模型配方
选择模型，检查运行时，并复制与你的机器匹配的命令。
267 models运行时数据来自仓库 manifest
模型任务运行时PythonCUDACheckpoint
AC3DCamera-Control VideoWorldFoundry vendors the official AC3D inference runtime under worldfoundry/synthesis/visual_generation/ac3d/ac3d_runtime; WorldFoundry owns path resolution, command construction, output loading, and pipeline/catalog exposure.
camera-control-videotext-to-video+1Runner verified独立环境3.1012.4—AnimateDiffguoywwVerified against the official repo default RealisticVision scripts.animate demo; file SHA256, decoded pixel SHA256, and artifact size match exactly.
text-to-videolegacy-unet-diffusionRunner verified统一环境3.1112.8guoyww/animatediffAstraEvanEternalIn-tree Astra inference route is registered under worldfoundry/synthesis/visual_generation/kling/astra_runtime.
image-to-videocamera-control+1Runner verified统一环境3.1112.8EvanEternal/AstraBeing-H0.5BeingBeyondWorldFoundry keeps the inference-only Being-H0.5 architecture and policy runtime directly under worldfoundry/synthesis/action_generation/being_h05.
vlavla.policy_rollout+2Runner verified独立环境3.10Cu118 Or NewerBeingBeyond/Being-H05-2BCameraCtrlCamera-Control VideoRuntime code is vendored in-tree; checkpoints remain external assets under the normal hfd/ckpt cache.
worldvideo+2Runner verified统一环境3.1011.3hehao13/CameraCtrlCogVideoXTHUDMHF cardData reports Apache-2.0 for 2B and 'other' for several larger variants.
text-to-videoimage-to-videoRunner verified统一环境3.1112.81137dacfcDepth Any PanoramasInsta360-ResearchIntegrated as an independent geometric prior at worldfoundry/base_models/three_dimensions/depth/depth_anything/depth_anything_v1.
panoramic-depth-estimationmetric-depth-estimation+1Runner verified统一环境3.1112.8Insta360-Research/DAP-weightsDepth Anything 3 Priordepth-anythingIntegrated as an independent depth prior at worldfoundry/base_models/three_dimensions/depth/depth_anything/depth_anything_v3.
monocular-depth-estimationmetric-depth-estimation+1Runner verified统一环境3.1112.8depth-anything/DA3METRIC-LARGEDepth Anything V1LiheYoungIn-tree Depth Anything V1 pipeline is registered for relative depth and geometry artifact export.
depthgeometry+1Runner verified统一环境3.1112.8LiheYoung/depth-anything-large-hfDepth Anything V2 Priordepth-anythingIntegrated as an independent depth prior at worldfoundry/base_models/three_dimensions/depth/depth_anything/depth_anything_v2.
monocular-depth-estimationmetric-depth-estimation+1Runner verified统一环境3.1112.8depth-anything/Depth-Anything-V2-LargeDualCamCtrlCamera-Control VideoRuntime code is integrated in-tree; checkpoints use the default Hugging Face/local cache loading path.
worldvideo+4Runner verified统一环境3.1112.8FayeHongfeiZhang/DualCamCtrlDUSt3RNAVER LABS EuropeWorldFoundry vendors DUSt3R under worldfoundry.base_models.three_dimensions.general_3d.dust3r.
feed-forward multi-view 3D reconstruction and geometry estimationRunner verified统一环境3.1112.8—DUSt3R Base ModelnaverBase-model profile for the DUSt3R catalog entry; exposed through the shared geometry-prior pipeline/operator contract.geometry3d-reconstruction+1Runner verified统一环境3.1112.8—FantasyWorldAMAP and Beijing University of Posts and Telecommunications authorsWorldFoundry exposes the in-tree FantasyWorld Wan2.1 and Wan2.2 camera-control pipelines.
geometry-consistent world modelingRunner verified统一环境3.1112.8acvlab/FantasyWorld-Wan2.1-I2V-14B-480PFlashWorldFlashWorld authorsWorldFoundry exposes the in-tree FlashWorld pipeline, operator, and representation runtime.
fast 3D scene generation from image or textRunner verified统一环境3.1112.8imlixinyang/FlashWorldGeoCalib PriorcvgIntegrated as an independent camera-calibration prior at worldfoundry/base_models/three_dimensions/general_3d/geocalib.
camera-calibrationgravity-estimation+1Runner verified统一环境3.1112.8—GigaBrain-0open-gigaaiWorldFoundry in-tree GigaBrain-0.1 runtime emitted a checkpoint-backed CUDA action_trace using converted LeRobot RobotWin stats.
vlavla.action_prediction+3Runner verified统一环境3.1112.8open-gigaai/GigaBrain-0-3.5B-BaseHunyuan Game CrafttencentIn-tree Hunyuan GameCraft pipeline is registered; official inference uses the HYGameCraft conda profile and torchrun dispatch.
interactive-world-modelimage-to-video+1Runner verified独立环境3.1012.4022caa869HunyuanVideotencentTencent-Hunyuan GitHub and tencent Hugging Face repos were both reachable.text-to-videoimage-to-videoRunner verified独立环境3.1012.46204ad6aeHunyuanVideo 1.5tencentAdded as a high-priority current open-weight video model. Native T2V and I2V infer-only `generate.py` paths are now ported in-tree and release-validation verified with local checkpoint assets.text-to-videoimage-to-videoRunner verified统一环境3.1112.89b49404b3HunyuanWorld-MirrorTencent Hunyuan / HY-WorldWorldFoundry exposes framework inference through the in-tree HunyuanMirrorPipeline and vendored inference-only WorldMirror model code.three_dimensionRunner verified统一环境3.1012.8tencent/HunyuanWorld-MirrorHunyuanWorld-VoyagerTencent Hunyuan / HY-WorldExisting in-tree runtime is now exposed through a BaseSynthesis subclass.worldvideo+1Runner verified统一环境3.1112.8ef280b832HY-WorldPlayTencent Hunyuan / HY-WorldExisting in-tree runtime is now exposed through a BaseSynthesis subclass.worldvideo+1Runner verified统一环境3.1112.8tencent/HY-WorldPlayInfinite-WorldMeiGen-AIIn-tree Infinite-World runtime resolves local or Hugging Face checkpoint layouts without relying on an external repository at inference time.
worldvideo+1Runner verified统一环境3.1112.8thuml/InfiniteWorldInSpatio-WorldinspatioThe in-tree InSpatio-World pipeline generated and visually validated a 297-frame video through Workspace on 2026-07-10.
video-conditioned-world-generationtrajectory-conditioned-video+1Runner verified统一环境3.1112.8inspatio/worldInSpatio-WorldFMInSpatioThe WorldFoundry runner supports the official meta.json demo in the unified environment profile with local WorldFM assets.worldvideo+1Runner verified统一环境3.1112.8—Kairos SensenovaEmbodied World ModelUses the in-tree Apache-2.0 official torchrun runtime under worldfoundry/synthesis/visual_generation/kairos/kairos_runtime.
text-to-videoimage-to-video+2Runner verified独立环境3.1012.8kairos-agi/kairos-sensenova-robot-4B-480P-distilledLingBot-WorldLingBotLingBot-World wrapper now only resolves local HFD or explicit checkpoint directories and no longer downloads at runtime.
worldcamera_control+1Runner verified独立环境3.1012.16fc824ffcLoGeRJunyi42In-tree LoGeR point-cloud navigation pipeline is registered for reconstruction and trajectory rendering.
3d-reconstructiontrajectory+1Runner verified统一环境3.1112.8Junyi42/LoGeRLyraNVIDIA Spatial Intelligence LabWorldFoundry exposes in-tree Lyra-1 and Lyra-2 pipelines backed by vendored runtime adapters.
open generative 3D and 4D world modelsRunner verified独立环境3.1012.4depth-anything/DA3NESTED-GIANT-LARGE-1.1Matrix-Game 2.0Matrix-GameMatrix-Game 2 already uses an in-tree runtime and no external repo_root dependency was found.
worldinteractive+1Runner verified统一环境3.1112.8f1729d99aMatrix-Game 3.0Matrix-GameMatrix-Game 3 runtime is vendored under worldfoundry/synthesis/visual_generation/matrix_game/matrix_game_3_runtime.worldinteractive+1Runner verified统一环境3.1112.8382aa382aMetric3D PriorJUGGHMIntegrated as an independent metric-depth prior at worldfoundry/base_models/three_dimensions/depth/metric3d.
monocular-depth-estimationmetric-depth-estimation+1Runner verified统一环境3.1112.8JUGGHM/Metric3DMolmoAct2allenaiWorldFoundry in-tree MolmoAct2 runtime lives under worldfoundry/synthesis/action_generation/molmoact2 and no longer depends on the external molmoact2 checkout at runtime.
vlavla.action_prediction+2Runner verified统一环境3.1112.8allenai/MolmoAct2-DROIDMonST3RMonST3R authorsWorldFoundry vendors the MonST3R infer-time components in-tree.geometry estimation in the presence of motionRunner verified统一环境3.1112.8—MotionCtrlCamera/Object Motion ControlRuntime code is vendored in-tree; checkpoints remain external assets under the normal hfd/ckpt cache.Runner verified统一环境
3.1112.8—NeoVerseCASIA and CreateAI authorsWorldFoundry exposes the in-tree NeoVerse pipeline and vendored DiffSynth runtime adapter.
4D world modeling from in-the-wild monocular videosRunner verified统一环境3.1112.8Yuppie1204/NeoVerseOpenVLAopenvlaIn-tree OpenVLA HF runtime is registered through the WorldFoundry pipeline binding.
vlavla.action_prediction+1Runner verified统一环境3.1011.347a0ec7fcPrior Depth AnythingRain729Integrated as an independent depth-completion prior at worldfoundry/base_models/three_dimensions/depth/priorda.
depth-completionmetric-depth-estimation+1Runner verified统一环境3.1112.8Rain729/Prior-Depth-AnythingReCamMasterKling Team / Kuaishou, Zhejiang University, CUHK, and HUST authorsWorldFoundry exposes the in-tree ReCamMaster pipeline, operator, and synthesis runtime under the Kling namespace.
camera-controlled video rerenderingRunner verified统一环境3.1112.84f3f73917RoboFlamingorobovlmsRoboFlamingo/OpenFlamingo architecture and inference source are vendored in-tree; public pipeline CUDA action_trace validation passed.
vlavla.policy_rollout+2Runner verified统一环境3.8Prepare Onlyrobovlms/RoboFlamingoSCOPEzizhaotongOfficial SCOPE inference source, DiffSynth fork, README, license, and examples are vendored under worldfoundry.synthesis.visual_generation.scope.scope_runtime.
interactive-world-modelgame-world-model+2Runner verified统一环境3.1012.8zizhaotong/SCOPEShow-oshowlabRuntime source is vendored in-tree; current WorldFoundry wrapper supports t2i image generation.
multimodal-generationtext-to-imageRunner verified统一环境3.1011.3c2c6f548cSkyReels V3SkyworkWorldFoundry registers SkyReels V3 through a model-specific pipeline backed by the in-tree runtime shim.reference-to-videoaudio-guided-video+2Runner verified统一环境3.1112.8Skywork/SkyReels-V3-R2V-14BSolarisnyu-visionxIn-tree inference-only Solaris runtime is packaged under worldfoundry/synthesis/visual_generation/solaris/solaris_runtime.
video_generationmulti-agent-world-model+1Runner verified统一环境3.1112.8nyu-visionx/solarisSplatt3RbrandonsmartRuntime code is vendored in-tree under worldfoundry/base_models/three_dimensions/general_3d/splatt3r/splatt3r_runtime; checkpoints remain external assets.
3d-reconstructiongaussian-splattingRunner verified统一环境3.1011.3brandonsmart/splatt3r_v1.0StarVLAStarVLAWorldFoundry runtime executed official StarVLA QwenOFT predict_action with local Qwen3-VL-OFT-LIBERO-4in1 checkpoint and Qwen3-VL-4B base VLM on an 80GB CUDA GPU.
vlawam+2Runner verified统一环境3.1012.81947454beVchitect-2.0 T2VVchitectUse id vchitect-2-t2v. The in-tree Vchitect runtime emits generated_video when local checkpoints and runtime dependencies are available.
text-to-videovideo-generationRunner verified统一环境3.1112.8379368187VGGT-OmegaHuggingfaceVGGT-Omega uses an official-architecture representation runner plus the existing WorldFoundry VGGTResult/save interface.
3d-reconstructiongaussian-splatting+1Runner verified统一环境3.1112.8facebook/VGGT-OmegaVMemliguang0115In-tree VMem runtime at worldfoundry/synthesis/visual_generation/vmem/vmem_runtime; no external checkout is required.
video-generationmemory+2Runner verified统一环境3.1112.8ac5921080Wan2.1Wan-AIOnly official Wan-Video GitHub and Wan-AI Hugging Face repos are listed.
text-to-videoimage-to-videoRunner verified统一环境3.1112.837ec51262Wan2.1 VACEWan-AIRuntime no longer depends on external local Wan source checkouts.video-editingvideo-to-video+2Runner verified统一环境3.1112.8Wan-AI/Wan2.1-VACE-14BWarp-as-Historyyyfz233Official Warp-as-History source, Helios diffusers runtime, Pi3 submodule code, CLI scripts, and demo assets are vendored under worldfoundry.synthesis.visual_generation.warp_as_history.warp_as_history_runtime.
image-to-videocamera-controlled-video-generationRunner verified独立环境3.1012.1yyfz233/warp-as-historyWorldCamKAIST / Adobe Research / MAUM AIWorldCam runtime code is in-tree and the wrapper now supports load_model=False planning without external checkout dependency.
worldcamera_control+1Runner verified统一环境3.1112.8worldcam/worldcamYUMEYUME authorsWorldFoundry exposes in-tree YUME and YUME-1.5 pipelines backed by vendored runtime adapters.
image/video conditioned interactive world video generationRunner verified统一环境3.1112.8stdstu123/Yume-I2V-540PZeroScopeHuggingfaceCC-BY-NC-4.0 requires benchmark policy review before commercial or automated benchmark use.
text-to-videolegacy-unet-diffusionRunner verified统一环境3.1112.8cerspense/zeroscope_v2_576wAction Chunking TransformertonyzhaozhThe ACT policy, DETR-VAE, transformer, ResNet backbone, preprocessing, and action-chunk aggregation execute from the flattened in-tree package.
action_chunking_policyrobot_policy+2Integrated独立环境3.10Cu118 Or Newer—AlayaWorldLTX-2.3 World ModelsThe runtime reuses WorldFoundry's existing LTX-2.3 Transformer, VAE, Gemma connector, RoPE, patchifier, attention dispatcher, and distributed helpers.image-to-videocamera-controlled-video+2Integrated统一环境3.1112.8e3f1adebeAllegrorhymes-aiGitHub verification succeeded after retry; the first attempt hit a transient TLS error.
text-to-videoimage-to-videoIntegrated统一环境3.1112.8c1b9207bbATI-Wan2.1-14BCamera-Control VideoOnly ATI's official inference delta is vendored; the unchanged Wan2.1 implementation is reused from worldfoundry/base_models.camera-controlled-videoimage-to-videoIntegrated统一环境3.1112.8bytedance-research/ATI每个配方记录什么
详情页直接由仓库 manifest 生成，而不是人工维护一张宣传表。每个配方会把下面几类声明分开呈现：
| 记录 | 可以确认什么 |
|---|---|
| Catalog | 稳定 model ID、任务、官方来源、别名和集成状态。 |
| Runtime profile | 输入契约、artifact 契约、backend stage 和运行时就绪状态。 |
| Environment | Python、CUDA tier、固定依赖、验证 import，以及独立/统一环境路由。 |
| Pipeline binding | WorldFoundry 使用的 runner 与具体 pipeline target。 |
| Evidence | 原生 demo 与 WorldFoundry runner 的 parity；只有 catalog 记录并不代表 GPU 已验证。 |
缺失值会明确显示为 未记录。没有 runtime 或 evidence 记录时，WorldFoundry 不会把 catalog 条目升级成更强的“已支持”声明。
常用命令
# 检查完整 catalog manifest。
worldfoundry-eval zoo model-show \
--model-id <model-id> \
--include-manifest \
--json
# 不下载，只解析 checkpoint、凭据与本地资产 blocker。
worldfoundry-eval zoo model-download \
--model-id <model-id> \
--check-local \
--json
# 为一个 runtime profile 解析并安装环境 manifest。
bash scripts/setup/model_env_install.sh --model <model-id>