YAML Metadata Warning:The pipeline tag "text-to-motion" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other
PRISM-KT HumanML3D
This repository contains the self-contained PRISM-KT checkpoint used by Motius. PRISM represents motion as a time-by-body latent grid and generates it with a Kinematic-Unit Flow Transformer. This variant uses kinematic-tree-derived KT-RoPE and enables the depth-driven KAFS inference schedule by default.
The artifact includes KU-FlowT, the causal Motion VAE, T5 tokenizer and text encoder, scheduler, motion statistics, and Motius metadata. Its native output is 138D SMPL-H body-22 motion at 30 fps. The VAE runs in fp32; KU-FlowT and T5 may run in bf16.
from motius.pipelines.prism import PRISMPipeline
pipe = PRISMPipeline.from_pretrained(
"ZeyuLing/motius-prism-kt-humanml3d",
bundle_kwargs={"device": "cuda"},
)
result = pipe.text_to_motion(
"a person walks forward and waves with the right hand",
num_frames=129,
seed=42,
)
HumanML3D T2M
Fixed selected-caption protocol, 20-repeat means:
| Evaluator | Samples | R@1 | R@2 | R@3 | FID | MM-Dist | Diversity |
|---|---|---|---|---|---|---|---|
| HumanML3D Official | 3,970 | 0.5448 | 0.7308 | 0.8176 | 0.2081 | 2.9067 | 9.4802 |
| MotionStreamer | 4,042 | 0.7408 | 0.8619 | 0.9050 | 19.9682 | 15.8072 | 27.2536 |
See the full PRISM model card for TP2M, physical diagnostics, representation details, and citation.