MotionLab

The Director's Syntax

Advanced configuration for the dual-stream engine.

01
🧠

The Cognitive Process

Thinking Tokens

MotionLab operates differently from standard diffusion models. Before generation begins, our engine utilizes a proprietary "Thinking Token" architecture.

•
How it works: Instead of reacting to keywords immediately, the model aggregates context to ensure semantic stability before rendering the first frame.

The Implication: You do not need to repeat yourself or "spam" the prompt with tags. State your intent clearly in natural language. The model "ponders" the relationship between elements to ensure the audio matches the visual physics.

02
🎬

The Screenplay Standard

Cinematic Structure

The engine is trained to understand cinematic structure. For optimal results, write your prompt as a screenplay scene using the Present Tense.

The Formula

[Context/Setting] + [Subject Action] + [Audio Atmosphere] + [Dialogue]

Bad Syntax

"Cool robot, fighting, explosions, loud noise, HD."

Correct Syntax

"INT. HANGAR. A rusted battle-droid repairs its own arm. Sparks fly with a metallic hissing sound. The droid looks up and whispers: 'System critical.'"

03
🔊

Joint Synthesis

Audio-Visual Sync

We utilize an Asymmetric Dual-Stream Architecture (14B Video / 5B Audio). This means audio is not an afterthought; it is generated simultaneously and shares timing with the video.

•
Bidirectional Logic: Visual cues (an object falling) will trigger auditory events (a thud sound). Conversely, describing a specific sound can influence the visual timing.

💬Dialogue & Lip-Sync

To trigger Lip-Sync, place text inside "quotation marks". Describe the emotion of the voice before the quote (e.g., 'Screaming in terror: "Watch out!"').

04
📷

Optical ControlCRITICAL

Camera

Do not describe camera movements in the text prompt.

•
Why: Text descriptions of camera moves (e.g., "Pan Left") pollute the semantic tokens and degrade quality.
•
The Solution: Use the dedicated Lens Modules (LoRA) in the interface to control trajectories (Pan, Tilt, Zoom, Truck). Let the text handle the content, and the interface handle the lens.
05
🔁

Loop Theory

VJ & Animation

MotionLab is specialized for continuous motion.

For VJ Loops

Use keywords like "Seamless loop," "Cyclic motion," and "Infinite flow."

Subject Isolation

Keep the background description minimal or solid if you intend to use the Extraction Module later for transparent overlays.

06
🚫

Exclusion Protocol

Negative Prompt

To ensure structural integrity, instruct the engine on what to reject.

Standard Void List

MorphingJitterDisjointed limbsText overlaysFrozen subjectSilenceAudio distortion