DLSS Runtime Setup (One Click)

Here's the thing nobody says in the node galleries: ComfyUI-DLSS5 doesn't do anything until you've run this node. It's not an image processor, it's the pack's installer wearing a node costume. Every other node in the pack - the upscalers, the neural renderer, the guides - quietly expects the VapourKit runtime and a nvngx_dlssnr.dll to exist somewhere, and this is the node that puts them there. If you skipped setup and jumped straight to a pipeline node, your first run will be an error about a missing runtime, and this is the node that error is pointing you back to.

The pack's premise is that DLSS runs through a VapourSynth wrapper, and the wrapper needs a full isolated Python environment with the NVIDIA runtime DLLs beside it. Installing that by hand means downloading a ~368 MB VapourKit build, extracting it, verifying hashes, copying proprietary DLLs into the right folder, and writing a config file. This node does all of that from inside ComfyUI, with one explicit confirmation gate.

How to use it

1. Install the pack via ComfyUI Manager (Git URL https://github.com/HECer/ComfyUI-DLSS5 until the registry listing is approved) and restart ComfyUI once so dependencies install.
2. Drop this node into any workflow. Set action to Check location and queue it.
3. The output tells you the exact runtime directory and whether the neural DLL is present. Copy your legally obtained nvngx_dlssnr.dll there - keep the exact filename, and this is the one file the installer will never fetch for you.
4. Set action to Install verified VapourKit, flip confirm_download on, and queue again.

That second run downloads the pinned VapourKit Windows nightly, verifies its published SHA-256, extracts it with py7zr into the local runtime directory, locates every wrapper and runtime, and writes runtime/config.json. Restart ComfyUI, then confirm with DLSS 5 Runtime Status.

The two inputs, explained

• action - Check location just reports and creates the runtime directory (it also tells you whether your nvngx_dlssnr.dll is in place). Install verified VapourKit does the real installation work.
• confirm_download - the safety gate, off by default. The node deliberately refuses to download until you've flipped it, because an installer that silently pulls a ~368 MB build from the internet on first queue is how you surprise people. The pinned source and hash are in the README for you to review first.

The output, setup_report, is a STRING that tells you what happened at each step. This node is also an output node, so it ends the branch cleanly without needing a preview attached.

What to watch for

The installer downloads VapourKit exactly once and never fetches the neural DLL - that part is on you, and the README is appropriately paranoid about where you get it. Do not publish runtime/config.json anywhere; it's full of machine-specific absolute paths. If Runtime Status later says NOT READY, rerun setup and check that file. And the honesty clause that governs this whole pack: it's an experimental alpha on Windows + NVIDIA D3D12, verified against an RTX 5090, and runtime compatibility depends on the exact DLL, driver, GPU, and wrapper build you end up with. Setup getting you to READY is the beginning of the troubleshooting, not the end of it.