windy/my-vault

Fork 0

Files

T

windyboy 0e4d21f721 vault backup: 2026-01-08 09:34:22

2026-01-08 09:34:22 +08:00

9.2 KiB

Raw Blame History

title, created, tags, source

title

created

AI Image Generation Prompt Engineering Tips

A comprehensive guide to crafting effective prompts for AI image generation models (Nano Banana Pro/Gemini/Imagen 2).

Core Principles

1. Be Specific and Detailed

Use precise technical parameters rather than vague descriptions
Specify exact camera models, lens types, and settings when relevant
Include lighting direction, quality, and temperature
Define exact aspect ratios and resolutions

2. Structure Your Prompts

Use clear sections for complex requests:

Subject - What/who is in the image
Environment - Setting, location, background
Lighting - Type, direction, mood
Camera/Style - Technical specs, artistic style
Details - Textures, colors, specific elements

3. Reference Real-World Standards

Mention specific film stocks (e.g., "Kodak Portra 400")
Reference camera models (e.g., "Sony A7III with 85mm f/1.4")
Cite artistic styles or movements
Use professional terminology (bokeh, depth of field, golden hour)

Technical Parameters

Camera Settings

- Camera: [Model] (e.g., Canon EOS R5, Hasselblad H6D-100c)
- Lens: [Focal length] [Aperture] (e.g., 85mm f/1.4, 35mm f/2.8)
- Aperture: f/1.8 to f/5.6 (shallow DoF) or f/8+ (deep DoF)
- ISO: 100-400 (clean), 800-1600 (grainy/documentary)
- Shutter speed: 1/60s (standard), 1/125s+ (action)

Lighting Specifications

- Type: Natural light, studio lighting, flash, ambient
- Direction: Key light, fill light, rim light, backlighting
- Quality: Soft/diffused vs. hard/direct
- Temperature: Warm (golden hour) vs. cool (blue hour)
- Time of day: Golden hour (sunset/sunrise), midday, blue hour

Style References

- Photography eras: 1990s digital camera, 2000s flash, film aesthetic
- Film stocks: Kodak Portra 400, Kodak Ektar 100, Fuji Velvia
- Artistic styles: Cinematic, editorial, documentary, fashion
- Processing: Vintage grain, clean digital, film texture

Consistency Techniques

Face/Identity Preservation

Key phrases for maintaining facial features:
- "Keep the facial features of the person exactly consistent"
- "Preserve original face 100% accurate from reference image"
- "Do not change the face, maintain exact facial structure"
- "Face consistency: preserve_original: true"

Texture Preservation

For maintaining surface textures:
- "Preserve the original fabric texture, color, and logos"
- "Maintain the aged, greasy, textured look"
- "Keep the same grain, focus depth, and lighting"
- "Match ambient lighting, color temperature, shadow direction"

JSON Format for Complex Prompts

For multi-parameter requests, use structured JSON:

{
  "subject": {
    "description": "Young woman...",
    "age": "early 20s",
    "expression": "confident and playful",
    "hair": {
      "color": "dark",
      "style": "long, voluminous waves"
    }
  },
  "photography": {
    "camera_style": "early-2000s digital camera aesthetic",
    "lighting": "harsh super-flash with bright highlights",
    "angle": "mirror selfie",
    "texture": "subtle grain, retro highlights"
  },
  "environment": {
    "setting": "bedroom",
    "elements": ["dresser", "posters", "vanity"]
  }
}

Category-Specific Tips

Portrait Photography

- Specify lens compression (85mm for flattering portraits)
- Define depth of field (shallow for portraits, deeper for groups)
- Include catchlights in eyes
- Mention skin texture preference (natural pores vs. smooth)
- Specify makeup and styling details

Product Photography

- Use "pure white background (RGB 255, 255, 255)"
- Specify "soft studio lighting, even illumination"
- Request "subtle contact shadow at base"
- Include "no harsh glare, photorealistic rendering"
- Define "high-resolution, 8k quality"

Vintage/Retro Aesthetics

- Reference specific years (e.g., "early-2000s aesthetic")
- Mention "film grain, authentic texture"
- Include era-specific elements (fashion, technology, decor)
- Specify "flash photography" or "disposable camera look"
- Use "nostalgic, vintage color grading"

3D/Isometric Renders

- Specify "Cinema 4D rendering" or "3D isometric view"
- Include "soft studio lighting, clean materials"
- Define "miniature diorama style" or "architectural visualization"
- Mention "rounded forms, pastel colors" for cute aesthetics
- Request "blind-box toy aesthetic" for collectible styles

Advanced Techniques

Multi-Image Consistency

For generating multiple related images:

"Create [number] images with:
- Same subject with exact facial features preserved
- Same lighting setup and color grading
- Same environment and props
- Different poses/angles only
- Maintain consistent wardrobe and styling"

Image-to-Image Translation

"Transform the uploaded image by:
- Preserving [specific elements]: face, pose, composition
- Changing [target elements]: background, lighting, style
- Maintaining [technical aspects]: resolution, aspect ratio
- Matching [aesthetic]: color palette, mood, era"

Text Integration

For adding text to images:
- "Overlay text in [font style]: bold, serif, handwritten"
- "Place text at [location] with [color] and [effects]"
- "Ensure perfect spelling and centered alignment"
- "Use drop shadow and outline for readability"
- "Integrate text naturally into image depth of field"

Background Manipulation

Removal: "Remove all people/objects in background, fill with..."
Extension: "Expand to 16:9, extend scenery naturally on both sides"
Replacement: "Replace background with [description] while preserving subject"
Context-aware: "Match original lighting, weather, and texture perfectly"

Common Pitfalls to Avoid

❌ Vague Descriptions

❌ "Make it look nice"
✅ "Soft diffused natural lighting, shallow depth of field, warm color grading"

❌ Missing Technical Context

❌ "A portrait photo"
✅ "Professional headshot, Sony A7III, 85mm f/1.8, three-point lighting"

❌ Ignoring Consistency

❌ "Multiple photos of the same person"
✅ "Same person, exact facial features preserved, different poses only"

❌ Unclear Hierarchy

❌ Long paragraph with everything mixed together
✅ Structured sections: Subject → Environment → Style → Technical specs

Workflow Strategies

1. Start Broad, Then Refine

First pass: "Portrait of a young woman, outdoor setting"
Refined: "Portrait of a young woman, 85mm lens, shallow DoF,
golden hour lighting, urban park background, soft focus bokeh"

2. Use Reference Layering

Base: "Fashion photography"
+ Style: "in the style of Annie Leibovitz"
+ Technical: "shot on Hasselblad H6D, 120mm macro lens"
+ Mood: "editorial, high-fashion, dramatic lighting"

3. Build Template Libraries

Create reusable templates for common needs:

Professional headshots
Product photography
Social media content
Vintage aesthetics
3D renders

Negative Prompts

What to exclude (if supported):

Common negatives:
- "no distorted faces, no extra limbs"
- "no blurry, no low quality, no artifacts"
- "no watermarks, no text overlays"
- "no unrealistic proportions"
- "no modern elements" (for vintage styles)

Quality Indicators

Add these for best results:

- "8K resolution, ultra-detailed"
- "Photorealistic, high-fidelity"
- "Professional photography, award-winning"
- "Sharp focus, crisp details"
- "Cinema-quality rendering"

Quick Reference Cheat Sheet

Portrait

Product

Vintage

3D Render

Resources

Official Guide: Google Prompting Tips
Advanced Guide: How to prompt Nano Banana Pro
Source Repository: ZeroLuawesome-nanobanana-pro 🚀 An awesome list of curated Nano Banana pro prompts and examples. Your go-to resource for mastering prompt engineering and exploring the creative potential of the Nano banana pro(Nano banana 2) AI image model.

Practice Exercises

Upgrade a basic prompt: Transform "photo of a cat" into a detailed, technical prompt
Style transfer: Take a prompt and adapt it for different eras (1980s → 2000s → modern)
Multi-parameter challenge: Create a JSON prompt with 5+ nested parameters
Consistency test: Write prompts for 3 related images maintaining exact subject consistency

Last updated: 2026-01-07 Source material: awesome-nanobanana-pro GitHub repository

9.2 KiB Raw Blame History