Rule-Based Camerawork Controller for Automatic Comic Generation from Game Log

Abstract

Abstract. We propose a rule-based camerawork controller for a recently proposed a comic generation system. Five camerawork rules are derived through an analysis of online-game webcomics about Lineage 2, one rule for each of the five event types: chatting, fighting, moving, approaching, and special. Each rule consists of three parts relating to the three camera parameters: camera angle, camera position, and zoom position. Each camera-parameter part contains multiples shot types whose value indicates the frequency of their usages in the analyzed webcomics. In this paper, comic frames generated with the proposed camerawork controller are shown and compared with those generated with our previous controller based on heuristic rules, confirming the effectiveness of the proposed camerawork controller.

Camerawork for Comics Generated from Visitors’ Experiences in a Virtual Museum

“…Previously, we proposed in [11] and [12] a camerawork editor and an automatic camerawork module, respectively. The former allows the viewer to manually edit the camerawork of a generated comic while the latter automatically decides the camerawork of each frame based on rules derived from an analysis of online-game web-comics.…”

Diversified Virtual Camera Composition

“…Thawonmas et al [22] identify variety of shots as a major problem in automatic generation of cinematics and they introduce a roulette-wheel selection mechanism to force variety in shot descriptions. However, by altering the shot properties, this approach does not only vary the shot visual aspect but potentially changes the shot meaning.…”

“…The properties for this problem include full visibility of all characters and a projection size equal to half of the screen for all characters and an horizontal angle of 90 degrees to the right of each character. The last problem is based on the chat scene by Thawonmas et al [22] and it includes three characters with one ideally chatting to the other two. The visibility and projection size properties are equal to the ones in the first problem but the camera is also expected to be on the back of the listening characters.…”

“…However, as pointed out by Thawonmas et al [22], one single best solution is often unsatisfactory. When filming a scene with little movement, such as a dialogue, selecting always the same solution will lead to a repetitive direction.…”