Wan2.1-I2V-14B-720P-Diffusers-fp16

An fp16 cast of Wan2.1-I2V-14B-720P-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-720P-Diffusers-fp16", torch_dtype=torch.float16
).to("cuda")

frames = pipe(
    image=load_image("still.png"),
    prompt="the camera pushes in slowly",
    height=720, width=1280, num_frames=81,
).frames[0]

License

Apache-2.0, inherited from Wan2.1.