Video · text-to-video · video-generation

Step-Video-T2V

stepfun-ai

Runtime source is vendored in-tree; full execution remains multi-GPU/service/checkpoint gated.

Planned统一环境step-video-t2v

构建运行命令

选择仓库中已记录的 variant，然后复制对应的准备、检查或运行命令。

worldfoundry.pipeline

Variant

任务text-to-video

环境worldfoundry-unified-cu128

设备Prepare Only

worldfoundry-eval evaluate \
  --mode model \
  --model-id step-video-t2v \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/step-video-t2v \
  --metric artifact_count \
  --json

兼容性与版本

由 Manifest 生成

集成状态: planned
Runner 证据: implemented multigpu services pending
环境: worldfoundry-unified-cu128
Python: 3.10
CUDA: Prepare Only
PyTorch: —
源码 revision: —
Checkpoint revision: 7a2b639ca2685350e87a4df7e4026285309f7fb6

Runtime profile: step-video-t2v
Pipeline binding: step-video-t2v
Runner: worldfoundry.pipeline
Pipeline target: worldfoundry.pipelines.component_pipelines:StepVideoT2VPipeline
Backend stage: in_tree_runtime_plan
Runtime 状态: in_tree_step_video_runtime_ported_multigpu_services_pending
Driver 状态: blocked_official_requires_torch25_cuda124_or_newer_driver
环境类型: 统一环境

安装环境

环境解析器会读取已记录 profile，并选择此处显示的统一或独立环境。

bash scripts/setup/model_env_install.sh --model step-video-t2v

依赖版本约束 13

accelerate
transformers
diffusers
sentencepiece
imageio
numpy
einops
aiohttp
flask
flask_restful
ffmpeg-python
requests
xfuser

Conda 依赖 3

python
pip
ffmpeg

Checkpoint 与资产

分配算力前先做本地检查；gated、private 与 license 字段直接来自 checkpoint manifest。

worldfoundry-eval zoo model-download --model-id step-video-t2v --check-local --json

stepfun-ai/stepvideo-t2v

Revision: 7a2b639ca2685350e87a4df7e4026285309f7fb6
License: mit
Gated: —
Private: —

运行与输出

生成的命令通过共享 evaluation 边界运行，并持久保存结果 manifest 与 artifact。

worldfoundry-eval evaluate \
  --mode model \
  --model-id step-video-t2v \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/step-video-t2v \
  --metric artifact_count \
  --json

输入契约

字段	已记录契约
`image`	Optional
`prompt`	Required

Artifact 契约

Artifact 类型	文件名 / 路径
`generated_video`	`—`
`generated_video`	`step_video_t2v.mp4`

证据与来源

Catalog 集成、原生 demo parity 与 runner parity 是三条独立记录。

集成状态: planned
Runner 证据: implemented multigpu services pending
原生 Demo 证据: 未记录
验证 Imports: transformers, diffusers, flask

配方溯源

worldfoundry/data/models/catalog/video/step-video-t2v.yamlGitHubhttps://github.com/stepfun-ai/Step-Video-T2V

构建运行命令

输入契约

Artifact 契约

相关配方