Generative AI and the Director's Vision — From Prompt to Pre-Production

A director sits with a cinematographer. The script is open. The scene is described: "A woman walks through a market at golden hour. She pauses at a flower stall. Something catches her attention off-screen."

From this text, two things must emerge: a visual plan (framing, movement, lens, lighting) and an emotional intention (nostalgia? surveillance? freedom?).

Traditionally, this translation — from text to visual and emotional plan — happens entirely in human conversation, mediated by talent, experience, and shared creative intuition.

Now, for the first time, AI can participate in that conversation.

Generative AI as Creative Interlocutor

The most productive way to think about generative AI in pre-production is not as a replacement for creative decisions, but as a creative interlocutor — a participant in the dialogue between concept and image.

Consider how a director might use a multimodal AI during pre-production:

Director: "Show me the market scene. Late afternoon. Shoot it like Wong Kar-wai — slow, intimate, saturated."

The AI generates five images. Each interprets the prompt differently — different market architectures, different compositions, different quality of light.

Director: "Number three is closest. But I want the flower stall more prominent. And the light should be coming from behind her — backlit."

New generations. Refinement continues. In ten minutes, the director has explored more visual possibilities than a concept artist could generate in two days — not because the AI is more talented, but because it is faster at the exploration stage.

The creative decision remains entirely human. The exploration velocity is artificially enhanced.

The Prompt as Creative Language

One of the most interesting developments in generative AI is the emergence of prompting as a creative language — a new way for directors to articulate visual intent.

A traditional director's brief might say: "I want this scene to feel claustrophobic but beautiful." A prompt for generative AI must translate that into specific visual parameters: "tight framing, warm color palette, shallow depth of field, textured walls close to the subject, volumetric light from a single practical source."

This translation process is itself creatively valuable. Prompting forces directors to articulate their visual intent with more precision than informal conversation typically demands. The AI doesn't accept vague intentions — it requires specific visual vocabulary.

In practice, this sharpens the creative vision. Directors who use generative AI in pre-production report that the prompting process itself helps them discover what they want — by iteratively defining what they don't want.

Storyboard Generation: The First Application

The most immediately practical application of generative AI in pre-production is storyboard generation.

Traditional storyboarding is expensive and slow. A feature film requires 800-2000 storyboard frames. A professional storyboard artist produces 15-30 finished frames per day. At $300-500 per day, a full storyboard costs $10,000-30,000 and takes 4-8 weeks.

AI storyboard generation can produce:

• Rough storyboards from scene descriptions in minutes
• Consistent character appearance across all frames (using reference images and style-locking techniques)
• Multiple visual approaches for the same scene, enabling rapid comparison
• Animated storyboards (animatics) by interpolating between generated frames

The human storyboard artist is not eliminated — they become a curator and refiner of AI-generated options, focusing their talent on creative judgment rather than repetitive rendering.

Casting Visualization

A controversial but increasingly used application: casting visualization. Before any actor is approached, generative AI can produce concept images of characters in different casting configurations.

This is not deepfake territory — no real person's likeness is used. Instead, the AI generates character archetypes that exploring age, ethnicity, physical type, and wardrobe options, allowing the creative team to develop a visual language for each character before casting begins.

Used responsibly, this can:

• Expand the casting imagination beyond default industry biases
• Test character visual relationships (do these two characters look like they belong in the same world?)
• Generate wardrobe and physical characterization options that inform costume design

Location Scouting Enhancement

Location scouts photograph potential sites. Generative AI can then transform those photographs into visual predictions of how the location will look under production conditions:

• "Show me this location at dawn with fog"
• "Add set dressing appropriate for 1940s Paris"
• "Show me the same street with and without period vehicles"
• "Light this interior with practicals only — no overhead lighting"

This allows directors to make location decisions based on the location's cinematic potential, not just its current appearance — dramatically improving the quality of location selection for productions that can't afford extensive set construction.

The Danger of Over-Specification

At Al-Haytham Labs, we are conscious of a risk inherent in generative AI pre-production: over-specification.

When a director generates a perfect image of every scene before shooting, the risk is that shooting becomes mere reproduction — an attempt to match the AI's output rather than to discover new possibilities in real space with real actors.

The best use of generative AI in pre-production is as a conversation starter, not a conversation ender. It should expand the visual imagination, not constrain it. It should provoke ideas, not specify outcomes.

For this reason, we advocate for what we call "loose generation" — deliberately imprecise AI outputs that establish mood, atmosphere, and emotional direction without defining exact compositions. The director arrives on set with a visual feeling, not a blueprint.

Because the greatest moments in cinema are rarely planned. They are discovered — in the gap between intention and reality, in the space where the world offers something the imagination hadn't considered.

The Director's Vision, Augmented

Generative AI will not replace the director's vision. It cannot — because vision is not the ability to generate images. It is the ability to know which images matter.

What AI does is lower the cost of visual exploration to nearly zero. Every visual possibility that a director can describe becomes an image they can see. Every atmospheric variation becomes testable. Every creative instinct becomes verifiable.

The director's vision remains the north star. AI is the telescope that brings more of the visual universe into view.

From prompt to production. From imagination to image. The creative process has not changed. The bandwidth has.