Nodes/VRGameDevGirl Video Enhancement Nodes/VRGDG LTX Preview XYZ Plot
ComfyUI Node

VRGDG LTX Preview XYZ Plot

Creates a final checkpoint comparison video from saved preview videos once training reaches the target step.

By vrgamegirl19·Created about a year ago·Updated 22 days ago· 533
VRGDG LTX Preview XYZ Plot
  • vhs_filenames
  • xyz_video_path
  • created
  • status
preview_folder
output_nameLTXChunkRun
completed_steps0
total_target_steps3000
cell_width0
cell_height0
label_tilestrue
output_fps24
render_backendNVIDIA GPU (h264_nvenc)
CategoryVRGDG/Training

Inputs (10)

NameTypeDefaultDescription
vhs_filenames*Trigger input from VideoHelperSuite Combine Video so this node runs after preview videos are written.
preview_folderSTRINGFolder containing the saved preview videos, or the VHS filename_prefix from the trainer node. If a filename_prefix is provided, this node automatically uses its parent output folder.
output_nameSTRINGLTXChunkRunLoRA/output name used to filter matching preview videos and to name the final XYZ video.
completed_stepsINT00–1000000Current completed training step from the trainer node.
total_target_stepsINT30001–1000000Final training target step. The XYZ compare video is only created when completed_steps reaches this value.
cell_widthINT00–4096Width of each tile in the final comparison grid. Use 0 to auto-detect from the first preview video. Examples: 320 for many checkpoints, 512 for a balanced layout, 768 for larger tiles.
cell_heightINT00–4096Height of each tile in the final comparison grid. Use 0 to auto-detect from the first preview video. If labels are enabled, the label band is added automatically.
label_tilesBOOLEANtrueAdds a label band above each tile using the preview filename, which should include the LoRA name and step.
output_fpsINT241–120FPS for the final comparison video. Example: 24 for standard preview playback.
render_backendCOMBONVIDIA GPU (h264_nvenc)Final XYZ export backend. CPU uses libx264. NVIDIA GPU uses NVENC for faster encoding when ffmpeg has NVENC support. Note: the grid compose filters still run in ffmpeg, so GPU mode mainly speeds up the encode stage.

Outputs (3)

NameTypeDescription
xyz_video_pathSTRING
createdBOOLEAN
statusSTRING