Distributed Placeholder

You read the name right: this node's entire output is a single black 1×1 pixel. No inputs, one image output, and the tensor it produces is torch.zeros(1, 1, 1, 3) sitting on the CPU. It is the most honest useless-looking node in the pack, and in a distributed workflow it's quietly load-bearing.

Here's the situation it exists for. In the master/worker model this pack uses, the master coordinates a job and may or may not render images itself - the Distributed panel has a "master rendering enabled" toggle, and when you turn it off the master's only job is to hand out work and collect results. The problem is that ComfyUI is a graph: nodes downstream of your sampler still want an IMAGE to flow through them, and if you rip the sampler out of the master's copy of the graph, the pipeline breaks. Placeholder is the dummy that keeps the graph executable without burning a single byte of VRAM. On the master's pruned copy of the workflow, it stands in for the whole generation chain - black, 1×1, instant - while the real sampling happens on the workers and lands in a DistributedCollector.

Think of it as the plumbing answer to "the master isn't allowed to render." You put it wherever an IMAGE is required on a segment that should do no GPU work, and it satisfies the graph type-check in microseconds. That's the same role the pack's video sibling DistributedVideoPlaceholder plays with the native VIDEO type.

There's no configuration, which is the whole point - the source comment calls it "lightweight CPU-only." If you're only ever running with the master rendering enabled, you'll probably never touch it. The moment you switch master rendering off to keep your coordinator machine free, this node is what stops your master-side workflow from dying on a missing image. Don't confuse it with a preview or a "put a black image here" utility; it's a structural shim for distributed execution, not something you'd ever wire into a real output.