Video-Reason/VBVR-Wan2.2
LoRA-style vs Wan-AI/Wan2.2-I2V-A14B-Diffusers. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of 0, 2, proj, to_k, to_q, to_v changed, in 40 of 40 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/240, Attention 320/640, MLP 80/160, other 0/43, Embeddings 0/12
2 derivatives checked
- LiconStudio/VBVR-wan2.2-comfy-bf16: Different architecture
- CCSSNE/LiconStudio-VBVR-wan2.2-comfy-bf16: Different architecture