Video · image-to-video · long-video-generation · depth-controlled-video · trajectory-controlled-video

LongVie 2

Vchitect

WorldFoundry vendors the official LongVie DiffSynth runtime under worldfoundry/synthesis/visual_generation/longvie/longvie_runtime.

Integrated统一环境longvie-2

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image-to-video

环境worldfoundry-unified-cu128

设备CUDA 12.8

worldfoundry-eval evaluate \
  --mode model \
  --model-id longvie-2 \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/longvie-2 \
  --metric artifact_count \
  --json

兼容性与版本

由 Manifest 生成

集成状态: integrated
Runner 证据: pending
环境: worldfoundry-unified-cu128
Python: 3.10
CUDA: CUDA 12.8
PyTorch: torch
源码 revision: —
Checkpoint revision: dd061bbd6d2084fa474d8d66bbbfbfb790d0fde3

Runtime profile: longvie-2
Pipeline binding: longvie-2
Runner: worldfoundry.pipeline
Pipeline target: worldfoundry.pipelines.longvie.pipeline_longvie:LongVie2Pipeline
Backend stage: in_tree_runtime
Runtime 状态: checkpoint_assets_pending_not_runnable_until_staged
Driver 状态: compatible
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 longvie-2

依赖版本约束 21

torch
torchvision
transformers>=4.57.0,<5
accelerate
peft
imageio
imageio[ffmpeg]
safetensors
einops
sentencepiece
protobuf
modelscope
ftfy
pynvml
pandas
decord
pillow
numpy
opencv-python
huggingface_hub
xfuser

Conda 依赖 5

python=3.10
pip
ffmpeg
setuptools
wheel

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 longvie-2 --check-local --json

Vchitect/LongVie2

Revision: —
License: —
Gated: —
Private: —

Wan-AI/Wan2.1-I2V-14B-480P

Revision: —
License: —
Gated: —
Private: —

Vchitect/LongVie2

Revision: dd061bbd6d2084fa474d8d66bbbfbfb790d0fde3
License: —
Gated: —
Private: —

Wan-AI/Wan2.1-I2V-14B-480P

Revision: 6b73f84e66371cdfe870c72acd6826e1d61cf279
License: —
Gated: —
Private: —

Wan-AI/Wan2.1-T2V-1.3B

Revision: 37ec512624d61f7aa208f7ea8140a131f93afc9a
License: —
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 longvie-2 \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/longvie-2 \
  --metric artifact_count \
  --json

输入契约

字段	已记录契约
`controls`	dense_depth_video_per_segment, sparse_pointmap_or_track_video_per_segment
`image`	Required
`prompt`	Required
`video`	Required

Artifact 契约

Artifact 类型	文件名 / 路径
`generated_video`	`—`
`generated_video`	`longvie-2.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: integrated
Runner 证据: pending
原生 Demo 证据: 未记录
验证 Imports: torch, torchvision, transformers>=4.57.0,<5, accelerate, peft, imageio, safetensors, einops, sentencepiece, modelscope, ftfy, pynvml, pandas, decord, PIL, numpy, cv2, xfuser

配方溯源

worldfoundry/data/models/catalog/video/longvie-2.yamlProjecthttps://vchitect.github.io/LongVie2-project/GitHubhttps://github.com/Vchitect/LongVieHugging Facehttps://huggingface.co/Vchitect/LongVie2Hugging Facehttps://huggingface.co/Wan-AI/Wan2.1-I2V-14B-480P

构建运行命令

输入契约

Artifact 契约

相关配方