FastWan2.2-TI2V-5B — GGUF Q6_K
GGUF Q6_K conversion of the transformer of
FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers
(revision 3e187042a324f6f5fb68fd22110a78725253de8f, file transformer/diffusion_pytorch_model.safetensors),
for stable-diffusion.cpp.
PatiPix Studio uses it to create short videos from a text, locally on the user's PC.
How it was made
- The diffusers tensor names were mapped back to the original Wan 2.2 layout; all 825 tensors were checked against the Wan2.2-TI2V-5B Q8_0 GGUF of QuantStack.
- Quantization with
sd-clifrom stable-diffusion.cpp (commit012c7f8):-M convert --type q6_K.
No training or fine-tuning: the weights are those of FastWan 2.2; only the file format and the quantization change.
File
| File | Size (bytes) | SHA-256 |
|---|---|---|
FastWan2.2-TI2V-5B-Q6_K.gguf |
4,141,808,800 | 219b7707993aab360e4b187e2a652a67831bab396f4c6e053a0dfaf40cce172f |
Use
With stable-diffusion.cpp: --diffusion-model FastWan2.2-TI2V-5B-Q6_K.gguf, together with the Wan 2.2 VAE
(wan2.2_vae.safetensors) and the umT5-XXL text encoder, a few sampling steps and --cfg-scale 1.0
(FastWan is a distilled model).
License and credits
Apache License 2.0, like the original model (see LICENSE).
Original works: Wan 2.2 by Wan-AI (Alibaba); FastWan 2.2 by the FastVideo team (Hao AI Lab).
- Downloads last month
- 178
6-bit