Dialogue / Character Film Prompt is a Claude AI skill — cinematically calibrated prompts for Sora, Runway, and Kling that produce footage with intentional visual grammar.
You've written the prompt. A woman sits at a kitchen table in the early morning. She's thinking about something difficult. The light is coming through the window. The footage comes back and it is technically correct in every way — a woman, a table, morning light, a contemplative expression. It looks like the establishing shot in ten thousand other films. It looks like AI.
The prompt wasn't wrong. It just gave the model nothing to work with beyond the narrative situation. It didn't specify whether the camera is at table level looking up, or mounted high looking down — two framings that produce completely different power dynamics for the character. It didn't specify whether the morning light is hard and directional, casting sharp shadows across the table surface, or diffuse and overcast, flattening the scene into something quieter. It didn't say whether the character is absolutely still or whether her fingers move slightly against the table. These are not details — they're the decisions that distinguish a scene with a point of view from footage that happens to contain a person.
The Dialogue / Character Film Prompt skill is built around a single principle: AI video tools don't fail because the technology is weak. They fail because the prompt was written without scene grammar.
Why Does AI Video Generation Produce Generic-Looking Footage?
Generic AI footage is the direct output of generic prompts — descriptions of narrative situations rather than visual specifications. When a prompt tells an AI video tool what happens in a scene, the model fills every unspecified visual decision with its training distribution average: the most common camera height, the most common lighting setup, the most common expression for the emotional state described. That average is what generic footage looks like. It is the visual equivalent of a sentence that is correct but says nothing.
A film prompt is not a scene description. It is a camera specification, a lighting specification, and a character physical specification simultaneously. Every element the prompt leaves unspecified gets filled by the model's training average — which is the definition of generic. Specificity is not polish; it is the actual content of the prompt.
The problem compounds for dialogue and character scenes specifically. Action sequences can absorb some visual ambiguity — the camera follows the action and the action provides the structure. Dialogue scenes have no inherent action to follow. Without precise specification of which character holds frame, where the cut falls, how physically close the characters are to each other, and what the camera relationship between them expresses about their dynamic, the model produces footage of two people talking that could belong to any scene in any film. The dialogue is in the writing. The scene is in the visual grammar — and visual grammar requires visual specification.
That gap is exactly what the Dialogue Character Film Prompt skill for Claude was built to close.
What Does Cinematic Calibration Change About the Prompt?
Cinematic calibration replaces narrative description with visual specification across the five layers that determine what AI video footage actually looks like. Before writing a word of the prompt, the Dialogue / Character Film Prompt skill researches current cinematic references in the genre and visual register you're working in, then builds the prompt from those five layers outward.
| Prompt layer | Generic prompt (narrative) | Calibrated prompt (visual specification) |
|---|---|---|
| Camera | Not specified — model chooses | "Handheld, medium close-up, slight push in, 50mm equivalent — camera stays with her face as she speaks" |
| Character body | "She looks nervous" or "she seems upset" | "Sits very still, weight shifted back, hands flat on the table — movement only in her eyes tracking left before she speaks" |
| Light | "Morning light through the window" | "Hard directional side light from frame left, warm 4500K, casting table shadow — her left side lit, right side in shadow" |
| Colour grade | Not specified | "Slightly desaturated, teal shadows, muted highlights — reference: A24 naturalistic drama palette" |
| Scene grammar | "Two people having an argument" | "Tight over-shoulder shot, camera axis 30 degrees off eyeline — the frame cuts off the other character at the shoulder, creating pressure without showing their reaction" |
The calibrated prompt is not longer for the sake of detail — it's more specific about the things that actually determine the visual output. A model given precise camera, character, lighting, colour, and scene grammar specifications produces footage with intentional visual decisions. A model given a narrative situation produces footage where every visual decision was made by averaging its training data.
The prompt is the direction. What you don't specify gets decided for you — by the model's training average, which is the most generic version of whatever you asked for.
What Does the Dialogue / Character Film Prompt Skill Produce?
Three complete prompt variants per scene, each with a different camera approach, plus a master character reference document for visual consistency across multiple shots.
Narrative Prompt vs Cinematically Calibrated Prompt
Same scene — a woman in her late thirties, alone in a kitchen, early morning. She's received news she hasn't processed yet. Two prompts; one written from narrative, one from visual specification.
The narrative prompt gives the model a situation. The calibrated prompt gives it a decision for every visual element that determines what the footage looks like: camera height (table level, looking slightly up), camera movement (handheld drift — neither static nor active), character physiology (completely still, fingers spread — specific physical choices), lighting (hard, directional, exact colour temperature, which side of the face), colour grade (teal shadows, desaturated highlights), duration (8 seconds), and a cinematic reference that anchors the whole frame in a specific visual tradition. The model still makes the final image. But its decisions are now constrained by intention rather than left to the training distribution average.
Is This Skill Right for Your Production?
The skill delivers the most value for character-driven scenes where the emotional content lives in physical specificity — the stillness, the eye movement, the weight — rather than in action. Dialogue scenes, contemplative moments, and character-establishing shots are where generic prompts fail most visibly and where calibrated prompts produce the largest improvement in output quality.
The next piece most people tackle from here is a podcast episode script with natural conversation flow. If you're working across the full Video & Pod workflow, the Video & Pod bundle covers everything in one place.
Frequently Asked Questions
Put this to work: the Dialogue Character Film Prompt skill for Claude turns everything above into one guided workflow you run in a normal Claude chat. Not ready to buy? Start with a free Claude skill and see how it works first.