AI avatar with hand gestures: make an avatar move or dance

To give an AI avatar specific hand gestures or dance moves, drive it with a video of those moves: a motion-control model copies the motion of a reference clip onto the avatar's still image. On Sume, Avatar 1.0 talking videos take no gesture or pose setting, so you send the avatar's handle and a driving clip of up to 30 seconds to Kling 3.0 Motion Control, and the output follows that clip's motion and length.

Facts come from Sume's Generate avatar video, Video generation, and Face swap docs and the motion control schema in the Sume API reference, read on 2026-09-27; anything called current behavior is read from Sume's code. The motion control API guide covers the endpoint in full.

Does an Avatar 1.0 video include hand gestures?

Only the ones generation adds. In current code, every clip's prompt asks for natural talking motion: facial expression changes, blinking, mouth movement, slight head movement, relaxed shoulders, and small hand gestures. You can't choose them. POST /v1/avatar-1.0/talking-video has no gesture, pose, or motion field, and any unlisted field fails with 400 ("… is not supported in this launch API contract.").

How do I make an avatar do specific gestures?

• Record or pick a driving clip of the gestures, at most 30 seconds long, at a fetchable public HTTPS URL.
• Send POST /v1/kling/3.0/motion-control with the avatar's avatar_handle, the clip as motion_video_url, and the clip's length as duration_seconds (1–30).
• Don't describe the gestures in prompt: motion and framing follow the driving video, and the prompt only steers appearance details.
• Set character_orientation to image to keep the avatar still's framing; the default, video, keeps the clip's.
• It bills per driving-video second, rounded up, plus a 5.5% agent fee by default; GET /v1/catalog lists the current rate.

curl -X POST https://api.sume.com/v1/kling/3.0/motion-control \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: avatar-gestures-001" \
  -d '{
    "avatar_handle": "product_host",
    "motion_video_url": "https://example.com/gestures.mp4",
    "duration_seconds": 8,
    "keep_original_sound": false,
    "character_orientation": "image"
  }'

How do I make an AI avatar dance?

Use a dance clip as the driving video. The output length follows the driving video, so one job covers up to 30 seconds of dancing. The clip's own audio track stays in the output unless you set keep_original_sound: false, which returns a silent clip.

No reference clip? Describe the moves instead. On POST /v1/videos, pass the avatar's preview_image_url, a public Sume-hosted image, as a first_frame in frame_images, and put the dance in prompt; the motion then comes from the text. Each model lists the frame types it accepts in supported_frame_images, and the image-to-video guide shows the request.

Can the avatar talk and gesture at the same time?

Not with gestures you choose, in one documented call. Avatar 1.0 talks with only the small gestures generation adds. The motion control request covers motion, framing, and whether to keep the clip's sound; it has no script or speech-audio input. VEED Fabric 1.0 lip syncs a still to audio but takes no motion input.

To put the avatar's face on footage you film yourself, Face swap is in Beta: it applies a ready avatar's face to a short public source video, which its docs currently plan at about 4–15 seconds with usable audio. See Avatar Face Swap API (Beta).

Which way should I make my avatar move?

Pick the route by where the movement should come from. The motion control API guide lists every motion control field and prices an example.