Format-Adaptive Rembrandt Portrait Lighting

Scope and subject fidelity

• Use the built-in image-generation editing workflow. If the source is a local file, inspect it with

  view_image

  first. Never overwrite the original.
• The subject can be one or more people, pets, or a mixed group. If no photo is supplied, ask the user to attach one; if they want only a prompt, provide it without generating an unrelated subject.
• Preserve the original subject, count, identity, expression, clothing, accessories, and visible anatomy. Preserve pose and viewpoint by default, but permit the controlled composition and posture refinements below when they make a static source feel like a credible new studio sitting. Do not mirror or radically re-pose merely to accommodate a lighting diagram.
• For people, preserve identity, apparent age, ethnicity, skin tone, facial proportions, hairstyle, hands, tattoos, jewelry, glasses, and wardrobe.
• For pets, preserve species, breed traits, coat color and markings, muzzle, eye color and shape, ear position, whiskers, collar, limbs, and tail. Do not humanize animal eyes, add makeup, alter paws, or invent a different animal.

Default: full low-key studio conversion

• Make the subject’s key-side planes the highest-value area. Render empty background as a smooth, noise-free deep charcoal that falls toward ink black across the frame. Use one extremely broad, low-contrast, asymmetric tonal transition aligned with the key so the set has depth without becoming visible scenery; never use a centered head halo, spotlight disc, gray haze, or uniformly clipped black canvas. The subject must read instantly even as a thumbnail.
• Treat the background as a clean low-ISO commercial digital capture: silky continuous-tone gradient, visually noise-free at 100% zoom, with no film emulation. Gradient is allowed and desirable; grain, texture, and random luminance variation are not.
• Use one dominant, image-coordinate key (

  key from image-left

  or

  key from image-right

  ), 40–60° off camera and 30–45° above the primary face/eyes. Never use an unqualified “left” or “right”.
• Describe a large, close, feathered, gridded soft source: directional enough for visible highlight-to-shadow modeling, but broad enough for a natural penumbra, gentle highlight roll-off, and believable light wrap at hair, fur, ears, and shoulders.
• Default to a 4:1–6:1 key-to-fill ratio (about two to two-and-a-half stops) with restrained black-flag negative fill on the shadow side. Reserve 8:1 for an explicitly severe theatrical result. Keep both eyes alive and preserve texture in the shadows; do not make the shadow side a dead black cutout.
• Add a restrained, physically plausible kicker only when the subject merges into the backdrop. It should define fur, hair, shoulders, or ears at lower intensity than the key—never a neon rim halo. Preserve irregular fine edge hairs and slight optical softness so the silhouette does not resemble a segmentation mask.
• Hold focus on the nearer eye(s) and facial/muzzle plane, then let sharpness fall naturally across ears, shoulders, and clothing as an 85 mm-equivalent studio portrait would. Preserve real skin, fur, whiskers, and fabric without amplifying pores, wrinkles, weave, or edge contrast. Ban clarity/texture boosting, crunchy microcontrast, plastic smoothing, and synthetic bokeh.

Camera, pose, and living presence

• Prefer a subtle three-quarter relationship: turn the virtual camera or the head/torso only a few degrees, keep both eyes plausible, and preserve facial geometry. Never reveal a substantially unseen side of the face or body.
• For people, allow a slight head turn or tilt, relaxed shoulder drop, asymmetric shoulder line, mild torso rotation, or natural weight shift. Keep the original expression unless the user requests another; do not manufacture a grin, change hand gestures, or invent hidden limbs.
• For pets, allow a small head/gaze turn, more attentive but anatomically natural ear posture, or a slight seated/standing weight adjustment. Preserve breed structure, ear shape, paws, tail, coat markings, and the original emotional character.
• Use an eye-level or subtly elevated portrait camera by default. A small camera-height or yaw refinement is acceptable; dramatic low angles, overhead views, wide-angle distortion, or major perspective changes require an explicit request.
• Build asymmetry with gaze room, shoulder direction, light direction, and negative space—not dutch angles or theatrical action. The result should feel alert, present, and recently photographed, not frozen or ceremonial.

Format and framing

• If the user specifies an aspect ratio or exact dimensions, obey them. If they specify only a use, choose a conventional ratio:

  1:1

  for avatars,

  4:5

  for social portrait posts,

  3:4

  for classic portrait/editorial,

  2:3

  for full-length portrait/print,

  3:2

  for photographic landscape/editorial,

  4:3

  for general landscape, or

  16:9

  for banners and cinematic covers.
• Preserve orientation unless the requested use clearly requires another orientation. Treat aspect ratio and pixel dimensions separately; never promise dimensions the generator cannot actually deliver.
• Reframe with the least destructive method. First remove expendable background. If cropping would clip hair, ears, chin, whiskers, hands, paws, tail, meaningful clothing, or the spatial relationship between multiple subjects, extend the studio field instead of cutting the subject.
• When extending, synthesize only plausible continuation of the dark studio set, existing clothing/body edges, and necessary contact shadows. Do not invent new anatomy, accessories, furniture, scenery, or a substantially wider camera viewpoint. Extension must match the key direction, lens perspective, depth of field, tonal falloff, and background cleanliness.
• Keep the face or pet head dominant without imposing a fixed head size on every source. Tight portraits may use head-and-shoulders framing; wider source photographs may retain upper body, full body, or group context when that better preserves the original photograph.
• Place the eye line and negative space according to the selected orientation and gaze direction. Do not center mechanically; leave breathing room toward the subject's gaze while keeping the subject visually dominant.
• Keep the eye(s) and face/muzzle as the first visual fixation through focus, local contrast, and controlled dodge-and-burn—not artificial glow or changed eye color.
• Use simple negative space and a clean silhouette. Do not add typography, badges, ornamental frames, fake lens flares, or props.
• Use the highest available native image quality. Do not claim “8K” unless the actual delivered dimensions are 8K; fake resolution language and oversharpening are not quality.

1:1

Subject-specific light pattern

Background and set continuity

• Preserve the source aspect ratio by default. Choose or change to a standard ratio only from an explicit dimension, aspect ratio, or use-case request, following the format rules above.
• Default to a clean, smooth, textureless deep-neutral-charcoal-to-ink studio field—not a uniformly clipped digital

  #000000

  canvas, charcoal cloth, painted wall, paper, visible cyclorama texture, or fog. Maintain a subtle broad asymmetric tonal falloff aligned with the key; it should create spatial depth with visually continuous, band-free tone. Background cleanliness is non-negotiable: no luminance noise, chroma noise, film grain, sensor grain, dither pattern, mottling, speckle, paper/fabric texture, smoke, dust, banding, JPEG blocks, or AI texture synthesis. Use absolute pure black only when explicitly requested.
• Background separation must come primarily from the directional key, natural light wrap, tonal depth, and only if essential a restrained edge kicker. Never place a gray glow, radial hotspot, obvious vignette, halo, or light pool behind the head to fake separation.
• The background’s direction, falloff, contact shadow, and color temperature must agree with the key. A bright source from the wrong side behind the subject is a defect.
• Respect difficult contours: hair, fur, whiskers, ears, transparent eyeglass frames, translucent fabric, and paws. Preserve natural contact and floor shadows when visible.
• Do not add speed trails, motion blur, wind, action effects, extra subjects, or duplicate limbs unless explicitly asked.

Avoid memorial-portrait cues

Prompt construction and iteration

identity-preserve

lighting-weather

Final check