This project focuses on the transition between several high-resolution, close-up photographs of eyes. Using a neural pipeline, I wanted to explore how one eye could morph into another while maintaining the specific "human" details that make a portrait feel real.

The Balance of Control

To manage the transition, I built an AnimateDiff and Stable Diffusion workflow that relied on a "weighted" system:

• ControlNet: This acted as the structural anchor. It ensured the AI respected the physical shape of the eye—the pupil, the iris, and the eyelids.

• IP-Adapter: This provided the creative influence. It allowed the AI to "dream" up the textures and colors that bridged the gap between the different eyes.

By adjusting the weights between these two tools, I could decide exactly how much the AI was allowed to experiment. If the ControlNet was too heavy, the morph felt stiff; if the IP-Adapter was too heavy, the eye would "hallucinate" and lose its human quality. This study was about finding the exact "sweet spot" where the morphing feels fluid but remains physically grounded.

A Digital Artifact

This specific workflow is now a "moment in time." The pipeline was built using custom nodes by Matteo Latini (the developer behind the original IP-Adapter integration for ComfyUI).

Because ComfyUI and the IP-Adapter software have been updated and discontinued in their original forms, this exact configuration is no longer supported. This makes the resulting video a digital artifact—a specific look captured during an early stage of AI development that can’t be reproduced with current, "updated" tools.