EnhanceMotion

Improves animation speed. For FastMix.

Boost1stStep

Enhances motion. For FastMix.

SelectModel

Switch between EasyWan22 standard FastMix model and base model.

TextEncoder

• Kijai: Can be used stably.
• Nag: Can use negative prompts even with Cfg 1 high-speed generation.
  • During use, CFG must be 1, resulting in different outcomes from other TextEncoders for better or worse.
• Native: Higher precision than Kijai with fp8_scaled, but prone to VRAM-related troubles.
  • Please use with sufficient VRAM margin.

TorchCompile

• Disable when older GPUs like Geforce RTX 20x0 and earlier cannot properly utilize compilation optimization.
  • In that case, you may also need to set attention_mode in ModelLoader to spda.

ModelLoader

• Lora(1~5)-(High|Low), Lora(1~5)-(High|Low)Strength
  • Set manually loaded LoRA and its strength.
  • Please input so-called trigger words in PositiveInput.
• BlocksToSwap
  • If you don't understand VRAM consumption, keep it at 40.
  • You can generate videos faster by reducing this value.
    • Whether to use surplus VRAM for high-precision models, increasing video resolution/duration, or speeding up generation is up to your preference.
  • offload_img_emb, offload_txt_emb, prefetch_blocks are also swap-related settings.
    • Setting prefetch_blocks to 1 may speed up some environments. It slowed down in my environment.
• base_precision
  • Changing from fp16_fast to fp16 will slow down but increase precision.
• quantization
  • When using fp8 models instead of GGUF, match the model's quantization.
• load_device
  • If you have sufficient VRAM, loading to main_device might speed things up.
• attention_mode
  • Change from sageattn to sdpa when older GPUs like Geforce RTX 20x0 and earlier cannot use SageAttention.
• FastMix-(High|Low)
  • You can download and use FastMix models with different precision using Download/diffusion_models/FastMix/Wan22-I2V-FastMix_*.bat.
• Base-(High|Low)
  • You can download and use Base models with different precision using Download/diffusion_models/Base/Wan2.2-I2V-A14B-*.bat.

FAQ

Want to use base model instead of FastMix

You can use it by selecting Base in SelectModel under Model.
More detailed settings are also possible in the Base-Sampler subgraph at the bottom left.

In my environment, parameter adjustments haven't been fully refined, and compatibility with EndImage hasn't been improved.
If you have recommended settings that can achieve good results with any combination of features, regardless of LoRA or EndImage presence, please let me know.