/image-master

02 — Physics, Light, Optics, Camera, Materials, Motion

Highest-leverage idea: AI models learn the statistical look of images, not the physics of light and 3D space. They have no internal scene geometry, so they fail exactly where a forensic analyst checks — shadow-to-light convergence, reflection vanishing points, consistent specular highlights. Your prompt substitutes photographic and physical causality for the model's default "glossy average." Specificity is the constraint that forces a physically plausible solution.

1. LIGHT PHYSICS

Rule AI breaks constantly: a point on an object, its shadow, and the light source lie on one straight line — those lines converge at the light. AI scatters that convergence → impossible multi-directional shadows. Name ONE source, its direction, and the resulting shadow behavior. Multi-light prompts confuse the model.

Direction / setup bank:

• Direction: light from camera-left, long soft shadows to the right · low-angle sun behind subject · top-down key · 45° key
• Named setups: Rembrandt lighting (triangle under the eye on the shaded side) · butterfly · rim light · backlight / contre-jour · silhouette · key + soft fill · bounced fill from below
• Quality: soft window light · overcast diffused, no harsh shadows · hard single light, deep chiaroscuro · practical lighting (lamps in scene) · studio softbox
• Time/color: golden hour, warm low rim light · blue hour · harsh noon overhead, short hard shadows · dusk · dust/breath catching the light

Why: naming one source + shadow direction gives a coherent target instead of averaging conflicting cues. "Golden hour rim light" reproduces well (millions of correct examples); "magical/HDR/dramatic" pulls toward stylized renders.

2. REFLECTIONS & OPTICS

Same geometric constraint: object-to-reflection lines share a vanishing point, and a reflection must contain the actual scene. AI's worst failure zone — mismatched/missing reflections, wrong-direction shadows inside reflections, and eye catchlights lacking crisp directional structure (a giveaway).

Optics bank:

• single sharp catchlight in each eye matching the key light · mirror reflects the actual room behind camera · wet asphalt with accurate streaked reflections of the neon signs · glass with subtle surface glare and depth, not a perfect mirror
• Tasteful artifacts (sparingly): slight chromatic aberration on high-contrast edges · gentle anamorphic lens flare · mild vignetting · specular highlights / caustics on rippling water

Why: telling the model what the reflection contains and constraining catchlight count/direction fights decorative, geometry-free reflections. Subtle aberration/vignetting signal a real glass element — flawless optics read as CGI.

3. CAMERA / LENS REALISM (the NatGeo look)

Naming a real body + lens + aperture is the strongest single realism lever — the model learned the specific optical signature. Always pair focal length with an f-stop (the aperture tells the model how much blur to apply).

• Aperture: f/1.4–f/1.8 blown bokeh · f/2.8 subject isolation · f/8 balanced · f/11–f/16 deep-focus landscape
• Bodies: Canon EOS R5 · Sony A1 / A7R V · Nikon Z9 · Hasselblad (medium format) · Leica
• Real lenses: NIKKOR Z 400mm f/2.8 TC VR S · Canon RF 85mm f/1.2 · Sony FE 70-200mm f/2.8
• Shutter/ISO/grain: 1/1000s frozen · 1/30s motion blur · ISO 800 · subtle film grain · slight sensor noise
• Film stocks: Kodak Portra 400 · Ektachrome E100 · Fujifilm Pro 400H · Tri-X · CineStill 800T

NatGeo wildlife template:

Realistic wildlife photograph, [animal + action] in [environment], Nikon Z9 + 400mm f/2.8 at f/4, 1/1000s, ISO 800, golden-hour warm rim light on fur, breath visible in cold air, shallow DoF, crisp focus on the eye, National Geographic editorial color grade

Why: generic quality tags ("8k, ultra realistic") trigger aesthetic bias → waxy, over-sharpened, flat-lit output. Real gear names invoke physically grounded distributions: compression, DoF falloff, grain that match a real exposure.

4. ATMOSPHERE & DEPTH

Flat AI images lack air. Force layers and atmospheric perspective (distant objects fade cooler/hazier) — the cue the eye uses for scale.

• Layering: [foreground: grass/rocks/water] · [midground: subject] · [background: distant peaks/horizon], layered depth, strong parallax
• Atmosphere: volumetric god rays through gaps · morning mist / valley fog · heat haze · cool blue haze on distant ridges · dust glowing in backlight
• Scale cues: tiny figure for scale · acacia silhouettes against vast sky

5. MATERIAL / TEXTURE PHYSICS

"Texture is revealed by how light interacts with surfaces" — describe the material AND its light behavior + wear. Default failure = over-smoothing → plastic/waxy.

• Skin: visible pores, fine lines, subtle redness, natural oiliness and tonal variation — nothing airbrushed
• Fabric: raw silk with visible slubs and sheen · heavy linen, coarse weave, natural creases · wool, fuzzy nap, visible knit stitches · velvet, directional pile sheen
• Metal: mirror-polished chrome, sharp speculars · brushed steel, directional grain, fingerprint smudges · oxidized copper, green verdigris · pitted rusty iron
• Fur/hair: individual backlit strands, anisotropic sheen, matted wet clumps
• Water/wetness: beaded droplets, specular glints, wet sheen and weight in soaked fabric
• Foliage/wood: leaf veins, sun-dappled translucency · weathered grain, cracks, dust
• Universal imperfection: scratches, dust, fingerprints, worn edges, natural micro-detail

6. MOTION & ENERGY

Match the verbal cue to a real shutter behavior.

• Freeze: 1/2000s, frozen mid-leap, crisp edges, frozen droplets and flying debris, individual dust particles sharp
• Blur: 1/30s motion blur on the legs, sense of speed
• Panning: panning shot, sharp subject, horizontally streaked motion-blurred background
• Energy: kicked-up dust trail, splashing water arc, wind-blown mane

7. ANTI-PATTERN LIST (kill these — they trigger aesthetic bias → CGI gloss)

Avoid: 8k, 4K, ultra HD, hyperrealistic, photorealistic (as a tag), masterpiece, award-winning, trending on ArtStation, octane render, unreal engine, cinematic, ultra-detailed, stunning, flawless, perfect, ethereal, dreamlike, hyper-saturated, HDR, neon/magical/fantasy lighting

Negative guardrails (where supported): no CGI, no 3D render, no game engine, no plastic/waxy skin, no airbrushing, no oversharpening, no oversaturation, no cartoon/illustration

Deepest recurring insight: believability comes from narrative causality, not descriptor density. "Flour on her sleeves, morning light through lace curtains" beats ten superlatives because it gives the scene a physical reason to exist. Real-life imperfection + one coherent light + named optics is the whole game.

Sources: AI Video Bootcamp 2026; Artlist (lighting); Aituts (camera prompts); Content Authenticity Initiative & Amped/Forensic Focus & IEEE Xplore (shadow/reflection forensics); Meri CreativAI (the #1 fake mistake); ZSky (textures, landscapes); media.io (wildlife); Photography Life (motion/panning); Miraflow; ZeroSkill; D5 Render (atmospheric perspective). Verified 2024–2026.