Add extra model paths for WAN

[Shellishack]

Motivation

Add additional extra model paths for WAN models to discover diffusion_models, audio_encoders, text_encoders.

Issues closed

#165

[TimPietruskyRunPod]

Useful addition — these three (diffusion_models, audio_encoders, text_encoders) are required for WAN, current FLUX dev-fp8 / Qwen-image, and audio models to be discovered from a network volume. Direction is good.

One change requested + one nit:

Change requested — loras is already present on line 9 of src/extra_model_paths.yaml:

  loras: models/loras/  # existing entry

Adding it again creates a duplicate YAML key, which is invalid in strict parsers and will either silently override or fail depending on the loader. Please remove the loras: models/loras/ line from your diff.

Nit — also adding models/-prefix conventional paths:

For consistency with the existing entries that use models/<dir>/, keep that pattern (which you already do). 👍

Changeset: please add .changeset/add-model-paths-for-wan-audio-textencoders.md:

---
"worker-comfyui": patch
---

fix: add `diffusion_models`, `audio_encoders`, and `text_encoders` to `extra_model_paths.yaml` so WAN, FLUX (UNETLoader), and audio-encoder models are discovered when stored on a network volume.

Once the duplicate loras is gone and a changeset is added, this is ready to merge.