Construction foreman late 30s, weathered skin with visible pore texture, short dark hair with grey temples at 30%, clear polycarbonate safety glasses with subtle green reflections, bright orange ANSI Class 2 reflective vest with silver stripes over faded indigo chambray work shirt with rolled sleeves showing forearm hair, tan canvas cargo trousers with reinforced black Cordura knee panels, scuffed brown leather work boots with Vibram soles, holding yellow hard hat under left arm and rolled architectural blueprints with visible grid lines in right hand, standing on crushed gravel construction site with dust at ankle level, massive concrete building frame with exposed vertical rebar and formwork gaps behind him, towering yellow tower crane with lattice boom silhouetted against 3200K sunset sky, warm amber light 2900K bleeding through structural gaps creating volumetric god rays, atmospheric haze with dust particles catching sidelight, lens flare from 15-degree low sun angle, dramatic 3/4 rear rim lighting on subject separating him from background, shot on ARRI Alexa 35 with Cooke S7/i 75mm lenses at T2.0, cinematic color grading with lifted shadows and warm midtones, shallow depth of field isolating subject from construction chaos, 4:3 aspect ratio, subtle 35mm film grain texture --ar 4:5 --style raw --s 250 --no oversaturated orange, plastic skin, flat lighting
Quick Tip: Click the prompt box above to select it, then press Ctrl+C (Cmd+C on Mac) to copy. Paste directly into Midjourney, DALL-E, or Stable Diffusion and customize to make it your own!
The golden hour is a color temperature instruction, not a mood. When a prompt says "golden hour," most diffusion models resolve that phrase into a global warm cast applied across the entire frame, and the result is not sunset light — it is a sepia overlay with a subject trapped inside it. The fix is not to remove the warmth. It is to specify two distinct temperatures and control the gap between them.
That gap is the whole mechanism. A real sunset scene contains at least two light sources with different color temperatures: a low sun at roughly 3000-3200K and a cooler fill from the sky or bounced ambient around 5500-6500K. When a model receives a single "warm light" instruction, it collapses both into one value and the physical relationship between key and fill disappears. Detail goes with it, because detail lives in the contrast between them.
In the construction site prompt, the two values are declared explicitly: "3200K sunset sky" and "warm amber light 2900K bleeding through structural gaps." That is a 300K differential. It sounds small, and that is exactly the point. A 300K gap reads as atmospheric color separation. A 1500K gap reads as a filter. The model treats the wider gap as a white balance error rather than an intentional contrast, and it corrects toward neutral in the shadows while pushing the highlights toward orange. Skin texture, chambray weave, and Cordura panel grain all flatten in the process.
Why the rim light is the load-bearing instruction
The most consequential line in this prompt is not the temperature. It is "dramatic 3/4 rear rim lighting on subject separating him from background." Rim lighting is a geometric instruction before it is a color instruction. It tells the model where the light source sits relative to the camera and the subject, which determines which surfaces receive direct illumination and which fall into shadow.
A 3/4 rear position means the sun is behind the subject and offset roughly 45 degrees from the camera axis. The lit edge runs along one shoulder, the side of the jaw, and the outer arm. The front of the vest stays in shadow but picks up bounced light from the gravel and the concrete formwork. This is what produces the separation: the subject's silhouette is defined by a bright edge against a mid-tone background, not by a brightness difference across the whole frame.
If the prompt omits the rear position and only says "golden hour," the model defaults to frontal or near-frontal illumination because that is the most common composition in its training data. The result is flat, evenly lit, and reads as a stock photo. The rim instruction is what forces the model to reason about occlusion — which parts of the subject the sun can reach and which parts the body itself blocks.
The 15-degree low sun angle compounds this. Low angles produce long shadows and short specular highlights on curved surfaces, which is why the safety glasses pick up "subtle green reflections" rather than a broad white glare. Combined with the rim position, the angle determines the length of the shadow the foreman casts across the gravel, and that shadow is one of the strongest depth cues in the frame. Prompts that specify a light temperature but not an elevation angle lose this cue entirely.
Volumetric god rays work on the same principle. "Warm amber light bleeding through structural gaps" only produces visible beams if the prompt also declares a participating medium: "atmospheric haze with dust particles catching sidelight." Light is invisible in a vacuum. The dust is not decoration — it is the physical mechanism that makes the beam a beam instead of a gradient. Removing the haze line while keeping the god ray line produces a soft glow with no defined edges.
Lens choice reinforces the separation. A 75mm at T2.0 on a full-frame sensor gives a shallow but not extreme depth of field, which keeps the subject's boots and face both acceptably sharp while the concrete frame and crane fall into soft focus. Wider apertures would blur the rebar into unrecognizable shapes and lose the structural context that makes the scene read as a construction site rather than a generic outdoor portrait. The dramatic feathered portrait lighting breakdown covers the same trade-off between subject isolation and environmental context in a studio setting.
The crane deserves specific attention because it is a silhouette, not a lit object. By declaring it "silhouetted against 3200K sunset sky," the prompt removes it from the lighting calculation entirely. Without that word, the model will try to illuminate the lattice boom and the yellow paint will compete with the vest for visual weight. Silhouetting the crane also solves a practical problem: the lattice structure is geometrically complex, and silhouettes hide the errors that appear when a model renders thin repeating structural elements in full light.
Material specificity does the same work on the subject. "Faded indigo chambray work shirt with rolled sleeves showing forearm hair" is not padding. Each noun constrains the render. Chambray has a visible warp-and-weft texture distinct from denim or poplin. Rolled sleeves create a hard fabric edge at the elbow and expose a forearm that must be rendered with skin detail consistent with the face. Forearm hair is a high-frequency texture that forces the model to commit to a level of detail it would otherwise skip.
The ANSI Class 2 vest specification matters more than it appears. ANSI Class 2 is a real standard with defined retroreflective stripe placement — two horizontal bands around the torso and two vertical bands over the shoulders. Naming the standard forces the stripe geometry to be correct. A generic "orange safety vest" produces stripes in arbitrary positions, and on a backlit subject those stripes are the second-brightest element in the frame after the rim light itself. Wrong stripe geometry breaks the silhouette reading.
The Cordura knee panels and Vibram soles serve a similar function at the bottom of the frame. Cordura is a matte, abrasion-resistant nylon with a distinctive dull sheen that reads differently from canvas under low-angle light. Vibram soles have a chunky lug pattern that catches rim light on the tread edges. These are small details, but they are the details that survive the shallow depth of field and tell the viewer the subject is a working foreman rather than someone in costume.
Color grading instructions interact with all of the above. "Lifted shadows and warm midtones" is a filmic look that keeps shadow detail visible, which is essential when a subject is backlit and the front of the vest sits in shadow. If the prompt instead said "crushed blacks," the vest would lose its texture and the subject would read as a flat orange shape against the sky. The lifted shadow instruction preserves the information the rim light is supposed to reveal.
"Subtle 35mm film grain texture" is a final quality signal rather than a compositional one. Grain tells the model to avoid the over-smoothed, plastic surface that diffusion models default to at high step counts. It is a mild instruction and its effect varies by model version — on some checkpoints it barely registers, on others it produces visible luminance noise in the sky gradient. This is one of the parameters where the honest answer is that results depend on the specific model and sampler, and it is worth testing at two or three grain values rather than assuming one works universally.
The same uncertainty applies to the temperature differential itself. A 300K gap between key and sky is a judgment call calibrated for warm-toned renders. Cooler-leaning checkpoints may need a wider gap to produce visible separation, while heavily warm-biased ones will need a narrower gap to avoid an orange wash. The principle holds — declare two temperatures and control the gap — but the number is a starting point, not a fixed value. The impasto night scene prompt runs the same two-source temperature logic in a completely different register, which is useful for calibrating how a given model handles declared color contrast.
What the prompt does not do is guarantee the result. Every parameter here constrains the model's interpretation, but none of them are deterministic. The value of writing the prompt this way is that failures become diagnosable: if the rim light is missing, the rear-position instruction was ignored. If skin looks plastic, the temperature gap is too wide. If the crane competes with the subject, the silhouette instruction failed. That diagnosability is the actual payoff of specificity, not the hope of a perfect first render.
What to verify before accepting the result
- Check that a distinct bright edge runs along one shoulder and jaw, not across the whole face. Uniform frontal light means the rim instruction was dropped.
- Confirm the front of the vest retains visible fabric and stripe detail in shadow. If it is a flat orange mass, shadow lift was insufficient or the temperature gap was too wide.
- Verify the crane reads as a dark silhouette with no competing yellow highlights. Partial illumination means the silhouette instruction failed.
- Inspect skin at cheek and forearm for pore or hair-level texture. Smooth plastic surfaces indicate the model corrected the color cast toward neutral.
- Check the shadow the subject casts on the gravel. A missing or very short shadow means the 15-degree sun angle was ignored.
- Confirm the visor stripe geometry matches ANSI Class 2 placement — two horizontal, two vertical. Arbitrary stripe positions break the silhouette read.
Label: Cinematic
Key Principle: A 2900K key against a 3200K sky is only a 300K gap. Anything wider turns rim light into a wash that erases skin texture. Keep the differential under 400K.
Comments
Post a Comment