Text Prompt Guide

In SayMotion, a Text Prompt serves as your guiding instructions for creating dynamic 3D animations, leveraging the power of Generative AI.

‍

Best Practices for Text Prompts to Generate 3D Motion Animation

For the best experience with SayMotion, craft precise prompts for SayMotion by treating your character like an actor, providing specific yet interpretable directions. Each animation currently can only be a maximum of 10 seconds, so plan your motions accordingly incorporating diverse elements like action type, direction, emotion, and more for nuanced and accurate animations.

How to Say More Words for Less Motions

Consider describing characteristics that convey emotion in the motion. Explore the descriptive elements below. Picture your motion in context. From what position is your actor starting their motion? How are they feeling? What arm or leg do they need to use? Are there any objects they are interacting with such as a chair or staircase?

Enhance the richness of your prompt by including these elements: 

• Action type: e.g. squat, kick, run
• Direction: e.g. forward, clockwise, left, right
• Body parts: e.g. arms, hip, left hand, right arm
• Velocity: e.g. fast, sprint
• Trajectory: e.g. clockwise circle/square, Z-shape
• Relative position: e.g. right hand side, over head
• Emotion: e.g. angry, joyful, spiritless
• Style: e.g. effortless, energetic, fluid, mechanical, frantic
• Distance: Exact distances are not currently supported. Lean into direction like 'forward' 'backward' etc.

‍

Detailed Prompt Example

"A person walks clockwise in an angry manner, and then reaches out with their left arm to grab the left arm of a chair, lowering themselves heavily into the seat."

Why this works: 

The motion is very specific notice that several motion types have been strung together to generate something unique and the animated character has emotion and depth.

‍

Good Prompt Example:

"A person walks forward and pokes something on the ground with a spear in their right hand."

‍

Why this works:

While not highly specific, the model understands the general motion of the character and location of the object they're interacting with infusing some depth to the movement.

‍

Bad Prompt Example:

"Stand walk sit"

‍

Why this doesn't work: 

Vagueness is the enemy here. Similar to giving unclear directions to an actor, the model needs specific actions to perform. The example lacks clarity on how the actions should be executed and when so the directions ended up too contradictory. For the model to generate the desired motion more specific and coherent actions are needed.