---
title: WanVideo_comfy
canonical_url: "https://www.modelscope.cn/models/Kijai/WanVideo_comfy"
md_url: "https://www.modelscope.cn/models/Kijai/WanVideo_comfy.md"
repository: Kijai/WanVideo_comfy
last_updated: 2026-06-18
pipeline_tag: text-to-video-synthesis
tasks:
  - text-to-video-synthesis
base_model:
  - Wan-AI/Wan2.1-VACE-14B
  - Wan-AI/Wan2.1-VACE-1.3B
base_model_relation: finetune
parameters: 997.5B
tensor_type:
  - F16
  - F32
  - BF16
  - BOOL
  - I64
  - F8_E4M3
  - F8_E5M2
  - U8
library_name:
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 3427645
stars: 126
tags:
  - diffusion-single-file
  - comfyui
---

# WanVideo_comfy

> WanVideo_comfy - Kijai 在 ModelScope 开源的模型。Combined and quantized models for WanVideo, originating from here:

Kijai/WanVideo_comfy 是 ModelScope 魔搭社区上的 997.5B 参数text-to-video-synthesis模型，基于 Wan-AI/Wan2.1-VACE-14B、Wan-AI/Wan2.1-VACE-1.3B 构建。

- **Repository**: Kijai/WanVideo_comfy
- **Tasks**: text-to-video-synthesis
- **Parameters**: 997.5B
- **Base model**: Wan-AI/Wan2.1-VACE-14B, Wan-AI/Wan2.1-VACE-1.3B
- **Tags**: diffusion-single-file, comfyui
- **Downloads**: 3427645
- **Stars**: 126
- **Last updated**: 2026-06-18

Source: https://www.modelscope.cn/models/Kijai/WanVideo_comfy

---

Combined and quantized models for WanVideo, originating from here:

https://huggingface.co/Wan-AI/

Can be used with: https://github.com/kijai/ComfyUI-WanVideoWrapper and ComfyUI native WanVideo nodes.

I've also started to do fp8_scaled versions over here: https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled

Other model sources:

TinyVAE from https://github.com/madebyollin/taehv

SkyReels: https://huggingface.co/collections/Skywork/skyreels-v2-6801b1b93df627d441d0d0d9

WanVideoFun: https://huggingface.co/collections/alibaba-pai/wan21-fun-v11-680f514c89fe7b4df9d44f17

---

Lightx2v:

CausVid 14B: https://huggingface.co/lightx2v/Wan2.1-T2V-14B-CausVid

CFG and Step distill 14B: https://huggingface.co/lightx2v/Wan2.1-T2V-14B-StepDistill-CfgDistill

---

CausVid 1.3B: https://huggingface.co/tianweiy/CausVid

AccVideo: https://huggingface.co/aejion/AccVideo-WanX-T2V-14B

Phantom: https://huggingface.co/bytedance-research/Phantom

ATI: https://huggingface.co/bytedance-research/ATI

MiniMaxRemover: https://huggingface.co/zibojia/minimax-remover

MAGREF: https://huggingface.co/MAGREF-Video/MAGREF

FantasyTalking: https://github.com/Fantasy-AMAP/fantasy-talking

MultiTalk: https://github.com/MeiGen-AI/MultiTalk

Anisora: https://huggingface.co/IndexTeam/Index-anisora/tree/main/14B

Pusa: https://huggingface.co/RaphaelLiu/PusaV1/tree/main

FastVideo: https://huggingface.co/FastVideo

EchoShot: https://github.com/D2I-ai/EchoShot

Wan22 5B Turbo: https://huggingface.co/quanhaol/Wan2.2-TI2V-5B-Turbo

Ovi: https://github.com/character-ai/Ovi

FlashVSR: https://huggingface.co/JunhaoZhuang/FlashVSR

rCM: https://huggingface.co/worstcoder/rcm-Wan/tree/main

---
CausVid LoRAs are experimental extractions from the CausVid finetunes, the aim with them is to benefit from the distillation in CausVid, rather than any actual causal inference.
---
v1 = direct extraction, has adverse effects on motion and introduces flashing artifact at full strength.

v1.5 = same as above, but without the first block which fixes the flashing at full strength.

v2 = further pruned version with only attention layers and no first block, fixes flashing and retains motion better, needs more steps and can also benefit from cfg.