Lukas Schmidt 1b49fc9c62 feat: DukeStereoVideoBatch with VHS meta_batch streaming
Adds a batch variant that keeps one ffmpeg writer alive across requeued
runs via meta_batch.outputs, so long videos stream to disk instead of
accumulating frames in RAM. Fixes the 48-byte truncated mp4 left behind
by an interrupted batch.
2026-11-07 18:15:09 +01:00
2026-02-09 21:48:55 +01:00
2026-02-08 18:08:16 +01:00
2026-02-10 08:15:42 +01:00
2026-02-09 21:29:26 +01:00

ComfyUI-DukeStereo

A ComfyUI custom node that converts video frames to side-by-side (SBS) stereo format for VR viewing.

Features

  • Polylines Fill - High-quality gap filling during stereo warping (from StereoImage Node)
  • Edge-aware depth blur - Smooths depth while preserving edges
  • Streaming output - RAM-efficient, writes frames directly to video
  • H.264/yuv420p - VR-player compatible output
  • Format-Strings - %date:yyyy-MM-dd%, %time:HH-mm-ss%, %divergence%, %width% etc.
  • Home-Expansion - ~/video/... paths supported
  • Auto-Increment - Existing files get _001, _002 suffix

Installation

cd ComfyUI/custom_nodes
git clone https://github.com/killertofu86/ComfyUI-DukeStereo.git
pip install -r ComfyUI-DukeStereo/requirements.txt

Requirements

  • torch, transformers, numpy, Pillow, numba, scipy, ffmpeg

Parameters

Parameter Default Description
output_path stereo_output.mp4 Output path with format strings
divergence 2.5 3D strength (higher = more depth)
separation 0.0 Additional horizontal offset
stereo_offset_exponent 2.0 Non-linear depth curve
convergence_point 1.0 Zero-parallax plane (0.0-1.0)
fill_technique polylines_soft Gap filling: polylines_soft, polylines_sharp, none
depth_blur 0 Depth blur kernel (0 = off)
depth_blur_edge_threshold 6.0 Edge preservation threshold
fps 24 Output framerate
crf 19 Quality (lower = better)

Inputs

  • images - Video frames (IMAGE)
  • depth_maps - Depth maps from LeReS or similar (IMAGE)

Output Path Examples

~/video/stereo_%date:yyyy-MM-dd%_%time:HH-mm-ss%.mp4
~/video/test_div%divergence%.mp4
/output/stereo_%width%x%height%.mp4

License

MIT

S
Description
No description provided
Readme MIT
227 KiB
Languages
Python 98.7%
JavaScript 1.3%