The LivePortraitLoadFaceAlignmentCropper node is part of the ComfyUI LivePortraitKJ extension. This node is specifically designed to load and configure a face alignment-based cropper for manipulating and processing live portrait images within ComfyUI workflows. It serves as a tool for providing facial cropping capabilities using advanced face detection and alignment techniques.
The LivePortraitLoadFaceAlignmentCropper node initializes a face alignment cropper suitable for portrait manipulation tasks. It offers a way to automatically detect and crop faces from images using pre-trained face detection and alignment models, thereby facilitating further processing in live portrait projects.
The node requires the following inputs to function efficiently:
Face Detector: A choice of face detection models, such as 'blazeface', 'blazeface_back_camera', or 'sfd'. This input defines which model to use to detect faces in images.
Landmarkrunner Device: Specifies the device on which to run the landmark detection operation. Options include CPU, CUDA, ROCM, CoreML, and torch_gpu. The choice here impacts the performance and speed of face detection.
Face Detector Device: The device to be used for running the face detector. Choices are 'cuda', 'cpu', and 'mps', which cater to different hardware capabilities.
Face Detector Data Type: Determines the data type for the face detector. Options include fp16, bf16, and fp32, affecting precision and performance.
Keep Model Loaded: A boolean input indicating whether the model should remain loaded in memory. This can optimize performance by reducing loading times in subsequent operations.
The node outputs a single item:
In a typical ComfyUI workflow, the LivePortraitLoadFaceAlignmentCropper node might be used at the beginning stage of a live portrait processing pipeline to prepare images for detailed facial manipulation tasks. It is ideal for scenarios where precise facial cropping is necessary before applying image transformations, retargeting, or blending techniques.
By setting the appropriate inputs, users can ensure that the face detection fits their specific hardware configurations and processing needs. The output cropper can then be used with other nodes, such as those for retargeting facial expressions or compositing final images.
Device and Precision Flexibility: The node offers adaptability through device and precision inputs, allowing it to be tailored to different environments and hardware setups.
Dynamic Model Loading: The option to keep the model loaded can be a significant performance booster in workflows where repeated facial detection is needed across multiple images.
Advanced Face Detection Models: The node incorporates sophisticated face detection models capable of handling various scenarios and face orientations, ensuring robustness in cropping tasks.
This node is a critical component for users who wish to leverage the full capabilities of facial manipulation in ComfyUI, providing a foundation for high-quality face alignment and detection workflows.