jackrr/eymo

AI Summary: This issue proposes enhancing the language to allow explicit input specification for command statements, rather than implicitly using the previous statement's output. This aims to prevent unintended artifacts from intermediate processing steps. Solutions include enabling parallel statements (DAG-like structure) or explicit input labeling, which will require refactoring to manage original and intermediate textures.

AI Summary: Implement support for MacOS virtual webcam output in the Eymo face-manipulation program. This involves researching and integrating a suitable MacOS virtual camera solution, likely involving CoreMedia, to allow users to stream the modified video output as a virtual webcam source.

AI Summary: Optimize the Eymo face transformation program by combining multiple individual shader runs into a single, more efficient GPU render pass. This involves restructuring the vertex data to include transformation metadata (brightness, saturation, channels, etc.) within the vertex struct itself, thereby avoiding redundant pipeline executions.

AI Summary: Investigate and implement optimizations to reduce the time spent decoding input images from the nokwha library in the Eymo face manipulation program. This involves exploring alternatives to the current decoding method or finding ways to speed up the existing process, aiming to reduce the 40% of compute time currently dedicated to this task.

AI Summary: Implement a runtime interface for the Eymo face transformation program, allowing users to add, remove, and modify transformations without restarting the application. This involves designing and implementing a communication mechanism (e.g., a socket or named pipe) to receive commands, parsing those commands, and dynamically updating the active transformations within the program's main loop.

AI Summary: The task is to improve the interaction between the `tile` operation and other image transformations in the Eymo face-manipulation program. Currently, enabling tiling causes other transformations to be skipped, and the goal is to make them work together correctly.

AI Summary: The issue is that the face landmark output from MediaPipe becomes inaccurate as the face tilts, affecting the accuracy of transformations applied by eymo. The task is to investigate and fix the inaccuracy in the landmark detection within the `landmarks.rs` file, specifically lines 56-57, potentially involving adjustments to the MediaPipe configuration or implementing a correction algorithm to compensate for tilt.

AI Summary: The task is to debug a stateful transform in the `eymo` face manipulation program. A timeout in frame processing is causing the reset of stateful transformations (like `drift` and `spin`), which should maintain their state across frames. The issue points to a specific line of code in `lang.rs` that needs to be reviewed and corrected to prevent this state reset.