Cloud is the primary source of avatars. Enter your credentials to continue, or load local models instead.
Place the camera and the avatar until the shot looks right: drag to orbit, right-drag to pan and scroll to zoom, or use the sliders. Apply this view then saves exactly what you see. Turn avatar to match moves the orbit into the body's turn instead, so the embedded widget gets the angle too.
The avatar's expression when nothing else drives the face. Spoken emotions come and go on top. Each emotion has two dots: pull them apart and the face rests at the lower one and comes up toward the higher one now and then, the way a smile comes and goes; together, it is held at that one strength.
Between turns the avatar keeps the mood of its last reply and shows short emotional moments that fit it.
Hold any Unreal FACS pose on the face. ★ marks the pose behind an LLM emotion tag.
Share of the time the eyes are on the viewer. 1.00 = always (at rest: only the odd brief glance away); lower adds looks away, e.g. 0.63 while speaking and 0.79 while listening is what people measure at in conversation.
How far the head turns, tilts and nods, how the eyes move around the viewer's face, and how the avatar notices the cursor (desktop) or a tap (touch). Visitors who ask for reduced motion get half of every range.
Checks the eye range and alignment. While on, W A S D move where the eyes look (Shift = 5° steps), R brings them back to the viewer, Esc stops. Past about 18° in the head the head turns to carry the rest.
In a voice session, speaking over the avatar stops it and starts your turn. Turn it off where loudspeakers or other voices in the room keep cutting the avatar off; a click or tap still interrupts.
Where the eyes really point against where the camera is. Start it, have the avatar say a reply, then read the figures.
The avatar rests on its main idle; every so often a variety clip kicks in, plays through and hands back. Clips always play to their end, and the blend into the next one starts that long before the last frame. Times are in seconds.
What Unreal adds on top of a clip: the hips travel over planted feet, and a turn of the hand or foot spreads along the limb. Both are solved live on this avatar's own proportions.
Every clip in the loaded library. Pick one to play it on the avatar.
Swap the clip library the avatar draws its idles and gestures from. Any library plays on any avatar; it only changes the motion style.
Open level = how wide at average speaking loudness · Mouth range = how much stressed syllables open past it.
Per-viseme coarticulation. Edits apply live during speech.
Added on top of everything else, every frame. Save to cloud covers these sliders and the blend shapes below.
Scalp hair only. Brows and beards are untouched.
Press 💾 on any reply to save its audio and motion here for token-free replay.