worldgen
Guides WorldGen workflows for generating 3D scenes from text, images, panoramas, or the interactive demo, including Gaussian-splat and mesh outputs.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- worldgen
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Guides WorldGen workflows for generating 3D scenes from text, images, panoramas, or the interactive demo, including Gaussian-splat and mesh outputs.
- disable-model-invocation
- true
02 资源文件 7
- 参考文档references/api-reference.md5.6 KB
- 参考文档references/cli-reference.md2.9 KB
- 参考文档references/repo-provenance.md1.2 KB
- 参考文档references/repo-routing-metadata.json405 B
- 参考文档references/troubleshooting.md6.1 KB
- 脚本scripts/check_worldgen_env.py1.9 KB
- 脚本scripts/worldgen_demo.py12.9 KB
03 SKILL.md 正文
WorldGen
WorldGen turns text prompts, source images, or equirectangular panoramas into 3D scenes. The main pipeline generates a 360 panorama, predicts depth, and converts the result into a Gaussian splat or an Open3D mesh; the bundled viewer adds browser-based free-view exploration and novel-view export.
Install and import check
WorldGen targets Python 3.11 and a CUDA-capable PyTorch environment for real scene generation:
python -m pip install -e .
python scripts/check_worldgen_env.py --demo-help
The package also needs the external DA-2/PyTorch3D/Nunchaku/Open3D/Viser stack
shown in references/troubleshooting.md. Keep
those compiled dependencies on a compatible Python, torch, and CUDA ABI.
Route by task
- Generate from text, an image, or a panorama; select splat or mesh output:
use
sub-skills/scene-generation/SKILL.md. - Run the interactive browser viewer or save novel views: use the same
scene-generation route and
references/cli-reference.md. - Inspect constructor, method, tensor, and output contracts: read
references/api-reference.md. - Diagnose installation, CUDA, model downloads, optional Sharp/inpainting, or
viewer failures: read
references/troubleshooting.mdand then the scene-generation troubleshooting reference. - Check freshness against the source checkout: read
references/repo-provenance.md.
Common starting points
- Begin with the scene-generation sub-skill for
WorldGen(mode="t2s"),WorldGen(mode="i2s"),_generate_world(pano_image), orreturn_mesh=True. - Use
scripts/check_worldgen_env.pybefore any long model download. It checks imports, public signatures, and a tiny CUDA allocation without loading weights. - Use
scripts/worldgen_demo.py --helpto inspect the self-contained interactive launcher. It does not depend on the original repository checkout. - Treat
use_sharp=Trueandinpaint_bg=Trueas optional experimental branches; validate the default splat or mesh path first.
Shared references and scripts
references/api-reference.mdcontains verified signatures, mode rules, output types, and lower-level conversion helpers.references/cli-reference.mddocuments the bundled demo flags, output files, viewer behavior, and safe preflight.references/troubleshooting.mdcovers install/import, CUDA, VRAM, model access, input validation, and viewer errors.scripts/check_worldgen_env.pyruns the read-only import/API/CUDA smoke check.scripts/worldgen_demo.pyis the bundled, self-contained Viser launcher adapted from the repo's public demo workflow.
Boundaries
This is a runtime user skill for WorldGen's generation and visualization surface. It does not cover model training, release engineering, repository maintenance, upstream DA-2/PyTorch3D development, or generic Open3D/Viser usage without WorldGen. Optional Sharp and background-inpainting guidance is included only as an explicitly unverified extension to the core generation route.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).