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)
audio - Optional ComfyUI AUDIO, e.g. the audio output of VHS Load Video

Both stereo output nodes support audio. Connect the video loader's audio output directly to the stereo node; audio bypasses image upscaling and depth processing. Without this connection, output remains silent.

The batch node retains the first batch's audio and muxes it once, after the last meta-batch. Supply the complete source audio, as VHS Load Video does, rather than individual audio chunks. Audio is encoded as AAC; the H.264 video is copied without re-encoding. Longer audio is trimmed to the video; shorter audio is padded with silence. Temporary audio files use the output directory and are removed after muxing. If muxing fails, the silent video is retained.

For a directory workflow, expose the path loader's AUDIO output from its group and route audio from the same source as the selected images. Output fps must match the loader's effective frame rate for synchronization. After updating, restart ComfyUI before starting a new meta-batch run.

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