Every command-line tool in the pipeline, grouped by what you are trying to do. One script, one job: nothing here is imported by image_to_3dlab/, and anything worth reusing was pulled out as a module-level function so it can be tested without Blender or a GPU.

Each summary below is the script's own first line. tests/test_scripts_registry.py fails if a script is missing from this file or if a summary has drifted from the docstring, so a new script is not finished until it is listed here.

Most tools print their real documentation with --help; the docstring at the top of each file explains why it exists, which is usually the part you need.

Run a backend end to end, or re-run part of one without paying for the whole thing again.

For the staged fitted metarig → proxy weights → reference pose → tunable gait → standing transition workflow, start with the quadruped pipeline guide. It includes commands, parameter definitions, failure checks, and an LLM-assisted tuning checklist. Manual fitting and visual weight/deformation review remain required.

For prop sheets (many props from one image in one run, then split and finished into game-ready LODs), start with the prop sheet guide.

A generation run costs 15-20 minutes. These cost seconds and are worth it first.

Everything here operates on a GLB or a cached decode and is headless.

The asset's surface, as opposed to its shape.

Numbers and renders to decide whether a change helped. Read docs/ before trusting an old one.

These talk to a running Blender over the execute_code socket on port 9876 unless they say headless.

Heavier edits that need Blender's own operators rather than trimesh.

The *_pose.py files are pure curve maths with no bpy, which is why they have tests.

Numbers about the project itself, not about any asset.

vendor/ is git-ignored, so every fix to someone else's checkout lives here as a re-appliable patch script. Each one asserts its anchor and is idempotent; re-running after a bootstrap is the intended workflow.