ComfyUI Node: 🩪 LTX Face Attention Anchor

🩪 LTX Face Attention Anchor:

The LTXFaceAttentionAnchor node is designed to enhance video processing by anchoring the identity of a face within a specific region of a video frame. This node is particularly useful in applications where maintaining the identity of a face across frames is crucial, such as in video editing or animation. It works by targeting the bounding box region of an anchor frame as the identity source, ensuring that the face's features remain consistent throughout the video. This node is part of the LTX2.x video DiT (v4.0) framework and is specifically tailored for face-region identity anchoring, providing a robust solution for face tracking and identity preservation in dynamic video content.

🩪 LTX Face Attention Anchor Input Parameters:

model

This parameter specifies the model to be used for processing. It is a required input and serves as the backbone for the node's operations, ensuring that the face anchoring is performed using the specified model's capabilities.

face_bbox_norm

This parameter defines the normalized bounding box coordinates for the face region in the format "x1,y1,x2,y2". The default value is "0.35,0.10,0.65,0.50", which represents a typical face region in normalized coordinates. It is crucial for determining the area of the frame that will be used as the identity source.

strength

This optional parameter controls the intensity of the face anchoring effect. It ranges from 0.0 to 5.0, with a default value of 0.10. A higher strength value increases the influence of the anchor on the face's identity, making it more prominent in the video.

inject_mode

This parameter determines the mode of injection for the face anchoring process. It offers options such as "tracked" and "tracked_correction", which affect how the face identity is maintained across frames. The choice of mode can impact the stability and accuracy of the face tracking.

track_threshold

This optional parameter sets the threshold for tracking the face identity. It ranges from 0.0 to 1.0, with a default value of 0.50. This threshold determines the sensitivity of the tracking process, influencing how easily the node can detect and maintain the face identity.

face_threshold

This parameter specifies the threshold for face detection within the bounding box. It ranges from 0.0 to 1.0, with a default value of 0.30. A lower threshold allows for more lenient face detection, which can be useful in challenging lighting or occlusion conditions.

identity_threshold

This parameter sets the threshold for maintaining the identity of the face. It ranges from 0.0 to 1.0, with a default value of 0.75. A higher threshold ensures that only strong matches to the original face identity are accepted, enhancing the consistency of the face's appearance.

depth_curve

This parameter allows you to select the depth curve for the face anchoring process. Options include "flat", "ramp_up", "ramp_down", "late_focus", and "middle", with "flat" as the default. The depth curve affects how the anchoring strength is distributed across the video frames.

spatial_prior

This parameter controls the spatial prior for the face anchoring, ranging from 0.0 to 1.0, with a default value of 0.50. It influences the spatial distribution of the anchoring effect, affecting how the face identity is preserved in relation to its position in the frame.

block_index_filter

This optional parameter allows you to specify a filter for block indices, which can be used to refine the anchoring process by focusing on specific parts of the video.

bypass

This boolean parameter, with a default value of False, allows you to bypass the face anchoring process if set to True. It can be useful for debugging or when you want to temporarily disable the node's effects.

debug

This boolean parameter, with a default value of False, enables debug mode, providing additional logging and insights into the face anchoring process. It is helpful for troubleshooting and understanding the node's behavior.

🩪 LTX Face Attention Anchor Output Parameters:

model

The output parameter model represents the processed model after the face anchoring has been applied. This output is crucial as it contains the updated model with the face identity anchoring effects, ready for further processing or rendering in video applications.

🩪 LTX Face Attention Anchor Usage Tips:

• To achieve optimal face tracking, adjust the strength parameter based on the video's lighting and motion conditions. Higher values may be necessary for fast-moving scenes.
• Use the inject_mode parameter to fine-tune the anchoring process. "Tracked" mode is generally suitable for stable scenes, while "tracked_correction" can help in correcting drift in more dynamic environments.
• Experiment with the depth_curve to see how different curves affect the anchoring over time. This can be particularly useful in scenes with varying depth or focus changes.

🩪 LTX Face Attention Anchor Common Errors and Solutions:

"Invalid bounding box coordinates"

• Explanation: The face_bbox_norm parameter contains invalid or improperly formatted coordinates.
• Solution: Ensure that the coordinates are in the correct format "x1,y1,x2,y2" and represent a valid region within the frame.

"Model not specified"

• Explanation: The required model parameter is missing or not properly set.
• Solution: Provide a valid model input to the node to enable the face anchoring process.

"Threshold value out of range"

• Explanation: One of the threshold parameters (track_threshold, face_threshold, or identity_threshold) is set outside its allowable range.
• Solution: Adjust the threshold values to be within the specified range of 0.0 to 1.0.