Build a repeatable wedding cinematography shot list with AI: coverage tiers, lighting plans, prompt structures, and review loops for any venue.
A wedding film is usually decided before the ceremony begins — in the ten minutes you spend working out where to stand when the light turns gold, and in the quiet hour the week before, imagining shots you have never actually seen. Cinematographers who work across unfamiliar venues live or die on that preparation. Generative video has finally become good enough to make that preparation visual: you can test compositions, lighting moods, and camera moves against a room you have only seen in photographs, then arrive on the day already knowing what you want.
What follows is a practical system for building an AI-assisted shot list that holds up in any location — a stone church in the rain, a rooftop under harsh noon sun, a barn lit by three working bulbs. It is not about replacing the person behind the camera. It is about turning a vague creative instinct into a document you can shoot against, edit against, and defend to a client who asks why the film looks the way it does.
Coverage logic: think in passes, not shot names
Most wedding shot lists are lists of events: rings, dress, first look, vows, kiss, dancing. Useful, but they describe what happens rather than what you see. A cinematography list describes coverage — the visual building blocks an editor needs to assemble a scene that holds together.
Three passes over the same moment solve most of it:
- Establishing pass. Where are we, what does the space feel like, what is the scale of it.
- Performance pass. Faces, hands, reactions, the emotional event itself.
- Texture pass. Details that carry meaning: the grip of a hand, shoes on gravel, a candle guttering.
A six-hour coverage day usually resolves into 12 to 18 setups, each with two or three variations — wide and tight, static and moving. That number is remarkably stable across venues. What changes is which setups are possible, and that is exactly what pre-visualization tells you early, while you can still change the plan.
| Tier | Purpose | Typical framing | Best use of generated reference |
|---|---|---|---|
| Establishing | Geography and mood | 24–35mm wide, slow push | See how the room reads at the ceremony hour |
| Performance | Emotion and story | 50–85mm, shallow depth | Check expressions and background compression |
| Texture | Meaning and rhythm | macro, odd angles | Test detail shots that may not exist on the day |
| Transition | Editing glue | whip pans, defocus, silhouettes | Test match cuts between locations |
Two things belong in the coverage plan but rarely get written down: audio and slack. Vows and room tone determine how long a shot has to hold, which affects how you shoot it. And a schedule with no gaps is a schedule that breaks the first time a grandmother wants a photograph. Reserve twenty percent of the day as unallocated time and treat it as a real part of the plan rather than a luxury.
Scout the venue like a location manager
Before you write a single prompt, collect facts. A recce is not a vibe check; it is data gathering. Photograph and note:
- Window orientation and the sun path at the ceremony hour, not the hour you happen to visit.
- Ceiling height and colour — a low cream ceiling bounces warm light back into faces.
- Existing practical lights, and whether they are dimmable or on one circuit.
- Power access and cable runs that avoid both frame and guest traffic.
- Ambient noise: kitchens, roads, air conditioning, a bar next door.
- The three cleanest backgrounds for portraits and the three most cluttered.
- Where the couple will actually stand, sit, walk, and exit — not where the planner's diagram says they will.
Those notes become a one-page location brief. Everything downstream references it: prompt language, lens choices, lighting plan, fallback plan. Skip this and you will generate beautiful frames that have nothing to do with the venue, which is worse than useless because it builds false confidence.
The two-column constraint test
Write two columns. Fixed: church rules, no flash, no crossing the aisle, hard 4pm ceremony, one usable window. Flexible: arrival time, portrait location, whether you can rig a light, whether you can steal ten minutes at golden hour.
Constraints drive creativity far more reliably than mood boards. Every setup on your final list should survive the fixed column. If a generated frame shows a light source that cannot exist in that room, it is a nice picture, not a plan.
A worked example from a difficult room
Picture a converted barn: dark timber walls, one glazed door at the far end, ceremony at 3pm, no rigging allowed. The recce tells you the door light will be flat and behind the couple, the walls will eat every bounce, and the only clean background is a pale stone wall to camera right.
The plan writes itself from those facts. One wide from the door side to establish the space and silhouette the couple. One tight 85mm from the stone wall side, where the ceremony light falls across a face. Texture coverage of hands, rings, and the order of service on a wooden bench. And a deliberate decision, made in advance, to bring one soft battery light for the later speeches because nothing in that room will help you. Every one of those decisions can be tested as a generated frame before you pack the car.
Ask the questions nobody volunteers
Three answers change plans more often than people expect: when the cleaners finish, whether the aisle carpet can be lifted, and what time the band loads in. Add a fourth — who else will be shooting. A documentary photographer with a 70–200mm will occupy angles you were counting on, and knowing that before you write the list prevents a day of quiet negotiation in the aisle.
Turn the recce into a one-page look bible
A look bible is a visual agreement with yourself, short enough that you actually reopen it on the day. It holds:
- Palette: three dominant colours and one accent, in plain words ("warm limestone, sage, and a bruised blue in the shadows").
- Contrast: milky and soft versus punchy and separated.
- Texture: grain, halation, softness in the corners.
- Lighting signature: how many sources, how soft, how directional.
- Motion signature: mostly locked off, or permanently drifting.
This matters because generative models drift. Twelve reference frames generated without a shared language give you twelve different films. Write the look bible first, then reuse its exact phrases in every prompt, and the outputs start to feel like one project rather than a Pinterest board.
If you would rather anchor with images than adjectives, generate a few reference stills in the AI image generator and keep the good ones as a small, named reference set. Three stills do more work than three paragraphs of description, because the model reads them as a fixed target instead of an interpretation.
Two disciplines keep the bible useful. First, keep it to a single page — a five-page treatment becomes a document you admire and never open. Second, describe light by source and direction rather than mood. "Soft window light from camera left, 4300K, shadow falling off the right cheek" survives a handoff to a second shooter. "Dreamy lighting" does not survive anything.
Write prompts that behave like shot cards
A prompt is not a wish. It is a shot card. The most reliable structure reads like a camera report:
Subject + action | framing | lens | camera movement | light source and quality |
colour and texture | duration and pace | what must not appear
Three examples, all for the same imaginary venue — a stone chapel with one tall window on the left:
Bride at the altar turning toward the groom | medium close-up, 85mm equivalent |
slow 10% push in | soft directional window light from camera left, 4300K,
shadow falloff on the right cheek | warm limestone, muted sage, fine 35mm grain |
4 seconds, unhurried | no lens flare, no visible crew, no modern signage
Empty chapel nave before guests arrive | wide, 28mm, low camera height |
static with slight handheld breath | single shaft of daylight across the pews,
dust visible, deep shadow in the apse | cool shadows, warm highlights |
5 seconds | no people, no flowers yet
Hands exchanging rings | macro, 100mm equivalent | locked off |
window light raking across skin, soft specular highlight on the band |
natural skin tones, background dissolved to bokeh | 3 seconds |
no exaggerated sparkle, no brand marks
Now the hard one — the shot that most often fails in planning, a backlit ceremony window:
Couple facing each other, silhouetted shoulders | medium wide, 40mm |
very slow lateral drift, 5% of frame | strong backlight from a tall window
behind them, soft wraparound fill from pale interior walls, exposed for
skin with the window just holding | warm haze in the highlights,
low contrast shadows | 6 seconds | no blown highlights, no artificial rim light
Four habits separate prompts that work from prompts that waste an afternoon:
- One idea per prompt. A lighting change plus a camera move plus a wardrobe change produces mush.
- Name the light source, not the mood. See above for why.
- State negatives explicitly. Models will happily add flares, drone moves, and extra guests.
- Fix the length. Four seconds at 24fps is 96 frames; a model asked to fill ten seconds will change the edit rhythm you planned for.
Keep your own venue-tested phrasing in a prompt library — a plain text file is enough — because most weddings share eight or nine lighting situations. You will write the reception speech reaction card once and reuse it thirty times a season, adjusting only the wardrobe and wall colour.
Give the model fewer decisions
The less a model has to invent, the closer the output lands to your intent. If the couple's wardrobe is fixed, say it once and reuse the same sentence. If the film lives at 35mm and 85mm, never generate a 24mm frame and then wonder why the animatic feels like a different movie. Every variable you remove frees attention for the variables you actually care about: light and faces.
Match generated references to the camera you will actually shoot
The most common failure in AI-assisted wedding work is a mismatch that only appears in the edit: reference clips that are cooler, cleaner, and sharper than anything the camera actually captured. Fix it during planning by deliberately degrading your references toward reality.
- Generate at the frame rate and shutter feel you will shoot.
- Ask for grain and slight softness if your lens is vintage or your sensor is small.
- Keep colour temperature consistent. If the venue runs on tungsten and candles, do not pre-visualize at 5600K daylight and call it a look.
- Test the ugly scenarios, not the hero ones: mixed reception light, the blown-out window, the high-ISO corner. Knowing how bad it can get tells you whether to bring a light at all.
Useful decision criteria before you generate anything expensive: does this frame tell me where to put a light, or does it only tell me what a nice wedding looks like? If it is the second, skip it. Pre-visualization pays for itself only when it changes a physical decision — position, lens, source, or time of day.
This is where the work becomes commercial, too. You can show a couple what "no additional lighting" looks like next to "one soft source bounced off the ceiling" before anyone commits, and let them choose with their eyes rather than their imagination.
Keep faces and palette consistent
Consistency is the hardest problem in AI-assisted pre-production, and the fix is boring: reduce variables.
- Two wardrobe colours maximum in generated references, matching the real couple.
- One reference frame per person, reused, rather than a fresh description each time.
- Fixed lens language across every prompt.
- A single grade sentence repeated verbatim.
If a reference drifts, throw it out instead of patching the prompt. Ten consistent frames beat forty near-misses, and the near-misses quietly poison your sense of the film.
Cut a short animatic instead of showing a paper list
A shot list on paper is abstract to clients and to a second shooter. Forty seconds of moving pictures is not. Generate eight to twelve short clips in sequence — arrival wide, dress texture, first look, ceremony wide, vows close, a reaction, the exit, reception wide — and cut them against the ceremony music.
What this actually achieves:
- Pacing problems surface immediately. Two wides in a row feel slow; three details in a row read as a montage whether you intended one or not.
- The couple can respond to a tone instead of a spreadsheet, which produces better feedback and fewer surprises in the edit.
- A second shooter understands the grammar of the day without a briefing call.
Build the pieces in the AI video generator using the exact look-bible phrases, then assemble them in any editor. If you want to see how a particular model handles natural light and skin tones before writing your own prompts, browse Seedance examples and compare, rather than guessing from marketing copy.
Label the animatic as tone, not storyboard
The one real risk with showing generated footage is that a couple reads it as a promise. Say plainly, in the email and on the call: this is coverage and mood, not a frame-by-frame plan. Clients who understand that give sharper feedback. Clients who do not will ask why the real cake looks different.
Score setups against a rubric so the standard compounds
The difference between "looks nice" and a repeatable standard is a rubric. Score each generated setup 1–5 on five axes:
| Axis | Question |
|---|---|
| Composition | Is there a clear subject, and does the frame have depth? |
| Light | Is the source motivated, and does it shape the face? |
| Motion | Does the movement serve the moment or decorate it? |
| Story | Would an editor actually use this shot? |
| Match | Does it sit with the rest of the film's palette and texture? |
Eighteen out of twenty-five is a workable threshold. Below that, either redesign the setup or cut it from the list. The rubric also ends arguments faster than taste does: "I don't like it" becomes "the light is unmotivated and the story score is two."
A typical first pass scores 20 on composition and 12 on light, because generated frames love a tidy arrangement and forget what is producing the illumination. That pattern is diagnostic: it tells you to spend the next study session on motivated sources rather than on shot variety.
Keep the scores across a season. They show which lighting situations you consistently misjudge — usually mixed reception light and backlit ceremony windows — and that is exactly where the next investment of time belongs.
Seven mistakes that quietly wreck the plan
- Pre-visualizing only hero shots. The problems live in transitions and ugly rooms.
- Generating at a different aspect ratio than delivery. Crop surprises cost compositions you already paid for in planning time.
- Over-planning to the minute. A shot list is a menu, not a timetable. Leave twenty percent slack.
- Ignoring audio. Vows and room tone decide how long a shot must hold.
- Reusing another wedding's look bible. Venue light is not transferable; the language is.
- Letting references replace scouting. A generated frame is a hypothesis, not evidence.
- Never reviewing. A list is only a standard if you score against it and revise.
FAQ
Do I need AI to make a good wedding film? No. Plenty of excellent films are shot on instinct and experience. Pre-visualization shortens the learning curve on unfamiliar venues and makes your intent legible to clients and crew — that is the whole claim, and it is enough.
How many generated frames and clips per wedding? Six to twelve stills and eight to twelve short clips is enough for a forty-second animatic. Beyond that you are making a different film rather than planning this one, and the extra work rarely changes a decision.
Will generated references make my footage look artificial? Only if you grade toward the references instead of toward the venue. Generate images that respect your camera's real limits — grain, softness, mixed colour temperature — and treat them as composition and lighting guides rather than as a delivery target.
What about venues where photography is restricted? Plan tighter, longer-lens coverage and lean on texture and reactions. Pre-visualization is more valuable there, not less, because you cannot freely test compositions on the day and every angle has to be chosen in advance.
How do I keep the list from feeling mechanical? Build in deliberate gaps. Mark two or three windows as unplanned observation and protect them. The list covers the predictable majority; the film's best moment usually comes from what you did not schedule.
Is an animatic worth showing clients? Usually yes, framed as tone and coverage rather than a storyboard to match frame for frame. It is also the cheapest way to discover that your couple actually wants something quieter than what you proposed.
How long does this workflow take? A few hours per wedding once the templates exist: recce notes, a look bible, a first list, one animatic cut, one review pass. The first two weddings take a day each; after that it becomes routine.
What if the venue is so dark that nothing is possible? Then the plan changes honestly. Commit to longer lenses, closer work, and one motivated source you control. Knowing the limit before the day is worth far more than hoping the room behaves differently than it did on the recce.
Build the shot list before you build the film
A cinematic wedding film is an argument about light, pacing, and attention, and arguments are stronger when they are prepared. Start with the venue, write the look bible, translate it into shot cards, generate a short animatic, and score the results so the standard improves each season instead of resetting every spring.
Orelon is built for exactly this stage: cinematic ideas in motion, tested before the day. Generate reference stills, run motion tests for the rooms that worry you, and cut a forty-second animatic your couple can genuinely respond to. Begin with the AI video generator, borrow structure for the animatic step from the video templates, or read more production workflows on the Orelon blog.

