Video Tile Merge (overlap soft)

Let's be straight about this one: Video Tile Merge (overlap soft) is a compatibility node, and the pack's own README says the main merge can do everything it does. It's here so older saved workflows keep loading, not because it's the one you should reach for on a fresh graph.

It exists to provide one specific merge behavior: always a weighted sum of the geometry feather weights - symmetric, order-independent overlap blending. That's the "soft" in the name. It's genuinely good at making seams disappear, and before the main merge gained its blend_mode selector it was the only way to get it. Today you get the identical result from Video Tile Merge by setting blend_mode=weighted_average, which also gives you feather_curve and the option to fall back to alpha_over painter-order when you want it.

The inputs

Same shape as the main merge: tile_config, tiles (the processed tile list), feather (same 0–0.5 fraction interpretation), and merge_device - note that here merge_device is a required input, not optional like on the main merge node. Output is a single merged IMAGE.

When to use it

Two honest cases:

1. You have a saved workflow that references it. It works, it won't break, leave it alone. The pack deliberately keeps it registered for exactly this.
2. You want to force weighted averaging with no knobs to accidentally flip. Some people like a node that can't be misconfigured.

For everyone else: just use Video Tile Merge with blend_mode=weighted_average. One less node in the graph, and the main merge is where the rest of the ecosystem's tutorials will point you anyway.

Install

It's part of the comfyui-video-tiler pack, so the install is the same as every other node in it:

cd ComfyUI/custom_nodes
git clone https://github.com/maDcaDDie2000/comfyui-video-tiler

Restart, find it under Video Tiler. No extra Python dependencies, no model downloads - PyTorch and NumPy are all it needs. Apache 2.0, vibe-coded for personal use, not actively maintained. That "not actively maintained" is worth repeating, because a compatibility node in a pack like this is exactly the kind of thing that lingers untouched: fine for the workflows that already use it, but for new work, build on the main merge node so you're on the actively shaped path.