Wan 2.2 + Lightx2v V2 | Ultra Fast T2V

What is the Wan 2.2 + Lightx2v V2 ComfyUI Workflow?

The Wan 2.2 + Lightx2v V2 ComfyUI workflow represents a breakthrough in ultra-fast text-to-video generation, combining Alibaba's Wan 2.2 14B model with the revolutionary Lightx2v V2 LoRA distillation technology. This workflow delivers professional-quality video generation at lightning speed - up to 4x faster than standard Wan 2.2 workflows while maintaining excellent visual quality.

Wan 2.2 uses a two-stage sampling process: a high-noise model that establishes the basic video structure, followed by a low-noise model that adds fine details and refinements. Lightx2v V2 is a specialized distillation LoRA that compresses this process dramatically - reducing the standard 40 steps (20+20) to just 8-14 steps while using low CFG settings and LCM sampling for maximum speed.

Think of this as having a turbocharged video generator that maintains the quality of the original two-stage process but cuts generation time by up to 75%. Perfect for rapid prototyping, content creation, and professional video workflows where speed matters.

Key Features and Benefits of Wan 2.2 + Lightx2v V2

Ultra-Fast Generation with Flexible Steps: Reduces standard 40-step sampling to just 8-14 steps using distillation techniques. Choose 8 steps (maximum speed), 12 steps (balanced), or 14 steps (optimal quality).

Advanced LoRA Acceleration Options: Full acceleration (both stages) for maximum speed or half-acceleration (refinement only) for better prompt control while maintaining significant performance gains.

Optimized for Lower Hardware Requirements: Reduced VRAM usage and computational cost through LCM sampling with low CFG settings, making high-quality video generation more accessible.

Professional Quality Output: Maintains Wan 2.2's high-quality video generation and motion coherence despite the dramatic speed improvements through advanced distillation technology.

How to Use Wan 2.2 + Lightx2v V2 in ComfyUI

Text-to-Video Generation Workflow

Configure your video parameters

• Use the EmptyHunyuanLatentVideo node to set your video dimensions:
  • Width: 704 pixels (optimized for Lightx2v V2)
  • Height: 544 pixels
  • Length: 81 frames for ~3.4 second videos at 24fps
  • Batch_size: Keep at 1 for optimal performance

Craft your text prompt

• In the CLIP Text Encode (Positive Prompt) node:
  • Write detailed, descriptive prompts for best results
  • Include motion descriptions and camera movements
  • Specify lighting, mood, and visual style preferences

Essential KSampler Settings (Flexible Configuration)

• High Noise KSampler (Foundation Stage):
  • Steps: 4-7 (adjust based on speed vs quality needs)
  • CFG: 1.0-3.5 (lower for speed, higher for prompt control)
  • Sampler: lcm (for acceleration) or euler (for quality retention)
  • Scheduler: simple
• Low Noise KSampler (Refinement Stage):
  • Steps: 8-14 total (adjust based on desired quality level)
  • CFG: 1.0-3.5 (coordinate with foundation stage settings)
  • Sampler: lcm (recommended for refinement acceleration)
  • Scheduler: simple

Parameter Guidelines

• Total Steps: 8-14 steps work well - adjust based on your speed vs quality priorities
• CFG Values: 1.0-3.5 range - lower values for faster generation, higher for better prompt following
• Mix and Match: Combine different settings to find your optimal balance

Universal LoRA Acceleration Settings (Both T2V and I2V)

LoRA Configuration Options

• LoraLoaderModelOnly nodes control the acceleration level:
  • T2V LoRA: lightx2v_T2V_14B_cfg_step_distill_v2_lora_rank64_bf16.safetensors
  • I2V LoRA: lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors
  • Full Acceleration: Apply LoRA to both high-noise (foundation) and low-noise (refinement) models with strength 1.0
  
  • Half Acceleration: Apply LoRA only to the low-noise (refinement) model with strength 1.0 - better prompt control
  

Speed vs Quality Comparison

• Maximum Speed Setup: LCM sampler, CFG 1.0, 8 steps total, LoRA on both stages
• Balanced Setup: Mixed samplers, CFG 1.0-3.5, 12 steps total, LoRA on refinement only
• Quality Priority Setup: Euler + LCM, CFG 3.5, 14 steps total, selective LoRA usage
• Custom Configuration: Adjust steps (8-14), CFG (1.0-3.5), and LoRA placement based on your specific needs

Image-to-Video Generation Workflow

Upload your input image

• Use the LoadImage node to import your starting frame

Apply Lightx2v V2 acceleration

• Follow the same T2V acceleration settings above for KSampler configuration, LoRA placement, and speed vs quality options

Acknowledgement

This Wan 2.2 + Lightx2v V2 ComfyUI workflow integrates Alibaba's Wan 2.2 foundation model with the innovative Lightx2v V2 LoRA distillation technology. Special recognition to Alibaba's Tongyi Wanxiang research team for developing the Wan 2.2 model series and to the community developers who created the Lightx2v V2 acceleration techniques. The workflow demonstrates how advanced distillation methods can dramatically improve generation speed while preserving visual quality for professional applications.

More Resources About Wan 2.2 + Lightx2v V2

Explore additional resources and documentation:

• HuggingFace Model Hub – Pre-trained Wan 2.2 model weights and Lightx2v V2 LoRA files.
• Lightx2v V2 LoRA Files – Distillation LoRA models for ultra-fast generation.