AI Image Generation Cheatsheet

Prompt Structure Formula

Basic formula: [Subject] + [Style] + [Setting] + [Lighting] + [Composition] + [Quality modifiers]

Example: "A golden retriever wearing a bowtie, watercolor painting style, sitting in a sunlit garden, soft warm lighting, centered composition, highly detailed"

Front-load the most important elements — AI image models weight the beginning of prompts more heavily.

Be specific about what you want, not what you don't want. "Red sports car" works better than "car, not blue, not sedan."

Include the medium: "oil painting", "digital illustration", "35mm photograph", "pencil sketch", "3D render".

Specify the mood: "moody", "cheerful", "dramatic", "serene", "dystopian", "whimsical".

Style Keywords That Work

Photorealistic: "DSLR photo", "35mm film", "Canon EOS R5", "shallow depth of field", "bokeh background"

Illustration: "digital art", "concept art", "character design", "flat illustration", "vector art"

Fine art: "oil painting", "watercolor", "impressionist", "art nouveau", "baroque", "renaissance"

Modern/graphic: "minimalist", "geometric", "isometric", "low poly", "vaporwave", "synthwave"

Quality boosters: "highly detailed", "8K resolution", "professional", "masterpiece", "award-winning"

Lighting keywords: "golden hour", "studio lighting", "dramatic rim light", "neon glow", "volumetric lighting", "chiaroscuro"

Tool-Specific Tips

DALL-E (ChatGPT): Best for quick iterations and edits. Describe what you want in natural language — it handles conversational prompts well.

Midjourney: Best image quality overall. Use "--ar 16:9" for widescreen, "--v 6" for latest model, "--style raw" for less stylized output.

Stable Diffusion: Most control and customization. Free and open-source. Steeper learning curve but supports ControlNet, LoRA, and inpainting.

Adobe Firefly: Best for commercial use — trained on licensed content. Integrates with Photoshop and Illustrator.

Leonardo.ai: Good free tier. Strong at consistent character design and game art styles.

Ideogram: Best for text in images — other models struggle to render readable text.

Common Use Cases

Blog and article headers: "Minimalist flat illustration of [topic], clean white background, modern color palette, suitable for a tech blog header"

Social media posts: "Eye-catching Instagram post graphic about [topic], bold typography area at top, vibrant colors, square format"

Product mockups: "Professional product photo of [item] on a marble surface, soft studio lighting, lifestyle setting, shallow depth of field"

Presentation slides: "Clean corporate illustration of [concept], isometric style, blue and white color scheme, no text, transparent background friendly"

Avatar/profile images: "Professional headshot illustration, [description], digital painting style, neutral background, warm lighting"

Icon sets: "Set of 6 matching flat icons for [category], consistent line weight, monochrome [color], minimal style, white background"

Common Mistakes to Avoid

Too vague: "a nice picture" gives random results. Be specific about subject, style, and mood.

Too complex: cramming 20 elements into one prompt creates chaos. Focus on 3-5 key elements.

Ignoring aspect ratio: default square format does not work for all uses. Specify landscape (16:9) or portrait (9:16) when needed.

Expecting perfect text: AI image generators still struggle with readable text. Add text in a design tool afterward.

Not iterating: your first generation is rarely the final one. Refine prompts based on what the AI gets wrong.

Copyright concerns: understand each tool's licensing before using AI images commercially. Some tools have clearer commercial rights than others.

Advanced Techniques

Image-to-image: Upload a reference image and ask the AI to generate variations or transform the style.

Inpainting: Edit specific parts of an image while keeping the rest intact. Great for fixing hands, faces, or backgrounds.

Consistent characters: Use detailed character descriptions saved as templates. Include specific features, clothing, and style in every prompt.

Negative prompts (Stable Diffusion/Midjourney): "--no blurry, distorted, low quality" helps avoid common artifacts.

Seed numbers: Use the same seed to create variations with consistent style. Useful for series or collections.

Upscaling: Generate at base resolution, then upscale with the model or dedicated tools like Topaz for print-ready quality.

Create stunning images with AI using the right prompts. This cheatsheet covers prompt formulas, style keywords, common mistakes, and tool-specific tips for DALL-E, Midjourney, Stable Diffusion, and other popular image generators.