Hi Maureen,
I understand why this would be desirable. You have hit on one of the areas of difficulty-
keywords on the asset instead of the annotation. If detection finds multiple animal annotations
in a single image, it could have multiple viewpoint keywords.
Further, the keyword does not have a record of the action that created it. There is no way to find a viewpoint keyword from detections to manipulate automatically and exclude other keywords created in other ways.
For now, the best way to deal with it is to remove previous viewpoint keywords and create new ones once you make a manual annotation.
The addition of keyword creation to manual annotation is more of a feature request. For our next generation platform Codex, we have certainly discussed issues like these and are starting out with keywords available on annotations.
Thanks!