshot instead.