Pose Recognition

<<< Quickly navigate using the tabs on the left side.

Add Pose saves the exact body posture you perform when you say “Capture.”

You can then use that pose to trigger the corresponding keyboard, mouse, or other output actions.

Tracking Options

When you click Add Pose Recognition, you’ll see several tracking options:

Dual-Hand Tracking (Recommended): Tracks both hands. Best for preventing false triggers, since typically one hand is in a fixed position while the other is performing an action.

Single-Hand Tracking: Useful when there are few skill bindings and you want poses to trigger more casually.

Torso Tracking: Enable this if your actions involve upper-body rotation or leaning.

Leg Tracking:Only available as dual-leg tracking. Since walk recognition is essentially a specialized form of leg pose recognition, you must place your camera at knee height and align with the human silhouette for better tracking. If you can’t align properly, at least ensure your feet are near the bottom edge of the camera view.

You’re free to try out different tracking options. Each tracking option will produce different similarity behavior—so you’ll need to experiment and find the one that works best for your pose and gameplay.

Capture Interface

Frame Rate

Shown in the top-left (landscape) or bottom-left (portrait). Slightly lower during training than in real gameplay.

Similarity

Generated once a pose is Captured. Indicates how similar your current posture is to the saved one, as a percentage.

Observe Similarity Fluctuation

Your pose may shift slightly during intense gameplay, so it’s important to observe the range of similarity values and allow some buffer. (Sometimes a one-frame similarity drop might not even be noticeable on screen.)

To end training, say “Save Data” or click manually to exit.

You can also say “Capture” to overwrite with a new pose, or “System Off” to cancel the current Pose Capture.

Pose Widget

After successfully Capturing a pose, you got a pose widget.

Trigger Similarity:

Determines how closely your current pose must match the saved one to trigger an output

Higher value = harder to trigger;

Lower value = easier to trigger.

Leave Similarity:

When your similarity drops below this value, The Link considers the pose “released.”

Resets all "Press Once” actions to allow re-trigger.

Releases any “Hold” actions.

Reverses state for all “Hold”-based Wrist To Cursor, Wrist To Mouse, Page Toggle, and Link Toggle.

For keyboard and mouse output, any keys set to "Press Again" will execute actions after the icon

For keyboard and mouse output, any keys set to "Spam" will stop.

Stay:

This value specifies how many frames the pose must be held to trigger.Use this to prevent accidental trigger during intense gameplay.

Forward and Backward Poses

Normal cameras see the world like a flat image. They can tell where something is up, down, left, or right, but they can’t tell how close it is to the camera.

To detect forward and backward movements, The Link uses size differences (closer looks bigger, farther looks smaller).

However, this method is not as accurate as horizontal movements. So try to design more horizontal gestures when possible.

Sometimes we really need a forward or backward pose(like punching or sword strikes), set a lower similarity value and add some stay.

Adding Multiple Pose/Inputs

Each Link can accept up to 3 inputs, including:

Pose

Hitbox

Voice

Walk/Jump

Arrow Down: Inputs are executed in order (Sequential).

The Link’s Pose Recognition can only recognize one pose at a time. But if you want to recognize a complete action (like a swing

punch or a double kick), you can split it into 2-3 key poses and add them as a sequence . The Link will do the outputs after you complete the action.

**Tip:**Most actions have a start pose and an end pose. So, in most cases, adding just these two poses is efficient and fast.

Scroll your mouse wheel over the indicator to switch input logic:

Parallel Lines(AND): All conditions must be met to trigger .

More precise pose control:

This feature allows you to split a pose into separate body parts, enabling independent similarity thresholds—for example, left hand and right hand can be tuned separately for higher precision. Each body part is evaluated individually for similarity, and only when all conditions are met will the action be triggered.

Splitting a single pose into multiple body segments:

You can also combine poses with voice commands, hitboxes, walk or jump detection—use your imagination!

Dashed Line (OR):Any input triggers .

Useful for making key skills highly reliable during intense movement.

You can add multiple versions of a pose to cover slight variations.

Pose not triggering?

Switch to Spam Mode:

Sometimes the problem isn’t that your pose wasn’t recognized—it’s that you missed the game’s input window.When playing with keyboard&mouse, you may unconsciously spam a key to make sure your skill or magic is cast as soon as possible.To simulate this behavior, you can set the key output mode to "Spam" in The Link. For More details Please Check Keyboard&Mouse, Similarity

Tip: Enable Show Keys in the floating window settings to check if The Link is actually recognizing the input.

Lower Similarity

The simplest solution to make a pose easier to trigger is to lower the trigger similarity.

Use Hitboxes

Hitboxes are one of the most stable input methods.You can combine a Hitbox with a low-similarity pose to ensure reliable triggering without frequent false positives.

More details about Hitbox please check: Hitbox

Multiple Poses For the Same Action

Refer to the multi-pose input above.

Add multiple variations of a pose that you might make during fast-paced gameplay.

Connect them with a dashed line (OR logic)—any one of the poses being recognized will trigger the action.

Pose Mistriggers

Add Stay

Most reliable way to prevent accidental triggers.You need to stay in the pose for a while to trigger.

Sometimes, with stay, the pose may fail to trigger frequently. In that case, try lowering the "exit similarity." This way, when you hold the pose, it won't accidentally exit and reset the stay count.

For poses that frequently misfire:

High Trigger Similarity: Further ensures that the pose is only triggered when you perform it correctly.

Higher Stay Frames: This ensures the pose won’t be accidentally triggered.

Low Exit Similarity: Prevents the stay count reset.

For a pose with high stays , first do the captured pose, then make small, subtle movements until it is triggered.

Try more precise pose control:

Refer to Add Multiple Poses → More Precise Pose Control.

Split a single pose into multiple body segments and connect them using AND logic, so you can independently adjust the similarity threshold for each part.

Use Dynamic Page/Link Toggle

For games with complex skills or multiple stances, use dynamic page or link toggle controls.

This limits pose recognition to only the relevant set at any given time, reducing misfires.

For more details, please check Toggle Page and Toggle Link

Combine Pose with a Hitbox

Hitbox acts as stable input constraints—you must touch (or leave) the drawn hitbox for the pose to trigger or deactivate.

This allows highly precise and tailored execution conditions.

Use parent link to temporarily activate a link

By using a Link as a trigger condition, you can make another Link active only for a short window after the first one activates.

This is useful for:

Create Combos

Reflective Combat abilities

Increasing input flexibility without increasing misfire risk

Back to previous directory

Wiki

Pose Recognition