WorldFoundry
English中文

T2V Turbo T2V

Method overview: T2V-Turbo’s consistency-distillation stack for few-step video generation.

T2V Turbo T2V from University of California, Santa Barbara: text-to-video; Static load only; T2V-Turbo.

Edit this pageGitHubMarkdown

T2V-Turbo is a fast text-to-video model distilled from VideoCrafter2 with consistency distillation plus reward feedback, generating watchable clips in 4-8 steps. In this catalog it is assembled from the shared native diffusion components: the VideoCrafter2 base model and the ordered T2V-Turbo LoRA remain separate declarative checkpoint inputs.

Official paper teaser: T2V-Turbo distilled text-to-video samples from the paper.
Official paper teaser: T2V-Turbo distilled text-to-video samples from the paper.
Method overview: T2V-Turbo’s consistency-distillation stack for few-step video generation.
Method overview: T2V-Turbo’s consistency-distillation stack for few-step video generation.

The recorded paper is T2V-Turbo: Breaking the Quality Bottleneck of Video Consistency Model with Mixed Reward Feedback (2024).

Project page: t2v-turbo.github.io. Code: Ji4chenLi/t2v-turbo.

The paper is T2V-Turbo: Breaking the Quality Bottleneck of Video Consistency Model with Mixed Reward Feedback. Project page: t2v-turbo.github.io. Code: Ji4chenLi/t2v-turbo.

Build your run

Pipeline
T2VTurboT2VPipelineworldfoundry.pipelines.t2v_turbo.pipeline_t2v_turbo_t2v
TaskInteractive Video interactive-video
Runnerworldfoundry.pipeline
Loadingfrom_pretrained
Invocationnative_or_standard_invocation
worldfoundry-eval run \
  t2v_turbo_t2v \
  --pipeline.task-profile interactive-video \
  --json

What it is for

Pick t2v_turbo_t2v when you want this family's text-to-video line in WorldFoundry: a prompt in, a generated clip out. Use it as the catalog T2V start rather than an editor or a world-model stepper.

It is published by University of California, Santa Barbara.

Catalog tasks include text-to-video. Catalog aliases include t2v-turbo-t2v and t2v-turbo. Documented modalities: text in, video out.

Recorded evaluation entry points include VBench (T2V-Turbo was tuned for exactly this suite's dimensions).

Runtime contract

Optional prompt, frames, fps, height, and width.

CLI task profiles: interactive-video. WorldFoundry binds T2VTurboT2VPipeline.

Family status: Static load only.

Runner status records native checkpoint validation: local VideoCrafter2 and T2V-Turbo VC2 assets are available for strict component and LoRA checks. Use it when you need many quick T2V drafts rather than maximum quality.

Uses the shared WorldFoundry native diffusion assembler, runner, checkpoint loader, offload policy, and LVDM components.

Running in WorldFoundry

Launch worldfoundry-eval run t2v_turbo_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.

The environment profile is not recorded.

Recorded weight repositories include jiachenli-ucsb/T2V-Turbo-VC2 and VideoCrafter/VideoCrafter2. Prepare the environment and check local assets with bash scripts/inference/prepare_model_infer.sh t2v_turbo_t2v. Add --download to fetch missing public checkpoints.

Status: Static load only.