AI Image Styles Cheatsheet: Prompts for Every Visual Style

May 2, 2026 · Editorial Team · 8 min read · ai-image prompts cheatsheet

Getting consistent visual styles out of AI image generators requires understanding both the vocabulary that works and the syntax quirks that differ across tools. What prompts Midjourney understands fluently may need rephrasing for Flux or Stable Diffusion, and vice versa.

This is a working reference, not a theory piece. Each section covers the style, what makes a prompt work for it, and tool-specific syntax notes where they matter.

Photorealism

Photorealism is the most requested style and the one with the widest quality gap between tools and prompts.

What works: Specificity about lighting, camera, and subject matter. Vague prompts like "a realistic photo of a person" produce average results across all tools. Specificity about the type of camera, lens, and lighting conditions triggers the model's understanding of photography aesthetics.

Core prompt structure:

[subject], photographed with a 50mm lens, f/1.8 aperture, natural window light, shallow depth of field, film grain, Canon EOS R5

Style modifiers that consistently help:

shot on 35mm film, adds grain and tonal range
golden hour lighting, warm, directional light
studio lighting setup, clean, even, professional
RAW photo, signals high detail, less post-processing look
hyperrealistic, works on most models

Tool-specific notes:

Flux Pro and Flux Dev are the current leaders in photorealism for portraits and products. The prompt can be written naturally, Flux handles long, descriptive text prompts better than Midjourney, which responds better to comma-separated keywords. For Flux, write the prompt as a description: "A woman in her 40s sitting at a cafe table, afternoon light through a window behind her, shallow depth of field, documentary photography style."

Midjourney responds better to: woman at cafe, afternoon window light, shallow dof, documentary photography, 50mm lens --ar 4:5 --style raw

The --style raw parameter in Midjourney reduces the default stylization and is almost always worth adding for photorealism work.

Anime and manga

Anime is one of the more forgiving styles for prompt beginners because the aesthetic tolerates variation well. The challenge is distinguishing between anime sub-styles, which vary dramatically.

Sub-styles worth naming explicitly:

Studio Ghibli style, painterly backgrounds, soft character outlines, pastoral scenes
shonen manga style, high energy, strong outlines, action poses
shoujo manga style, soft features, detailed eyes, floral elements
cyberpunk anime, neon, urban, high contrast
chibi style, simplified proportions, oversized heads, cute aesthetic

Core prompt structure:

[character/scene], anime style, [sub-style reference], cel shading, clean line art, vibrant colors

Useful modifiers:

cel shading, the flat, animation-cell coloring of traditional anime
key visual, signals a polished, promotional-quality composition
manga panel, black and white, ink line art
anime screencap, suggests in-episode quality

Tool-specific notes:

For anime style, Stable Diffusion with anime-specific fine-tuned models (Anything V5, AbyssOrangeMix, or CounterfeitXL) significantly outperforms base Midjourney or Flux for authentic anime aesthetics. These models are available through Civitai and can run locally or via some hosted platforms.

If you're using Midjourney without fine-tuned models, the --niji 6 parameter (Niji mode) switches to a Midjourney model trained specifically on anime and illustration styles. It's a meaningful quality jump for this style.

Leonardo.AI has several anime-specific models in its model library (including AlbedoBase XL and DreamShaper) that produce strong anime outputs without local setup.

Watercolor

Watercolor is a tricky style to prompt well. Default AI outputs for "watercolor" often look like a photo with a watercolor filter applied rather than genuine watercolor media characteristics.

What makes real watercolor look different:

Color bleeding at edges
Visible paper texture
Transparent overlapping washes
Uneven color saturation
Unpainted "white" areas (left as paper)

Core prompt structure:

[subject], traditional watercolor painting, paper texture visible, transparent washes, loose brushwork, color bleeding at edges, white paper showing through

Useful modifiers:

loose watercolor sketch, reduces overworked feel
wet on wet technique, soft, bleeding edges
botanical illustration style, controlled, detailed watercolor
plein air watercolor, outdoor scenes, natural light
Winsor and Newton watercolors, signals material authenticity

Tool-specific notes:

Midjourney handles watercolor reasonably well with detailed prompts. Adding artist references helps significantly: in the style of John Singer Sargent watercolors or reminiscent of Albrecht Durer botanical watercolors produces more authentic media characteristics than generic watercolor descriptors.

For Stable Diffusion, the AnythingXL or realistic models with "watercolor" in the prompt often need negative prompts to avoid photographic qualities: add digital art, 3d render, photography, sharp edges to your negative prompt.

Pixel art

Pixel art has high demand for game assets, retro aesthetics, and indie game branding. It's also one of the more technically demanding styles for AI because pixel art has specific technical constraints (fixed resolution, limited color palettes) that most models don't respect naturally.

Core prompt structure:

[subject], pixel art, 16x16 sprite, limited color palette, retro game aesthetic, NES style

Resolution and era references that help:

8-bit pixel art, NES/Game Boy era, very limited palette
16-bit pixel art, SNES/Sega Genesis era, richer colors
32x32 pixel sprite, specifying resolution helps
isometric pixel art, for game environments
RPG Maker style, specific game aesthetic many models recognize

Tool-specific notes:

Most standard diffusion models struggle with true pixel art because they blur and anti-alias at a level that destroys the sharp, aliased pixel edges that define the style. Better results come from:

Using a pixel-art specific LoRA or fine-tuned model in Stable Diffusion (there are many on Civitai)
Generating at low resolution (256x256 or 512x512) and accepting the coarser result
Post-processing with Aseprite or pixel art-specific upscalers

For Midjourney, pixel art as a style keyword works reasonably well for the general aesthetic, but the output won't be strict grid-aligned pixels. It's more of a pixel-art-influenced illustration. Useful for branding and art, less so for actual game sprites.

3D render

The "3D render" style covers a wide range of aesthetics from clean product visualization to stylized Pixar-adjacent character art to architectural visualization.

Core prompt structure:

[subject], 3D render, Blender, octane render, global illumination, studio lighting, product shot

Sub-style variations:

Pixar 3D style, stylized, warm, character-focused
architectural visualization, precise, photorealistic spaces
product render, clean backgrounds, precise lighting
low poly 3D, geometric facets, minimal polygons
claymation 3D, tactile, handmade surface quality
octane render or cycles render, signals specific rendering engines with distinct looks

Tool-specific notes:

3D render prompts work well across Midjourney, Flux, and DALL-E 3. Midjourney is particularly strong on stylized 3D (Pixar-style characters, clean product renders). Flux performs well on photorealistic product visualization.

For the Pixar/animated 3D look specifically: 3D animated movie style, volumetric lighting, subsurface scattering, character design, vibrant colors produces consistently good results in Midjourney.

Sketch and line art

Sketch and line art styles are useful for storyboarding, character design, illustration, and editorial contexts.

Core prompt structure:

[subject], pencil sketch, hand-drawn, loose gestural lines, graphite on paper

Variations by use case:

architectural sketch, technical drawing quality, precise lines
gesture drawing, loose, quick, expressive
ink illustration, clean, deliberate, publishing quality
comic book inking, bold outlines, hatching, graphic
blueprint style, technical lines, white on blue

Useful modifiers:

cross-hatching, adds depth and shadow through line patterns
loose gestural strokes, reduces tight, overworked feel
rough sketch, work in progress, signals intentional looseness
clean lineart, no shading, useful for coloring later

Tool-specific notes:

For clean line art with no shading, Stable Diffusion with a LoRA specifically trained on line art (available on Civitai) produces cleaner results than base models. Midjourney and Flux can generate strong sketch aesthetics but rarely produce truly clean, gap-free lineart for use as coloring pages or design templates without post-processing.

Oil painting

Oil painting prompts work well across most modern tools when the lighting and composition are specified.

Core prompt structure:

[subject], oil painting, visible brushstrokes, impasto technique, canvas texture, [lighting description]

Artist references that help orient the style:

in the style of Rembrandt, dramatic chiaroscuro, portrait focused
impressionist oil painting, Monet/Renoir aesthetic, soft light, outdoors
baroque oil painting, formal, theatrical lighting
contemporary realism, oil on canvas, modern figurative painting
plein air oil painting, outdoor, loose, natural light

Useful modifiers:

impasto, thick, visible paint application
glazing technique, layered, luminous color
alla prima, wet on wet, painted in one session
museum quality, signals fine art production values

Tool-specific notes:

Oil painting style is one of Midjourney's stronger areas. The default Midjourney stylization tendency toward painterly aesthetics works in favor of this style rather than against it. Flux produces good oil painting results with detailed prompts but benefits more from artist references than Midjourney does.

For Stable Diffusion, the base SDXL model handles oil painting reasonably. Adding oil paint texture, canvas grain to prompts and using a painting-focused fine-tuned model improves results significantly.

Quick-reference table

Style	Key modifiers	Best tool	Avoid
Photorealism	`50mm lens`, `f/1.8`, `RAW photo`, `--style raw`	Flux Pro, Midjourney	Over-specifying lighting
Anime	`cel shading`, `key visual`, `--niji 6`	SD + anime model, Midjourney Niji	Generic "anime style"
Watercolor	`transparent washes`, `paper texture`, `loose`	Midjourney + artist ref	`watercolor filter`
Pixel art	`8-bit`, `16x16 sprite`, `limited palette`	SD + pixel LoRA	Base SDXL (blurs pixels)
3D render	`octane render`, `studio lighting`, `Blender`	Midjourney, Flux	Over-describing every element
Sketch	`pencil`, `loose gestural`, `graphite on paper`	SD + line art LoRA	Expecting clean lineart from base models
Oil painting	`impasto`, `visible brushstrokes`, `alla prima`	Midjourney	Photographic lighting terms

For tool selection guidance, the AI image generators comparison covers how Midjourney, Flux, Leonardo, and Stable Diffusion compare on quality and pricing. If you're choosing an image tool primarily based on photorealism for commercial work, Flux Pro via the Flux agent page has the pricing breakdown.