Skip to content

Image Generation Prompts

A comprehensive library of 72 optimized image generation prompt styles across 12 categories, each with multiple contextual prompt variations tailored for real-world use cases. Every style doc provides copy-paste prompts for Nano Banana 2 (featured), ChatGPT, Midjourney, and Stable Diffusion — simply replace [PLACEHOLDERS] with your own content.


If you just want to generate something great right now:

  1. Pick a style from the categories below.
  2. Open the style doc and go to the Nano Banana 2 section (listed first).
  3. Choose the variation that matches your use case (e.g., “Social Media Post,” “Desktop Wallpaper,” “Character Portrait”).
  4. Copy the prompt, replace the [PLACEHOLDERS], and paste into Nano Banana.
  5. Iterate — use follow-up messages to refine colors, composition, lighting, or details.

Have a reference photo? Jump to the Image-to-Image Transformations section to transform existing images into any style.


Each prompt is provided in four platform-specific variants. Understanding their strengths will help you get the best results.

Section titled “🔵 Nano Banana 2 (Featured — Best Overall Quality)”

Nano Banana 2 (powered by Gemini 3.1 Flash Image) is a major upgrade over the original. Treat it as a full-featured creative engine, not a quick-filter tool.

  • Style: Detailed, structured natural language. Front-load the asset type and primary subject, then layer in style, lighting, and composition details.

  • Strengths: Combines pro-model quality with flash-model speed. Supports multi-turn conversational refinement, search grounding (reference real-world subjects by name), accurate text rendering, character consistency (up to 5 characters), and output up to 4K resolution.

  • Why It’s Featured: NB2 consistently produces the most detailed, accurate, and stylistically faithful results across all 57 styles in this library. Its conversational refinement loop means you can iterate toward perfection without re-writing your entire prompt.

  • Key Tips:

    TipDetails
    Be specific, not vagueNB2’s reasoning engine rewards precise descriptors for textures, colors, and lighting over generic adjectives
    Front-load important detailsPlace the asset type and primary subject at the beginning of your prompt
    Specify aspect ratiosExplicitly state the format: 16:9 for wallpapers, 3:4 for Instagram, 9:16 for stories, 1:1 for profile pictures
    Iterate conversationallyRefine with follow-ups: “Change the palette to warmer tones,” “Move the subject left,” “Add more texture to the background”
    Leverage search groundingFor real-world subjects, ask the model to reference accurate imagery: “Use image search to find accurate reference for [specific subject]“
    Use positive framingInstead of “no cars,” describe the scene you do want: “empty, deserted street”
    Specify camera/lensFor photographic styles, include lens (85mm, 35mm, macro), aperture (f/1.8, f/8), and camera angle (low, eye-level, bird’s eye)
    Name your lightingUse specific lighting terms: “Rembrandt lighting,” “golden hour,” “tungsten warm glow,” “volumetric fog”

Replaces the standalone DALL-E 3 model, which is deprecated as of May 2026.

  • Style: Fully conversational, natural language. Describe your scene as if briefing an artist.
  • Strengths: Excellent scene comprehension, precise spatial relationships, and iterative refinement through follow-up messages.
  • Tips: Be specific about subject placement, lighting direction, and mood. You can assign it a persona (e.g., “Act as a professional product photographer”) for higher fidelity. Use follow-up messages to adjust individual elements without re-describing the whole scene.

V7 is the current default model. Parameters like --v 6.0 are no longer needed.

  • Style: Natural language with parameters appended at the end. V7 has moved away from keyword stuffing toward descriptive sentences.

  • Key Parameters:

    ParameterPurposeExample
    --arAspect ratio--ar 16:9, --ar 4:5
    --sStylization (0–1000). Low = literal, high = artistic--s 250
    --noExclude elements (replaces negative prompts)--no text, blur
    --crefCharacter reference — maintains identity across images--cref [URL]
    --srefStyle reference — applies the aesthetic of a reference image--sref [URL]
    --iwImage weight for reference influence (0–2)--iw 1.5
    --chaosVariation in initial grid (0–100)--chaos 30
  • Tips: Place the most important visual information first. Use --niji for anime and illustration styles.

  • Style: Supports both keyword tags and natural language. Modern models (SDXL, SD 3.5) work best with descriptive sentences, structured as Subject → Action → Environment → Style/Lighting.
  • Negative Prompts: Use sparingly and only for specific unwanted elements. Bloated negative prompt lists degrade output quality on modern models. Try generating without negatives first, then add targeted exclusions as needed.
  • Tips: Avoid over-weighting (e.g., (keyword:1.5)) — use it to nudge, not force. Ensure your resolution matches the model’s native aspect ratio.

Quick reference for common output formats. Note that aspect ratio capabilities vary by platform:

  • Nano Banana 2: Fixed presets (16:9, 4:3, 1:1, 3:4, 9:16).
  • ChatGPT: Fixed presets (1:1, 3:2, 2:3).
  • Midjourney & Stable Diffusion: Fully dynamic support for precise pixel dimensions and custom ratios.
Use CaseRatioPixels (Recommended)Notes
Instagram Post1:11080×1080Square feed post
Instagram Portrait4:51080×1350Tallest feed format, maximum real estate
Instagram / TikTok Story9:161080×1920Full-screen vertical
LinkedIn / Twitter Header3:11500×500Wide banner
Desktop Wallpaper16:93840×2160Standard widescreen / 4K
Ultrawide Wallpaper21:93440×1440Ultrawide monitors
Phone Wallpaper9:19.51290×2796iPhone Pro Max
Profile Picture / App Icon1:1512×512 or 1024×1024Square, tight crop
Pinterest Pin2:31000×1500Tall vertical
YouTube Thumbnail16:91280×720Standard video thumbnail
Poster / Print (A-series)1:1.4142480×3508A4 at 300 DPI
Book Cover2:31600×2400Standard paperback

🖼️ Using Reference Images (Image-to-Image)

Section titled “🖼️ Using Reference Images (Image-to-Image)”

When you want to stylize an existing photo (e.g., turning a portrait into a Pixar character or transforming a landscape into a watercolor), each platform handles reference images differently. Each style doc includes a dedicated 🔄 Image-to-Image Transformations section with platform-specific prompts.

Upload your image to the conversation, then describe the transformation:

"Using the attached image as the base, transform it into [STYLE]. Preserve the subject's appearance and pose, but apply [SPECIFIC STYLE DETAILS]. Use a [ASPECT RATIO] format."

Conversational refinement — After the initial result, iterate with follow-ups:

  • “Increase the intensity of the style — push it further from photorealism”
  • “Keep the subject’s face more recognizable”
  • “Change the background to [NEW ENVIRONMENT] while keeping the style”
  • “Apply the style only to the background — keep the subject photorealistic”

Upload the image to the conversation, then prompt:

"Using the attached image as a reference for the subject's appearance, create a [STYLE] version. Preserve the person's likeness, clothing, and pose. [ADDITIONAL STYLE DETAILS]."

Place the image URL at the start of your prompt:

[IMAGE_URL] [STYLE DESCRIPTION] --iw [WEIGHT] --ar [RATIO]
ParameterRecommended ValuePurpose
--iw1.0–1.5 for style transfer, 1.5–2.0 for close preservationControls how closely the output follows the reference
--crefUse for character consistency across multiple imagesMaintains identity (face, features) across generations
--srefUse to apply a style from a separate referenceApplies the aesthetic of another image

Use the Img2Img pipeline with these recommended settings:

SettingLight StylizationMedium TransformationHeavy Restyling
Denoising Strength0.3–0.450.45–0.650.65–0.85
Use CaseSubtle filter effectBalanced style transferNew artwork based on composition
  • ControlNet alternative: For preserving exact pose/composition while completely changing style, use ControlNet with the canny or openpose preprocessor.
  • Prompt: Describe the desired output style. The original image provides composition/structure.
  • Negative Prompt: Target specific unwanted artifacts only.

Not sure which style to pick? Start from what you need:

Use CaseRecommended Styles
LinkedIn / Professional headshotProfessional Headshots, Cinematic Headshots
Social media profile pictureModern Avatars, Pixar, The Simpsons
App icon / thumbnail3D Isometric Resin, Minimalist Notion, Pixelated 16-bit
Desktop wallpaperSolarpunk, Cyberpunk Noir, Bioluminescent Underwater, Double Exposure
Instagram / social media postWes Anderson, Studio Ghibli, Claymation
Poster / printArt Deco, Concert Poster, Soviet Constructivist
Album / book coverVinyl Album Cover, Double Exposure, Glitch Art
Children’s contentYarn / Amigurumi, Claymation, Paper Cutout
Educational / scientificBotanical Illustration, Blueprint, Electron Microscope
Holiday / greeting cardStained Glass, Illuminated Manuscript, Embroidered
Game asset / concept artIsometric Pixel City, Holographic UI, Steampunk
Brand / marketing assetArt Nouveau, Minimalist Notion, Wes Anderson
Photo-to-character conversionPixar, The Simpsons, Lego, Studio Ghibli, Collectible Figurines
Turning a photo into fine artUkiyo-e Woodblock, Art Nouveau, Byzantine Mosaic
Product marketing / mockupGlass Embossed 3D, Exploded Product View, Brand Concept Renders
Movie / TV fan artMovie Diorama Cube, Collectible Figurines, Superhero Comic Book
Travel / city contentFridge Magnet Knolling, Landmark Dioramas, Tilt-Shift Miniature
Moody editorial / album artFrosted Silhouette, Double Exposure, Daguerreotype / Tintype

  1. Find a style from the categories below or use the use-case guide above.
  2. Click through to the individual style doc.
  3. Pick a variation that matches your intended output (profile pic, wallpaper, poster, etc.).
  4. Copy the prompt for your platform — Nano Banana 2 is listed first and recommended.
  5. Replace placeholders[SUBJECT], [ENVIRONMENT], [COLOR], etc. — with your desired content.
  6. Iterate — adjust details, swap environments, or refine lighting based on initial results.

Contribute! Generated a great image with one of these prompts? Submit a PR to replace the placeholder image in the style doc with your result!