RopeWrapper_OptionNode

RopeWrapper_OptionNode is the settings bag for this whole pack. It exists for one reason: the standalone Rope GUI had an enormous panel of sliders and switches for tuning a swap, and a node graph can't hold thirty widgets on one box. So this node collects them all into a single ROPE_OPTION blob that you wire into the SwapNode or FaceRestore node's ROPE_Options input. It's a pure data node - no models, no GPU, no heavy lifting. It just packages.

If you skip it and leave ROPE_Options disconnected, those swap nodes quietly try to load a custom_nodes/Rope/saved_parameters.json file you almost certainly don't have, and fall back to defaults. So for anything beyond a bare swap, you want one of these in your graph.

The knobs that actually matter

There are about thirty inputs here, and most map one-to-one to Rope's own parameters. A few are worth your attention on day one:

• RestorerSwitch plus RestorerTypeTextSel (CF, GFPGAN, GPEN256, GPEN512) and RestorerSlider - the face-enhancement pass after the swap. GFPGAN is the community's old standby for cleaning up faces; CodeFormer (CF) tends to hold up better on badly degraded input. The slider is strength.
• SwapperTypeTextSel (128, 256, 512) - the resolution the inswapper works at. 128 is the model's native resolution and the default; 256 and 512 look sharper but cost, and Rope's own benchmarks show 512 runs roughly six times slower than 128. Start at 128, bump only if you're happy with the swap itself.
• BlendSlider - edge blending between the swapped face and the rest of the frame (default 5). Higher feathers the border more, which hides the swap seam.
• OrientSwitch and OrientSlider - rotate the swap when the source face isn't upright, which fixes some misaligned swaps without re-detecting.
• ColorSwitch with the three color sliders and ColorGammaSlider - color correction so the swapped face matches the scene's lighting instead of glaring.
• CLIPSwitch, CLIPTextEntry, CLIPSlider - Rope's optional CLIP-guided tweak; a text hint plus a strength.

Everything else - DiffSwitch/DiffSlider, OccluderSwitch/OccluderSlider, FaceParserSlider, MouthParserSlider, Border*, FaceAdjSwitch with the KPS* sliders - is Rope's advanced toolbox: differential motion, occlusion handling, and manual face-geometry nudging. Leave them off until you have a specific problem. Each has a master switch plus a slider, so the pattern is "flip the switch, then tune."

How the wiring works

Internally the node builds a Python dict keyed exactly like Rope's saved-parameters file, sets every field you touched, and fills anything you didn't expose with Rope's defaults (things like thread count and the RetinaFace detector choice are hardcoded in). The VideoManager reads vm.parameters, so once the dict reaches SwapNode, the swap runs exactly as if you'd clicked those buttons in the old GUI.

Install

Install once for the whole pack - ComfyUI Manager → search "ComfyUI_RopeWrapper", or:

cd ComfyUI/custom_nodes
git clone https://github.com/fssorc/ComfyUI_RopeWrapper

Restart ComfyUI. The README requires ComfyUI-VideoHelperSuite and ComfyUI-KJNodes as companion packs, and the model files go in ComfyUI/custom_nodes/ComfyUI_RopeWrapper/models/ (see the LoadModels article for the full list).

Gotchas

The one trap: this node can't tell you a model is missing. If you flip RestorerSwitch on but never downloaded GFPGANv1.4.onnx, the swap node will fail at run time, not here. And note the FaceRestore node forces the restorer on no matter what this node says - the OptionNode only recommends settings to the nodes that read it.