Keeping AI Film Settings Consistent Across Shots —— Character Routes and Scene References in Poison Cat

In Poison Cat, A He arrives at a village in a sedan chair, passes through a grain-drying square, and reaches her dwelling. These places need to remain connected across shots. A path may first appear in the distance and later become the main setting for her movement. As the camera moves closer, the images need to clarify where she has come from and where she is going. We faced two linked tasks. Cultural references had to guide the design of buildings, decorations, and props. The paths and homes also had to form one village across different views. We used A He’s route to organize this work. Here, a character route means the sequence of places that a character passes through in the story.

Research on film continuity explains how visual attention and expectations shape the experience of a cut (Smith). This directs attention to the action and visual focus on each side of a cut. In this case, the practical concern is A He’s movement toward a destination. Image generation research also addresses visual control. ControlNet uses inputs such as edges, depth, and segmentation maps to guide image layout (Zhang et al. 3836). StoryDiffusion changes how information is shared within the model to help a series of images retain consistent content (Zhou et al.). These studies address layout within an image and connections across images. We focus on how we organized scene references for narrative shots. We mainly used GPT, Gemini, and Higgsfield to create images. We draw on our production notes, scene images, and shot stills to explain how we developed the settings. We compare recurring buildings and paths across the character’s route, different viewpoints, and lighting variants.

We collected references for Qiang buildings, streets, clothing, patterns, and props. Photographs of village settings supplied details of stone walls, timber fronts, and paths. We also identified the places that A He needed to visit and revised the village layout around her movement. Figure 1 shows the main village locations and A He’s route. A He’s route connects the mountain approach, village gate, grain-drying square, residential path, and dwelling. The square contains the sedan-chair stop and the grandmother’s storytelling area. This arrangement gives recurring places a role in different scenes. It also gives us a location reference when preparing another shot. We generated the overall setting before developing lanes, courtyards, and entrances. Figure 2 arranges existing images at these scales. The overview establishes the village, the lane views describe the walking environment, and the entrance view identifies a destination. We can select a reference that matches the distance and purpose of a shot.

After setting the layout, we refined materials and decorations. We turned each correction into a specific visual task and supplied a matching reference. For example, we replaced hanging corn with flags required by the film’s design. We then added references for the flags’ shape and patterns. This process turned a broad request for cultural style into concrete image changes. We prepared several views of the same setting. The courtyard images in figure 3(a–c) retain stone buildings, a paved open area, a timber frontage, and an exterior stair. Higher views reveal more roofs and courtyard edges. These recurring features give us recognizable points for checking the relationships between buildings and paths. We then refined the visual style and prepared sunny and overcast versions. The lane images in figure 3(d–f) retain a passage between stone buildings while changing the sky and the light on the walls. Some paving patterns and building proportions also change. This observation makes layout review part of choosing a weather version: we need to check the paths and buildings as well as the atmosphere. We summarize the scene comparisons as three review questions in table 1. Landmarks help identify a place. Nearby structures help check a viewpoint. Route connections place each image within the character’s movement.

Figure 4 presents three stills from the first scene. The lane view places A He inside the village. The elevated view shows her position among the buildings. The doorway view shifts attention to her destination. Each image serves a distinct purpose: establishing the setting, locating the character, or showing an entrance. This process connects scene preparation with shot design. When choosing a reference, we need to consider the viewing direction, the character’s position, and her destination. A path can sit in the background of a wide shot and fill the foreground of a closer shot. Its connection to the dwelling needs to carry through both views so that the change in shot size continues the same action. Cultural detail and place relationships work together here. Stone walls, clothing, and decorations establish the film’s visual setting. Paths, stairs, and entrances organize the action. Planning the connections first gives later detail changes a clear purpose. When another shot returns to a location, we can choose a suitable reference from the existing scene images.

The scene workflow in Poison Cat follows A He’s route. We first connected the gate, square, lanes, and dwelling. We then added cultural details and prepared views under different lighting and weather. The workflow links character movement, scene layout, and shot selection. Scene consistency depends on a place keeping its spatial role as the images change. Our working order is clear: the route defines the connections, references develop the details, and shots follow the character’s action.