This page is an example of how I approach AI-assisted filmmaking — not as a shortcut, but as a production process.
Case Study:
A Single Shot, Built Properly
A Single Shot, Built Properly
This case study demonstrates how I approach complex, cinematic camera movement using a filmmaker-first mindset — designing physically plausible motion, locking continuity, and building the shot so it survives editorial, VFX, and delivery.
AI is used as a controlled execution layer, not a shortcut. Every decision starts with lens choice, camera constraints, and narrative intent.
AI is used as a controlled execution layer, not a shortcut. Every decision starts with lens choice, camera constraints, and narrative intent.
The Challenge
The Shot: Camera starts tight on a smartphone screen (clean plate for later UI insert), then slowly pulls back and booms down to reveal a mechanic on a creeper under a car. The mechanic slides out, sits up, grabs the phone, and looks at it
The Ask: Explain, step-by-step and in as much detail as possible, exactly how you would generate this full shot today with current tools to achieve maximum consistency, cinematic quality, and precise camera control.
The Ask: Explain, step-by-step and in as much detail as possible, exactly how you would generate this full shot today with current tools to achieve maximum consistency, cinematic quality, and precise camera control.
image description here.
Why This Shot Matters
This isn’t a flashy idea.
It’s a control problem.
A shot like this tests:
Most AI video falls apart somewhere in that list.
So instead of “prompting harder,”
I approached this like a real shoot.
This isn’t a flashy idea.
It’s a control problem.
A shot like this tests:
- camera consistency
- spatial logic
- character stability
- lighting continuity
- emotional clarity
Most AI video falls apart somewhere in that list.
So instead of “prompting harder,”
I approached this like a real shoot.
Philosophy and Concept
"
I don’t believe AI replaces filmmaking.
I believe it removes friction from it.
The job is still the same:
plan the shot, control the frame, respect the audience.
Tools change. Taste doesn’t.
"
My Process
Plan the shot before generating anything
Camera type. Lens feel. Lighting direction.
This opening frame is the most important image in the sequence.
If that’s wrong, everything downstream is noise.
Separate elements that need control
The phone screen is treated as a clean plate so the UI can be handled later — the same way real commercials are made.
That separation keeps flexibility and avoids locking mistakes into the video.
Lock the environment before moving the camera
The car, the creeper, the space, the lighting — all of that needs to feel real and consistent.
If a mechanic watches this, it should feel correct.
Treat the character like a real casting decision
The mechanic isn’t a one-off face. He’s a repeatable character.
That consistency makes the rest of the work possible.
Design the camera move intentionally
This isn’t random AI motion.
It’s a controlled pullback and jib-style move that communicates scale and realism.
Focus on the reaction, not the trick
The moment the mechanic looks at the phone is where the shot actually lives.
Everything before it is setup.
EXECUTIVE / TECHNICAL
BREAKDOWN
BREAKDOWN
For those who want the full technical breakdown, below is the step-by-step construction of the shot.
When I read “camera starts tight on a smartphone screen (clean plate)”, my first consideration is cinematography, not AI.
I immediately define:
For the clean plate, I would not assume a green screen by default. I would research best practices for screen replacement in AI-assisted pipelines. Ideally, the phone screen is generated as a neutral, reflection-accurate blank surface, with UI handled as a separate, isolated asset.
My preferred approach:
I immediately define:
- Camera type and sensor feel
- Lens choice (macro vs standard close-focus)
- Lighting direction, intensity, and falloff
- Emotional intention of the opening frame
For the clean plate, I would not assume a green screen by default. I would research best practices for screen replacement in AI-assisted pipelines. Ideally, the phone screen is generated as a neutral, reflection-accurate blank surface, with UI handled as a separate, isolated asset.
My preferred approach:
- Generate or ingest a still UI mockup in a separate project
- Version and approve UI independently
- Composite or insert the UI later for maximum control and flexibility
Once the starting frame is locked, I focus on camera movement continuity.
The phrase “booms down” would be translated into cinematic language. In film terms, this is a jib move, not an audio boom. That distinction matters — both for human collaborators and for AI interpretation.
By specifying a jib, I’m communicating:
The phrase “booms down” would be translated into cinematic language. In film terms, this is a jib move, not an audio boom. That distinction matters — both for human collaborators and for AI interpretation.
By specifying a jib, I’m communicating:
- A crane-based camera move
- Physical space requirements
- A controlled, cinematic arc rather than a handheld or drone move
- A jib arm of ~12–14 feet
- A set footprint large enough to justify that motion
- A grounded, professional production environment
Next, I validate domain accuracy.
The mechanic’s creeper isn’t just “a creeper” — it’s a specific object with cultural and professional meaning. I would research:
The vehicle matters just as much:
I would also lock the final frame of the previous shot before generating movement, ensuring spatial and visual continuity across generations.
The mechanic’s creeper isn’t just “a creeper” — it’s a specific object with cultural and professional meaning. I would research:
- Common brands
- Typical colors (e.g., red)
- Whether the terminology aligns with real-world mechanic language
The vehicle matters just as much:
- Most common car colors
- Most common vehicle types
- Whether the setting is suburban, industrial, municipal, or blue-collar urban
- A sedan
- A pickup
- A work truck
- A municipal or fleet vehicle
I would also lock the final frame of the previous shot before generating movement, ensuring spatial and visual continuity across generations.
Before animation, I establish the mechanic as a repeatable character.
I would:
Versioning isn’t overhead — it’s production infrastructure.
I would:
- Generate multiple isolated images or short clips
- Lock facial features, age, body type, and wardrobe
- Treat the character as an asset, not a one-off
- Narrative continuity
- Future reuse
- Expansion into episodics, BTS, or longer arcs
Versioning isn’t overhead — it’s production infrastructure.
At this stage, consistency is already solved.
Because the character, environment, phone, UI, lighting, and camera language are locked:
Because the character, environment, phone, UI, lighting, and camera language are locked:
- The focus shifts to performance and cinematography
- The work becomes about quality, not correction
- Does the mechanic slide out fully into frame?
- Is it one continuous move or segmented cuts?
- Do we rack focus from the phone to the mechanic?
- Do we reveal detail — worn hands, grease, phone case wear — to imply lived experience?
When the mechanic looks at the phone, the shot lives or dies on emotion.
I would define:
I would define:
- The exact facial expression
- The emotional question being asked of the audience
- Whether the reaction is subtle, deadpan, concerned, relieved, or conflicted
Results
This case study demonstrates how I approach AI-assisted filmmaking as a production process, not a prompt experiment.
The goal was to design a single, technically demanding shot and execute it using current consumer-access AI tools while maintaining cinematic logic, spatial realism, and narrative intent.
The process successfully established and stabilized all required production components:
Current publicly available AI image and video models still struggle with multi-object spatial consistency, scale accuracy, and occlusion across iterations. These limitations are not solvable through prompting alone without paid access, internal tooling, or compositing pipelines.
Rather than forcing an artificial “final result,” this case study stops at the point where professional judgment would intervene in a real production.
The goal was to design a single, technically demanding shot and execute it using current consumer-access AI tools while maintaining cinematic logic, spatial realism, and narrative intent.
The process successfully established and stabilized all required production components:
- A repeatable, grounded mechanic character suitable for reuse
- A physically plausible auto-repair environment with correct spatial logic
- Separate, controllable assets (vehicle, creeper, phone)
- Cinematic camera intent defined using real-world filmmaking language
- A step-by-step workflow that mirrors traditional commercial production
Current publicly available AI image and video models still struggle with multi-object spatial consistency, scale accuracy, and occlusion across iterations. These limitations are not solvable through prompting alone without paid access, internal tooling, or compositing pipelines.
Rather than forcing an artificial “final result,” this case study stops at the point where professional judgment would intervene in a real production.
What This Example Represents
This isn’t about one shot.
The job is still the same: plan the shot, control the frame, respect the audience.
This isn’t about one shot.
It’s about:
- plan shots before generating anything
- design for control and consistency
- understand where AI works — and where it breaks
- apply traditional filmmaking logic to modern tools
AI is just another tool in the chain.
The job is still the same: plan the shot, control the frame, respect the audience.
Versioning Example
Final Words
- KennyCreative Director / Filmmaker / AI Curator"Thank you for visiting my Case Study on AI Filmmaking! If you want the full technical breakdown (step-by-step), you can download it below."
Steps
The actual work.