Layer Rebuild
Give a flat image another working life.
A local image-processing experiment that extracts selected elements, attempts background repair, and produces a layered PSD draft for further design work.
THE DESIGN QUESTION
How can a designer recover useful, editable structure from an image that only exists as a flat file?
Combine an interactive Claude Code skill with a parameterized image pipeline. Keep selection and external-processing choices visible, and inspect the output before accepting it.
- My contribution
- Workflow design, a custom Claude Code skill, and a local interface for planning and reviewing layer extraction.
- Current scope
- Local alpha · SAM3 Cutout pipeline
- Exploring
- Agent Skills · Tool orchestration · Human control
THE WORKFLOW
Select
Choose the image and identify the elements that must be separated.
Extract
Route regions through segmentation or local extraction methods.
Repair
Choose whether and how to fill exposed background areas.
Inspect
Review layers and export a PSD draft with its recipe.
A CLOSER LOOK
From pixels to a reviewable layer stack.
Workflow illustration · no original product image or application screenshot
“Separate the main object and the label.”
Required elements become explicit extraction requests rather than an implied promise to separate everything.
A recipe records the chosen operations.
The image, selected elements, repair method, and output paths form a repeatable processing configuration.
Inspect the foreground and the background separately.
A valid PSD does not prove that all objects are correctly separated or repaired.
Export a layered draft and its recipe.
The local export also includes a preview and a manifest describing unresolved limitations.
The current local UI does not reconstruct hidden object parts. Its text layers are raster pixels; the advanced pipeline has separate, manually configured text reconstruction.
UNDER THE SURFACE
Make the process legible.
Make the boundaries explicit.
Skill + tools
A custom Claude Code skill guides choices and prepares a JSON recipe for segmentation, edge refinement, repair, and PSD assembly.
Execution controls
The local interface requests separate consent for external segmentation and generative fill, and fails required-object extraction when it cannot be fulfilled.
Output validation
Checks confirm the PSD exists and contains layers; visual separation and repair quality still require human inspection.
TODAY
Local prototype
The parameterized pipeline and local alpha interface run locally. Exports are explicitly treated as drafts.
WHAT’S NEXT
Still exploring
Improve per-object selection and the quality of background repair before preparing a hosted trial.
Built around the SAM3 Cutout pipeline, using existing segmentation and inpainting models alongside local image-processing methods.