crop-and-resize

name	crop-and-resize
description	Use when the user wants ID-style 3×4 photos, an A4 print sheet of faces, or batch resizing of images for printing. Triggers on "fotos 3x4", "passport photo", "ID photo", "tile faces on A4", "resize for print".

crop-and-resize

End-to-end workflow for face-aware ID-photo cropping and A4 tiling. Repo-local skill — assumes image-playground layout.

When to use

Pick this skill if the user wants any of:

• Crop one or more faces from photos into 3×4 cm format
• Build an A4 print sheet of faces (with multiple copies, or one of each)
• Batch resize images to a fixed square size for printing

Inputs

The user typically provides:

1. Source photos — already in samples/, or paths/URLs to copy in.
2. Selection — which detected faces to keep (numbered after first run).
3. Layout choice — fill the sheet with copies, or place one of each.
4. Spacing — gap between photos (default 0.5 cm).

Workflow

digraph crop_resize_flow {
    "Receive request" [shape=doublecircle];
    "Source images in samples/?" [shape=diamond];
    "Copy/download to samples/" [shape=box];
    "Run script (no flags)" [shape=box];
    "Show contact_sheet.jpg" [shape=box];
    "User confirms which faces to keep" [shape=box];
    "Run with --keep=... [--unique]" [shape=box];
    "Show a4_sheet.jpg" [shape=box];
    "Adjust spacing/layout?" [shape=diamond];
    "Edit SPACING_CM, re-run" [shape=box];
    "Done" [shape=doublecircle];

    "Receive request" -> "Source images in samples/?";
    "Source images in samples/?" -> "Run script (no flags)" [label="yes"];
    "Source images in samples/?" -> "Copy/download to samples/" [label="no"];
    "Copy/download to samples/" -> "Run script (no flags)";
    "Run script (no flags)" -> "Show contact_sheet.jpg";
    "Show contact_sheet.jpg" -> "User confirms which faces to keep";
    "User confirms which faces to keep" -> "Run with --keep=... [--unique]";
    "Run with --keep=... [--unique]" -> "Show a4_sheet.jpg";
    "Show a4_sheet.jpg" -> "Adjust spacing/layout?";
    "Adjust spacing/layout?" -> "Edit SPACING_CM, re-run" [label="yes"];
    "Adjust spacing/layout?" -> "Done" [label="no"];
    "Edit SPACING_CM, re-run" -> "Show a4_sheet.jpg";
}

Step-by-step

1. Verify the venv

[ -d .venv ] || python3 -m venv .venv
.venv/bin/pip install -q -r requirements.txt

2. First pass — detect all faces

.venv/bin/python scripts/face_to_a4.py

Produces:

• output/individual_crops/face_NN.jpg — every detected face cropped to 3×4
• output/contact_sheet.jpg — numbered preview, show this to the user
• output/a4_sheet.jpg — first-pass A4 with all faces tiled

3. Show the contact sheet

Read output/contact_sheet.jpg and let the user identify which numbered faces to keep (duplicates across photos, false positives at edges, bystanders).

4. Rebuild with selection

.venv/bin/python scripts/face_to_a4.py --keep=1,3,5,7

Each kept face is tiled to fill the A4 (copies distributed evenly).

For literal one-of-each (no duplicates, mostly empty sheet):

.venv/bin/python scripts/face_to_a4.py --keep=1,3,5,7 --unique

5. Spacing and tunables

Top of scripts/face_to_a4.py:

Constant	Default	Meaning
DPI	300	Print resolution
PHOTO_CM	(3, 4)	Crop size (width, height) in cm
A4_CM	(21, 29.7)	Sheet size
SPACING_CM	0.5	Gap between cells (cut-line)
MIN_FACE_PX	130	Reject smaller detections (background)

Edit in place and re-run; no flags for these (intentional — change is rare).

6. Batch resize (alternative path)

If the user just wants square crops for printing (no face detection):

./scripts/resize.sh -s 8 -u cm samples/*.jpg          # 8×8 cm
./scripts/resize.sh -s 4 -u inch -o output/ *.jpg     # 4×4 inch into output/
./scripts/resize.sh -h                                # all options

Common pitfalls

• Multiple detections of one person across photos. The script can't dedupe by identity (no face_recognition). Always show the contact sheet and let the user pick.
• Tight crops cutting off hair. Tune photo_h = fh * 1.7 in crop_3x4() upward (e.g., 1.9) for more headroom.
• Bystanders detected. Raise MIN_FACE_PX from 130 to 180+ to drop background faces.
• Dropped a personal photo into samples/. Move it elsewhere — samples/ is committed to git. Use output/ for anything ephemeral.