UGR | 2025 | Puppet Design ‐ Blended Space (Spring Semester) - CHI-CityTech/BSP-graphic-imagery GitHub Wiki

By Samuel Cheung

Behind every puppet is a story. When I was younger, I saw a shadow puppet performance that stuck with me—watching artists and puppeteers bring characters to life through careful design and movement. That memory continues to shape how I work on our current productions, especially as we blend traditional and digital methods.

This spring, I focused on improving how we create visual assets for our puppet show, especially using a more organized approach with AI tools. All of our puppets started as original hand-drawn designs. These drawings were used to build the physical puppets and also served as visual references when we began generating digital versions or additional poses using AI.

We reused specific prompts and worked within a fixed style to keep characters visually consistent across different scenes. In earlier semesters, the AI would often generate images that didn’t match from one version to the next. This time, by sticking to one base design and refining the prompt language, we were able to keep the look of our characters steady.

Before designing the characters for this semester, we spent time revisiting and refining some of the puppets we built last year. One of the key examples was the archer puppet. The original concept for this character came from one of my own drawings, but due to time constraints, we weren’t able to convert it into a full puppet in time. As a result, the version used in our first show had limited mobility and didn’t perform the way it was intended to.

Over time, I’ve learned that AI has its limits when it comes to character design—especially in something as hands-on as shadow puppetry. While AI can help fill in details or generate references, the process of turning an image into a working puppet still depends on human effort. Every step—from adjusting the proportions to making sure the joints move correctly—requires a person behind the scenes guiding the design.

The figure above shows a comparison between the original puppet concept and its updated version after being processed through an AI model and refined by hand.

Fig 1. The original sketch for this character was drawn in FireAlpaca as a concept for a shadow puppet.

Fig 2. The sketch was imported into ChatGPT’s image workflow, where a character sprite was generated based on the original design.

Fig 3. After processing the image through Scenario AI for refinement, we compared several variations. In the end, the original drawing was chosen as the stronger option for puppet conversion. It was passed through a threshold converter to create a solid black silhouette with basic outlines. The final cut and design details—such as the holes used for light and movement—were revised in FireAlpaca to ensure they supported both visual appeal and storytelling on stage.

This semester, our puppets begin as concepts and ideas, shaped by the script and the characters needed for the production. After reviewing the storyline, we identify the potential characters that will appear and explore their visual and narrative roles.

As part of the design process, we utilize AI tools like ChatGPT to assist in innovating and refining character designs based on existing artwork. Some of the newer characters, however, were created through prompt engineering, allowing for fresh and unique designs.

After refining character designs with ChatGPT, Scenario AI is used to further enhance the artwork. Once the most suitable image is selected, it is saved and processed through a background remover.

As part of the preparation process, the character's structure is modified by removing the arms to create space for ball joints, allowing for articulated movement. Once the background is removed, an online thresholding tool converts the image into a high-contrast black-and-white format.

The finalized image is then imported into FireAlpaca for additional refinements. At this stage, the design is adjusted to ensure it is cuttable, durable, and detailed enough for laser cutting.

This section showcases puppets that may appear in the show, highlighting their design inspirations, assembly process, and final appearance. It also includes design templates and explains how the puppets are constructed. Additionally, this space explores their integration into the blended puppet environment, combining traditional craftsmanship with digital techniques.

This virtual puppet is part of a new generation created using an AI-generated template. Designed with both human and dragon forms, subtle details hint at Kara’s draconic nature. Her original human template, generated by Scenario AI, was converted into black and white before being refined in Fire Alpaca. Each limb was individually sized and adjusted for proper proportions, and a dragon-like crest was added to her sleeve to reflect her serpent heritage.

The Kunti puppet remains unchanged from its original design. However, one of its major flaws was the excessive number of holes, which caused the puppet to become structurally unstable. Due to time constraints, there was not enough time to fully reconstruct it. Instead, the puppet served as a template for the creation of the Karna puppet. While the original Kunti puppet lacks stability, future plans will address these issues to ensure its safe and effective use during the live performance.

Karna 2.0 is an upgraded version of the character from our previous production. In last year’s show, Karna’s puppet was designed using FireAlpaca, but due to tight deadlines, the final output felt rushed—resulting in a wardrobe malfunction during the live performance.

This time around, we approached the design with much more care. Starting with the Kunti puppet as a base, we reworked it to create Karna’s new form. The updated design now shares subtle visual traits with Kunti, reinforcing the fact that they’re brothers, while also reflecting Karna’s individual personality and his dream of becoming an archer.

From the refined image cuts to the chipboard shaping and the puppet prototype, every step was given more attention. We reduced the number of cutout holes in the puppet’s body, which helped improve its durability and stage performance. Thanks to these adjustments, Karna 2.0 is more stable, smoother in motion, and far less likely to fall apart during a live show.

The Dronacharya puppet was the last character chosen for the refinement process. Because of the rushed schedule last semester, there wasn’t much time to fully develop the puppet before the show. Toward the final stages, some of the limbs started to fall apart, and parts of the puppet had limited mobility during the performance.

This year, more time and care were put into the production process. Dronacharya’s design was first refined using ChatGPT, then brought into FireAlpaca for final adjustments—transforming the AI-generated image into a working puppet layout. The hole placements were planned in advance to avoid weakening key parts of the structure. By being more intentional with the design, we were able to keep the visual storytelling strong without compromising the puppet’s durability on stage.

Above is the theoretical design for the dragon puppet. The inspiration behind this design was originated from Nagikara's design.

For more information about this character click this link:

However, this version of the puppet does not fully achieve the intended look, as the serpent is meant to appear much larger. To address this, the puppet design was modified by removing the bottom coils of the serpent while keeping the individual head section. The coils were then replaced with separate coil segments positioned in the background, creating the illusion of a much larger serpent.

The dragon's design was later adjusted to emphasize its heads, while the coils were redistributed into different segments for better movement. Since the dragon would be in constant motion throughout the play, keeping a coil that made it look stationary didn’t make sense. To maintain a more fluid and dynamic appearance, the original base design was ultimately scrapped in favor of a more adaptable structure.

This is a virtual representation of the tiger puppet. Like the Nagikara serpent puppet, it was designed with stage adaptability in mind while preserving the essence of its character. The original concept featured multiple joints, but the design was later refined to incorporate adjustable legs for greater flexibility. To address concerns about structural fragility due to the intricate details, the virtual model includes reinforced elements. A transparent material, such as acrylic, is proposed for the back of the puppet to enhance durability during performances while maintaining its visual integrity.