This is a Workflow for the Wan2.1 14B FusionX GGUF Model

I use this Workflow on my linux+amd setup (RX 7900XT) and get 480p 5s videos in under 10 minutes without using TeaCache.

If you're on linux+amd you have to delete the sageattention node; This is for the NVidia Users only.

This workflow utilizes Florence2Run to concatanate a description of the input image into your prompt so you don't have to write it yourself.

Image resizing all happens automatically according the the resolution slider in the settings.

Needed Files:

1. ⬇️QuantStack/Wan2.1_I2V_14B_FusionX-GGUF
   Chose matching quant size for your GPU; On my RX 7900XT (20GB VRAM) i chose the Q4_K_S
   📂Put in /models/unet

2. ⬇️city96/umt5-xxl-encoder-gguf
   Chose same quant size as your base model
   📂Put in /models/text-encoders

3. ⬇️Kijai/WanVideo_comfy
   📂Put in /models/vae

4. ⬇️Kijai/WanVideo_comfy
   📂Put in /models/lora

5. ⬇️MiaoshouAI/Florence-2-base-PromptGen-v2.0
   Download whole archive
   📂Put in /models/LLM/Florence-2-base-PromptGen-v2.0

Optional File:

1. ⬇️Wan14B_Detailer/Enhancer_T2V
   📂Put in /models/lora