👀 SBS Image Uploader by SamSeen

On its own, this is just a LoadImage with a resize option and an upload button. You might wonder why it exists at all. In this pack's workflow it's the difference between a 30-second depth pass and a three-minute one.

What it is

The SBS Image Uploader loads a single image from ComfyUI/input, optionally resizes it, and outputs a standard batched IMAGE tensor. The dropdown lists every image file already sitting in your input folder (png, jpg, jpeg, webp, bmp), and the pack's web extension adds an Upload image button so you can push a file in from the browser instead of manually dropping it into the folder. It's the still-image sibling of the pack's Video Uploader.

Why it exists

The stereoscopic shift - and the depth pass before it - is a per-pixel loop. Feed in a 4K photo and you're burning GPU time on pixels the 3D effect doesn't need. The node's default max_width of 512 keeps that loop cheap and the VRAM bounded. If you're feeding a single image into the SBS node to test a look, or prepping a frame before it goes into a batch, capping the resolution here is the difference between instant feedback and watching a progress bar crawl.

Inputs and output

• image - dropdown of files in your input directory.
• max_width (64–4096, default 512, step 8) - cap the width.
• max_height (0–4096, default 0 = auto) - leave at 0 to scale height proportionally; set both if you want a hard bounding box.
• Output - one IMAGE tensor.

Resizing preserves aspect ratio and uses LANCZOS, and sizes step in multiples of 8, which video encoders appreciate.

The gotchas (both real, both from the source)

• Empty input folder. If there are no images to list, the dropdown shows "none" and the node silently returns a black 512×512 tensor. First time it happens you'll think the pack is broken. Put an image in your input folder (or use the Upload button).
• Soft output that isn't the stereo node's fault. max_width defaults to 512, so that 4K photo is downscaled on load. If your SBS result looks soft, check whether you actually changed this before blaming the depth or the shifting. Bump the cap when you need detail.

One more note: this node is RGB-only and doesn't expose a mask output like ComfyUI's built-in LoadImage. For most of this pack's use that's fine - you're feeding the SBS node, not doing compositing.

Install

No separate install; it ships with the pack and brings no dependencies of its own beyond the pack's:

cd ComfyUI/custom_nodes
git clone https://github.com/SamSeenX/ComfyUI_SSStereoscope.git
pip install -r ComfyUI_SSStereoscope/requirements.txt
# restart ComfyUI

Or search "SideBySide_Stereoscope" in ComfyUI Manager.

Honest verdict: you already have LoadImage, and for general work it's better. Reach for this one when you're inside the SS Stereoscope workflow and you want the resize control (and the in-browser upload) without adding a separate downscale node to the graph. The resize is the whole game - it's what keeps the per-pixel depth and shift loops from being the slowest thing in your day.