Match the starting state closely enough that the inputs do not describe contradictory movements. Compare body facing, visible hands, limb coverage and weight distribution. You do not need to copy every joint angle blindly, but a seated character and a standing jump recording create a different problem from two similar standing poses.
Compare the first clear state
Use the character image and the first clearly visible performance frame side by side. If the clip begins in a blur or transition, find whether a usable preparation exists before the main event. Keep that preparation when choosing the excerpt.
Check five relationships: torso direction, head direction, hand location, supporting leg and visible body coverage. Write differences in ordinary terms, such as “image faces the viewer; recording begins in profile.” Avoid describing a hidden hand's exact position when you cannot see it.
This comparison identifies ambiguity. It is not a measurement of the model's internal pose representation.
Keep identity separate from body preparation
A reference can establish facial appearance and clothing while still being a poor motion starting point. Create two review lists. The identity list covers recognizable face, hair and approved clothing. The pose list covers the visible starting arrangement.
Do not repair the pose by accepting a newly altered face. If you create another character reference, compare identity before using it. For a real person, prefer an authorized photograph of the needed posture when available. For an original fictional character, preserve the approved design while changing only the preparation that the shot requires.
Act-Two specifically recommends a similar starting pose for gestures. That advice concerns its performance workflow; it does not establish one required pose for every dance model.
Resolve the largest conflict first
| Input conflict | Practical first correction |
|---|---|
| Opposing facing directions | Select a compatible source or documented orientation mode |
| Hands hidden in one input | Use a clearer reference showing the gesture's starting state |
| Seated versus standing | Re-record or redesign the intended action |
| Tight crop versus wide arm movement | Prepare a wider character composition |
| Different props in the hands | Clarify the prop state before motion |
Change one major conflict, then repeat the comparison. If you alter the crop, costume, camera and action together, an improvement becomes difficult to attribute.
Test the transition out of the starting pose
Choose a segment that leaves the preparation visibly: a hand rising, a shoulder turn or a shift toward one side. Watch whether the result begins with that transition or jumps directly to a later state.
Inspect the body before the face. A successful facial likeness can distract from a wrist that emerges from the wrong location. Then review normal playback for an abrupt opening or a brief pose reset.
Keep both inputs and the comparison notes with the test result. They make a later revision more precise than remembering that the previous version “felt awkward.”
When similarity is not enough
Should both inputs face exactly the same direction?
Compatibility is the goal. Check any documented orientation setting and inspect the result; matching direction does not guarantee correct hands or weight transfer.
Can a prompt fix a badly mismatched pose?
A clear brief can express the intended start, but it cannot prove missing body evidence. When the visible contradiction is large, improve the reference or performance before expanding the prompt.