Same Character, Different Shot

A recurring character needs room for another angle, another setting, another episode. What can change while the character still reads as itself?

We propose a CESC study asking which visual changes affect same-character judgments, and how well automated identity scores track those responses. The aim is to connect a production question to viewer evidence.

Research on learning real faces from varied images motivates testing the introduction as well as the new appearance. Transfer to stylized characters remains a hypothesis. Brief experimental exposure would not reproduce years of audience familiarity.

Build a character for the study.

We would create a new, owned test character unrelated to CESC’s private properties. The first study would use still images and distinct distractor characters sharing selected cues.

A feasibility pilot would select changes in silhouette, proportions, palette, facial details, costume, prop, or rendering style. These would be graded changes, with unintended differences recorded rather than assumed away.

Candidate comparisons

Condition	What it would test
Context control	Preserve the design while varying pose, angle, or lighting.
Attribute variant	Change a specified attribute; inspect what else changed.
Distractor	Show a different character sharing selected cues.

Introduce it in different ways.

Consenting adults would be randomly assigned to narrow-variation or varied-appearance introductions, matched in image count and viewing time. The varied set would change pose, angle, or lighting while preserving the design.

This tests exposure variation. Both groups would then see held-out appearances, with randomized order and counterbalanced variants to limit learning from repeated versions of an image.

We would ask same, different, or unsure, alongside confidence. Portrayal fit would be a separate judgment. A higher tendency to answer “same” would not automatically mean better recognition; controls and distractors would help reveal that tendency.

For altered variants, the outcome would be the proportion of same-character judgments, rather than an accuracy score defined by the creator’s intention. Neither recognition nor portrayal fit would establish attachment or actual return.

Sources and scope

The exposure question draws on Ritchie & Burton’s real-face research. The proposed stylized-character study has not run. Sample size, response rules, primary comparisons, and analysis would be fixed before main collection. One character and style would supply case evidence; broader recommendations would require replication.