Performance direction
“Cross the room with slow, hesitant steps, shoulders hunched, pausing halfway.”
A useful motion prompt does more than name an action. It describes how the performance should unfold: what happens, what happens next, how quickly it moves, and what the body does along the way.
Use this guide to direct sequence, pacing, posture, weight shifts, and movement style with Text-to-Motion 3.0.
An action label gives the model a category. Performance direction gives it information about how the motion should read.
The second prompt adds visible direction: pace, step quality, posture, and a pause. It does not guarantee an exact result, but it gives the model more useful information about the intended performance.
Longer prompts are not automatically better. Add details that should change what the viewer sees in the body.
Most useful text-to-motion prompts combine some or all of five elements. Treat these as a drafting checklist, not required syntax.
Start with what the character is doing: walking, lifting, turning, throwing, crouching, reaching, or another full-body action.
Describe the order of events. Words such as then, before, after, pauses, turns, and stops help define how the performance develops over time.
Describe whether the movement should feel slow, quick, hesitant, urgent, tired, relaxed, or forceful. These words direct a generative interpretation; they are not numeric speed controls.
Name the physical cues that matter: shoulders hunched, weight on the back foot, head turned left, arms held high, torso leaning forward, or feet planted in a wide stance.
Style is often clearer when it is expressed through the body. Instead of relying only on a word such as sadly or angrily, describe the pace, posture, stride, arms, shoulders, and weight that should communicate it.
For example:
Use only the components that affect the performance you want. This framework is a checklist, not a magic formula.
The examples below pair exact prompts with corresponding Text-to-Motion 3.0 outputs. Watch for what changes in sequence, pace, posture, and body direction.
A prompt can describe movement around a box, bow, table, person, or environment. Text-to-Motion generates the subject’s skeletal motion—not the missing object, setting, or additional character.
Demonstrations use Text-to-Motion 3.0 and are shown as generated. Uthana generates body motion rather than complete characters, props, environments, or rendered scenes.
Better approach: describe the physical behavior that should communicate anger—pace, stride, posture, shoulders, arms, and weight.
Words such as slowly, quickly, or five steps can shape the generated performance. They should not be treated as deterministic speed, distance, timing, or step-count controls.
The prompt may mention an object, environment, or another character to give the action context. The output is still the subject’s skeletal motion. Focus most of the prompt on what the body should do.
A prompt can imply lifting, pushing, reaching, or setting something down. It does not guarantee exact collision or contact against supplied scene geometry.
Longer is not automatically better. Keep details that should change the visible performance and remove details that only describe an unseen story or setting.
Use when you want a generative interpretation of a described action or performance.
Explore Text-to-MotionUse when you already have a reference performance and want to reconstruct its timing and body movement as editable 3D motion.
Convert video to motionUse when direction, speed, stride count, repeatability, and movement style should come from explicit travel controls.
Create controllable locomotionThe prompts below are drafting examples. Unless an example is paired with a generated video elsewhere on this page, it should not be presented as a tested result or promise of exact output.
A useful prompt describes the performance rather than only naming an action. Include the sequence, pacing, posture, body direction, weight shifts, or movement style that should be visible in the result.
Not necessarily. Useful physical direction matters more than word count. Include details that should change the visible motion and remove details that do not affect the performance.
Yes. An object can provide context for an implied interaction, but Text-to-Motion generates the subject’s skeletal motion rather than the object or surrounding environment.
You can describe a step count or ending state as qualitative direction, but language is not a deterministic numeric or geometric controller. For explicit direction, speed, stride count, and repeatability, use Locomotion.
Describe what the character should do and how the performance should unfold, then generate editable full-body motion with Text-to-Motion 3.0.
