Comfyui Wan

Created by HearmemanAI https://www.hearmemanai.com

Troubleshooting guide in case you encounter any errors:

[click here](https://docs.google.com/document/d/1822H-x7AevWz2T_jzMu8-9e5UlQ-zrH0FhCFmQ6FtRc/edit?usp=sharing)

v12 — Major Refactor (May 2026)

⚠️ **Breaking change for existing users.** The old per-resolution / per-feature flags are gone. If you're updating an existing pod template, **delete** the removed env vars below and **add** the new ones — otherwise nothing will download.

⚠️ **CUDA 12.8 required.** Before deploying, open Additional Filters and select CUDA 12.8. Fallback (older CUDA): https://runpod.io/console/deploy?template=0b7tk92912&ref=uyjfcrgy

---

What's new in v12

• Workflow-driven downloads. The pod now reads the workflows you choose and downloads only the models those workflows actually reference. No more redundant multi-gigabyte downloads.
• Faster HF transfers. Hugging Face's hf_xet accelerator is enabled image-wide (HF_XET_HIGH_PERFORMANCE=1), typically 1.5–3× faster than aria2 on HF.
• Live download manager. A 3-way pool with an append-only status snapshot every 10s (active %, speed, ETA; queued list; done count). No more interleaved aria2 spam.
• Faster boot. comfy-aimdo / comfy-kitchen now install in the background instead of blocking startup; redundant apt/pip steps removed.
• Slimmer workflow set. Obsolete Wan-Fun-SDXL helper, legacy VACE preview, and the unused Video Extend workflow have been removed.

---

⚙️ Environment Variables

Set the flags you want to **`true`** (lowercase, as strings). Multiple may be enabled at once — shared models are deduplicated automatically.

Download flags (new in v12)

Variable	Downloads + provisions
download_wan21	Wan 2.1 (incl. VACE) + Infinite Talk workflows + their models
download_wan22	Wan 2.2 + SVI Video Extension workflows + their models
download_wan_animate	Wan Animate workflows + models
`download