scene-generation
Generate WorldGen 3D scenes from text, images, or equirectangular panoramas, choose splat or mesh output, and explore results in a local Viser viewer.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- scene-generation
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Generate WorldGen 3D scenes from text, images, or equirectangular panoramas, choose splat or mesh output, and explore results in a local Viser viewer.
- disable-model-invocation
- true
02 资源文件 2
- 参考文档references/troubleshooting.md2.8 KB
- 参考文档references/workflows.md4.0 KB
03 SKILL.md 正文
Scene generation
Use this route when the task asks for a WorldGen scene from a text prompt, source image, or 360 panorama; wants a Gaussian splat or Open3D mesh; needs low-VRAM or experimental Sharp/inpainting options; or wants to explore and export novel views in a browser.
Choose the input path
- Text-to-scene: construct
WorldGen(mode="t2s", ...)and callgenerate_world(prompt). - Image-to-scene: construct
WorldGen(mode="i2s", ...), load an RGB PIL image, and callgenerate_world(prompt, image=image). - Panorama-to-scene: load a 2:1 equirectangular PIL image and call
_generate_world(pano_image, return_mesh=...)to skip panorama diffusion. - Interactive viewer: run the bundled
scripts/worldgen_demo.py; it starts Viser athttp://localhost:8080, renders the scene, and exposes camera path and novel-view export controls.
Read references/workflows.md for concrete code and
command recipes. Read references/troubleshooting.md
for workflow-specific failures after the package-level checks in the root
references/troubleshooting.md.
Select the output
- Leave
return_mesh=Falsefor the defaultSplatFileGaussian scene. Save it withscene.save("scene.ply"). - Set
return_mesh=Truefor an Open3D triangle mesh and save it withopen3d.io.write_triangle_mesh("scene.glb", mesh). - Do not combine
return_mesh=Trueandinpaint_bg=Truein the bundled demo. The inpainting implementation merges foreground/background splats.
Select runtime options
- Use
low_vram=Truewhen the GPU has limited memory or when model loading causes an OOM. The bundled launcher auto-enables it below 24 GB. - Use
use_sharp=Trueonly after the base pipeline works and the optional Sharp package/checkpoint is available. - Use
inpaint_bg=Trueonly for an explicitly experimental background repair; it adds segmentation and LaMa model dependencies and extra downloads. - Use
resolutionto trade panorama detail against memory and latency. Start lower when debugging an environment or an input contract.
Safe preflight
Before a long model download or generation run:
python scripts/check_worldgen_env.py --demo-help
This confirms the installed public API and a tiny CUDA allocation without loading FLUX, DA-2, Sharp, or LaMa weights. For the full flag surface, run:
python scripts/worldgen_demo.py --help
Boundaries
This sub-skill owns user-facing generation, output conversion, the bundled viewer helper, and optional generation modes. It does not own package installation diagnosis, upstream DA-2/PyTorch3D development, model training, release engineering, or repository maintenance. Route those questions to the root references or a more appropriate skill.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.