⛧Load Images Conditional

ComfyUI's stock LoadImage loads one image and is permanently on. Load Images Conditional is the version with a switch - the "condition" in the name is just a boolean called enabled. Flip it off and the node returns an empty batch without touching your disk, which lets you gate an entire image-loading branch on a workflow setting instead of deleting nodes or rerouting wires.

So why bother? In this pack's Wan loop workflows, this is the workhorse loader: it's what Load Image Folders calls internally for every numbered folder, and it's what you'd use directly when you want a folder's worth of frames loaded as one batch - for example, feeding a chunk of previously rendered frames back into a video continuation pass.

What it takes, what it gives

• folder_path - the directory to load. Relative paths are resolved against ComfyUI's input/ folder; absolute paths work as-is.
• enabled - true loads, false returns an empty batch instantly. Default on.

Outputs:

• images - a single IMAGE tensor stacking every image in the folder, sorted by filename ascending. One batch, not one-per-file.
• count - an INT, how many images made it into the batch. This is the part people underuse: wire count into a conditional downstream and you can branch on "did this folder have the expected number of frames?"

The mechanism, briefly

Enabled path: it scans the folder (os.scandir, which is faster than listdir), sorts by name, and converts each file to an RGB float tensor. Corrupt or unreadable files are silently skipped - so a half-written PNG from a crashed render doesn't kill your queue, it just lowers count. Disabled path: returns torch.zeros((0, 64, 64, 3)) and 0, no disk access. Missing folder, empty folder, or zero valid images all raise a ValueError with a readable message.

The empty-batch behavior is worth understanding: when disabled, you get a zero-length batch, not a black frame. Downstream nodes that assume at least one frame will still see a batch - just an empty one. If that breaks something, branch on count instead of relying on the empty tensor to behave like nothing.

Install and context

The node lives in ComfyUI-HeresyNodes, heresyCoder's small pack of personal support nodes for Wan video loop pipelines:

cd ComfyUI/custom_nodes
git clone https://github.com/heresyCoder/ComfyUI-HeresyNodes

Restart afterward; ComfyUI Manager also finds it under "ComfyUI-HeresyNodes". No model downloads, no extra Python dependencies.

The honest take

This is a loader with a toggle - if you never need the toggle, ComfyUI's built-in nodes already cover you. But the conditional-load pattern is genuinely useful once you're automating reruns: same graph, different enabled setting, and a whole reference-image branch switches on or off. Combined with the count output, it also doubles as a folder validator for exactly the kind of resume-friendly pipelines this pack is built around. Reach for it when you want the load itself to be part of the workflow's logic, not a fixed step.