Sign In

WAN 2.2 High and Low | How to add new LoRAs in ComfyUI with Wan 2.2

29

Aug 22, 2025

tool guide
WAN 2.2 High and Low | How to add new LoRAs in ComfyUI with Wan 2.2

How to add new LoRAs in ComfyUI with Wan 2.2

  1. Load the base model

    • On the left you start with the Load Diffusion Model node
      (wan2.2_i2v_high_noise... or wan2.2_i2v_low_noise...).

    • This is your base Wan model.

  2. Attach the first LoRA

    • Right after that, add a LoraLoaderModelOnly node.

    • Select your LoRA (.safetensors) inside.

    • The strength_model slider controls how strong the LoRA is applied (common values: 1.0 – 2.0).

  3. Stack more LoRAs

    • To use multiple LoRAs, just chain more LoraLoaderModelOnly nodes one after another.

    • The order can affect the result.

  4. Connect to ModelSamplingSD3

    • At the end of the LoRA chain, connect everything into a ModelSamplingSD3 node.

    • This prepares the final model for the KSampler.

  5. Two base model paths (High Noise & Low Noise)

    • Wan 2.2 has two base models:

      • wan2.2:i2v_high_noise

      • wan2.2:i2v_low_noise

    • Each can have its own LoRAs stacked, and both outputs can be sent into the KSampler.

To be precise:

  • The high noise model chain (base model + LoRAs) must end in its own ModelSamplingSD3 β†’ then connect to its own KSampler.

  • The low noise model chain (base model + LoRAs) must also end in its own ModelSamplingSD3 β†’ then into a separate KSampler.

So you do not merge them into one ModelSamplingSD3 or one KSampler.
They are two parallel pipelines β€” one for high noise, one for low noise.

πŸ‘‰ This way you can render with both high-noise and low-noise versions independently.


⚠️ Important:
Always use LoraLoaderModelOnly (not the normal LoraLoader).

  • LoraLoader replaces the entire model.

  • LoraLoaderModelOnly stacks the LoRA on top of the Wan base model β€” that’s what you want.

 β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
 β”‚   Base Model (High)   β”‚  wan2.2:i2v_high_noise
 β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
           β”‚
    [LoraLoaderModelOnly]
           β”‚
    [LoraLoaderModelOnly]
           β”‚
    [LoraLoaderModelOnly]
           β”‚
 β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
 β”‚   ModelSamplingSD3    β”‚   (High path)
 β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
           β”‚
 β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
 β”‚      KSampler         β”‚   (High path)
 β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜


 β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
 β”‚   Base Model (Low)    β”‚  wan2.2:i2v_low_noise
 β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
           β”‚
    [LoraLoaderModelOnly]
           β”‚
    [LoraLoaderModelOnly]
           β”‚
    [LoraLoaderModelOnly]
           β”‚
 β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
 β”‚   ModelSamplingSD3    β”‚   (Low path)
 β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
           β”‚
 β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
 β”‚      KSampler         β”‚   (Low path)
 β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

29

Comments