INITIALIZING BUNKROS IDENTITY LAB
LOC UNDERGROUND
SYS --:--:--
01 / 14
Masterclass

Video Is Not Image + Time

It's storytelling under probabilistic motion constraints.

AI video generation fundamentally differs from image creation. You're not generating frames—you're orchestrating temporal coherence, motion physics, and camera intent across time.

Begin Your Journey
02 / 14
Fundamentals

How Video Models See Time

Video models predict the next frame based on the previous ones. They don't "understand" motion—they approximate it statistically.

Frame Prediction Timeline

Scrub through to see how temporal coherence works:

F1
F2
F3
F4
F5
F6
F7
F8
F9
F10

Current frame: 4 — The model uses all previous frames to predict the next

Coherence

Frame-to-frame consistency

Continuity

Motion flow prediction

03 / 14
2026 Landscape

The Model Ecosystem

Each model has its own personality, biases, and strengths. Understanding these helps you choose the right tool.

🎬

Kling

Cinematic quality, strong motion physics

🌊

Hailuo (MiniMax)

Fluid motion, excellent for organic scenes

Luma Dream Machine

Dreamlike quality, creative interpretations

🚀

Runway Gen-3

Balanced realism, reliable outputs

Pika

Fast iteration, stylized results

🔓

Open-Source

Full control, requires infrastructure

04 / 14
Core Skill

Anatomy of a Video Prompt

Great prompts are structured. Each component controls a different aspect of generation.

In a forest clearing, a lone dancer spinning slowly. Camera: slow dolly in. Golden hour lighting, cinematic style.
05 / 14
Deep Dive

Motion & Physics Control

Motion is where video generation fails first. Understanding physics constraints helps you write better prompts.

Motion Intensity 50%

Low = subtle movements • High = dramatic action

Realism Level 70%

Low = stylized physics • High = natural physics

Speed 1.0x
❌ Stiff Motion
✅ Natural Flow
06 / 14
Cinematography

Camera Language

Your camera prompt is often more important than your subject prompt. Define intent, not just movement.

Selected: Pan

Horizontal camera rotation from a fixed point. Great for revealing environments or following action.

⚠️ What if you forget to define the camera?

The model will choose randomly—often resulting in unstable, erratic movement that breaks immersion.

07 / 14
Advanced

Scene Consistency & Identity

Keeping characters and environments consistent across generations is the hardest challenge in AI video.

Identity Anchors

Techniques to maintain consistency:

  • 1 Detailed descriptors: "woman with red curly hair, green eyes, wearing a blue jacket"
  • 2 Reference frames: Use the same starting image across generations
  • 3 Prompt chaining: Build on previous outputs systematically
  • 4 Repetition: Repeat key descriptors in every prompt
Without Anchors
With Anchors
08 / 14
Process

Iteration as a System

The best results come from systematic refinement, not single attempts. Version your prompts like code.

v1.0
Initial Output
Score: 4/10
v1.1
+ Camera Fix
Score: 6/10
v1.2
+ Motion Adj
Score: 7/10
v2.0
Full Rewrite
Score: 9/10

Failure Analysis

Common failure patterns and fixes:

09 / 14
Expert

Model-Specific Prompting

The same prompt produces wildly different results across models. Learn each model's "language."

Same Prompt, Different Models

"A woman walking through a rainy city at night, neon reflections, cinematic"
🎬

Kling Interprets

Dramatic lighting, slower pace, film grain

Luma Interprets

Ethereal glow, dreamlike movement, soft focus

🚀

Runway Interprets

Balanced realism, natural walking speed

10 / 14
Practice

Prompt Debugging Lab

Identify what's missing in these broken prompts. Click to reveal the fix.

"A cat jumps on the table"
Missing: Camera definition, lighting, style, motion quality. The model doesn't know HOW the cat jumps or from what angle to show it.

Better: "A tabby cat gracefully leaps onto a wooden kitchen table, landing softly. Low angle locked shot, warm afternoon light through window, photorealistic"
"Cinematic forest scene with birds"
Missing: What do the birds DO? "Cinematic" alone doesn't guide motion. No temporal intent.

Better: "A flock of starlings swirling through a misty pine forest at dawn, moving as one fluid mass, slow motion, drone shot pulling back to reveal scale"
"Man walking down street, 4K"
Missing: "4K" isn't a style—it's a resolution. No emotion, atmosphere, or camera work. Could be anything.

Better: "A businessman in a grey suit walks purposefully down a rain-slicked Tokyo street at night, handheld tracking shot from behind, neon reflections, melancholic mood"
11 / 14
Creative

Creative Storytelling

AI video excels at mood, atmosphere, and abstract expression. Prompt for emotion, not just detail.

🎵 Music Visuals

Sync motion to rhythm and mood

🎭 Mood Pieces

Abstract emotional narratives

🎬 Concept Trailers

Visualize ideas before production

💫 Dream Sequences

Surreal, physics-defying scenes

Prompting for Emotion

Instead of describing what happens, describe how it FEELS:

❌ "A person sits by a window looking outside"

✅ "The weight of absence. A figure silhouetted against grey light, motionless. Dust particles drift. Time stretches. Locked shot, shallow focus on the figure's hands."
12 / 14
Critical

Ethics & Representation

With great generative power comes responsibility. Consider the implications of what you create.

Scenario: Creating a "historical" video

You want to create a video showing a historical figure speaking.

Create it with clear "AI Generated" watermark
Avoid creating it—risk of misinformation
Use stylized/artistic interpretation only

Key Questions to Ask

  • Could this be mistaken for real footage?
  • Am I perpetuating harmful stereotypes?
  • Do I have consent for likenesses used?
  • What if this went viral out of context?
13 / 14
Resources

Prompt Library

Curated, annotated prompts to learn from. Understand WHY they work, then adapt.

"Underwater light rays dance through kelp forest, sea turtle gliding peacefully left to right, camera slowly rising, dreamy ethereal quality, blue-green color palette"
Why it works: Clear subject motion (left to right), defined camera move (rising), specific atmosphere (dreamy), constrained palette
What could break: Adding too many subjects, conflicting motions, unrealistic physics
"Ink drops falling into water in extreme slow motion, swirling and expanding, abstract patterns forming, macro lens locked shot, black ink on white milk background, hypnotic meditation"
Why it works: Simple physics (ink diffusion), locked camera (stability), high contrast (clarity), clear mood (hypnotic)
14 / 14
Mastery

Your Mental Checklist

Before every generation, run through these fundamentals:

Keep Learning

Models evolve weekly. Stay curious. Experiment responsibly. Share what you learn.

Start Again