How to Storyboard a Dialogue Scene for AI Video

A practical guide to planning shots for a two-person conversation so the AI-generated clips actually cut together

Two chairs, one line still to be spoken. Every dialogue scene starts as a decision about distance.

A dialogue scene storyboard for AI video needs one thing most people skip: a plan for who the camera is with at every moment, not a list of pretty shots. That means picking a wide to establish the space, over-the-shoulder or single shots for each speaker, and knowing in advance when you’ll cut to a two-shot that holds them together. Decide that before you generate a single frame and the clips actually cut together into a conversation. Skip it, and you get technically clean shots that never feel like two people talking to each other.

This is the guide I wish existed when I started running dialogue scenes through AI video tools. It covers what a storyboard panel for dialogue actually has to decide, how many shots a conversation needs, why the 180-degree rule still applies even though there’s no camera on a physical set, and the mistakes that quietly wreck a scene before you notice.

What does a dialogue scene storyboard actually need to decide?

A storyboard panel for dialogue is not a picture of what a line sounds like. It’s a decision about distance and point of view: how far the camera is from each character, and whose experience of the moment we’re sharing.

John Huston, who directed The Maltese Falcon and The Treasure of the Sierra Madre, wrote about this directly in his autobiography, An Open Book. He described two ways to shoot a scene. The first is to cover everything (a wide master, then mediums, then close-ups of everyone) and sort it out later in the edit. The second is to find the one shot that introduces the scene and let the rest follow from it. “Once you write your first declarative sentence, the narration flows,” he wrote. “Understanding the syntax of a scene implies that you already know the way the scene will be cut together, so you shoot only what’s required.”

For AI video, that second method is the only one that works. You can’t generate forty takes and sort it out later the way a film crew can. Every shot is a separate generation, and every one you don’t plan for is a shot you’ll have to write, prompt, and probably regenerate. Deciding the shot pattern before you generate anything isn’t a nice-to-have. It’s what keeps the scene from costing you three times over.

How many shots does a two-person conversation actually need?

Three, most of the time: a wide, a set of singles or over-the-shoulders for each character, and a two-shot you save for the moment that matters.

The wide tells the audience where they are and how close or far apart the two people stand. The singles carry the actual back-and-forth. Sidney Lumet, in his book Making Movies, pushed back on the idea that dialogue is somehow less cinematic than action: “Dialogue is not uncinematic,” he wrote. “So many movies of the 30s and 40s we adore are constant streams of dialogue.” A conversation shot well, mostly in singles that let you read each face in turn, carries a scene as well as any chase.

The two-shot is the one beginners overuse. Save it. Huston’s own rule of thumb was that the image should match the real distance and stakes between the characters. Two people a few feet apart having an ordinary exchange should read as an ordinary exchange. A two-shot that lingers on both faces at once claims something is happening between them right now. Spend that shot on the line that earns it: the agreement, the lie, the thing neither one says out loud.

What is the 180-degree rule, and why does it matter for AI video?

The 180-degree rule says you pick an imaginary line running through the two characters and keep the camera on one side of it for the whole scene. Cross it, and the characters appear to swap places, so a person who was facing right is suddenly facing left for no reason.

On a physical set this is a blocking problem. In AI video it’s a prompting problem, and it’s easier to get wrong because nothing on screen reminds you the line exists. Each shot is generated on its own, often from a text description that says nothing about where the other character is standing. If your wide shows character A on the left and character B on the right, every single, OTS, and two-shot after that needs to keep that same left-right relationship, or the geography falls apart the moment you cut between them.

A five-step process for storyboarding a dialogue scene

1. Read the scene once for where the temperature changes. Before you draw or describe a single panel, mark the two or three moments where the tone shifts: where it goes from small talk to real, or calm to angry. Those are your two-shot candidates.
2. Shoot the wide first, on paper. Decide where the characters are relative to each other and to the camera. Note it (character A screen-left, character B screen-right) so every later shot in the scene can reference it.
3. Assign a shot to every beat, not every line. A four-line exchange where nothing changes emotionally can live in one OTS pair. Don’t burn a generation on a new angle for every sentence.
4. Mark whose shot each panel is. For every single or OTS, write down whose experience the camera is sharing at that moment. Huston put it plainly: the camera identifies with one character and sees the other through their eyes. That’s a decision, not a default.
5. Place your two-shot at the beats you marked in step one. Nowhere else. If you’re tempted to add a second one, make sure it’s earning its place and not just filling space.

Three mistakes that make AI dialogue scenes fall apart

The first is generating a new angle for every line of dialogue. It multiplies your generations, and it also fights the scene: constant cutting reads as nervous energy even when the conversation is calm.

The second is ignoring the 180-degree line once you’ve set it. It’s easy to lose track across separate generations, especially if you come back to a scene a day later and forget which side each character was on.

The third is treating the two-shot as the default instead of the exception. It feels like the safe choice because it shows both people, but a scene told entirely in two-shots has no rhythm. Nothing gets emphasized because everything looks the same.

A better way to keep it all straight

The reason these mistakes happen isn’t a lack of skill. It’s that the plan for a scene, who’s screen-left, which beat gets the two-shot, whose point of view a shot belongs to, usually lives in a separate document from the tool doing the generating. By the third or fourth shot, that context is gone, and you’re re-deciding things you already decided.

That’s the specific problem I built ScreenWeaver to close: it keeps the storyboard panel, its shot logic, and the script it came from in one place, so the reasoning behind a shot travels with it into generation instead of getting left behind in a notes app.

A dialogue scene is one of the oldest problems in filmmaking, and it was solved with paper and a stopwatch long before anyone had a render queue. The tools changed. The plan still has to come first.

What’s the dialogue scene you’re planning right now, and where does it change temperature?

FAQ

Do I need a two-shot in every dialogue scene?

No. Some scenes work entirely in singles and OTS shots. Use a two-shot only for the moment the scene is actually about.

How do I keep the 180-degree line straight across separate AI generations?

Write it down before you generate anything: which character is screen-left, which is screen-right, and refer back to that note for every shot in the scene.

What’s the difference between an OTS and a single for AI video prompting?

An OTS includes a hint of the other character’s shoulder or profile in frame, which helps signal proximity. A single is just the one character, usually used when you want the audience fully with them.

ScreenWeaver keeps a scene’s shot plan attached to the script it came from, from the first wide to the last two-shot: screenweaver.ai