Skip to content

Instantly share code, notes, and snippets.

Show Gist options
  • Select an option

  • Save horushe93/046d7e9882ecdf13335ff2108e89e135 to your computer and use it in GitHub Desktop.

Select an option

Save horushe93/046d7e9882ecdf13335ff2108e89e135 to your computer and use it in GitHub Desktop.
AI Image Prompt Engineering Cheat Sheet (2026)

AI Image Prompt Engineering Cheat Sheet (2026)

Last updated: February 2026 | A practical reference for writing prompts that work across every major AI image model.

Writing a good prompt is the difference between a throwaway image and something you'd actually use. This cheat sheet breaks down the anatomy of an effective prompt, covers model-specific syntax, and gives you ready-to-paste templates for the most common use cases.


The Universal Prompt Formula

Every strong AI image prompt follows the same basic structure. Not every field is required — but the more specific you are, the closer the output matches what you had in mind.

[Subject] + [Style/Medium] + [Lighting] + [Composition/Camera] + [Color/Mood] + [Details]
Element What to specify Example keywords
Subject The main focus of the image. Use concrete nouns — avoid abstract concepts. a golden retriever, a ceramic coffee mug, a woman in her 30s
Style / Medium Art style or visual medium. oil painting, 35mm film photography, watercolor, 3D render, anime, cyberpunk
Lighting How the scene is lit. This single element has the biggest impact on realism. soft diffused light, golden hour, rim light, chiaroscuro, studio lighting, neon glow
Composition / Camera Framing and perspective. close-up, wide shot, bird's eye view, shallow depth of field, macro, low angle
Color / Mood Palette and emotional tone. warm earth tones, muted pastels, high contrast, monochrome, vibrant saturated
Details Anything else that matters — texture, background, specific objects. on a marble countertop, with rain streaks on the window, wearing a denim jacket

Quick examples using the formula

Product shot:

A minimalist white sneaker on a pure white background, studio lighting with soft
shadows, front three-quarter angle, clean edges, commercial product photography

YouTube thumbnail:

A man with an exaggerated surprised expression, mouth open, pointing at a glowing
laptop screen, bold vibrant colors, close-up portrait, high contrast, energetic mood

Fantasy illustration:

A lone knight standing at the edge of a cliff overlooking a vast canyon filled with
clouds, golden hour light from behind, oil painting style, muted warm palette,
cinematic wide shot

Lighting Reference

Lighting is the most underrated part of a prompt. Default AI lighting tends to be flat and generic. Adding one or two lighting keywords dramatically improves output quality.

Keyword Effect Best for
studio lighting Even, controlled, professional Product photos, portraits
soft diffused light Gentle, no harsh shadows Lifestyle, fashion
golden hour Warm, directional, long shadows Outdoor scenes, portraits
rim light Bright edge outline separating subject from background Dramatic portraits, product hero shots
chiaroscuro Strong contrast between light and dark Moody portraits, fine art
volumetric lighting Visible light rays through atmosphere Fantasy, cinematic scenes
neon glow Colored artificial light sources Cyberpunk, nightlife, sci-fi
overcast / flat light Soft, even, no directional shadows Documentary, street photography
backlight / silhouette Subject dark against bright background Dramatic, editorial
candlelight / firelight Warm, flickering, intimate Cozy scenes, period settings

Camera & Composition Reference

Keyword Effect
close-up Tight framing on face or detail
extreme close-up / macro Very tight — texture, eyes, small objects
medium shot Waist-up framing
wide shot / establishing shot Full scene with environment
bird's eye view Looking straight down
low angle Looking up at the subject (makes things look powerful)
Dutch angle Tilted frame (tension, unease)
over-the-shoulder POV from behind another person
shallow depth of field Sharp subject, blurred background (bokeh)
deep focus Everything sharp from foreground to background
rule of thirds Subject placed off-center
symmetrical composition Centered, balanced framing
leading lines Lines in scene drawing the eye to the subject

Art Style Reference

Category Keywords
Photorealistic photorealistic, 35mm film, DSLR, RAW photo, hyperrealistic
Illustration digital illustration, concept art, matte painting, children's book illustration
Painting oil painting, watercolor, acrylic, impressionism, expressionism
Anime / Manga anime style, manga, chibi, Studio Ghibli style, cel shading
3D 3D render, Pixar style, isometric, clay render, unreal engine
Retro retro, vintage, polaroid, 1970s aesthetic, vaporwave, synthwave
Graphic Design flat design, vector art, pop art, minimalist, infographic style
Dark / Moody dark fantasy, gothic, noir, cyberpunk, dystopian

Model-Specific Tips

Different models respond to prompts differently. Here's what works best on each major platform in 2026.

Gemini (Nano Banana / Nano Banana Pro)

Gemini models work well with natural language — you can write prompts like you're talking to a person. Nano Banana Pro is especially strong at following complex, multi-step instructions because it reasons about the composition before generating.

What works:

  • Conversational, paragraph-style prompts
  • Multi-turn refinement ("make the background warmer", "add a second person on the right")
  • Explicit text instructions ("the sign reads 'OPEN 24 HOURS'")
  • Reference images — upload up to 8 images for character/style consistency

Syntax notes:

  • Specify resolution: output in 4K or generate at 2K resolution
  • Aspect ratio: 16:9 landscape or 9:16 vertical
  • For best text rendering, use Nano Banana Pro with thinking mode enabled

Template — e-commerce product shot:

Professional product photo of [PRODUCT] centered on a pure white background.
Studio lighting with soft, even illumination and subtle shadows beneath the product.
The product fills approximately 85% of the frame. Sharp focus, clean edges,
no props or text. Output at 4K resolution, 1:1 square aspect ratio.

Template — social media post:

Eye-catching social media graphic for [PLATFORM]. Show [SUBJECT/SCENE] with
[MOOD] energy. Bold, saturated colors that pop on a mobile screen. Include the
text "[YOUR HEADLINE]" in large, clean sans-serif font at the top. 4:5 portrait
aspect ratio.

Template — illustration with text:

A colorful infographic-style illustration explaining [TOPIC]. Include labeled
sections with the following text: "[LABEL 1]", "[LABEL 2]", "[LABEL 3]".
Use a clean, modern flat design style with a light background. Text should be
crisp and perfectly readable. 16:9 landscape format.

Midjourney (v7 / v8)

Midjourney excels at mood, atmosphere, and artistic style. It prefers short, punchy prompts over long paragraphs. Think of it like a mood board — evocative keywords work better than detailed instructions.

What works:

  • Short, comma-separated keyword phrases
  • Artistic and emotional descriptors ("ethereal", "brooding", "whimsical")
  • Style references using --sref and character references using --cref
  • Negative prompts via --no to exclude unwanted elements

What doesn't work:

  • Text rendering — Midjourney still struggles with readable text in images
  • Overly long, complex instructions

Key parameters:

--ar 16:9       Aspect ratio
--v 7           Model version
--s 250         Stylization (0-1000, higher = more artistic interpretation)
--c 15          Chaos (0-100, higher = more variation between outputs)
--q 2           Quality (1 or 2, higher = more detail, slower)
--no trees      Exclude specific elements
--sref [URL]    Style reference image
--cref [URL]    Character reference image
--cw 50         Character weight (0-100, lower = face only, higher = full appearance)

Template — cinematic scene:

A lone astronaut standing on a crimson desert planet, two moons on the horizon,
volumetric dust particles, cinematic widescreen, muted sci-fi palette --ar 21:9
--v 7 --s 400

Template — portrait:

Portrait of an elderly fisherman, deep weathered skin, kind eyes, soft morning
light, shallow depth of field, 35mm film grain --ar 4:5 --v 7 --s 200

Template — stylized brand asset:

Isometric 3D illustration of a cozy home office, warm afternoon light through
window, plants, books, coffee cup, soft pastel colors, clean vector aesthetic
--ar 1:1 --v 7 --s 300 --no people text

DALL-E / GPT Image (via ChatGPT)

GPT Image (the model that replaced DALL-E 3 in ChatGPT) responds well to detailed, natural-language descriptions. Its biggest strength is following precise instructions and rendering readable text.

What works:

  • Detailed, descriptive paragraphs
  • Explicit text content ("the poster says...")
  • Iterative editing through conversation
  • Combining generation with editing in the same session

Template — poster with text:

Design a modern event poster for a music festival called "AURORA 2026". The poster
features a gradient sky transitioning from deep purple at the top to coral at the
bottom, with geometric mountain silhouettes. The festival name "AURORA 2026" is
displayed in bold, clean sans-serif typography at the center. Below it reads
"June 15-17 | Riverside Park". Style: modern minimalist with a premium feel.

Template — product lifestyle:

A lifestyle photograph of [PRODUCT] being used naturally in a modern kitchen. Morning
sunlight streaming through a window, creating warm highlights. The product is in sharp
focus while the background has a gentle bokeh. Styled to feel authentic and aspirational,
not overly staged. Shot on a Canon R5, 50mm f/1.4.

Flux (Flux 2 Pro / Flux 2 Max)

Flux is open-source and highly customizable. It handles structured, weighted prompts well and produces excellent photorealistic output. Flux 2 also renders in-image text better than most competitors.

What works:

  • Structured keyword prompts with clear hierarchies
  • Photography-specific language (lens types, film stocks)
  • LoRA models for custom styles and characters
  • Detailed scene descriptions

Template — photorealistic portrait:

Portrait photograph of a young woman with curly auburn hair, wearing a cream
turtleneck sweater, sitting in a sunlit cafe. Shot on Fujifilm X-T5, 56mm f/1.2,
natural window light, shallow depth of field with creamy bokeh. Warm color grade,
subtle film grain. Editorial style.

Template — product on textured surface:

Overhead flat lay of a leather journal, fountain pen, and espresso cup arranged on
a dark slate surface. Dramatic side lighting creating long shadows. Rich warm tones,
high detail on leather texture and coffee crema. Shot at f/8 for deep focus.
Commercial still life photography.

Common Mistakes to Avoid

Mistake Why it fails Fix
Using abstract subjects ("love", "freedom") AI needs concrete visual elements to render Describe a scene that represents the concept
No lighting specified You get flat, default lighting Add at least one lighting keyword
Prompt too long (100+ words) Models lose focus; competing instructions cancel out Keep it under 75 words for Midjourney; up to 150 for Gemini/GPT
Prompt too vague ("a nice photo") AI has nothing specific to work with Add subject, style, lighting, and composition
Conflicting instructions "minimalist" + "intricate ornate details" confuse the model Pick one direction per prompt
Ignoring aspect ratio Default square crops may cut off important elements Always specify ratio for your target platform
Expecting perfect text on the first try Text rendering varies by model; some need iteration Use Nano Banana Pro or Ideogram for text-heavy images; review and refine

Quick-Copy Templates by Use Case

E-Commerce

Professional product photo of a [PRODUCT] on a clean white background. Even studio
lighting, subtle drop shadow, product centered and filling 85% of the frame. Sharp
focus, no reflections, no props. 1:1 square, 4K resolution.
Lifestyle shot of [PRODUCT] in a modern [SETTING — kitchen/office/bedroom]. Natural
daylight from a large window, shallow depth of field with the product in sharp focus.
Styled to look effortless and aspirational. 4:5 portrait ratio.

YouTube Thumbnails

Extreme close-up of a person with a shocked, wide-eyed expression, mouth slightly
open, looking directly at camera. Bold [COLOR] background. High contrast, vibrant
saturation. Space on the right side for text overlay. 16:9 ratio, 1280x720.
Split composition: left side shows [BEFORE STATE], right side shows [AFTER STATE],
divided by a dramatic lightning bolt or slash effect. Bold contrasting colors,
high energy. 16:9 ratio.

Social Media

Instagram-ready flat lay of [ITEMS] arranged neatly on a [SURFACE]. Soft overhead
lighting, subtle shadows, cohesive [COLOR PALETTE] color scheme. Clean, editorial
aesthetic. 4:5 portrait ratio.
TikTok/Reels cover image: [SUBJECT] in a dynamic pose against a [simple/gradient]
background. Bold, saturated colors that pop on mobile. Leave space at top and bottom
for platform UI elements. 9:16 vertical ratio.

Education

Clear educational diagram illustrating [CONCEPT]. Labeled sections with the text:
"[LABEL 1]", "[LABEL 2]", "[LABEL 3]". Clean flat design, high contrast for
readability, light background. All text must be perfectly legible.
16:9 landscape ratio.

Branding

Minimalist logo concept for a brand called "[BRAND NAME]". The design incorporates
[SYMBOL/ELEMENT] in a clean, modern style. Flat vector aesthetic, works on both
light and dark backgrounds. Simple enough to be recognizable at small sizes.
1:1 square ratio.

Resolution & Aspect Ratio Quick Reference

Platform Recommended Size Aspect Ratio
Instagram Feed 1080 x 1350 px 4:5
Instagram Stories / Reels 1080 x 1920 px 9:16
TikTok 1080 x 1920 px 9:16
YouTube Thumbnail 1280 x 720 px 16:9
Facebook Feed 1200 x 630 px ~1.91:1
LinkedIn 1200 x 1200 px 1:1
Pinterest 1000 x 1500 px 2:3
X (Twitter) 1200 x 675 px 16:9
E-Commerce (Amazon main) 2000 x 2000 px 1:1
Print (poster/flyer) 3840 x 2160 px+ Varies

Try It Without the Prompt Engineering

This cheat sheet gives you precise control when you need it. But if writing structured prompts isn't your thing, chat-based AI image generators let you describe what you want in plain language and refine through conversation — no formulas or parameters required.

Banana AI uses this approach: you describe the image, it generates options, and you iterate by chatting ("make it warmer", "remove the background", "add text that says..."). It supports Nano Banana, Nano Banana Pro (4K, accurate text rendering), and Flux Fast across 7 aspect ratio presets.


See Also

  • 50 AI Image Prompts for Every Marketing Scenario — ready-to-use prompt library sorted by use case
  • Social Media Image Size & Aspect Ratio Cheat Sheet (2026) — full platform-by-platform sizing guide
  • AI Image Style Guide: Photography Terms for Better Prompts — deep-dive into lighting, composition, and style vocabulary

Created by the Banana AI team — a chat-based AI image generator built on Nano Banana models.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment