There are several things you can keep in mind to help your results improve:
- A few consistent training frames will provide smoother results than a lot of inconsistent ones. The more training frames that are added for similar shapes, the more it can confuse the software and the more likely it is that there will be jittering.
- Use whatever parts of the face that you can consistently see to place landmarks accurately from frame to frame. Landmark placement doesn't have to be the same in every shot as long as it's the same within the current shot. Use whatever you can see to help with the placement, whether it's a hair, a wrinkle, a birthmark, etc.
- Track in small sections. If there's a particular section of the video that is posing a problem, track just that section by selecting that frame range by holding CTRL and dragging on the timeline. The use the "Train Model- Selected" and "Track - Selected" functions to work just in that section.