How to use
- Copy the installation command and review the source package before installing it in your own agent environment.
- Check the source document for required tools, credentials and supported hosts.
- Describe your input and desired deliverable; review the output against the source material.
npx skills add https://github.com/nphu0811/pmg-skills --skill pmg-image-decomposerBefore you start
- Installation does not include a model subscription, API allowance or permission to access your files.
- Repository stars and installation counts describe activity, not an independent quality or security test.
Rendered from the source repository. Headings, examples, code, tables, links, and referenced images are preserved.
PMG Image Decomposer
pmg-image-decomposer converts flat images into structured, reusable collections of independent visual objects consumable by the canonical PMG engine. Integrated via pmg.image_decomposer.ImageDecomposer and pmg.layers.LayerTree.
Core Decomposition Modes
- Mode 1 — Semantic Decomposition:
- Isolates meaningful brand/visual elements (e.g., Agribank green background, red diagonal bands, white separators, wheat stem and grains).
- Classifies editability:
native_vector(OpenXML<a:custGeom>freeform polygons) vs.raster_layer(transparent PNG). - Generates hierarchical grouping (
wheat_group->Grain_01..N).
- Mode 2 — Geometric Slicing:
- Partitions images into geometric segments for shutter, mosaic, tile, and carousel reveals.
- Patterns:
vertical,horizontal,grid,tiles,puzzle,radial/circular_sectors,triangles,custompolygons. - Guarantees complete pixel coverage without dropped content.
- Mode 3 — Depth Decomposition:
- Partitions scenes into foreground, midground, and background spatial planes.
- Supports 2.5D camera parallax and tracking shots.
- Explicitly flags inferred or inpainted occluded areas.
Reassembly Validation Gate
Every decomposition executes an automatic reassembly verification:
- Renders reconstructed layers and measures pixel similarity (
>= 0.94tolerance) and contour alignment against the original asset. - Reports raster-only vs. vector components.
- Layer trees with failed reassembly are rejected before PowerPoint compilation.
Native PowerPoint Compilation
Layer trees compile into native PowerPoint objects via pmg.native_objects.compile_layer_tree:
- Groups become PowerPoint
!!group shapes. - Native vectors become editable
<a:custGeom>shapes with custom paths. - Transparent raster segments become positioned and cropped picture shapes.
- Objects retain stable
!!identities across slides for Morph transitions.
Python API Usage
from pmg.image_decomposer import ImageDecomposer
from pmg.native_objects import compile_layer_tree
from pmg.layers import LayerTree
decomposer = ImageDecomposer()
# 1. Semantic decomposition
tree = decomposer.decompose(
source_path="path/to/logo.png",
output_dir="layers/logo",
mode="semantic",
granular=True,
)
assert tree.reassembly["status"] == "PASS"
# 2. Geometric slicing
shutter_tree = decomposer.decompose(
source_path="path/to/photo.jpg",
output_dir="layers/shutter",
mode="geometric",
pattern="vertical",
count=8,
)
# 3. Compile to native PowerPoint slide
# slide is a python-pptx Slide object
mapping = compile_layer_tree(slide, tree, bounds=(1.0, 1.5, 11.3, 5.0))
