TO DAY: Directorial Practice in AI-Assisted Filmmaking

TO DAY is a science-fiction short film based on a mysterious solar radiation pulse that causes humanity to lose the ability to form new long-term memories. Although time continues to pass, people can no longer remember yesterday, leading to the gradual breakdown of social order. The film follows three narrative stages: Fracture, Reconstruction, and Transcendence. The first depicts the collapse of society caused by memory loss. The second presents humanity’s attempt to preserve knowledge and maintain cooperation through recording. The third extends the story into deep-space exploration, where external memory systems allow humanity to continue its development. The central question is: If individuals cannot preserve today, can civilization still have a tomorrow? The concept of the “Reverse Tower of Babel” also inspired the project. Rather than losing language, humanity loses the continuity of experience. Written records consequently become a means of preserving shared knowledge and rebuilding trust. Through this concept, the film explores how civilization can continue when individual memory is no longer reliable. Figure 1、2 presents an AI-generated visual example of the film’s abandoned-world setting.

In generative AI filmmaking, prompts function not merely as technical instructions but also as a means of translating creative intentions into visual compositions and cinematic scenes (Ha, 2026). [1] As the director and screenwriter, the author was responsible for the story, screenplay, visual direction, and overall organization of the film. AI tools were used to assist with image generation, video production, and music composition. ChatGPT was used to assist in generating static images. Higgsfield and Kling were then used to transform these images into dynamic shots through prompts and a first-and-last-frame workflow. By defining the beginning and ending frames, the author could guide the intended visual development of each shot. Suno was used to generate music, which was integrated into the film during post-production. Figures 3、4、5、6 present examples of AI-generated visual materials developed for the film. The production workflow consisted of screenplay development, image generation, first-and-last-frame design, video generation, music creation, and final editing. This process allowed the author to create a science-fiction film without conventional location shooting. However, it also required repeated adjustments to images and generated footage to maintain visual consistency and narrative continuity.

The production demonstrated that generative AI can support the creation of individual cinematic shots and the assembly of a complete narrative film. However, generating visually convincing images does not necessarily guarantee coherent motion. A major challenge was the dependence of first-and-last-frame generation on the quality of the source images. If either frame contained visual defects, or if the two frames differed significantly in character appearance, composition, or spatial relationships, the resulting video could exhibit unnatural movement, changes in character appearance, inconsistent backgrounds, or visibly fragmented transitions. These problems indicate that image generation and video generation are closely connected stages. The quality of the initial images directly affects the possibilities and limitations of the subsequent motion. Therefore, careful image preparation, prompt refinement, and selective regeneration are necessary to improve the final result. Figure 4 shows one example of a generated detail that required visual consistency with the surrounding shots. In addition, AI-generated footage may contain inconsistencies that become more noticeable when individual shots are assembled. Post-production editing is therefore important for maintaining pacing, continuity, and the overall visual coherence of the film.

In AI-assisted filmmaking, directorial authorship can be understood through the establishment of creative intentions, the selection and editing of generated materials, and the organization of visual elements (Jung, 2026). [2]The production of TO DAY demonstrates that generative AI can expand the creative possibilities available to filmmakers, but it does not eliminate the need for directorial control. Although AI tools can generate images, motion, and music, the results do not always match the creator’s intentions. The director must evaluate the generated material, identify technical problems, decide which shots require regeneration, and organize the selected footage into a coherent narrative. The first-and-last-frame workflow makes this responsibility particularly clear. The quality and consistency of the source images influence the continuity of the resulting motion, meaning that visual planning remains essential even when AI performs much of the image generation. As the director and screenwriter, the author concludes that AI-assisted filmmaking depends not only on the capabilities of the tools but also on the creator’s ability to guide and evaluate their outputs. AI provides new methods of visual production, while the director remains responsible for connecting these methods to narrative intention and artistic expression. Through TO DAY, the author explored both the creative potential and the technical limitations of generative AI. Future development in AI filmmaking should focus not only on improving image generation but also on achieving greater consistency and control across the entire filmmaking process.