三次未命名项目 Three Acts of Silence

• 

  poster

Creative Approach Three Acts of an Untitled Project replaces plot with structure and dialogue with imagery. It builds a looping narrative about female experience in non-linear time. What we focus on is not how women act, but how they are seen, named, and disciplined. Three ruptured relationships across different historical moments are edited into a recurring, inescapable pattern. AI-generated visuals intertwine with live-action footage not to create spectacle, but to create distance—a way of seeing and a form of control.

This is a minimal-resource production. With a very small budget, we push for strong visual impact through a cold, fragmented, compressed rhythm. The goal is not to recreate reality, but to construct a new one; not to tell a story, but to dismantle the structure that stories rely on.

AI Workflow The film is built on a “live-action first, AI as extension” model. Live action carries the core narrative and performances. AI is used in three areas: visual transitions, surreal sequences, and replacing shots that would otherwise require a high budget.

We primarily use Jimeng AI, supported by Veo3, Higgsfield, and other video-generation tools. Jimeng serves two key functions: style-driven generation for transitions and dreamlike scenes, and frame expansion or detail replacement, such as extending crowd scenes or building simulated aerial shots.

In production, the storyboard clearly marks which shots are live-action and which are AI. We shoot all essential performances first, then feed the footage into AI models to generate dream states or large-scale environments. The final images are unified in DaVinci Resolve and After Effects through compositing and color work.

The overall idea is to use AI to enhance and substitute—not replace—live action. This keeps the tactile realism of real footage while unlocking imaginative scale and reducing the cost of large setups and aerial scenes, creating a world where the real and the surreal merge.

Built With

• aigc