YAML Metadata Warning: empty or missing yaml metadata in repo card (https://huggingface.co/docs/hub/model-cards#model-card-metadata)

WAN 2.2 I2V A14B - LightX2V 4-Step Distillation LoRAs (Rank-Reduced)

SVD rank-reduced versions of the lightx2v/Wan2.2-Distill-Loras I2V A14B 4-step LoRAs.

Variants

Rank Avg Reconstruction Error Energy Retained Size (HN/LN) Speed
32 ~40% ~83% 307 / 358 MB ~2-3 min (matches Comfy-Org)
48 ~27% ~93% 455 / 532 MB ~4-5 min
64 (original) 0% 100% 606 / 706 MB ~6-11 min

Processing

  • Rank reduction via exact SVD (QR + inner matrix SVD)
  • Stripped non-standard keys: .diff_m, norm_k_img, .diff (1D norm), .diff_b
  • Compatible with WGP / mmgp LoRA loader

Usage

Each variant has high_noise and low_noise LoRAs for the two-phase distillation schedule.

Source

Original LoRAs from lightx2v/Wan2.2-Distill-Loras:

  • wan2.2_i2v_A14b_high_noise_lora_rank64_lightx2v_4step_1022.safetensors
  • wan2.2_i2v_A14b_low_noise_lora_rank64_lightx2v_4step_1022.safetensors
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support