# Deferred work

Work intentionally not done in the initial build, with enough context to pick it
up cold. Each file states the goal, the recommended approach, the hard parts, and
a rough effort estimate.

## Index

| File                   | Topic                                           | Priority |
| ---------------------- | ----------------------------------------------- | -------- |
| background-removal.md  | Remove/replace background (phase 2)             | High     |
| face-aware-passport.md | Auto-frame faces for ID photos (phase 2)        | High     |
| free-form-crop.md      | True corner-drag crop + layers/text (Konva)     | Medium   |
| enhancements.md        | Undo/redo, persistence, batch, more adjustments | Medium   |
| known-limitations.md   | Current behavioral limits and caveats           | n/a      |

## The one cross-cutting tension to know about

The app sets Cross-Origin-Embedder-Policy: require-corp so SharedArrayBuffer is
available for threaded WASM (wasm-vips, jSquash mt). Under require-corp, any
cross-origin subresource (for example an ML model fetched from a CDN) is blocked
unless it carries a Cross-Origin-Resource-Policy header.

Consequence: features that pull large models (background removal, face
detection) must self-host their model and wasm assets same-origin, or the page
must drop cross-origin isolation (and therefore the vips fast path). Plan to
self-host the assets. This is called out again in the relevant files.

## Status of the data model for phase 2

The Template type already carries an optional `framing` field
({ faceHeightRatio: [min, max], eyeLineFromTop }) and the US passport template
populates it. EditorCanvas already renders a manual framing guide from it. The
remaining phase-2 work is to compute the crop automatically from a detected face.