Foreshortening is challenging for humans and AI alike. In both, it is one of the most reliable ways to expose how well
an artist or an image model understands depth, proportion and spatial order.
Humans struggle to find the right "shapes" that define receding objects, but we work intuitively, without
description: we know a foot near the lens must expand while the torso recedes. Most AI systems still treat bodies as 2D
assemblies of learned silhouettes. Asked for extreme perspective, they often misfire as they assemble from training data:
stretching limbs inconsistently, flattening torsos, or confusing what overlaps what.
This is a quick Sora recipe book designed to test and guide that behaviour, leaning on ordered descriptions, explicit
light cues and floor-plane anchors to help the model maintain depth. These are not guaranteed fixes, and results can be
unpredictable, but each prompt pressures the model's causal sense of space rather than just its stylistic memory.
Frame of mind
Most AI imagery fails here because it builds a scene as a single flattened snapshot. The formulas here introduce time
into the spatial reading. It is no longer what is depicted, but how it is revealed. That is a major unlock for any
generative system trying to simulate intentional foreshortening, spatial hierarchy or visual storytelling.
Think of the description as leading through the space of the body, foot to leg to torso to head, like travelling the
spine of a figure. Each part builds on the previous one's scale and overlap cues. The image stops being static and starts
behaving like a process. This unlocks an underused tool: progressive anchoring. Each stage gives the next something
to push against, so by the time the description reaches the head, the image has earned its depth.
Shift the mindset from "romantic scene → reaching hand toward a basket of bread" to:
wine glass → red liquid catching light → dinner plate behind glass → table surface continues → breadbasket visible →
reaching hand → knuckle detail clear → forearm follows hand, watch visible on wrist → fabric of sleeve, scale compressing
toward elbow → arm and shoulder emerge → torso compressed in perspective → far shoulder and neck visible, scale smaller →
face farthest behind reaching arm → expression engaged, soft focus.
Sequence capsules
Foreshortened figure: sequential depth capsule
"A single continuous shot, no cuts. Naturalistic studio lighting, 50–70 mm lens feel. Render the subject as a full-body figure reclining toward the viewer, built step by step:
Foreground: the soles and toes are nearest, dominating the lower third of frame.
Continue back through shins and knees, preserving strong scale decay and overlap.
Reveal thighs and pelvis, grounded with cast shadow contact.
Show ribcage and abdomen, compressed in perspective.
Add shoulders and neck, tapering along the depth axis.
Finally, resolve the head farthest from camera — smallest by visual angle, sharp yet least saturated.
Do not render later stages unless prior ones are correctly established. No montage, no disjointed limbs, no symbolic insertions."
This exploits token-chaining priority: the order and the gating words in the prompt set up a causal
scaffold the engine tries to satisfy in sequence. That sequence then maps to spatial structure: near to far, part to
whole, detail to field.
Tree canopy: vertical sequence capsule
"A single continuous shot looking upward, 24–28 mm wide-angle feel. Build the composition in vertical depth order:
Start with the tree trunk closest to camera, bark texture crisp and detailed.
Extend into primary thick branches splitting from the trunk with strong occlusion.
Continue into secondary thinner branches, creating a radial lattice.
Add a semi-translucent leaf canopy forming an overhead plane.
Finally, reveal a small sunburst peeking through the gaps, flare controlled and leaf detail intact.
No sun until the canopy exists. Flare must not overpower detail. All stages should appear continuous and physically connected."
This sequence mimics real embodied experience (look up → branch → canopy → sky), so the composition feels
natural even when the image is highly constructed.
Pencil on desk: linear progression capsule
"A single static shot, shallow depth of field, natural studio light. Sequence the object from nearest to farthest:
Foreground: the graphite tip in sharp focus, bevels and chips visible.
Progress into the sharpened wooden cone, clean helical cut.
Extend along the yellow hexagonal barrel, receding with proper perspective lines aligned to desk grain.
Show the metal ferrule with knurling and specular reflections.
End with the pink eraser, farthest and slightly softer.
Barrel must read continuous. Ferrule cannot appear before barrel. Only one pencil — no duplicates or clones."
The pencil is a microcosm of the same principle. From tip to eraser, the eye meets scale decay and
material contrast in order: sharpness, length, metal, softness. That small journey builds object intimacy and weight, and
sets up a compositional payoff: the pencil is no longer just an object, it becomes a path the viewer travels along.
Choreography of the image
This is not about choreographing perception. It is about the model's token sequencing under causal load: the prompt
pressures the generative process, not the viewer's eye.
Below are easy foreshortening prompts, which can be used as they are or built upon.
Simple foreshortening
The recline (classic)
"Low front camera. Nearest foot or knee dominates lower frame. Body reclines away, each joint overlapping the next. Light from upper-left, floor plane visible. No motion blur, no floating limbs."
The reach
"Single still frame. One hand or arm reaching toward camera, fingers sharp. Forearm, shoulder, and face recede along one line. Depth shown by focus falloff and scale, not blur streaks."
The stack (objects)
"Series of identical solid objects (boxes, books, chairs). Foremost fully visible, each behind smaller, same axis. Floor shadows aligned, no floating. 6–9 units, last merges with haze."
The coil (organic)
"Snake-like or rope-like form looping toward camera. Overlaps alternate light/dark; contact shadows visible. Haze increases along coil."
Spine capsules
Reclining figure: classic perspective capsule
"A single still frame, neutral studio light, 50 mm–70 mm lens feel. (1) Foreground: nearest foot dominates lower third, sharp texture and edge. (2) Shins and knees overlap the foot; scale decay clear. (3) Thighs and pelvis continue backward; cast shadow anchors to floor. (4) Abdomen compresses in perspective; lighting consistent. (5) Chest and shoulders taper; DOF softens slightly. (6) Farthest: head small by angle, least saturation, gaze off-frame."
Crouched figure: diagonal compression
"Single still frame, 35 mm lens feel. (1) Foremost: bent knee and hand planted near camera, in crisp focus. (2) Torso twists back, forming a diagonal path toward far shoulder. (3) Far knee and foot partly hidden, scale reduced. (4) Face and upper torso align to same perspective; light continuous. DOF shallow near knee, deeper toward head; shadows unify across floor."
Running motion (freeze-frame)
"Single exposure, 85 mm lens, low front angle. (1) Foremost: reaching hand extended toward camera, sharp. (2–3) Arm and torso recede; depth shown by foreshortened limb lengths. (4–5) Rear leg trails into haze; ground contact shadow continuous. Capture energy through linear compression, not motion blur."
Crouched figure: diagonal compression
"Single still frame, 35 mm lens feel. (1) Foremost: bent knee and hand planted near camera, in crisp focus. (2) Torso twists back, forming a diagonal path toward far shoulder. (3) Far knee and foot partly hidden, scale reduced. (4) Face and upper torso align to same perspective; light continuous. DOF shallow near knee, deeper toward head; shadows unify across floor."
Environments
Subway seat
"Low-front perspective of a person sitting on a subway bench, one boot or shoe nearest to camera, filling the foreground. Legs and torso taper toward back seats; train interior repeats evenly behind. Fluorescent light strips create linear perspective; reflections fade along ceiling. Ban fisheye distortion."
Alley run
"Narrow alleyway, low front camera. Foreground: running figure mid-stride, nearest foot in sharp focus, torso receding, far leg soft. Brick walls on both sides converge toward a single vanishing point. Morning light filtering in from behind; mild atmospheric haze."
Instrumental trumpet
"Single still frame, neutral studio light, 70 mm lens. (1) Bell of trumpet nearest, dominating lower third, metallic reflections crisp. (2–4) Tubing and valves recede; highlight streaks continuous. (5) Player's face farthest, framed by instrument; focus softens gradually. Gates: brass reflections aligned; no duplicate bells."
Artist's hand toward canvas
"Single frame, shallow DOF. (1) Foreground hand with brush nearly touches lens, crisp detail. (2) Brush shaft recedes along depth line. (3) Canvas surface occupies mid-plane; strokes visible. (4) Artist's face farthest, half-blurred. Gates: same light source for hand and face; no duplicated tools."
Start being creative
Doorway lean
"Low-angle shot inside a dim room. Foreground: a hand resting on a doorframe edge near camera. Torso and face lean outward into brighter hallway light. Depth shown by the shift in temperature — warm interior to cool exterior. Keep the doorframe vertical and consistent perspective lines on floor."
Workshop scene
"Interior with workbench. Foreground: extended arm holding a tool toward camera; focus on hand and metal surface. Mid-ground: torso leaning forward; background shelves soft and warm. Shadows align along bench plane; single neutral light source."
Painting of a woman in a hallway
"A painting of a woman sitting against a corridor wall, one knee bent and nearest to the camera, foot oversized from perspective. The hallway recedes into cool light, vanishing point centered behind the figure. Shadows run along the floor, light from an overhead source. Background walls parallel; tone softens with distance."
Pirate scene
"Far out on a plank, on a pirate boat, a snake-like or rope-like form looping toward camera next to a pirate's boots. Another pirate stands far in the background. Overlaps alternate light/dark; contact shadows visible. Haze increases along coil. Dark and moody."
Motorcycle jump at sunset
"Low-front camera, 35 mm wide feel. Motorcycle leaning into a turn speeding launching off a cliff. (1) Nearest wheel rim crisp, dominating frame. As if stopped in motion (2) Forks and body shrink toward seat; reflections aligned. Motorcyclist leaning forward, wind passing fast. (3) Rider's torso and helmet farthest, reduced contrast. Lighting sunset; single VP centered; ground shadow continuous. Light reflecting into sun glare. Background diffused with speed motion. Light continuous. DOF shallow near knee, deeper toward head; shadows unify across floor. tilt motorcycle to one side, add torque to rider"
Degas painting
"Single degas style painting of a dancer (1) Foremost: bent knee and hand planted near viewer, in crisp focus. (2) Torso twists back, forming a diagonal path toward the far shoulder. (3) Far knee and foot partly hidden, scale reduced. (4) Face and upper torso align to same perspective; light continuous. DOF shallow near knee, deeper toward head; shadows unify across floor."
Summary
If the figure holds shape and contact through three or four overlapping planes, it is already in high territory. If it
keeps rhythm and vanishing-point coherence through seven or more, the model is not just rendering; it is reasoning
spatially. And if the image starts to blur or repeat instead of distort, that is not a failure. It is the system
protecting itself from hallucinating geometry it cannot sustain.
Foreshortening is the edge where perception meets physics. When AI gets it wrong, it reveals the delta of what you are
working against.
For the full recipe book, with more tips, tricks and prompts: