Polyhedron LoRA Stack — ComfyUI Custom Nodes + WAN 2.2 Starter Workflows — v313

This ComfyUI workflow is for applying multiple LoRAs to Wan Video 2.2 T2V-A14B: it takes LoRAs and prompt concepts, groups and merges them into one model, then passes that model to a sampler or HIGH/LOW samplers.

Group LoRAs before merging them so LoRAs for different parts of the base model can be handled together.

The pack includes Bridge nodes for the WanVideo sampler, a multi-model node with numbered switches, example workflows, and analysis tools.

Use prompt concepts for elements such as skin, hair, faces, clothing, and lighting; these concepts can trigger specific LoRAs.

After grouping and merging, the workflow sends one combined model to a sampler or to HIGH/LOW sampler paths.

Install the pack in ComfyUI-Manager by searching for "polyhedron", or run `comfy node install polyhedron-lora-stack`.

Full documentation is available at https://github.com/PolyhedronAI/ComfyUI-PolyhedronLoRAStack.

Install ComfyUI-KJNodes 1.0.6 and ComfyUI-Custom-Scripts 1.2.5.

Prepare these model files: `umt5_xxl_fp8_e4m3fn_scaled.safetensors`, `wan_2.1_vae.safetensors`, `wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors`, and `wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors`.

WanVideo variants additionally need ComfyUI-WanVideoWrapper (kijai).

Group LoRAs by the element they affect—such as skin, hair, clothing, or lighting—before merging.

Suggestion · non vérifié

For several models, use the multi-model node's numbered switches and the HIGH/LOW sampler paths.

Suggestion · non vérifié

Before use, verify these exact filenames: `wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors` and `wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors`.

Suggestion · non vérifié

No license is listed on the source page; check its usage terms before redistributing the pack.

Suggestion · non vérifié