WorldFoundry
English中文

Vchitect-2.0 T2V

Method overview: Fig. 2 : Overview of the parallel transformer architecture used in Vchitect-2.0. The model combines text and video features using a unified framework. The Text Encoder processes input prompts, generating textual.

Vchitect-2.0 T2V from Shanghai AI Laboratory (Vchitect): text-to-video and video-generation; Static load only.

Edit this pageGitHubMarkdown

Vchitect-2.0 is Shanghai AI Lab's open text-to-video diffusion transformer; this entry serves the 2B checkpoint through a fully native assembly. It carries an unusually strong structural guarantee: the native DiT matches all 1019 checkpoint tensors and the native SD3 decoder matches all 138 decoder tensors. Aliases include vchitect, vchitect-2, and vchitect2-t2v.

Official paper teaser: Fig. 1 : High quality videos generated by our 2B model, demonstrating its capabilities in generating videos with high-fidelity and consistent actions.
Official paper teaser: Fig. 1 : High quality videos generated by our 2B model, demonstrating its capabilities in generating videos with high-fidelity and consistent actions.
Method overview: Fig. 2 : Overview of the parallel transformer architecture used in Vchitect-2.0. The model combines text and video features using a unified framework. The Text Encoder processes input prompts, generating textual.
Method overview: Fig. 2 : Overview of the parallel transformer architecture used in Vchitect-2.0. The model combines text and video features using a unified framework. The Text Encoder processes input prompts, generating textual.

The recorded paper is arXiv 2501.08453.

Code: Vchitect/Vchitect-2.0.

Build your run

Pipeline
Vchitect2T2VPipelineworldfoundry.pipelines.vchitect.pipeline_vchitect_2_t2v
TaskInteractive Video interactive-video
Environmentworldfoundry-unified-cu128
DeviceCUDA 12.8
Runnerworldfoundry.pipeline
Loadingfrom_pretrained
Invocationnative_or_standard_invocation
worldfoundry-eval run \
  vchitect-2-t2v \
  --pipeline.task-profile interactive-video \
  --json

What it is for

vchitect-2-t2v is the recorded prompt-to-video route — not image editing and not a policy. Start here when the job is a clip from text.

A parallel diffusion transformer paired with an SD3-family VAE decoder, assembled from role-based DiT, conditioner, scheduler, initializer, and decoder components. The pinned official source revision is recorded in the profile (Vchitect-2.0 @ be56767f). It is published by Shanghai AI Laboratory (Vchitect).

Catalog tasks include text-to-video and video-generation. Catalog aliases include vchitect, vchitect-2, vchitect2-t2v, and vchitect_2_t2v. Documented modalities: text in, video out.

Recorded evaluation entry points include VBench (Standard T2V quality suite).

Runtime contract

Optional prompt, negative_prompt, frames, fps, and height; artifacts generated_video and vchitect_2_t2v_public_runner.mp4.

CLI task profiles: interactive-video. WorldFoundry binds Vchitect2T2VPipeline.

Family status: Static load only.

Runner status is Static load only: the default path loads the shared native pipeline and does not emit blocked plan artifacts. Weights are apache-2.0 at Vchitect/Vchitect-2.0-2B. Use id vchitect-2-t2v.

Use id vchitect-2-t2v. The public pipeline is assembled from native role-based components.

Running in WorldFoundry

Launch worldfoundry-eval run vchitect-2-t2v --pipeline.task-profile interactive-video to sample the clip. The command builder above switches --pipeline.task-profile among recorded tasks and still emits worldfoundry-eval run.

It runs in the unified environment worldfoundry-unified-cu128 (Python 3.11, CUDA 12.8).

Recorded weight repositories include Vchitect/Vchitect-2.0-2B. Prepare the environment and check local assets with bash scripts/inference/prepare_model_infer.sh vchitect-2-t2v. Add --download to fetch missing public checkpoints.

Recorded licenses include apache-2.0; status: Static load only.