Two characters in one AI video scene: reference slots, coverage and credit cost

One character is solved. Two is where scenes fall apart.

Holding one face across a scene is close to routine now. Put a second character in the same frame and the problems come back: faces blend into each other, outfits swap, the wrong mouth speaks the line. The advice on the first page of Google is right as far as it goes. Make one reference sheet per character, tag each one, shoot coverage instead of one wide take. What it never says is how two characters fit inside each engine's reference limit, which engine to pick for a two-person cast, or what the coverage costs.

We cut two-hander scenes every week for our series, so this is the version with the numbers in it.

Step 1: count your reference slots before you write the scene

Every engine takes a fixed number of references per render. Two characters split that number between them, and anything left over goes to a location or a prop. These are the caps our studio enforces, per render:

engine	reference cap	how it is addressed in the prompt	two characters leaves
Seedance 2.5	30 images	@Image1, @Image2 ... by position	up to 15 images each, or 12 each plus 6 for the location
Seedance 2.0	9 images	@Image1 ... by position	4 each plus 1 location frame
Kling O3 Pro	5 elements	@Element1, @Element2	2 characters plus 3 spare elements
Veo 3.1 Fast	3 images	"image 1", "image 2" in plain words	2 for one, 1 for the other, nothing spare

The last row is the one that catches people. Veo 3.1 Fast holds three images in total, so in a two-shot one of your characters is running on a single picture. That is fine for a wide where both are small in frame. It is not fine for a close-up.

On Kling O3 Pro a character is one element, and an element carries its own images, so two characters cost two of the five slots and the face stays well fed. On Seedance the images are addressed by their position in the list, which means the order matters: if the hero is @Image1 in one shot and @Image3 in the next, you have told the engine two different things. We keep the order fixed for the whole series. Our element vs image reference guide covers the syntax in detail.

If a scene names more characters than the engine takes, our studio refuses the render before it bills and names the roomier engine to switch to. That refusal is cheaper than finding out from a blended face.

Step 2: build two sheets that do not look alike

Engines blend characters that share features. Two brown-haired men of the same build in similar jackets will trade faces by the third shot. Before you render a frame, make the pair easy to tell apart: different hair colour or length, a different silhouette, one bold colour each in the outfit. This is a design decision, not a prompt trick, and it does more than any wording.

Each character gets its own sheet: front, three-quarter left, three-quarter right, one full body. A sheet in our studio is 30 credits, so the pair is 60. The character reference sheet guide shows the layout we use.

Step 3: shoot coverage, not one long two-shot

Real film crews do not shoot a conversation as one wide take, and neither should you. We use a three-shot pattern for almost every exchange:

1. The wide. Both characters in frame, establishing where they are. Short, usually 4 seconds. Both references attached.
2. Single on A. Character A speaks, framed medium or close. Only A's references attached.
3. Single on B. Character B answers. Only B's references attached.

The singles are the important part. When only one character is in the shot, that character gets every reference slot. On Veo 3.1 Fast that is all three images for one face instead of a split. On Seedance 2.0 it is nine images for one face instead of four. The face you see in close-up is the best-fed face you can give the engine.

It also fixes the wrong-mouth problem. When there is one person in frame, there is one mouth to move. In the wide, keep dialogue out or keep it to one line, and name the speaker in the prompt: "@Element1 speaks, @Element2 listens."

What the three-shot pattern costs

Here is a 12-second exchange cut as a 4-second wide plus two 4-second singles, voiced, at 720p, with the rates our studio charges:

engine	wide (4s)	single A (4s)	single B (4s)	total
Kling O3 Pro	120	120	120	360
Seedance 2.0	180	180	180	540
Seedance 2.5	280	280	280	840
Veo 3.1 Fast, with references	240	240	240	720

The Veo row needs a note. A Veo 3.1 Fast render with a character reference is always 8 seconds, so each of those three shots is billed as 8 seconds at 30 credits, not 4. You get longer clips to trim, but you pay for them. For short dialogue singles, that makes Kling O3 Pro half the price of Veo with references.

At the Lite rate of 1,000 credits for $10, the Kling version of this scene is $3.60. Add the two sheets once, 60 credits, and they carry over to every scene after.

Why coverage costs no more than one take

A single 12-second voiced Kling O3 Pro take of the same two characters is also 360 credits. The seconds are the same, so the bill is the same. The difference shows up when something goes wrong. If B's face drifts in a single, you re-render one 4-second shot for 120 credits. If B's face drifts at second nine of a 12-second two-shot, you re-render all 12 seconds for 360. Over a series, coverage is cheaper because retakes are smaller.

Which engine we use for two-person scenes

• Kling O3 Pro for most dialogue. Two elements, three spare, and the cheapest voiced second we sell at 30 credits. Up to 15 seconds per shot.
• Seedance 2.5 when the scene needs both characters close together in one rich, moving frame, a fight or a dance, because 30 image slots give both faces plenty to hold on. Voiced and silent cost the same on Seedance.
• Veo 3.1 Fast for singles of one character, where its three images all go to one face. We avoid it for the two-shot itself.
• Seedance 2.0 when the budget matters and the characters are well apart in frame. Nine slots, 45 credits a second at 720p.

Mixing engines inside one scene works as long as each engine gets the same sheets. We often shoot the wide on Seedance 2.5 and the singles on Kling O3 Pro.

Touching is the hard part

The weak point on every engine is physical contact. A hug, a handshake, a shoulder grab in close-up: that is where faces and hands merge. We frame contact wide or from behind one character, and cut to singles for the reaction. If the story needs the contact up close, render it silent and short, check it, then decide whether to voice it.

The whole pattern, wide plus singles, fixed reference order, one speaker per shot, is how we cut two-hander scenes in the studio. It is also the shape of the consistent characters guide, scaled up from one character to two.

FAQ

How do I put two consistent characters in one AI video?

Make a separate reference sheet for each character, attach both to the wide shot, and attach only one character's references to each close single. Keep the reference order identical in every prompt. On ReelDouble, Kling O3 Pro takes 5 elements and Seedance 2.5 takes 30 images per render.

Why do my two AI characters swap faces?

The engine blends characters that look alike or that touch in a close frame. Give each character a clearly different hair colour, silhouette and outfit colour, and frame physical contact wide. Rendering each speaker as a single with only their own references stops most swaps.

Which AI video engine is best for two characters talking?

We use Kling O3 Pro for most dialogue: two characters use two of its five element slots, and a voiced second costs 30 credits on ReelDouble. Seedance 2.5 suits busy two-person action because it takes 30 reference images.

How much does a two-character AI video scene cost?

On ReelDouble a 12-second voiced exchange cut as a wide plus two singles is 360 credits on Kling O3 Pro, 540 on Seedance 2.0 and 840 on Seedance 2.5 at 720p. That is $3.60 for the Kling version at the Lite rate of 1,000 credits for $10.

Can Veo 3.1 handle two characters in one shot?

It can, but Veo 3.1 Fast takes 3 reference images in total, so one character gets a single image. We use it for one-character singles, where all three images hold one face, and keep the two-shot on Kling O3 Pro or Seedance.