Experimental DLSS — Easy Upscale & Render

The rest of this pack makes you build a guide chain by hand: a depth node here, a motion node there, a stabilizer in the middle, all feeding a pipeline node with sixteen knobs. This node says no to all of that. You hand it an image or a clip, pick a scenario and a look, and it quietly builds the whole thing for you - depth, motion, stabilization, upscale, neural render - then hands back an image and a plain-English report of what it actually ran.

That makes it the correct first node for almost everyone. The pack is experimental alpha, Windows-only, NVIDIA-only, and the full guided graph is where you go when you outgrow the defaults. Start here.

What it does behind the curtain

Read the source and you'll see the Easy node is really a dispatcher. Depending on your scenario it:

• runs Depth Anything V2 Small for depth,
• picks Optical Flow (fastest) for stills or RAFT for video (Large for quality, Small for long clips),
• stabilizes the depth temporally whenever there's more than one frame,
• then routes through the SR-only, neural-render-only, or combined pipeline.

The scenario presets do the thinking for you. Auto (recommended) looks at your frame count: one frame → still mode, up to 96 frames → Short video / best quality (RAFT Large, persistent context), anything longer → Long video / memory efficient (RAFT Small, bounded overlap-add). Fast preview drops to CPU optical flow and small windows when you just want to see if the idea works.

The inputs a beginner actually sets

• scenario - start with Auto; override when you know better.
• operation - Upscale + neural rendering is the full experience. Upscale only is a great smoke test that doesn't need the user-supplied nvngx_dlssnr.dll. Neural rendering only leaves resolution alone and just applies the look.
• scale - 2x, 3x, or 4x. The README's own tests ran at 2x; 4x is where quality presets start to sweat.
• look - Neutral / faithful is the honest evaluation setting and the safest for faces. Realistic detail and Strong detail lean on style index 2 with more aggressive structure and skin settings.
• effect_strength (default 0.85) - the global blend for the neural-render pass. Turn it down when the look is too there.

One gap worth knowing: this node's quality menu has four entries (Quality through Ultra Performance), not the six you'll find on the raw Super Resolution node. The fancier Ultra Quality and DLAA presets live on the guided path only.

Outputs

You get image (the result, at your chosen scale) and runtime_report - a STRING that's genuinely worth reading. It tells you which scenario preset fired, which flow model and processing mode ran, and per-window SR/NR details. When something looks off, that report is your first clue whether the clip was treated as persistent or bounded.

The catch that isn't in the node

"Easy" refers to the graph, not the prerequisites. This pack only runs after you've installed the runtime: add DLSS Runtime Setup (One Click), point it at a legally obtained nvngx_dlssnr.dll, and let it fetch the ~368 MB VapourKit build. Upscale only sidesteps the neural DLL, but the full operation needs it. Memory is the other wall - persistent video processing can chew through many gigabytes of RAM and temp space. Keep clips short for the first few runs. Install via ComfyUI Manager (Git URL https://github.com/HECer/ComfyUI-DLSS5), restart for dependencies, set up the runtime, and only then expect the Easy node to feel easy.