ZeyuLing/motius-motionclr-humanml3d

MotionCLR HumanML3D for Motius

This repository packages the official MotionCLR HumanML3D checkpoint as a self-contained Motius artifact. It includes the denoising network, frozen OpenAI CLIP ViT-B/32 text encoder, HumanML3D statistics, source revision, and license attribution.

from motius.pipelines.motionclr import MotionCLRPipeline

pipe = MotionCLRPipeline.from_pretrained(
    "ZeyuLing/motius-motionclr-humanml3d",
    device="cuda",
)
motion = pipe.infer_t2m(
    ["a person walks forward and waves"],
    [120],
)[0]

The output is an unnormalized (T, 263) HumanML3D motion at 20 fps. See the Motius MotionCLR model card for evaluation and representation details.

MotionCLR remains under the included IDEA License 1.0. Original source: https://github.com/IDEA-Research/MotionCLR