Orelon logoOrelon
料金

Documentary Lighting Techniques for AI Video: A Practical Guide

2026年11月17日 · Orelon Team 著

AI動画テンプレートを見る

着想のためにコミュニティ作品をいくつか閲覧し、任意のテンプレートを開いて Orelon で作成を続けましょう。

Learn how documentary lighting techniques translate into AI video prompts: contrast ratios, motivated light, color temperature, and a repeatable shot workflow.

Documentary cinematography is the most honest lighting school available to anyone making AI video. Nothing is hidden behind a set wall, nothing is faked with a bank of strobes, and the camera usually has one chance to get it right. That constraint produces a visual grammar you can study frame by frame, then describe precisely enough for a generative model to reproduce.

This guide is about that translation. We will look at how real non-fiction films use motivated light, measurable contrast, and color temperature to create mood, then turn those observations into prompt language and a repeatable workflow you can run inside Orelon or any other generative video tool. No abstract theory, no citation scaffolding — just practical decisions you can make on your next shot.

Why Documentary Lighting Is a Better Teacher Than Fiction Lighting

Narrative film lighting is designed to be invisible support for a story. Documentary lighting has to solve a harder problem: it must look credible while still shaping emotion, and it must do so in spaces the filmmaker does not control. A kitchen at 7 a.m., a hospital corridor, a fishing boat at dusk — these are not sets. The cinematographer negotiates with what exists.

That negotiation is what makes documentary lighting so useful as a reference library. Every frame carries a visible answer to the question "where is the light coming from, and why?" When you train your eye on those answers, you stop writing prompts like "cinematic lighting, dramatic" and start writing prompts like "single north-facing window as key, cool daylight 5600K, deep falloff into the back of the room, one warm practical lamp at 2700K as fill." The second prompt is repeatable. The first is a lottery ticket.

Reading Real Frames Like a Cinematographer

Before you write a prompt, spend ten minutes on a paused frame. Ask three questions: where is the key, how far does it fall, and what color is it? Almost every documentary look can be described through those three variables.

Motivated light and the honesty test

Motivated light means the audience can identify a plausible source. A window, a doorway, a fire, a phone screen, a car headlight. Documentary filmmakers lean on motivation because they cannot afford to invent sources, and the result is a visual logic that reads as truthful even when it is heavily shaped.

When you prompt an AI model, name the source. "Lit by a single interview key at 45 degrees camera left, softbox, with a 2-stop falloff across the subject's face" is far more controllable than "dramatic portrait lighting." Naming the source also gives the model a physical anchor, which reduces the weird floating-light artifacts that appear when a scene has no plausible illumination logic.

Contrast ratios you can actually measure

A lighting ratio is the difference in exposure between the lit side of a face and the shadow side. In documentary interview work, you will commonly see ratios in the 3:1 to 6:1 range — enough shape to look intentional, not so much that the subject feels interrogated. Harsher ratios, 8:1 and beyond, read as dramatic or ominous. Flatter ratios, 2:1 or lower, read as neutral and observational.

You can estimate a ratio by eye once you have practiced: compare the brightest patch of skin on the key side with the darkest patch still holding detail. Then describe it in the prompt as a relationship, not a number. "Soft key with gentle shadow side, roughly 4:1, shadow detail retained" gives a model a target it can interpret. Numbers alone sometimes push a model into extreme stylization because it does not know where the middle is.

Color temperature and the sense of place

Kelvin values do more emotional work in documentary than most people realize. Daylight at 5600K in a cold climate reads as clinical or brisk. Tungsten practicals at 2700K in a living room read as intimate. Mixed temperature — cool window light against a warm desk lamp — is the single most reliable way to make an interior feel real rather than rendered.

A practical habit: write the key temperature and the ambient temperature separately. "Cool ambient daylight 5600K from frame right, warm 2800K practical lamp in the background" produces a far more convincing interior than a single global color instruction. If you want a full breakdown of how color and camera language interact, the Orelon blog covers related topics in more depth.

Translating Observation Into Prompt Language

Most prompt failure is not a model limitation; it is an underspecified scene. The fix is a short, structured description that covers the same four things a gaffer would ask about.

The four-part lighting prompt

Write every prompt with these four blocks in this order:

  1. Source — what physically emits the light, and where it sits in the frame.
  2. Quality — hard or soft, and how large the source appears relative to the subject.
  3. Direction and falloff — where the shadow goes, and how quickly the light dies off.
  4. Color — key temperature, ambient temperature, and any practicals.

An example for a documentary-style portrait:

Medium close-up of a boat builder in a workshop. Key is a large north-facing window at camera left, soft and directional, falling off two stops toward the back wall. Cool daylight 5600K on the face, warm 2900K work lamp visible as a practical in the background. Shallow depth of field, 50mm equivalent, gentle grain.

The same blocks work for landscape and street scenes. Swap "window" for "overcast sky" and the structure holds.

Negative constraints do real work

Tell the model what should not be there. Documentary realism breaks quickly when a model adds rim light from nowhere, lens flares, exaggerated HDR skies, or plastic skin. Adding "no rim light, no flare, no HDR sky, natural shadow density" to your prompt costs eight words and saves a dozen regenerations.

Keep the constraint list short — three to five items. Long negative lists tend to cancel each other out.

A Repeatable Workflow for Documentary-Style AI Shots

Structure beats inspiration when you are producing a sequence rather than a single image.

Step 1: Build a reference board. Pull 8–12 frames from documentaries you admire. Group them by lighting situation — window key, overcast exterior, single practical, mixed temperature interior. You now have a palette you can return to instead of guessing.

Step 2: Write a lighting bible. One page. Define your key temperature, your contrast target, your preferred focal length, and your grain or texture preference. Keep it beside your prompt window. Consistency across shots is what makes AI footage feel like a film rather than a slideshow.

Step 3: Test with stills first. Generate a handful of images of the same scene before committing to video. The AI image generator is a fast, cheap way to lock lighting direction, ratio, and color before you spend time on motion. Once a still looks right, you already know your prompt is calibrated.

Step 4: Animate with minimal motion. Documentary cameras move rarely and deliberately. Slow push-ins, small handheld drift, a gentle pan. Ask for motion in the prompt in plain language and resist the urge to add camera moves to every shot.

Step 5: Generate in short beats. Produce 4–6 second clips and cut them together. Short beats hide imperfection and mirror how documentary coverage actually works — you rarely hold a shot longer than the idea requires.

Step 6: Grade for unity. A light contrast and saturation pass across all clips will do more for coherence than another round of generation. Match black levels and skin tones first; everything else is taste.

You can see how this plays out with actual sequences in the Seedance 2.5 examples, and the prompt library has ready-made structures for interview, street, and landscape lighting.

Five Documentary Lighting Archetypes and How to Prompt Them

Archetype What it looks like Prompt shorthand
Window key interview Soft directional daylight, deep falloff, dark background "Large window key camera left, 4:1 ratio, background falls to near black"
Overcast exterior Flat, even, cool, low contrast "Overcast sky as soft dome light, 5600K, no visible shadows, muted palette"
Single practical interior One warm lamp, heavy darkness elsewhere "One 2800K table lamp as key, rest of room in shadow, warm falloff"
Mixed temperature Cool ambient plus warm accent "Cool 5600K ambient from frame right, warm 2700K practical behind subject"
Golden-hour vérité Low warm sun, long shadows, handheld "Low sun at 3200K backlighting dust, long shadows, handheld drift, natural flare"

Each row is a starting point, not a formula. The value comes from naming the source, the quality, the falloff, and the color — which is exactly what the shorthand does.

Mistakes That Break the Documentary Look

Over-lighting. The most common failure is a scene where every corner is visible. Documentary images embrace darkness because real rooms have it. Let the background fall off.

Telephoto everything. AI models love shallow depth of field, but documentary work mixes 24mm, 35mm, and 50mm equivalents freely. Specify focal length in the prompt to avoid a uniform look.

Clean, modern color. Slightly muted saturation and a touch of grain read as footage. Over-saturated, razor-sharp output reads as advertisement.

Constant camera motion. A slow drift in every shot becomes exhausting. Alternate locked-off frames with subtle movement.

Ignoring skin tone. In mixed-temperature scenes, check that faces stay plausible. If a model pushes skin too orange or too blue, restate the key temperature explicitly.

Prompt drift across a sequence. Copy-paste your lighting block between prompts instead of retyping it. Small wording changes create large visual jumps.

Directing Sequences, Not Just Shots

A documentary scene is usually three shots: an establishing wide that explains the space, a medium that carries the action, and a close-up that carries the emotion. Plan AI generation the same way and you will get something that cuts together.

Match light direction across the three. If the wide has a window at frame left, the close-up should have its key from frame left too. Audiences do not consciously notice when this breaks, but they feel it — the scene reads as assembled rather than filmed. Write the light direction into a fixed block at the top of every prompt in that scene, and only change the shot size and subject.

Also plan for reaction shots and cutaways. Hands, tools, weather, doorways. These are cheap to generate, they hide continuity gaps, and they are the texture that makes non-fiction editing feel alive.

A Quick Quality Checklist Before You Accept a Clip

Run through six checks: Does the light have an identifiable source? Does the shadow side hold detail? Is there a plausible reason the background is dark? Are skin tones consistent with the previous shot? Is the camera movement small enough for the subject? Does the frame contain one clear focal point?

If a clip fails two or more, regenerate rather than fix in post. Grading can rescue contrast, but it cannot invent motivation.

Where AI Still Falls Short

Complex practical light interactions — a face lit by a fire that also flickers on a wall — remain inconsistent across frames. Long takes with continuous camera movement still drift in geometry. And true documentary spontaneity, the unplanned moment, is not something a prompt can request.

Work around these by keeping shots short, avoiding elaborate multi-source setups in motion, and treating AI as a way to produce coverage rather than a single perfect take. The AI video generator is strongest when you feed it a clear, physically grounded description and let it handle texture, grain, and atmosphere.

FAQ

Do I need real cinematography experience to use these techniques? No, but you need vocabulary. Learn four terms — key, ratio, falloff, color temperature — and you can describe almost any documentary look precisely enough to reproduce it.

How many lighting details should a prompt contain? Three to five concrete details beat a paragraph of adjectives. Source, quality, direction, and color is usually enough; add a negative constraint list of three to five items and stop.

Why does my AI footage look like a commercial instead of a documentary? Usually over-lighting, over-saturation, and constant camera motion. Reduce fill, mute color slightly, add grain, and lock the camera down in half your shots.

Can I match lighting across multiple clips? Yes. Keep a fixed lighting block that you paste into every prompt in a scene, and grade all clips together at the end. Consistency is a documentation problem more than a generation problem.

What is the best way to learn documentary lighting quickly? Pause frames from films you admire and write down the key source, estimated ratio, and color temperature for each. Twenty frames will teach you more than twenty tutorials.

Does aspect ratio affect the perception of lighting? It affects composition more than lighting, but wide frames expose more background, which means your falloff and practical placement matter more. Test your lighting block in the aspect ratio you plan to deliver.

Start With One Frame

Pick a single documentary frame you love, describe its light in four lines, and generate it. Then generate it again with a different ratio, then again with a different color temperature. That small exercise builds the muscle that separates usable AI footage from generative noise — and it takes less time than reading another tutorial.

When you are ready to put a sequence together, Orelon gives you a cinematic workspace to move from stills to motion without losing the lighting logic you worked to establish. Browse the video templates for structured starting points, or dive straight into the AI video generator and write your first four-part lighting prompt today.