Resolve whether apparent discrepancies arise from user execution or generation artifacts

Ascertain whether discrepancies between a user's physical materials or actions and a generated visual target result from the user's execution or from an artifact of visual generation.

Background

Contextualized visuals can closely resemble the user's workspace, which helps users compare the guidance with their materials but can also make anomalous changes appear meaningful. When a generated target is plausible yet differs from the physical result, users may be unable to determine whether they made a mistake or whether the model produced an unreliable depiction.

The paper identifies this attribution problem as an unresolved interpretive issue. Resolving it would help users decide whether to correct their physical actions, disregard the generated discrepancy, or seek alternative guidance.