MakeFrame

GetKeyFrames:

This node intelligently extracts keyframes from a sequence of images by first detecting scene changes and then selecting representative frames from within each scene.

Workflow:

1. Scene Detection: The node calculates the difference between consecutive frames. Any difference exceeding the threshold marks a potential scene cut.
2. Priority Filtering: If the number of detected scenes is greater than num_keyframes, the node automatically prioritizes and selects only the scenes separated by the most significant changes. This ensures that the final output is focused on the most important moments, even with a low threshold.
3. Dynamic Allocation: The total num_keyframes are intelligently distributed among these top-priority scenes.
4. Frame Extraction: Within each scene, the allocated number of keyframes are sampled uniformly to represent the entire scene segment.

Parameters:

• scene_cut_method: The algorithm for comparing frames to detect scene cuts.
  • MSE: Fast, good for sharp cuts.
  • SSIM: Slower, but closer to human perception.
  • pHash: Robust against minor edits.
• scene_select_method: The algorithm for selecting keyframes within each detected scene. All non-uniform methods use a peak-finding algorithm to ensure the selected frames are distributed and represent distinct moments of change.
  • Uniform: Default. Selects frames evenly distributed throughout the scene.
  • Peak_Difference: Selects frames that are most different from the start of the scene, capturing the climax of the action.
  • Edge_Change_Rate: A sophisticated method that focuses on the central area of the frame, detecting the rate of change in structural edges to find moments of highest action.
  • Highest_Contrast: Selects frames with the highest visual contrast, often corresponding to the most visually striking moments.
• num_keyframes: The total number of keyframes you want to extract across all scenes.
• threshold: The core parameter for scene detection, acting on a normalized 0.0-1.0 scale for all methods. It defines the minimum relative change required to be considered a scene cut. A value of 0.1 means only changes in the top 90% of intensity are considered. A good starting point for all methods is 0.1 to 0.2.
• central_focus_ratio: (Only used when scene_select_method is Edge_Change_Rate). Defines the size of the central area to analyze (e.g., 0.3 means the central 30% of the image).

MakeGrid:

Plots given images on a grid. Calculates number of rows and columns for most even aspect ratio.

• For concurrent frame generation. Creates potentially massive image.
• Images are resized to be evenly-divisible by rows, columns, and 8 (to prevent "resize wobble").
• Empty cells are padded with repeats of last image. Black space frustrates generation.

BreakGrid:

Breaks given image grid(s) back into individual images. Good practice to use rows/columns output from MakeGrid.

---

Acknowledgement

This project is a modified version based on the original work by LonicaMewinsky. The original repository can be found at https://github.com/LonicaMewinsky/ComfyUI-MakeFrame.