<aside> 👉🏻

This is an LLM-generated, hand-fixed summary of the #wan-chatter channel on the Banodoco Discord.

Generated on April 7, 2025.

Created by Adrien Toupet: https://www.ainvfx.com/ Ported to Notion by Nathan Shipley: https://www.nathanshipley.com/

Thanks and all credit for content to Adrien and members of the Banodoco community who shared their work and workflow!

</aside>

<aside> ⚠️

Some images are missing.

Also, Adrien notes: “Please note I didn't verify all the information, so take it as a general overview that might contain error and not as a strict guide.”

</aside>

Back to:

Nathan’s Notes on Open Source AI

Banodoco discord #wan_chatter summary

Last updated: April 7, 2025 - This content was generated using AI. Accuracy is not guaranteed, information may contain errors or omissions.

📕 Table of Contents:

1. Introduction to Wan

Wan 2.1 (originally called WanX) is an open-source AI video generation model suite released by Alibaba Cloud starting February 2025. It represents a significant advancement in open-source video generation capabilities, offering high-quality video generation that rivals some closed-source alternatives. The suite includes Text-to-Video (T2V) and Image-to-Video (I2V) models, along with specialized "Fun" versions for control and inpainting, and the powerful VACE module.

Key Features

License: Apache 2.0, fully open-source and available for commercial use.
Frame Rate: 16 FPS native output.
Default Length: 81 frames (approximately 5 seconds) is optimal for quality and consistency without context windows.
Text Encoder: UM-T5 XXL (multilingual).
VAE: Uses a specific Wan VAE, crucial for correct results.

2. Model Variants

Core Models (Wan 2.1)