WAS Node Suite v3 Reborn: 463 Nodes, Zero Dependencies for ComfyUI

WAS Node Suite has been around since 2023 and counts over a million downloads. The v2 pack bundled 220 nodes but dragged in around 20 Python packages on install, which made it a frequent source of broken environments. Author WASasquatch (Jordan Thompson) calls the release "WAS-NS Reborn": the pack has been rebuilt for modern ComfyUI, requiring only ComfyUI 0.14.0+ and Python 3.10+ with nothing installed, downloaded or built at any point.

What changed from v2 to v3

216 of the 220 v2 node IDs survive unchanged, and the four retired nodes (Image Crop Face, Image Gradient Map, Load Lora, Number to Text) are auto-swapped by ComfyUI's missing-node dialog when you open an old workflow. Old saved settings, widgets and wires carry across on load.

The key architectural change: face detection, gradient maps, background removal, frame interpolation, seamless textures, color matching, levels, palettes, masks and deconvolution all run on torch now, on ComfyUI's device. OpenCV, numba, rembg, timm, scipy, scikit-image, scikit-learn and matplotlib are no longer installed. The only optional dependency group is document_export (DOC/ODT/PDF output), which you enable explicitly in config.yaml.

247 new nodes

The headline additions:

• Three.js scenes (43 nodes): build a 3D scene out of wires, with geometry, materials, lights, cameras, model loading, rendering and path tracing. Off by default behind a feature gate.
• Files, archives and documents (28): zip archives on the wire, documents in six formats, whole folders through a graph keeping original filenames.
• Logic and flow (24): condition chains, typed switches, for and while loops, plus an Execution Gate that stops a branch from evaluating at all.
• Layers (20): a layer stack on the wire with effects, arrangement and a canvas drawn on the node.
• HDR and linear light (9): decode above white, carry range through filters, read and write EXR and DNG.
• Latent Affine sampling (6): scale and offset a latent through a mask, including during sampling, with content-aware patterns that follow the subject. The README includes an example workflow running affine inside a sampler on a MiniMax H3 video latent.
• Temporal Noise Hold: carries each video frame's noise pattern into the next frame instead of drawing fresh noise, filling out scenes from the starting noise on video models like MiniMax H3.

194 nodes draw a panel on themselves: histograms, color ramps, 3D views, text editors, and a contact sheet you click to pick a look. Compare Video plays two renders side by side under a draggable divider, synced frame to frame, which is a quick way to judge a workflow change against a baseline render.

The Content Viewer renders anything wired into it (markdown, HTML, SVG, documents, JSON, CSV, logs, image canvas) directly on the node, and accepts installable view extensions such as an image search panel and the OpenReel video editor.

Availability

WAS Node Suite v3 is MIT licensed. Install from ComfyUI Manager by searching "WAS Node Suite v3", or clone WASasquatch/was-node-suite-comfyui into custom_nodes. If you prefer the legacy v2 behavior, the author points to drltdata's "WAS-NS Revised" fork on ComfyUI Manager. Runnable example workflows for 3D scenes, HDR, layers, loops and the affine samplers ship in the repo's docs/workflows/ directory.