10 Tips for Writing Better AI Image Prompts

The prompt is everything. A great prompt turns a mediocre AI image into a stunning one. Here are 10 techniques we've learned from generating thousands of images.

1. Start with the Subject, Then Add Details

Structure your prompt from most important to least important. Start with what you want, then add how it should look.

❌ Weak: "A beautiful scene with nice lighting and a cat"

✅ Strong: "A tabby cat sitting on a windowsill, golden hour sunlight streaming through sheer curtains, soft bokeh background, warm tones"

2. Specify the Medium and Style

Tell the AI what kind of image you want. "Photo", "oil painting", "3D render", "vector illustration", "pencil sketch" — these style keywords dramatically change the output.

Example: "A medieval castle on a cliff, watercolor painting style, soft washes of color, loose brushstrokes"

3. Describe Lighting Explicitly

Lighting makes or breaks an image. Don't leave it to chance.

• Golden hour — warm, directional, long shadows
• Studio lighting — clean, professional, controlled
• Dramatic side lighting — high contrast, moody
• Backlit / rim lighting — silhouettes, glowing edges
• Neon lighting — colorful, cyberpunk, vibrant

4. Use Camera and Lens Terms

AI models understand photography terminology. Use it to control perspective and depth.

• Wide-angle lens — dramatic perspective, expansive scenes
• 85mm portrait lens — flattering compression, shallow depth of field
• Macro photography — extreme close-up, tiny details
• Aerial / drone shot — top-down or elevated perspective
• Tilt-shift — miniature effect

5. Include Negative Context

Sometimes saying what you don't want is as important as what you do. While not all models support formal negative prompts, you can use phrasing like:

Example: "A clean, professional headshot, plain white background, no distracting elements, no text"

6. Reference Specific Art Styles and Eras

Be specific about artistic influences:

• "Art Nouveau poster style" (Mucha-like organic forms)
• "Bauhaus geometric composition"
• "Ukiyo-e woodblock print style"
• "1980s retro synthwave aesthetic"
• "Scandinavian minimalist design"

7. Specify Colors and Mood

Don't rely on the model to guess your color palette.

Example: "An autumn forest path, deep oranges and burnt reds, misty atmosphere, melancholic mood, shot on medium format film"

8. Use Concrete Details Instead of Abstractions

Replace vague words with specific ones:

• ❌ "beautiful" → ✅ "symmetrical, well-composed, vibrant colors"
• ❌ "nice house" → ✅ "Victorian townhouse with bay windows and a wrought-iron fence"
• ❌ "cool car" → ✅ "1967 Ford Mustang Shelby GT500, metallic blue, low angle shot"

9. Iterate and Refine

Your first prompt is rarely your best. Generate, evaluate, and adjust:

1. Start with a basic prompt to test the concept
2. Identify what's missing or off
3. Add or modify specific details
4. Use ImageForge's refinement feature to tweak the result further

On ImageForge, you can also try the same prompt across multiple models to see which interpretation you prefer.

10. Match Model to Task

Different models interpret prompts differently. Use the right tool:

• Need text in the image? → Ideogram V4, FLUX.2 [dev], or Recraft V4
• Creative illustration? → Seedream 5 or Grok Imagine 2.0
• Clean, fast result? → Z-Image Turbo or FLUX.2 [klein] 9B
• Vector art or logos? → Recraft V4
• Maximum quality? → FLUX.2 [max] or Nano Banana Pro
• Photorealistic scenes? → Hunyuan Image 3.0, Kling Image O3, or FLUX.2 [pro]

Bonus: Prompt Template

Here's a template that works well across most models:

[Subject] in [setting/environment], [style/medium], [lighting], [mood/atmosphere], [technical details]

Example: "A wise old owl perched on a stack of ancient books in a candlelit library, digital painting, warm amber lighting, mysterious atmosphere, highly detailed, 4K"