The Complete Guide to Writing AI Image Prompts

Chapter 1: Why Prompts Are the Core Skill of AI Image Generation

In the world of AI image generation, your prompt is your paintbrush. The same model can produce a stunning commercial poster or an unusable mess — all depending on how you write your prompt. Industry consensus is clear: prompt quality accounts for 80% of AI image output quality.

This guide systematically breaks down the prompt-writing methods and technique differences across the five mainstream AI image generation models available on Viddo.ai. Whether you're a complete beginner or an advanced user migrating between models, you'll find ready-to-use frameworks and real-world examples here.

Who this guide is for:

· Creators who want to move beyond random keyword dumping

· Marketers and designers producing commercial-grade visuals

· Anyone switching between AI models on Viddo.ai

 

Chapter 2: The Universal 5-Layer Formula for AI Prompts

Through extensive testing, we've distilled a universal 5-layer prompt structure that works across every model on Viddo.ai. Once you master this formula, you can write high-quality prompts for any model in seconds:

 

In practice, you don't always need all five layers. For quick creative experiments, the first two layers are enough. For commercial-grade output, we recommend covering all five. The key principle: write the most important visual elements first, then layer in supporting details.

2.1 Layer 1: Medium & Style

Tell the AI what this image is — a photograph, digital illustration, oil painting, 3D render, or concept art? Style anchoring sets the entire tone.

 

2.2 Layer 2: Subject & Details

The subject is the focal point of your image. The more specific, the better — "a girl" is far less effective than "an 18-year-old American girl with twin ponytails wearing a red hoodie." Details help the AI narrow its imagination and reduce randomness.

2.3 Layer 3: Environment & Scene

Environment creates narrative depth. "In a forest" and "In a misty coniferous forest after rain, with moss and fallen leaves covering the ground" produce two completely different worlds.

2.4 Layer 4: Lighting & Mood

Lighting is the soul of your image's emotion. Common descriptors:

· golden hour — warm, cinematic, romantic

· blue hour — cool, melancholic, mysterious

· Rembrandt lighting — dramatic, classic portraiture

· neon light — cyberpunk, urban nightlife

· studio softbox — clean, commercial, professional

2.5 Layer 5: Technical & Quality

The final layer is quality enhancement. Note: different models respond very differently to quality keywords. Common technical terms: 8K, ultra detailed, sharp focus, depth of field, bokeh.

 

 

Chapter 3: Viddo.ai Model Comparison — 5 Models Side by Side

Viddo.ai integrates five mainstream AI image generation models, each with its own "prompt language." Here's a core comparison:

 

One-line summary: Nano Banana likes "instructions to a photographer," Grok Imagine likes "creative briefs," GPT Image likes "task-oriented structured constraints," Seedream likes "priority-ordered subjects," and Midjourney likes "atmospheric adjective stacking." Use the right language style and your output quality jumps immediately.

Chapter 4: Nano Banana Prompt Writing Techniques

Nano Banana is Google's AI image generation model (including Nano Banana 2, Pro, and base versions), known for its strong natural language understanding, multi-image fusion, and text rendering capabilities.

4.1 Core Principle: Brief a Photographer

Nano Banana prompts are essentially "describe the scene in natural language" — not keyword dumping. The best mental model: you're giving a photographer a shooting brief.

Universal formula:

Who (subject) + What (action) + Where (environment) + How to shoot (composition) + What lighting (mood) + What style (output)

4.2 Five High-Frequency Templates

Template 1 — Image Generation (Most Common):

Subject + Action + Scene + Style + Composition

Example: "A fashion model wearing a minimalist beige outfit, standing in a clean studio with white background, full body shot, high-end fashion editorial style."

Template 2 — Image Editing:

Core principle: What to change + What to keep

Example: "Remove the background crowd. Keep the main subject unchanged. Maintain original lighting and composition."

Template 3 — Multi-Image Fusion (Nano Banana's Strength):

Use for: outfit swaps, scene changes, character compositing.

Example: "Use the first image as the character reference, the second image for outfit design, and the third image as background. Blend them naturally into a realistic scene."

Template 4 — Poster Text Generation:

Text must be in quotes. Specify font and position.

Example: "Create a poster with the text 'SUMMER SALE' in bold white font, centered on a bright orange background."

Template 5 — Professional Photography Grade:

Add lens, lighting, depth of field, and color grading.

Example: "Portrait shot with a 50mm lens, shallow depth of field, soft cinematic lighting, warm color grading."

4.3 Advanced Tips

· Use sentences, not keywords — "A car driving through a rainy city street at night" vastly outperforms "car, night, neon, rain"

· Optimize in steps — Establish the scene first, then add details; this is more stable than writing everything at once

· State exclusions explicitly — "no text, no watermark, no logo" — not writing it ≠ not getting it

· Think like a photographer — wide shot / close-up / depth of field / cinematic lighting

· Precise editing — "Keep the face and pose unchanged. Replace the outfit with a black leather jacket."

 

Chapter 5: Grok Imagine Prompt Writing Techniques

Grok Imagine is xAI's AI image generation model, excelling at photorealistic photography, cinematic scenes, and creative concept art.

5.1 Core Principle: Write a Concise Creative Brief

Grok Imagine works best with a "creative brief" approach — specify subject, environment, action, style, camera, lighting, mood, details, and quality expectations.

Formula:

Subject: [main subject]
Environment: [setting]
Action: [behavior/activity]
Style: [photorealistic / cinematic / editorial / anime]
Camera: [close-up / portrait / wide shot / aerial]
Lighting: [soft daylight / golden hour / studio / neon]
Mood: [luxury / dramatic / cheerful / futuristic]
Details: [materials, textures, colors]
Quality: highly detailed, realistic composition, professional quality

5.2 Copy-Ready Examples

· A luxury perfume bottle on polished marble, golden hour lighting, soft reflections, cinematic commercial photography, shallow depth of field, high-end beauty ad.

· A futuristic electric sports car driving through neon-lit city streets at night, reflections on wet road surface, dynamic motion blur, photorealistic cinematic style.

· A professional chef preparing sushi in an upscale restaurant kitchen, natural lighting, authentic food textures, documentary photography style.

· A mountain lake at sunrise, mist rolling across the water, hyper-realistic landscape photography, vivid natural colors.

· Fashion portrait: model walking through a luxury shopping district, soft sunlight, magazine-grade photography, authentic fabric details.

5.3 Key Techniques

· Maintain one primary subject and one visual priority — don't crowd too many focal points into a single image

· If the output fails, change only one variable at a time (lighting, lens, background, material)

· Grok Imagine responds extremely well to photography terminology: focal length, aperture, and light type all significantly affect results

· Ideal for visual exploration and pre-production planning — generate concept images first, then refine

 

Chapter 6: GPT Image Prompt Writing Techniques

The GPT Image series on Viddo.ai includes GPT Image 2, GPT 1.5 Image, and GPT 4o. GPT Image 2 is the 2026 latest version and represents a paradigm shift in prompt engineering.

6.1 GPT Image 2: From Tag Dumping to Task Description

GPT Image 2 uses an autoregressive architecture, meaning its prompt strategy is fundamentally different from diffusion models:

 

6.2 GPT Image 2 Five-Layer Structure

Task type + Subject anchor + Structural constraints + Lighting/material/color + Text & language requirements + Retention / exclusion items

Full Template:

Create a [image type] for [use case].
Main subject: [specific subject and visible details].
Exact text (if any): "[copy that must appear]".
Composition: [framing, layout, negative space, subject placement].
Style and lighting: [visual language, medium, mood, light direction].
Constraints: [what cannot change, no extra text, no watermark].
Output format: [aspect ratio, transparent background, etc.].

6.3 Three Unique Techniques

Technique: Name the work type before the style.

Start with the output type: poster, product ad, app screenshot, character reference sheet. GPT Image 2 performs better when it knows the success criteria.

Technique: Treat text as locked assets.

If the image needs text, wrap it in quotes and specify font, color, and position. Don't ask it to "write a slogan" — unless you want the model to invent one.

Technique: Give the model camera and layout instructions.

Close-up / wide angle / overhead, left third / right third, clean grid — GPT Image 2 can follow composition instructions precisely.

 

Chapter 7: Seedream 4.5 Prompt Writing Techniques

Seedream 4.5 is ByteDance's AI image generation model, excelling at photorealistic portraits, cinematic scenes, and e-commerce product images. It natively supports Chinese prompts.

7.1 Core Principle: Order by Priority

Seedream 4.5 has its own "comprehension order" — what you place first carries the highest weight:

Subject → Style → Composition → Lighting & Atmosphere → Camera / Technical Details

Follow this sequence and the model "grasps your intent" more reliably.

7.2 Five Core Techniques

· Keep prompts between 30–100 words — too short is vague, too long causes internal conflicts

· Only describe what matters — pick 3-5 key adjectives; don't dump 20

· Use negative prompts to "fix bugs" — "no extra limbs," "no blurry details," "no cluttered background"

· Place the subject first — the first 5-8 words are critical

· Change only one variable per iteration — altering style + lighting + composition simultaneously makes it impossible to identify what worked

7.3 Ten Ready-to-Use Templates

1. Clean Portrait: Young woman, photorealistic portrait under soft natural light, studio portrait style, 85mm lens, shallow depth of field, restrained expression, clean soft-focus background.

2. Fashion Editorial: Male model full-body shot, minimalist street background, fashion editorial style, strong sunlight and shadows, 50mm lens, clear fabric texture.

3. E-Commerce Product: A matte black water bottle, pure white seamless background, commercial product photography, soft studio lighting, ultra-clear details, slight overhead angle.

4. Cinematic Scene: A lone warrior standing in a rain-soaked neon alley at night, cinematography style, 35mm lens, cold blue key light, neon reflections in puddles on the ground.

5. Fantasy Character: Elven knight, silver armor with glowing runes, fantasy concept art style, full-body composition, soft divine light streaming through, intricate metalwork details.

6. Anime: Anime girl, long blue hair, cel shading style, clean linework, bright but not overexposed colors, soft pastel background.

7. Environment Scene: An ancient castle floating above a sea of clouds, wide-angle establishing shot, concept art matte painting style, sunrise backlight.

8. Macro Hyperrealism: Macro photography of a green leaf with dewdrops, extreme close-up, visible texture details, natural soft light.

9. Brand Visual: Minimalist brand visual concept, geometric shapes, modern graphic design style, bold color palette, soft shadows, centered composition.

10. Mixed Media: Double exposure portrait: male silhouette merged with forest landscape, art photography style, black and white or low-saturation monochrome, high contrast.

 

Chapter 8: Midjourney Prompt Writing Techniques

Midjourney is currently the most artistically expressive AI image generator, excelling at fantasy, sci-fi, concept design, and cinematic scenes.

8.1 Syntax Structure

[Subject description], [Environment/scene], [Style/medium], [Lighting/mood], [Composition/camera] --ar [ratio] --v [version] --style [style]

Example: "ethereal forest spirit, bioluminescent flora, cinematic lighting, concept art --ar 3:2 --v 6.1 --style raw"

8.2 Core Parameter Quick Reference

 

8.3 Writing Techniques

· Place the most important visual elements first — Midjourney assigns stronger weight to words at the beginning

· Use rich descriptive adjectives: ethereal, dramatic, moody, luminous, intricate

· Always set --ar to match your output purpose (1:1 for social media, 16:9 for banners)

· Add --style raw for more faithful literal interpretation

· Use --chaos 20-40 when exploring new concepts; reduce to 0-10 once the direction is set

· Use --sref and --cref to maintain consistency across a series

 

Chapter 9: Universal Advanced Techniques — Composition, Lighting, and Style Mixing

9.1 Composition Commands

 

9.2 Style Mixing Techniques

Combining two seemingly unrelated styles often produces stunning results:

 

9.3 Emotional Temperature Control

 

Chapter 10: Negative Prompts — Practical Handbook

10.1 Seedream 4.5 Negative Prompts

Seedream 4.5 supports a dedicated negative prompt field. Build yours in categories:

 

10.2 Exclusion Techniques for Natural Language Models

For models like Nano Banana, Grok Imagine, and GPT Image that don't have a dedicated negative field, append exclusions at the end of your prompt in natural language:

"... No text, no watermark, no brand logos. Avoid modern elements. Do not include any people in the background."

Chapter 11: Real-World Case — Same Scene, Five Model Approaches

Using one theme — "Cozy Café on a Rainy Day" — we wrote prompts for all five models on Viddo.ai:

A young woman around 25, wearing a beige knit sweater, sitting at a dark wooden table by the café window. Medium rain falls outside, with droplets slowly trailing down the glass. She holds a pen in her right hand, writing in a leather-bound notebook, the pen tip leaving smooth strokes on the paper. A steaming latte with a clear rosetta sits at the corner of the table. The background features warm-toned brick walls and rows of dark brown wooden bookshelves filled with old books. A warm pendant lamp casts soft light and shadow on her face. The overall color palette is warm, the atmosphere cozy and serene, with a sense of healing.

11.1 Nano Banana Version

A young woman in her mid-twenties, wearing a soft and cozy beige knit sweater, sitting by a large rain-streaked window inside a warm café. She holds a pen in her right hand, writing in a leather-bound diary, the pen tip gliding smoothly across the paper. A steaming latte with delicate latte art sits at one corner of the dark wooden table. Outside the window, rain continues to trail down the glass, blurring the gray city streets beyond. Inside, the walls are exposed brick, and dark wooden bookshelves are filled with old books. A warm-toned pendant lamp casts a soft golden glow on her face. Medium shot with a slightly overhead angle, shallow depth of field, focus on the rain-streaked window and the woman's face; warm and soft color palette, cinematic soft light, photorealistic photography style. No text, no watermark.

 

11.2 Grok Imagine Version

Subject: Young woman in her mid-twenties, wearing a beige knit sweater, holding a pen and writing in a diary
Environment: Cozy café interior, exposed brick walls, dark wooden bookshelves filled with old books; a large rain-streaked window beside her, with gray city streets visible through the wet glass
Action: Sitting at a dark wooden table, writing in a leather-bound diary; a steaming latte with clear latte art at the corner of the table
Style: Photorealistic, cinematic
Camera: Medium shot, slightly overhead angle, shallow depth of field
Lighting: Warm pendant lamp casting soft golden light on her face; warm indoor ambient light contrasting with the cool blue-gray daylight filtering through the rainy window
Mood: Cozy, serene, healing, atmospheric
Details: Rain trailing down glass panes, pen tip gliding on paper, steam rising from the latte, soft texture of the knit sweater, spine textures of old books on shelves, leather diary cover texture
Quality: Rich detail, realistic composition, professional image quality, warm color grading, soft bokeh between rain scene and background

 

11.3 GPT Image 2 Version

Create a lifestyle photography image.

Main subject: A young woman around 25, wearing a beige knit sweater, sitting at a dark wooden table by the café window. She holds a pen in her right hand, writing in a leather-bound notebook. A steaming latte with a clear rosetta sits at the corner of the table.

Composition: Medium shot, slightly overhead angle, subject positioned at the right third of the frame, with visual space left on the left for the rainy window. Shallow depth of field, focus on the face and the rain-trailing window glass, with the bookshelf naturally blurred in the background.

Style and lighting: Photorealistic photography style, warm pendant lamp as the main light source, casting soft shadows on the subject's face. Outside the window, blue-gray rainy-day light creates a warm-indoor vs. cool-outdoor contrast. Overall warm palette with subtle cinematic color grading.

Environment details: Background features warm-toned brick walls and dark brown wooden bookshelves filled with old books. Outside the window, rain trails slowly down the glass, with a blurred gray cityscape faintly visible. The pen tip's sheen on the paper surface is clearly rendered.

Constraints: No text, watermarks, or logos in the image. No extra decorative elements. Style must be consistently photorealistic — no cartoon or over-stylized rendering.

 

11.4 Seedream 4.5 Version

Young woman, beige knit sweater, sitting by a café window, writing with a pen in a leather notebook, photorealistic photography style, medium shot composition, slightly overhead angle. Rainy day outside the window, rain trailing down the glass, warm indoor pendant lamp lighting, shallow depth of field, warm color palette, soft cinematic light and shadow, brick wall and wooden bookshelves in the background, a steaming latte at the corner of the table, cozy and serene atmosphere.

Negative prompt: No extra limbs, clear face not blurry, no deformed hands, no cluttered background, no text, no watermark, no logos, no cartoon style, colors not oversaturated

 

11.5 Midjourney Version

A young woman in a beige knit sweater, writing in her diary by a rain-streaked window inside a cozy café; leather-bound diary open on the table, a steaming latte with latte art at the table corner. Warm pendant lamp casting a golden glow, exposed brick walls and dark wooden bookshelves filled with old books in the background. Rain trailing down the glass window, gray cityscape outside blurred and hazy. Shallow depth of field, warm color grading, soft cinematic lighting, quiet and healing intimate atmosphere, photorealistic style --style raw

 

Chapter 12: The Iterative Prompt Optimization Workflow

Professional AI image creators never expect to write the perfect prompt on the first try. Iterative optimization is the standard workflow:

Step 1 — Rough Draft: Write a first-version prompt using the 5-layer formula. Generate 4-8 images. Select the base image closest to your vision.

Step 2 — Problem Diagnosis: Analyze the base image's shortcomings: Is the composition off? Lighting wrong? Style drifted? Extra elements? Record each issue.

Step 3 — Precision Adjustment: Change only one dimension per round (e.g., only lighting, or only composition). Avoid multiple simultaneous changes.

Step 4 — Style Lock-in: Once satisfied, use --sref/--cref (Midjourney) or reference images (Nano Banana) to lock the style and batch-produce.

Viddo.ai Workflow Tip: On Viddo.ai, you can quickly switch between models to compare results and generate batch variations. We recommend generating at least 4 variations per iteration, then selecting the best result for the next round of refinement.

 

Chapter 13: Common Mistakes and How to Avoid Them

 

Conclusion

Writing AI image prompts is a craft that blends visual aesthetics, language engineering, and model understanding. There's no single "magic prompt" that works for every scenario — but once you've mastered the 5-layer formula, the model-specific techniques, and the iterative methods in this guide, you're equipped to produce high-quality output on any model on Viddo.ai.

Three core principles to remember:

1. Precision beats vagueness — Every specific descriptor narrows the AI's imagination space.

2. Each model has its own personality — Understanding your model's "preferred language" matters more than blindly stacking keywords.

3. Iteration is the norm — Professional creators follow a cycle of "generate → diagnose → adjust → regenerate."

 

Start creating on Viddo.ai — apply the prompt techniques from this tutorial to kick off your AI creation journey.