ComfyUI Node: 🐇 RIFE VFI FPS Resample

🐇 RIFE VFI FPS Resample:

RIFE_FPS_Resample is a powerful node designed to convert video clips from one frame rate to another using the RIFE (Real-Time Intermediate Flow Estimation) model. This node is particularly beneficial for AI artists and video creators who need to adjust the frame rate of their videos for various purposes, such as creating slow-motion effects or ensuring smooth playback on different devices. By leveraging advanced interpolation techniques, RIFE_FPS_Resample can synthesize intermediate frames, allowing for non-integer frame rate conversions that maintain visual continuity and reduce flicker. Additionally, the node includes optional stabilization features to protect edges and enhance the overall quality of the output video. This makes it an essential tool for anyone looking to achieve professional-grade video transformations with minimal effort.

🐇 RIFE VFI FPS Resample Input Parameters:

ckpt_name

This parameter specifies the name of the RIFE model checkpoint to be used for the frame rate conversion. It determines the version and architecture of the model, which can affect the quality and speed of the interpolation process. There are no specific minimum or maximum values, but it should match a valid model name available in your setup.

frames

This parameter is a tensor containing the video frames to be processed. It serves as the input data for the node, and its quality and resolution will directly impact the final output. The frames should be provided in a format compatible with PyTorch tensors.

fps_in

This parameter represents the input frame rate of the video. It is a floating-point value that indicates how many frames per second the input video currently has. The minimum value is greater than 0, as a frame rate of 0 or less is invalid.

fps_out

This parameter specifies the desired output frame rate for the video. Like fps_in, it is a floating-point value and must be greater than 0. The node will adjust the video to match this frame rate, either by synthesizing new frames or decimating existing ones.

scale_factor

This parameter controls the scaling of the interpolation process. It is a floating-point value with a minimum of 0.25 and a maximum of 4.0, with a default of 1.0. Adjusting this factor can enhance or reduce the detail in the interpolated frames.

ensemble

This boolean parameter determines whether to use ensemble techniques during interpolation. When set to True, it can improve the robustness of the output by combining multiple predictions, but it may increase processing time.

linearize

This boolean parameter, when enabled, applies linearization to the interpolation process. It can help in achieving smoother transitions between frames, especially in videos with varying lighting conditions.

lf_guardrail

This boolean parameter activates a low-frequency guardrail, which helps in stabilizing the interpolation by reducing low-frequency noise. It is useful for maintaining the integrity of the video content.

lf_sigma

This parameter sets the sigma value for the low-frequency guardrail, influencing the strength of the noise reduction. It is a floating-point value with a default of 13.0.

source_pair_match

This boolean parameter, when enabled, ensures that source frame pairs are matched more accurately during interpolation. It can improve the consistency of the output, especially in complex scenes.

match_a_cap

This parameter sets a cap on the scale of matching during interpolation. It is a floating-point value with a default of 0.02, controlling the maximum allowable scaling difference between frames.

match_b_cap

This parameter sets a cap on the offset of matching during interpolation. It is a floating-point value with a default of 2.0/255.0, controlling the maximum allowable offset difference between frames.

edge_band_lock

This boolean parameter, when enabled, locks the edges of the video frames during interpolation. It helps in preserving the sharpness and clarity of edges, reducing artifacts.

tau_low

This parameter sets the low threshold for edge detection during interpolation. It is a floating-point value with a default of 1.5/255.0, influencing the sensitivity of edge preservation.

tau_high

This parameter sets the high threshold for edge detection during interpolation. It is a floating-point value with a default of 6.0/255.0, influencing the sensitivity of edge preservation.

band_radius

This integer parameter defines the radius of the edge band during interpolation. It affects the area around edges that will be preserved, with a default value of 4.

band_soft_sigma

This parameter sets the sigma value for softening the edge band during interpolation. It is a floating-point value with a default of 2.0, controlling the smoothness of the edge transition.

clear_cache_after_n_frames

This integer parameter specifies the number of frames after which the cache should be cleared. It helps in managing memory usage during long processing tasks, with a default value of 10.

🐇 RIFE VFI FPS Resample Output Parameters:

frames

The output parameter is a tensor containing the resampled video frames. This tensor reflects the new frame rate specified by fps_out, with any necessary intermediate frames synthesized to ensure smooth playback. The quality of the output depends on the input parameters and the RIFE model used, providing a seamless transition between frames and maintaining the visual integrity of the original video.

🐇 RIFE VFI FPS Resample Usage Tips:

• To achieve the best results, ensure that the ckpt_name matches a model version that suits your specific needs, as different versions may offer varying levels of performance and quality.
• Experiment with the scale_factor to find the optimal balance between detail and processing time, especially when working with high-resolution videos.

🐇 RIFE VFI FPS Resample Common Errors and Solutions:

"fps_in and fps_out must be > 0"

• Explanation: This error occurs when either the input or output frame rate is set to 0 or a negative value, which is invalid.
• Solution: Ensure that both fps_in and fps_out are set to positive values greater than 0.

"Invalid model name"

• Explanation: This error indicates that the specified ckpt_name does not match any available RIFE model checkpoints.
• Solution: Verify that the ckpt_name corresponds to a valid and accessible model in your setup.