ComfyUI Node: MiniMax H3 Face Refine Per-Frame Denoise / 逐帧去噪 (Advanced)

MiniMax H3 Face Refine Per-Frame Denoise / 逐帧去噪 (Advanced):

The MiniMaxH3FaceRefinePerFrameDenoiseT8Advanced node is designed to enhance video quality by applying a sophisticated denoising technique specifically tailored for face refinement on a per-frame basis. This node utilizes an audited upstream face-size curve, which is calculated by dividing the crop height by the crop factor, to adjust the video noise mask. This approach ensures that the denoising process is sensitive to the size of the face in each frame, providing a more nuanced and effective noise reduction. The default settings map a 30px face to a noise reduction factor of 0.8 and a 120px face to 0.35, with a smoothing effect applied over 9 frames. Importantly, the audio mask remains unaffected, maintaining its original state. This node is particularly beneficial for projects requiring high-quality face refinement in video sequences, as it offers a targeted approach to noise reduction that preserves essential facial details while minimizing unwanted noise.

MiniMax H3 Face Refine Per-Frame Denoise / 逐帧去噪 (Advanced) Input Parameters:

face_size_curve

This parameter represents the audited upstream face-size curve, which is used to determine the appropriate noise reduction factor based on the size of the face in each frame. It is calculated by dividing the crop height by the crop factor. The curve helps in mapping different face sizes to specific noise reduction values, ensuring that smaller faces receive more denoising compared to larger ones. The default mapping is set to 0.8 for a 30px face and 0.35 for a 120px face. Adjusting this curve can significantly impact the effectiveness of the denoising process.

smoothing_frames

This parameter defines the number of frames over which the smoothing effect is applied. The default value is 9 frames, which helps in creating a seamless transition between frames, reducing the appearance of noise while maintaining the integrity of facial features. Increasing the number of smoothing frames can lead to a more gradual noise reduction effect, while decreasing it may result in a more abrupt change between frames.

MiniMax H3 Face Refine Per-Frame Denoise / 逐帧去噪 (Advanced) Output Parameters:

denoised_video

The output parameter denoised_video provides the video sequence with the applied per-frame denoising effect. This output is crucial as it represents the enhanced video quality, with reduced noise levels specifically around facial areas. The denoised video maintains the original audio track, ensuring that only the visual component is altered. This output is particularly valuable for projects that require high-quality visual refinement without compromising the audio integrity.

MiniMax H3 Face Refine Per-Frame Denoise / 逐帧去噪 (Advanced) Usage Tips:

• To achieve optimal results, adjust the face_size_curve parameter to match the specific face sizes present in your video. This customization can enhance the effectiveness of the denoising process.
• Experiment with the smoothing_frames parameter to find the right balance between noise reduction and frame continuity. A higher number of smoothing frames can provide a more natural look, especially in videos with rapid motion.

MiniMax H3 Face Refine Per-Frame Denoise / 逐帧去噪 (Advanced) Common Errors and Solutions:

Error: "Invalid face size curve"

• Explanation: This error occurs when the face size curve is not properly defined or falls outside the expected range.
• Solution: Ensure that the face size curve is correctly calculated by dividing the crop height by the crop factor. Verify that the values are within the expected range for your specific video.

Error: "Smoothing frames out of range"

• Explanation: This error indicates that the number of smoothing frames specified is not within the acceptable range.
• Solution: Adjust the smoothing_frames parameter to a value that is within the recommended range, typically between 5 and 15 frames, to ensure smooth transitions between frames.