Clone Club: Multiplying Five People Into a Street Crowd

Every no-budget short eventually needs a crowd. A market, a gig queue, a protest, a bus queue outside a Wetherspoons. And every no-budget filmmaker eventually discovers that the going rate for a background artist, multiplied by thirty and a Saturday, is roughly the cost of a used car.

The workaround is old and reliable: shoot the same small cast over and over against green, then layer them into a plate. It looks terrible when it is done badly, and it looks completely fine when it is done carefully. The difference is almost entirely in what happens on set, not in the edit.

Sort out where you are allowed to be first

This is the step people skip, and it is the one that ends the day early.

Publicly accessible is not the same as free to film. Councils across the UK run film offices — Bristol, Leeds, Manchester, Glasgow and most London boroughs all have one — and a quick email will usually tell you whether a small crew with a tripod needs anything formal. For five people and a locked-off camera, most will say you are fine on a quiet weekday morning, provided you do not block a pavement or a fire exit.

Things that need a specific ask, every time:

• Station forecourts and anything inside the barriers (Transport for London and Network Rail both have filming policies, and bus operators have their own).
• Privately owned shopping centres, retail parks and car parks.
• National Trust and Forestry England land, which have permit schemes.
• Anything with a market trader in shot — they are self-employed and their stall is their livelihood.

Keep the footprint tiny, avoid a tripod in the middle of a busy pavement, and bring a hi-vis if you are anywhere near a road. It also helps enormously to shoot a plate with almost no one in it, because then the background itself is easy to license and easy to repeat.

Lock the camera. Then lock it again

The single biggest tell in a duplicated crowd is a background plate that drifts. If the camera moves between passes — even a millimetre of tripod settle, even the tiniest change in focus breathing — every layer will shift relative to the background and the illusion collapses.

So: heavy tripod, legs spiked or weighted, no image stabilisation (turn IBIS off — it hunts between takes and warms up differently), manual focus, manual iris, manual white balance, manual everything. Then tape the focus ring. Then photograph the setup on your phone so you can rebuild it if someone knocks it.

Shoot the plate first, with the frame empty or containing only the cast members who appear in every layer. Shoot every performance pass from the same position without touching the camera. If you have to move it, reshoot the plate.

The shooting order that saves you in the edit

Do not shoot five people five times randomly. Give the edit a structure.

1. Empty plate. Two minutes of clean background, no one in frame, ideally with nothing moving in it either.
2. Core pass. Your main characters, positioned where they need to be.
3. Individual passes. One person at a time, each in a different spot, each performing a different action — walking, waiting, on a phone, eating. One at a time is far easier to key than a group with overlapping limbs.
4. Depth passes. Two or three people shot much further from the camera, and one or two much closer, so you have layers at different scales to play with.
5. Shadow pass. Each performer stands on a plain grey or green surface and you shoot their feet and the shadow they cast. You will use this more than you expect.

Mark positions on the ground with tape and number them. When you are stacking eight copies of the same person, you need to know which take goes where.

Scale, eyeline and the depth-of-field lie

A copied figure looks wrong for one of three reasons: they are the wrong size, they are looking at the wrong horizon, or they are in impossibly sharp focus.

Match the camera height to the background plate and keep it there all day. If the background was shot at chest height, every performer must be shot at chest height too, or their eyeline will float.

For scale, work out your real-world conversions before you shoot. If the background street is four metres wide and your green screen is two metres wide, then a full-body shot framing the screen edge-to-edge represents a person twice life size in that world. Complicated, but worth ten minutes with a tape measure.

Depth of field is the easiest thing to forget. Background figures in a real crowd are rarely tack sharp. Add a small amount of defocus to your far layers in the edit, and a light haze or grade shift so distant clones do not pop with the same contrast as foreground ones.

Shadows sell it, and missing shadows kill it

Nothing floats like a composited person with no contact shadow. A hard-edged cutout standing on a pavement reads as a sticker, no matter how good the key is.

Three fixes, in order of effort:

• Bake it in. Have each performer cast a real shadow onto the green, and key the shadow along with them using a separate, softer key. Risky, because green shadows key unevenly.
• Shoot a shadow pass. A soft blob of shadow on grey, shot from the same angle, then screened or multiplied under the figure.
• Paint it. A soft black ellipse at 20-40% opacity, warped and blurred, sits under the feet. Crude and surprisingly effective at crowd distance.

Match the direction of the background's light. If the plate has sun coming from camera left, every clone's shadow goes right, and every clone's key light should have come from the left on the day.

The keying mistakes that scream duplicate

• Green spill on shoes and shoulders. Bounce a little of the background colour onto the subject, or use a spill suppression pass rather than cranking desaturation.
• Razor edges. Feather a pixel or two. Real people do not have a one-pixel alpha channel.
• Grain mismatch. Add the same film grain or noise to every layer, and to the background, at the end. This one trick does more for believability than any keyer setting.
• Identical performances. Same walk cycle, same jacket, same posture on four clones in a row. Change a hat, a bag, a gait.
• Perfect rows. Real crowds do not space themselves evenly. Overlap some, cluster others, leave gaps.

The free software stack

DaVinci Resolve's free version handles almost all of this: the Delta Keyer for the mattes, Magic Mask for the awkward bits, planar tracking for the layers, and enough node-based grading to match every clone to the background. Blender's compositor is there if you want to push layers around in 2.5D. Kdenlive and Shotcut will do the basics on a laptop that is struggling.

Shoot at 4K even if you deliver at 1080p. That headroom lets you scale clones down for distance, reframe the odd layer, and rescue a take where someone drifted out of position.

What five people can realistically become

Five performers, four passes each, plus a couple of depth layers, gives you twenty-odd bodies on screen — enough for a queue, a market, a bus stop at rush hour. Beyond about thirty, the repetition starts to show, and you are better off adding a blurred crowd matte or shooting your plate in a genuinely busy spot and layering only your principals into it.

It is not a Marvel crowd. It is a convincing one, made on a Saturday, for the price of five friends' lunches and a shopping trolley of charity-shop coats.