Repository navigation
Video2Scene v0.1.0 · Public preview
Pre-releaseVideo2Scene v0.1.0 — Public preview
Video2Scene turns RGB video into editable indoor scenes through a local workbench: review the video and estimated layout, approve a whitebox, refine objects with agents, and export saved scenes to Blender and GLB.
Included in this preview
- Persistent object refinement with a rolling task queue (default 3, configurable 1–4), shared model/Blender limits, serial scene saves and task recovery.
- A 30-second synthetic living-room comparison, linked from the README. The recorded case completed 18 modeling targets and reopened its Blender/GLB exports. The comparison is qualitative; camera and lighting conditions are documented in the case notes.
- Chinese and English READMEs, installation/configuration guides, a public workflow guide, a standalone workflow HTML page, contribution guidance and third-party notices.
- An original tabletop demo for editing, saving and exporting without model credentials or GPU inference. Blender is required for saving and export.
Start here
- README / quick start
- English README
- Workflow
- Living-room case and comparison conditions
- Installation
Validation
Local regression: Windows 506 passed / 16 skipped, Linux 496 passed / 27 skipped. Final scoped Windows checks passed after portability and release-tool changes. Browser checks covered 50 workbench cases, 5 real Blender demo cases and 6 public-document checks. Node transport/retry and editor-state checks also passed. Scope and skip conditions.
These checks do not call live models. The complete GPU/model environment was not reinstalled from scratch for this release; the living-room showcase is a separate existing end-to-end run. This is an early personal project for local use and feedback.
GitHub Actions for this release.
Attachments
video2scene-0.1.0-public.zip: reviewed source snapshot, includingRELEASE_FILES.jsonwith individual file hashes.livingroom-comparison.mp4: 30 seconds, 1920 × 792, H.264; reference scene beside agent reconstruction.SHA256SUMS.txt: SHA-256 checksums for the two attachments above.
Original project code is Apache-2.0. Bundled components, separately installed models and services retain their own terms; see third-party notices.