Step-by-step workflow
1
Download Wan 2.1 model
1.3B for 8GB VRAM, 14B for 12GB+ VRAM.
# 8GB VRAM (good quality):
huggingface-cli download Wan-AI/Wan2.1-T2V-1.3B --local-dir ./wan21_1b
# 12GB+ VRAM (best quality):
huggingface-cli download Wan-AI/Wan2.1-T2V-14B --local-dir ./wan21_14b
# Place in: ComfyUI/models/diffusion_models/
2
Install ComfyUI-WanVideoWrapper
Manager → search "WanVideo" → install → restart.
3
Write your prompt
Formula: [Lighting] + [Subject] + [Action] + [Environment] + [Camera] + [Style]
"Warm golden hour lighting, street food vendor in Old Delhi making jalebi, batter swirling into hot oil, traditional marketplace, medium close-up, documentary style, 4K"
# Settings: Resolution: 832x480 (1.3B) / 1280x720 (14B)
# Frames: 81, Steps: 30, Sampler: UniPC
4
Add ControlNet LoRA for camera control
Optional: adds precise camera movement control.
# Download: Wan-AI/Wan2.1-Camera-Control-LoRA
# Place in: ComfyUI/models/loras/
Pro tips
→
1.3B model is surprisingly good for social media
→
Wan 2.1 is better than LTX for human faces and subjects
→
Best workflow: Flux image → Wan 2.1 I2V = best results
Why this matters for India
// india context
Create product demo videos, restaurant content, fashion lookbooks — professional quality free