Google AI Studio Media Project

Generated Output

The AI-generated video

"Ignorance only becomes bliss when it saves you from madness." — Andrew Laeddis walks off screen as the camera refocuses on the lighthouse.

Project Overview

Business Problem

Studios and content creators need to produce cinematic scene extensions, previsualization, or concept footage without full production crews. This project explores whether AI Studio can faithfully extend a specific visual style and emotional tone from reference images into generated video — a practical use case for pre-production and storyboarding.

One-Sentence Summary

Used Google AI Studio to generate a 15-second cinematic video sequence by prompting the model with three reference stills from Shutter Island, a written scene context document, a custom line of dialogue, and specific camera motion instructions.

My Role

Wrote the full scene context document analyzing cinematography, color grading, period wardrobe, and camera specs. Engineered the prompts iteratively — including multi-round refinements to fix character continuity (a background figure disappearing mid-shot) and direct the final camera refocus on the lighthouse.

Biggest Challenge

Character consistency across frames. In early generations, the figure behind Andrew would disappear as the camera panned. Fixing it required multi-round prompt engineering — explicitly instructing the model to maintain all characters in frame until they naturally exit behind Andrew before the final lighthouse focus.

What I Learned

AI video generation is deeply sensitive to visual anchors — the more specific and technical the reference (color temperature, lens type, film grain), the closer the output matches. It also taught me that prompt engineering for visual media is a different skill from text prompting: you're directing a camera, not describing words.

Tools Used

Google AI Studio Veo (Video Generation) Prompt Engineering Cinematic Reference Scene Context Writing

Key Features

Multi-round iterative prompt refinement to achieve character continuity across frames. Written scene context document (color grading, lens specs, film grain, wardrobe period) fed as background context. Three reference stills used as visual anchors. Custom dialogue line integrated into the generated scene. Camera direction for handheld motion and final lighthouse refocus. Output: 15-second cinematic video clip.

GitHub & Demo

No code repository — this is a generative media project using Google AI Studio's web interface. The deliverable is the generated video (embedded above). The full prompt engineering document and scene context write-up are available on request.

Reference Stills

The three input images

These screenshots from Shutter Island (2010) were provided as visual references to ground the AI's output in the correct color palette, character staging, and atmospheric conditions. Click to enlarge.

Shot 1 — The Doctor's arrival

Shot 2 — Reverse medium (Andrew's POV)

Shot 3 — The final decision

Prompt Engineering

How the video was directed

1

Scene Context Document

Before any video prompting, I wrote a detailed "Scene Context and Narrative" document analyzing the cinematography — lens type (35mm prime), color grading (cold, desaturated, slate grays), film grain texture, wardrobe period accuracy, and the emotional meaning of each visual element. This document was fed to the model as background context.

"The images utilize a cinematic 35mm film aesthetic... a cold, desaturated palette dominated by slate grays, muted browns, and mossy greens."

2

Initial Generation Prompt

The base prompt instructed AI Studio to generate a 15-second video continuation — Andrew sitting on the steps, cigarette in hand, as the doctor watches from above. Specified the lighthouse as a background element, the stone steps environment, and the period wardrobe.

3

Dialogue + Camera Motion Refinement

Second prompt added the custom line of dialogue and specific camera movement — Andrew delivers the line, then walks off screen. The camera slowly defocuses from him and refocuses on the distant lighthouse.

"Have him say the line 'ignorance only becomes bliss when it saves you from madness' and walk off the screen as the camera unfocuses on him and focuses on the lighthouse."

4

Character Continuity Fix

Early generations had a continuity error — the background figure would disappear mid-generation. Required additional prompting to maintain all characters in frame until they exit naturally behind Andrew, before the final camera transition.

Technical Specifications

Cinematic parameters matched

Tool	Google AI Studio — Video Generation (Veo)
Reference Material	3 screenshots from Shutter Island (2010, dir. Martin Scorsese)
Output Duration	15 seconds
Simulated Lens	35mm prime — naturalistic perspective, soft background blur
Color Grading	Cold and desaturated — slate grays, muted browns, mossy greens
Film Aesthetic	High-grain, "lived-in" texture consistent with 1950s period pieces
Camera Motion	Static → slow defocus from character → refocus on lighthouse
Custom Dialogue	"Ignorance only becomes bliss when it saves you from madness."
Prompt Iterations	4 rounds of refinement