interior-home-3d-reconstruction-from-videos
Made by JeffReconstruct editable home interiors and connected utility digital twins from video. Use when modeling rooms, recovering cameras, tracing home systems, or preparing evidenced simulation inputs.
Works With
Quality and Utility come from canonical skill metadata. Community ratings appear separately as stars when subscriber reviews exist.
Subscribe to download
Public previews are available, but downloads, CLI sync, and premium content require an active subscription.
Sign In to Subscribe- Full skill content with deterministic versioning
- CLI sync + local search index
- Access to Jeffrey’s premium library
Save to your account
Saved skills live in your cloud list. Run jsm sync on each machine to pull them locally.
What's New with this Skill
2Updates, enhancements, and release notes for this skill.
Updated: interior-home, a .blend that opens without the source disk, a sync proof clip, and browser viewer checks
The delivered Blender file now carries what it needs: blender_split.py restore packs linked libraries and used images into the .blend and refuses to save if any dependency is missing. A new sync.py proof command renders a silent side-by-side clip of the exact source frames that a clock solution pairs across two to four cameras. The digital-twin reference grows from 932 to 2,338 lines, with checks for staged browser viewers, cutaway section caps, overlapping surfaces and foliage.
- blender_split.py restore packs linked libraries and used images into the delivered .blend and refuses to save if any dependency is missing
- New sync.py proof command: a silent side-by-side clip of the exact source frames a clock solution pairs across 2 to 4 cameras
- The digital-twin reference grows from 932 to 2,338 lines: staged browser viewers, section caps, overlap audits and foliage
These are lessons from taking a reconstruction through to a model you can walk through in a browser.
- A .blend that opens anywhere.
blender_split.py restorenow packs linked libraries and every used external image into the delivered file, and stops if anything is missing or still unpacked. The skill asks for a reopen with the original asset paths unavailable and a comparison of object, mesh and packed-resource fingerprints, because a save that succeeds while the source disk is mounted says nothing about portability. A new test covers the packing.
New skill: interior-home-3d-reconstruction-from-videos, walkthrough videos to an editable Blender interior with its utilities traced
An agent turns ordinary walkthrough videos of a home (a phone, optionally camera glasses, with spoken room names) into an editable whole-home Blender interior, placed by default inside the exterior model from /drone-flyover-video-to-home-model-in-blender. Rooms are fitted like jigsaw pieces against the exterior shell, every placement is checked against the original frames before anything is built, and wiring, fuel lines and pipes are traced as equipment joined at typed ports.
- Input is ordinary walkthrough video: a phone, optionally plus camera glasses, with narration naming rooms and features. It needs no capture app, IMU, scanner app or GPU inference.
- Maps are cut into 10 to 30 s slices at pose breaks and wherever two maps disagree about the camera's turn. On the real house, 39 of 102 slices contained breaks.
- Every room needs two marks on features the exterior shows (windows, window ACs, skylights). With one, it can still swing about that point and stays UNPLACED.
- assembly.py fits all slices against the exterior shell in one exact search. PLACED is a claim: certify.py records a look at the source frame, and only CERTIFIED or LOCATED pieces are built.
- Floors and walls come from measured points, and doors are proposed where the walk crossed between rooms; house_checks.py checks the shell, window claims, stairs and connectivity. Unsupported areas stay unknown.
- Utilities are equipment joined at typed ports: panel to circuits to loads, tank to burner, with fuel kept apart from the flue. Unseen routes stay hypotheses, and branch cuts show which loads lose their path.
- A foliage-heavy drone exterior is split into a light working copy, and split_verify.py proves a lossless restore for every deliverable. Facade doors are cut without touching the mesh.
- Delivery: the reopened .blend with architecture and contents views, camera replay, source/render comparisons, a browser viewer that walks inside, camera movies and verified GLB exports.
- 40 operator cards, each pairing a real failure with the test that guards it, and 41 helper scripts with 41 test files.
The skill was developed on a real house, where one global camera solve of a whole walkthrough never held: the map bends and folds. What placed the rooms was many small maps. slices.py cuts each map into 10 to 30 s pieces at pose breaks and wherever two maps disagree about how the camera turned, and drops featureless slices under about 15 cameras (39 of 102 real slices contained breaks). Each slice then gets marks on features the exterior also shows: windows, window air conditioners and skylights, as points whose rays meet at 3° or more and reproject within 3 px, or as bearings when depth is