Two ComfyUI custom nodes that convert 2D images or videos into a simulated 3D side-by-side (SBS) format, viewable on compatible devices such as VR/AR headsets like Meta Quest, Apple Vision Pro, HTC Vive, or even those crappy cardboard ones.
- Open
ComfyUI Manager- →
Custom Nodes Manager - → Search for
Y7orY7 SBS. - Install.
- Restart Restart ComfyUI
- →
-
Clone this repository into your ComfyUI custom_nodes directory:
cd /path/to/ComfyUI/custom_nodes git clone https://github.com/yushan777/ComfyUI-Y7-SBS-2Dto3D comfyui-y7-sbs-2dto3d -
Install Dependencies
# activate your venv (if you have one) # Linux/macOS source venv/bin/activate or # Windows venv/Scripts/activate pip install -r requirements.txt
-
Restart ComfyUI if it's already running.
Takes a single image as input, and its associated depth map and will produce an SBS formatted stereoscopic 3D image.
ℹ️ See More Information
Inputs/Widgets
- base_image: The image you wish to convert to 3D
- depth_map: The depth map of the base image
- method: Select the 3D rendering method:
- mesh_warping:
- grid_sampling:
- depth_scale: Controls the strength of the 3D effect (default: 30). Higher values create more pronounced depth but at the cost of tearing and artefacts
- mode:
- Parallel: For parallel viewing (left eye sees left image, right sees right image)
- Cross-eyed: For cross-eyed viewing (left eye sees right image and vice versa) - useful to check the effect if you do not have a 3D device.
- output_type
- SBS: Outputs a standard side-by-side stereoscopic image with separate left and right views. Suitable for VR headsets, 3D TVs, cross-eyed or parallel free-viewing. The image is split in half horizontally: left eye image on the left, right eye image on the right (or reversed if using cross-eyed mode).
- Anaglyph: Outputs a single red-cyan composite image that encodes stereo depth using color channels. The left eye is mapped to red, and the right eye is mapped to green/blue (cyan). This format is viewable with red-cyan 3D glasses and is ideal for quick previewing on standard monitors without special hardware.
- depth_blur_strength: Controls how much to blur the depth map transitions (3-33, odd values only).
- Lower values for sharper depth separation between objects or layers.
- Higher values for smoother transitions between depth planes, though this may introduce some distortion in some images.
Takes a video's frames as input, and their associated depth maps and and will produce an SBS formatted stereoscopic 3D video.
ℹ️ See More Information
Inputs/Widgets
- frames: The frames of the video you wish to convert to 3D
- depth_maps: The depth maps of video frames
- method: Select the 3D rendering method:
- mesh_warping:
- grid_sampling:
- depth_scale: Controls the strength of the 3D effect (default: 30). Higher values create more pronounced depth but at the cost of tearing and artefacts
- mode:
- Parallel: For parallel viewing (left eye sees left image, right sees right image)
- Cross-eyed: For cross-eyed viewing (left eye sees right image and vice versa) - useful to check the effect if you do not have a 3D device.
- output_type
- SBS: Outputs a standard side-by-side stereoscopic image with separate left and right views. Suitable for VR headsets, 3D TVs, cross-eyed or parallel free-viewing. The image is split in half horizontally: left eye image on the left, right eye image on the right (or reversed if using cross-eyed mode).
- Anaglyph: Outputs a single red-cyan composite image that encodes stereo depth using color channels. The left eye is mapped to red, and the right eye is mapped to green/blue (cyan). This format is viewable with red-cyan 3D glasses and is ideal for quick previewing on standard monitors without special hardware.
- depth_blur_strength: Controls how much to blur the depth map transitions (3-33, odd values only).
- Lower values for sharper depth separation between objects or layers.
- Higher values for smoother transitions between depth planes, though this may introduce some distortion in some images.
- temporal_smoothing: Controls smoothing between frames (0.0-0.5). Higher values create more consistent depth perception between frames but may reduce responsiveness to rapid depth changes.
- batch_size: Number of video frames to process at once. higher = more memory usage.
- Cross-eyed mode is primarily used for viewing stereoscopic 3D images on a regular 2D screen without needing any special equipment. Here's how it is used:
- With Parallel, the left-eye image feeds the left eye, and the right-eye image feeds the right eye.
- But with Cross-eyed, this is flipped it places the left-eye image on the right side and the right-eye image on the left side.
- When you cross your eyes, each eye ends up looking at the correct image, and your brain fuses them into a 3D image which will appear centered. If done correctly, that middle image will appear 3D without the need for a VR headset or 3D glasses.
- More Information here
I would highly recommend using depth maps generated by Depth-Anything V2. You can use MiDaS but DepthAnything V2 produces more dense, higher-fidelity depth maps and tends to preserve sharper edges and fine structures better which I feel are better suited for situations where accurate object boundaries and separation matter more.
The included workflows both use Kijai's DepthAnythingV2 custom nodes. As a starting point I would recommend using the depth_anything_v2_vitl_fp16 variant of the model which isn't too big at 1.3GB
Note that some 3D viewer apps will interpret square aspect ratios per eye as VR180 format. For example in ImmerGallery, you can force the format by suffixing the filename with _LR, or you can modify the image via its menu.
You can find and download sample SBS images and videos generated from these custom nodes here, Transfer them to your device and check them out. Below are some examples.
Image credits: Pixabay
If you are creating sbs images or video clips then it is assumed that you already have a viewing device and software. In any case, what I have been using is the Quest 3 with the immerGallery software.