Motion Control: puppet a character with a reference video.

Most AI video tutorials focus on prompts. The most powerful generation modes don't start with a prompt at all — they start with a reference video. Motion Control is the standout. You shoot a take with your phone, feed it to the model, and the AI generates a fully styled new character performing the same blocking, the same gestures, the same timing.

It is the closest thing to performance capture we've ever had on consumer hardware, and it solves the single hardest problem in AI video — making characters move like people instead of like algorithms. This guide walks through how to use it well.

What Motion Control actually does

Motion Control takes two inputs:

1. A reference video (you, an actor, anyone) performing the action
2. A description of the new character or style

The model extracts the motion — body movement, head tilt, hand gestures, even subtle weight shifts — and re-renders that motion onto a new character that matches your description. Your reference video becomes a skeleton; the AI dresses it.

Why this is a game changer

Generative AI is great at appearance and bad at performance. It can render a flawless 1980s detective in a rain-soaked alley — and then have him gesture like a stock-photo executive. Motion Control closes that gap. You provide the performance from a real human (you), and the AI handles only the things it's actually good at: lighting, wardrobe, environment, and style.

The result feels filmed. Because half of it is.

How to record a reference that works

Motion Control's quality depends almost entirely on the reference video. A clean reference produces a clean output; a noisy reference produces drift and artifacts. Follow these rules:

1. Frame your full body. The model can extract face, hands and full-body motion — but only what's in frame.
2. Use a static camera. Lock your phone to a tripod, a stack of books, a window ledge. Camera movement adds noise that the model will try to interpret as character motion.
3. Even, soft lighting. Window light an hour after sunrise or before sunset is ideal. Avoid harsh overhead light or backlight.
4. Plain background. A wall, a door, a curtain. The fewer distractions in the reference, the better the motion extraction.
5. Wear simple clothes. Loose fabric, complex patterns and reflective material confuse the motion estimator. A plain t-shirt and pants work best.
6. Keep it under 10 seconds. Long references compound small errors. Land the performance, then end the take.

Performance rules

Block your action like you'd block a stage scene. Hit a specific mark to start, perform clear and committed gestures, finish on a clean stop. Vague, drifty performances translate to vague, drifty AI output. Crisp performances translate to crisp AI output.

Writing the character description

Once you have the reference, the character description does the rest of the work. Lead with character, then wardrobe, then environment. A simple template:

[Character archetype, age, vibe], wearing [wardrobe], in [environment]. [Lighting]. [Style anchor].

Examples that consistently land:

• "A 1970s detective in a rumpled trench coat, in a rain-soaked alley at night. Practical neon. Film noir."
• "An astronaut in a clean white suit, on the deck of a research station. Cold halogen overheads. Hard sci-fi."
• "A flamenco dancer in a deep red dress, on a cobblestone street at sunset. Warm directional light. Editorial photography."

"You stop directing the AI and start directing yourself. The AI is just the wardrobe department now." — AI Studio Editorial

Three creative uses we keep coming back to

1. Multi-character dialogue scenes. Record both halves of a conversation yourself, generate two different characters, cut the result. You become every actor in the scene.
2. Branded social content with consistent talent. Lock a character anchor, then film yourself doing different actions for different posts. The "talent" stays identical across the series.
3. Pre-vis for live shoots. Block a scene with yourself in your living room, generate a stylized version, share with the crew. You've replaced the storyboard with playable previs.

Common pitfalls and how to fix them

Drift in the face. Your reference is too small in frame, or your character description is at odds with your face shape. Move closer to camera; or describe a character with similar bone structure to yours.

Limbs warping. Your reference has motion blur or the lighting is uneven. Re-shoot in better light, with cleaner motion.

Weird wardrobe. You wore a complex pattern. Re-shoot in solid colors and re-render.

Stiff motion. Your performance was too restrained. The AI mirrors what you give it. Be bigger.

The bottom line

Motion Control is the mode that makes AI characters move like real people, because real people are the source. If you've been frustrated by stiff, algorithm-shaped performances in your generations — this is the fix. Open AI Studio, set up your phone, record one good take, and watch the rest.