Pick the wrong style keyword and your render fights you the whole way. Pick the right one and the model snaps into focus on the first try. AI art styles are the single biggest lever you have over a generation, bigger than subject, bigger than lighting, often bigger than the model itself. This is a working reference of 50+ style keywords I actually keep in rotation, grouped so you can grab one and go.

Everything below works across Midjourney, Stable Diffusion, DALL·E, and Flux. Syntax differs a little, the vocabulary doesn’t. A “chiaroscuro oil painting” reads the same to every model that’s seen enough Caravaggio. Steal freely, mix two or three, and keep notes on what sticks.

How to use style keywords (the 10-second version) #

Style is one slot in a prompt, not the whole thing. The shape I lean on: subject, action, setting, style, lighting, camera or render note. Drop a style keyword into that fourth slot and let the rest carry the content. One strong style word usually beats three weak ones. Stack a movement with a medium — “Art Nouveau, stained glass” — and you get something specific instead of mush.

A quick warning before the lists: living-artist names are a coin flip. Some models nerf them, some ignore them, and leaning on a working artist’s name is a bad habit anyway. So this reference favors movements, media, eras, and techniques. You get the look without copying a person’s signature.

Art movements (the backbone of most AI art styles) #

Movements are dense. One word drags in a whole palette, brushwork, and set of subjects. These are the workhorses:

  • Art Nouveau — organic linework, floral borders, gold accents
  • Art Deco — geometric, symmetrical, brass-and-black glamour
  • Impressionism — loose, visible strokes, broken light
  • Surrealism — dream logic, impossible scale, melting form
  • Bauhaus — primary colors, hard geometry, function-first
  • Ukiyo-e — flat color, bold outline, Japanese woodblock
  • Baroque — heavy drama, deep shadow, ornate detail
  • Cubism — fractured planes, multiple viewpoints at once
  • Pop Art — flat blocks, halftone dots, loud color
  • Brutalism — raw concrete, blunt mass, no decoration
  • Romanticism — sublime landscapes, tiny figure against huge nature
  • Constructivism — bold diagonals, red-and-black propaganda poster energy
Midjourney:
a lighthouse on a cliff, Romanticism, sublime storm light, tiny figure for scale --ar 3:2 --stylize 250

Stable Diffusion:
a lighthouse on a cliff, romanticism oil painting, sublime stormy sky, dramatic scale, dramatic clouds
Negative: flat lighting, modern, photo, low detail

Media and technique (what the image is “made of”) #

Movements tell the model what era. Media tells it what physical stuff. This is where you control texture — the difference between a flat digital fill and visible canvas tooth.

  • Oil painting — thick, glossy, blendable
  • Watercolor — bleeding edges, paper showing through
  • Gouache — matte, opaque, flat poster color
  • Charcoal sketch — smudged grayscale, raw gesture
  • Ink wash — tonal, calligraphic, lots of white space
  • Linocut / woodcut — carved high-contrast print
  • Pastel — soft, chalky, gentle gradients
  • Stained glass — lead lines, glowing color blocks
  • Papercut / kirigami — layered paper, hard shadow depth
  • Claymation / stop-motion — fingerprinted, tactile, slightly lumpy
  • Pixel art — low-res grid, limited palette
  • Vector illustration — clean flat shapes, crisp edges
  • Etching / engraving — fine cross-hatched lines, antique print feel
  • Collage / mixed media — torn paper, layered textures, found imagery
  • Scratchboard — white line carved out of solid black
Midjourney:
a red fox curled asleep, gouache illustration, flat matte color, limited palette --ar 1:1

Stable Diffusion:
a red fox curled asleep, (gouache:1.2), matte poster colors, flat shapes, textured paper
Negative: glossy, 3d render, photorealistic, gradient noise

Style by era and technique (skip the living-artist name) #

Want a specific look without naming a person? Describe the era and technique that look came from. This gets you 90% of the way and keeps you out of trouble. Each line below is a descriptor you can drop straight in.

  • Golden-age comic ink — bold outline, flat CMYK, halftone shading
  • Mid-century children’s-book illustration — textured, warm, limited palette
  • 1970s sci-fi paperback cover — airbrushed chrome, deep space gradients
  • Dutch Golden Age portrait — dark background, single light source, rich fabric
  • Edo-period woodblock — flat planes, wave motifs, hand-printed grain
  • Renaissance fresco — soft earth tones, plaster texture, balanced composition
  • Soviet-era poster — flat geometric figures, red and cream, heavy type space
  • Victorian botanical plate — precise linework, labeled engraving feel
  • 1990s anime cel — hand-painted gradients, film grain, soft glow
  • Art-house storyboard — loose marker, gestural, quick value blocks
Midjourney:
an astronaut floating past a ringed planet, 1970s sci-fi paperback cover style, airbrushed chrome, deep gradient sky --ar 2:3

Stable Diffusion:
an astronaut and a ringed planet, retro 1970s sci-fi book cover, airbrushed, chrome highlights, star field
Negative: modern, photo, harsh shadows, text

Render engines and 3D looks #

For anything that should read as 3D, CGI, or game art, the engine keyword does heavy lifting. Models have absorbed enough renders tagged with these terms that the words now carry real meaning — specific lighting, specific material falloff. These five cover most of what you’ll want.

  • Octane render — clean studio CGI, glossy, strong reflections
  • Unreal Engine — game-cinematic, volumetric light, big depth
  • Blender / Cycles — neutral, physically-based, soft global illumination
  • Pixar-style 3D — rounded, expressive, soft subsurface skin
  • Claymorphism / soft 3D — matte putty surfaces, pastel, rounded edges

One rule: don’t stack three engines. “Octane render, Unreal Engine, Blender” cancels itself into generic CGI. Pick one, then push it with material words like “subsurface scattering,” “matte clay,” or “polished metal.”

Midjourney:
a tiny robot watering a potted plant, soft 3D, claymorphism, pastel studio lighting, octane render --ar 1:1 --stylize 200

Stable Diffusion:
a tiny cute robot watering a plant, 3d render, soft clay material, pastel palette, studio lighting, subsurface scattering
Negative: 2d, flat, sketch, harsh shadows, photoreal

Lighting keywords that change everything #

Lighting isn’t a style on its own, but it’s the multiplier. The same subject in the same style swings from cheap to gallery-worthy on the lighting word alone. Keep these close:

  • Golden hour — warm, low, long shadows
  • Chiaroscuro — extreme light-to-dark, single hard source
  • Rim lighting — bright edge separating subject from background
  • Volumetric / god rays — visible light shafts through haze
  • Soft studio light — even, flattering, low shadow
  • Neon / cyberpunk glow — saturated colored light, wet reflections
  • Backlit silhouette — subject dark against a bright source
  • Bioluminescent glow — soft light coming from within the scene
Midjourney:
a wolf standing in fog at dawn, oil painting, volumetric god rays, golden hour --ar 16:9 --stylize 300

Stable Diffusion:
a wolf in morning fog, (volumetric lighting:1.3), god rays, golden hour, atmospheric, cinematic
Negative: flat lighting, overexposed, dull, low contrast

Putting it together #

The move is to combine one item from a few groups: a movement, a medium, and a lighting word. “Art Deco, stained glass, rim lighting” is a complete visual idea before you’ve even named a subject. Start there, generate four, and adjust the slot that’s weakest. If the texture’s wrong, change the medium. If the mood’s flat, change the lighting. If the whole thing reads generic, the movement is doing too little — swap it for something with more teeth.

Save the keywords that work for you into a personal swipe file. Over a few weeks you’ll build a shortlist of maybe fifteen that fit your taste, and you’ll stop reaching for the full fifty every time. That shortlist is worth more than any single mega-prompt. If you want a running bank to pull from, the brand’s prompt library is built exactly for this.