WorldFoundry
English中文

WorldFoundry

Run, inspect, and evaluate world models in one reproducible workflow.

Explore model recipesStart the quickstart
267 manifest-backed model recipes

Configure a run

Choose a real catalog entry. Runtime and version facts come from the repository manifests, and the command stays copyable.

Selected modelHunyuanVideo 1.5tencent
Runtimeworldfoundry-unified-cu128Python 3.11 · CUDA 12.8
ReadinessRunner verifiedSeparate from runner evidence
Command preview
worldfoundry-eval evaluate \
  --mode model \
  --model-id hunyuanvideo-1.5 \
  --model-runner worldfoundry:pipeline \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/hunyuanvideo-1.5 \
  --metric artifact_count \
  --json

Why WorldFoundry

Infrastructure for operating world models.

Shared

Environment setup, checkpoints, input shaping, previews, and reporting stay shared while model-specific behavior stays explicit.

Durable

Expensive outputs survive the process that created them, so they can be inspected, rescored, compared, or audited without rerunning inference.

Honest

Catalog coverage, runnable integration, official validation, and leaderboard eligibility stay separate—not collapsed into one ambiguous “supported” label.

Read the benefits and tradeoffs

Universal Compatibility

One engine, endless possibilities. Browse integrated models and benchmarks.

225 models64 benchmarks

Video

Text-, image-, and video-conditioned generation, editing, and audio-video systems.

52 models

Operate end to end

Catalog, runtime, workspace, and evaluation share identities and durable contracts.

Know what exists

Manifests expose stable IDs, sources, capabilities, assets, runtime bindings, readiness, and blockers before compute is allocated.

See what is included

Run through shared boundaries

Pipelines preserve model-native I/O while TUI, CLI, Studio, Python, and MCP reuse the same execution contracts.

Understand the design

Inspect before scoring

Videos, geometry, actions, trajectories, and traces stay visible on disk and in Studio instead of disappearing inside scripts.

Explore Studio

Evidence, not headlines

Benchmark runners preserve per-sample outcomes, coverage, provenance, blockers, reports, and scorecards.

Read about evaluation

Five-stage pipeline

How it works

Every surface follows the same sequence. The artifact is the handoff between model execution and benchmark evaluation.

  1. Discover

    Choose a model and benchmark

    Catalog manifests expose stable IDs, capabilities, readiness, assets, and blockers.

  2. Prepare

    Stage runtime and assets

    Resolve the conda profile, checkpoints, datasets, metric weights, and credentials.

  3. Run

    Generate through one contract

    TUI, CLI, scripts, and Studio dispatch to a pipeline and operator.

  4. Inspect

    Review normalized artifacts

    Videos, geometry, actions, traces, and metadata remain visible and reusable.

  5. Evaluate

    Produce reviewable evidence

    Metrics and official runners write reports, blockers, and a normalized scorecard.

Artifacts are the handoff: model runtimes do not own benchmark logic, and benchmark runners do not load model checkpoints.

Everyone welcome

Got questions? We're here to help.

Whether you're just getting started or debugging a complex run, the community is open to everyone. No question is too basic.

  • Fast & friendly responses
  • Active maintainers
WorldFoundry WeChat community QR code
WeChat CommunityScan to join. The QR code is updated periodically if it expires.

Resources

Notes, events, and lab context around the project.

Blog

Updates, technical notes, and release highlights.

Events

Meetups, demos, benchmark sprints, and milestones.

OpenEnvision

The lab organization behind WorldFoundry.

Ecosystem

Sibling projects and source links in the OpenEnvision ecosystem.