Wan 2.1 T2V 1.3B — GGUF

GGUF conversions of Wan 2.1 T2V 1.3B for on-device video generation with stable-diffusion.cpp. Converted with sd.cpp's -M convert.

Files

• Wan2.1-T2V-1.3B-calibrated-Q4_K_S.gguf — calibrated Q4 quantization.
• Wan2.1-T2V-1.3B-Q4_K_S.gguf — smaller Q4 quantization.
• Wan2.1-T2V-1.3B-Q8_0.gguf — Q8 quantization.

Q4_K_S in a file name: stable-diffusion.cpp wrote the file with --type q4_K. Its weight tensors are Q4_K and the other tensors stay F16/BF16/F32. Q4_K_S is the closest standard GGUF label, so Hugging Face can list the file.

Required companion files, from off-grid-ai/Video-Companions-GGUF:

• umt5-xxl-encoder-Q4_K_M.gguf — UMT5-XXL text encoder, 3.655 GB.
• wan_2.1_vae.safetensors — Wan 2.1 VAE, 0.254 GB.

Q4 optimized uses the calibrated Q4 file with --fa --vae-conv-direct --vae-tiling.

Use with stable-diffusion.cpp

Download one generator and the companion files into one folder. Use a stable-diffusion.cpp build that supports this model.

hf download off-grid-ai/Wan2.1-T2V-1.3B-GGUF Wan2.1-T2V-1.3B-calibrated-Q4_K_S.gguf --local-dir .
hf download off-grid-ai/Video-Companions-GGUF umt5-xxl-encoder-Q4_K_M.gguf wan_2.1_vae.safetensors --local-dir .
hf download off-grid-ai/Wan2.1-T2V-1.3B-GGUF --include 'commands/*' --local-dir .
bash commands/q4-optimized.sh

Examples, measurements, and full settings. Timings are recorded runs, not controlled speed comparisons; shared GPU work can affect them.

Earlier file names

Files were renamed on 2026-10-21 so that Hugging Face lists every quantization and does not read a companion file as the model. File contents are unchanged (same SHA-256). Links pinned to an earlier commit still work.

Credit & license

Original model: Wan-AI/Wan2.1-T2V-1.3B — created by its respective authors.

License: apache-2.0 (carried over from the original; its use restrictions apply). See NOTICE for attribution and conversion notices.

This is a format conversion (quantization) only; all credit for the model belongs to the original creators.