Look-at for mesh bodies: directional head animations, and look-at through Pose Studio sync

sky_line commented

2026-10-21 00:40:36 +00:00

Owner

Problem

The built-in look-at turns the avatar's head and eyes toward what you look at, using LL's head/eye joint motion on the standard mHead / mNeck / mEyes bones. Many mesh bodies and heads don't show it, or show it wrongly:

Different bones: non-human bodies (quadrupeds, ponies and the like) rig the head to other bones, or use extra neck bones (Bento), so turning mHead alone looks wrong or does nothing.
Animations override it: an AO or pose keeps the head locked.
Others never see it: look-at isn't an animation, so other viewers never receive it. Only the eye/head look-at target is sent, and each viewer applies it to the standard skeleton.

Goal: look-at that works for mesh bodies. Two methods, which can coexist.

Method 1: directional animations (works for everyone, in every viewer)

Drive look-at with real animation assets that turn the head and neck chain toward a direction, so every viewer shows it.

Animation set: a set of directional poses (left, right, up, down and diagonals; e.g. a 3x3 or 5x3 grid of yaw and pitch, plus centre). Low-priority, head/neck-only animations that body AOs don't override. Where possible, reuse the owner's existing Pose HUD animations (the per-direction head poses) as the source set.
Selection: the viewer picks the grid cell nearest the look-at direction (camera target, chat partner, hovered object, or another avatar), with hysteresis and a dwell time so it doesn't flicker. It starts and stops the animation via AgentAnimation, the same channel gestures use. Optionally it cross-fades two neighbouring cells, if the sim's priority and blend behaviour allows.
Configuration: per body, a choice of animation set (bundled set, user set from inventory, or none), a list of bones the set drives, and a maximum yaw and pitch.
Trade-offs: the animations are coarse, stepped by grid cells, and each change costs a network message, so changes are rate-limited. They also need the animation assets to exist for the body's skeleton.

Method 2: viewer-to-viewer look-at through Pose Studio sync (smooth, Slipstream only)

Use the existing Pose Studio sharing protocol (pose channel, consent rules, helper bridge) to send a continuous look-at that other Slipstream viewers apply directly to the bones.

Data: a compact look-at message carrying a target direction or point, a weight, and a per-body bone map (which neck, head and eye bones, and how the rotation splits along the chain). It is sent at a low rate and smoothed on the receiver.
Receivers: they apply it as a procedural layer on top of animations, the same way Pose Studio drives bones, respecting per-bone limits.
Scope: only Slipstream viewers see it. Others fall back to method 1, or to nothing.
Rules: it follows the pose-sharing consent and security rules. A wearer's own look-at needs no consent from others, but receivers can mute it.

Shared pieces

Look-at target choice: reuse LL's look-at targets (camera focus, selected object, chat) with smoothing. Add a setting for whether look-at follows the camera, and keep privacy in mind: some users hide where they look, so there must be an off switch, and it should respect LL's existing look-at privacy settings.
Bone maps: per-body bone-map presets (standard human, Bento neck chain, common quadruped and pony rigs). These are editable in Pose Studio.

Acceptance

With method 1, other viewers see a mesh quadruped's head turn roughly toward what the wearer looks at, in coarse steps, with no flicker.
With method 2, other Slipstream viewers see it turn smoothly and continuously.
It can be turned off entirely.

## Problem The built-in look-at turns the avatar's head and eyes toward what you look at, using LL's head/eye joint motion on the standard mHead / mNeck / mEyes bones. Many mesh bodies and heads don't show it, or show it wrongly: - **Different bones:** non-human bodies (quadrupeds, ponies and the like) rig the head to other bones, or use extra neck bones (Bento), so turning mHead alone looks wrong or does nothing. - **Animations override it:** an AO or pose keeps the head locked. - **Others never see it:** look-at isn't an animation, so other viewers never receive it. Only the eye/head look-at target is sent, and each viewer applies it to the standard skeleton. Goal: look-at that works for mesh bodies. Two methods, which can coexist. ## Method 1: directional animations (works for everyone, in every viewer) Drive look-at with real animation assets that turn the head and neck chain toward a direction, so every viewer shows it. - **Animation set:** a set of directional poses (left, right, up, down and diagonals; e.g. a 3x3 or 5x3 grid of yaw and pitch, plus centre). Low-priority, head/neck-only animations that body AOs don't override. Where possible, reuse the owner's existing Pose HUD animations (the per-direction head poses) as the source set. - **Selection:** the viewer picks the grid cell nearest the look-at direction (camera target, chat partner, hovered object, or another avatar), with hysteresis and a dwell time so it doesn't flicker. It starts and stops the animation via AgentAnimation, the same channel gestures use. Optionally it cross-fades two neighbouring cells, if the sim's priority and blend behaviour allows. - **Configuration:** per body, a choice of animation set (bundled set, user set from inventory, or none), a list of bones the set drives, and a maximum yaw and pitch. - **Trade-offs:** the animations are coarse, stepped by grid cells, and each change costs a network message, so changes are rate-limited. They also need the animation assets to exist for the body's skeleton. ## Method 2: viewer-to-viewer look-at through Pose Studio sync (smooth, Slipstream only) Use the existing Pose Studio sharing protocol (pose channel, consent rules, helper bridge) to send a continuous look-at that other Slipstream viewers apply directly to the bones. - **Data:** a compact look-at message carrying a target direction or point, a weight, and a per-body bone map (which neck, head and eye bones, and how the rotation splits along the chain). It is sent at a low rate and smoothed on the receiver. - **Receivers:** they apply it as a procedural layer on top of animations, the same way Pose Studio drives bones, respecting per-bone limits. - **Scope:** only Slipstream viewers see it. Others fall back to method 1, or to nothing. - **Rules:** it follows the pose-sharing consent and security rules. A wearer's own look-at needs no consent from others, but receivers can mute it. ## Shared pieces - **Look-at target choice:** reuse LL's look-at targets (camera focus, selected object, chat) with smoothing. Add a setting for whether look-at follows the camera, and keep privacy in mind: some users hide where they look, so there must be an off switch, and it should respect LL's existing look-at privacy settings. - **Bone maps:** per-body bone-map presets (standard human, Bento neck chain, common quadruped and pony rigs). These are editable in Pose Studio. ## Acceptance - With method 1, other viewers see a mesh quadruped's head turn roughly toward what the wearer looks at, in coarse steps, with no flicker. - With method 2, other Slipstream viewers see it turn smoothly and continuously. - It can be turned off entirely.

sky_line added the

feature

avatars

labels

2026-10-21 00:40:36 +00:00

sky_line added this to the Slipstream project

2026-10-21 00:40:36 +00:00

sky_line referenced this issue from a commit

2026-10-22 08:20:13 +00:00

Look-at sync for mesh bodies through Pose Studio pose sync (#902)

sky_line referenced this issue from a commit

2026-10-22 08:20:13 +00:00