Quantized weights we publish
Collection
FP8, INT8, NF4 and precision-cast variants of open models. Each card says exactly which tensors changed and which were left alone. • 16 items • Updated
Image-to-Video
Diffusers
Safetensors
English
Chinese
WanImageToVideoPipeline
wan
wan2.1
video-generation
fp16
import torch
from diffusers import DiffusionPipeline
from diffusers.utils import load_image, export_to_video
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("wavespeed/Wan2.1-I2V-14B-480P-Diffusers-fp16", dtype=torch.bfloat16, device_map="cuda")
pipe.to("cuda")
prompt = "A man with short gray hair plays a red electric guitar."
image = load_image(
"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/guitar-man.png"
)
output = pipe(image=image, prompt=prompt).frames[0]
export_to_video(output, "output.mp4")Wan2.1-I2V-14B-480P-Diffusers-fp16
An fp16 cast of Wan2.1-I2V-14B-480P-Diffusers.
What was changed
Upstream ships the transformer in fp32. Here every component — transformer, UMT5 text encoder, CLIP image encoder and VAE — is stored in fp16. The transformer goes from ~65.6 GB to ~32.8 GB. Tensor names and shapes are unchanged and nothing is quantized; this is a straight dtype cast.
Usage
import torch
from diffusers import WanImageToVideoPipeline
from diffusers.utils import load_image
pipe = WanImageToVideoPipeline.from_pretrained(
"wavespeed/Wan2.1-I2V-14B-480P-Diffusers-fp16", torch_dtype=torch.float16
).to("cuda")
frames = pipe(
image=load_image("still.png"),
prompt="the camera drifts slowly to the left",
height=480, width=832, num_frames=81,
).frames[0]
License
Apache-2.0, inherited from Wan2.1.
- Downloads last month
- 24
Model tree for wavespeed/Wan2.1-I2V-14B-480P-Diffusers-fp16
Base model
Wan-AI/Wan2.1-I2V-14B-480P-Diffusers