FastWan 2.1 T2V 1.3B — hostmix GGUF

Mixed-precision GGUF quantization of FastWan2.1-T2V-1.3B (bidirectional DMD distillation of Wan2.1-T2V-1.3B, both Apache-2.0).

• Transformer blocks: q4_0
• Everything else (embedders, heads, norms): f16
• Converted with stable-diffusion.cpp (sd-cli -M convert --type f16 --tensor-type-rules "blocks\.=q4_0"); the VSA gate tensors are dropped (dense attention single-frame inference).

Built for stable-diffusion.cpp single-frame (T2I-style) inference at 3 steps / cfg 1. Loading with other GGUF runtimes (e.g. ComfyUI-GGUF) is untested.