How WITCH Came to Life: From Inspiration to AI Filmmaking

WITCH is an AI-generated short film that tells the story of a mother and daughter who share a legacy of herbal healing. The mother is accused of witchcraft and burned for allegedly cursing her village. Years later, when the Black Death emerges, the daughter uses the same knowledge and remedies to save lives, and is celebrated as a hero. The film reveals that nothing changed in their actions, only in how they were perceived, exposing the dangers of fear, bias, and the long history of misunderstanding women in medicine.

With “Open Your Eyes” as the theme for the MIT AI Film Hack 2026, I was drawn to telling a story about witches and two generations of women. I found inspiration in Chloé Zhao’s Hamnet, particularly in the portrayal of Agnes Hathaway and her connection to land, nature, medicine, and mysticism. This depiction reflects a powerful, intuitive relationship between women and healing. One that has historically been misread and distorted, especially during the era of witch hunts. In this context, the hackathon’s theme felt especially meaningful. “Open Your Eyes” became an opportunity to challenge these inherited perceptions and reveal how fear and bias have shaped the way women’s knowledge of medicine is understood. This idea became the starting point for creating the short film WITCH.

By setting the story in the mid-century and centering it on two female characters, I aimed to develop a visual language that blends symbolism and mysticism. This led me to the work of Gustav Klimt, whose paintings explore themes of life, death, and the female form through rich gold tones and intricate ornamentation. His aesthetic felt deeply aligned with the emotional and symbolic layers of the story, and could be closely woven into its visual narrative. To translate this into the film, I collected and uploaded Klimt’s paintings into Flick using the “create my style” feature as visual references.

One key challenge during the generation process was that figures in oil paintings often exist in an ambiguous space between realism and stylization, making it difficult for AI to render consistent characters. AI tends to incorporate real human features into the generation process. To address this, I curated AI-generated images that achieved a more precise balance,retaining the painterly qualities of oil while maintaining coherent human features. Then, I added them back into the style set to further refine the visual standard. Additionally, I found it essential to include a range of references for both characters and environments across different shot types, ensuring greater consistency and control over the film’s overall visual output.

With the assistance of AI generation tools and the style function in Flick, I was able to produce high-quality visual assets for the film. However, generating images is only the first step. Each frame must be intentionally shaped into a cinematic shot. For example, an initial image depicted a woman with visible blemishes and a worried expression, framed in a straightforward front-facing view. While descriptive, the shot lacked intention and did not meaningfully contribute to the storytelling. To address this, I re-staged the scene by placing the woman in front of a mirror, allowing the audience to see her face through its reflection. This adjustment introduced a more layered composition while also grounding her emotion in action. Her concern becomes legible through the act of looking at herself. The mirror not only enhances visual complexity but also transforms the image into a purposeful, narrative-driven moment.

Although AI has lowered the technical and financial barriers to filmmaking, it remains a challenge to develop the cinematic vocabulary needed to design shots in a more intentional and thoughtful way since many AI filmmakers do not come from a professional filmmaking background. One powerful feature in Flick that helps address this is the “Get Inspirations” function. Users can either browse a wide range of film scene references directly through the interface or use the chatbot to upload generated images and receive relevant visual references. This makes it easier to connect AI-generated outputs with established cinematic language. I have included an example below to demonstrate how this function informed more intentional decision-making during the generation process and resulted in a more curated shot.

WITCH emerged from an exploration of how perception shapes meaning, and how the same knowledge can be judged as either danger or salvation depending on context. Through AI filmmaking tools, I was able to translate this idea into a visual narrative that blends historical imagination with symbolic imagery. While AI lowered the barriers to production, the creative process also required careful curation, intentional shot design, and continuous refinement of cinematic language. Ultimately, this project is not only about telling a story of two women across generations, but also about learning how to “open our eyes” to the biases embedded in history, imagery, and the act of seeing itself.