| name | photo-book-builder |
| description | Automates high-resolution photo book creation and correction passes, including chronological sorting, orientation balancing, print-ready full-bleed masonry layout generation, existing layout crop corrections, contact-sheet/cell mapping, and cover art upscaling. Use this skill when asked to: (1) Organize photos into page folders, (2) Generate or regenerate 12" x 30" open spreads (9000 x 3600 px at 300 DPI) with gutter margins, (3) Create or adapt full-bleed masonry grid layouts with project-specific photo counts, (4) Correct existing exports by raising/lowering images or adjusting per-image crop framing without overwriting originals, (5) Upscale front, back, or briefcase box covers using high-quality LANCZOS interpolation.
|
Photo Book Builder
This skill provides a complete system for organizing, structuring, and generating print-ready, high-resolution photo books. It is designed for 12" × 30" open spreads (15" × 12" closed book size) at 300 DPI, enforcing strict gutter margins to prevent content from being lost in the binding.
Core Capabilities
The photo book builder supports five main stages:
graph TD
A[Stage 1: Prep & Organize] -->|organize_photos.py| B[Page Folders & Mapping JSON]
B --> C[Stage 2 & 3: Layout Generation]
C -->|create_full_bleed_layouts.py| D[9000x3600 px Print Spreads]
E[Stage 4: Cover Upscaling] -->|upscale_covers.py| F[Print-Ready Cover JPEGs]
D --> G[Stage 5: Correction Pass]
G -->|custom renderer or patched generator| H[New Corrected Export Folder]
Stage-by-Stage Workflow
Stage 1: Preparation & Balanced Page Sorting
Chronological sequence must be preserved at a macro level, but photos should be scrambled within local windows (chunks) to balance the number of landscape ("L") and portrait ("P") images per page folder.
- Target Pages: 20 page folders (
page 1 to page 20).
- Balanced Distribution: Target 11 or 12 photos per page (e.g. 9 Landscapes and 2-3 Portraits per page).
- JSON Map: A mapping file
photo_organization_map.json is generated to track which photos belong to which page.
Execution:
Run the organizer script, specifying the source folder containing the flat list of JPEG images and the target output directory:
python scripts/organize_photos.py --dir "path/to/photos" --out "path/to/workspace" --seed 42
Note: If running on Windows, use python -Xutf8 to avoid console encoding issues with status output.
Stage 2: Grid Layout & Coordinates
Spreads are generated at $9000 \times 3600$ px (300 DPI). Each page half gets an active grid area of $4275 \times 3300$ px after subtracting margins and gutter seams:
- Outer Margins: 150 px top, bottom, and outside edges.
- Vertical Center Gutter: 150 px total gutter seam (no image content between $x = 4425$ and $x = 4575$).
- Image Gaps: 40 px spacing horizontally and vertically.
Layout configurations are determined dynamically based on photo counts and orientations:
- Template 5A (5 Photos: 4L, 1P): One large landscape col, one portrait + landscape col.
- Template 6A (6 Photos: 5L, 1P): Two landscape rows (4 total), one portrait + landscape col.
- Template 6B (6 Photos: 4L, 2P): Two large landscape rows (2 total), one double-portrait + double-landscape col.
These templates are defaults, not hard limits. When a project already has a custom generator, page folders, or spreads with a different number of photos (for example 23-24 images per spread), preserve that project's generator and layout rules. Do not force the default 5/6-photo half-page templates onto an existing custom book.
For a complete coordinate reference and details on column configurations, see layout_templates.md.
Stage 3: Spread Assembly & Rendering
The layout compiler reads the page directories, detects photo orientations on the fly, applies the coordinate grids, crops photos to fill cells (using centered LANCZOS fit-scale), and merges them into $9000 \times 3600$ px JPEG spreads.
- Page Swapping: Alternating layout orientation avoids monotonous designs. Left page columns swap if
page_num % 2 == 0, and right page columns swap if page_num % 3 == 0.
When regenerating an existing project, first inspect local scripts in the delivery/project folder. If a project-specific script exists, prefer adapting it over the bundled default scripts. Avoid rerunning any organizer/reorganizer that flattens or moves page folders unless the user explicitly asks for a new organization pass.
Execution:
Compile the print-ready layouts:
python scripts/create_full_bleed_layouts.py --workspace-dir "path/to/workspace" --out-dir "path/to/output/layouts" --quality 95
Stage 4: Cover Art Upscaling
AI-generated cover art and presentation box lid files need to be upscaled to final high-resolution print dimensions using high-quality LANCZOS interpolation.
- Front/Back Cover size: $4500 \times 3600$ px.
- Presentation Briefcase Box size: $4950 \times 4050$ px.
Execution:
Upscale cover background textures or complete cover designs:
python scripts/upscale_covers.py --src "path/to/cover.png" --out "path/to/upscaled_cover.jpg" --width 4500 --height 3600 --quality 95
Stage 5: Existing Layout Correction Pass
Use this when the user provides screenshots or notes such as "blue means raise the image" and "red means take it down" for already-exported spreads.
Correction workflow:
- Identify the current source workspace, existing export folder, and correction screenshots.
- Inspect the local layout generator and
photo_organization_map.json if present.
- Create a contact sheet or map preview that overlays each cell with its source filename and rectangle. This prevents guessing from screenshots.
- Translate annotations into per-image corrections:
- Blue / "raise image up": move the visible subject up in the cell. In crop-ratio terms this usually means increasing vertical crop offset when there is extra source height below the crop.
- Red / "take it down": move the visible subject down in the cell. In crop-ratio terms this usually means decreasing vertical crop offset when there is extra source height above the crop.
- If the crop is already pinned at the top or bottom, use a small top-anchored or bottom-anchored overscale instead of adding filler or painting pixels. Start tiny, e.g.
1.006 to 1.012; larger values can overshoot.
- Render to a new export folder. Never overwrite the original export unless the user explicitly asks.
- Verify file count, output dimensions, and a reduced review sheet of corrected spreads.
Per-image correction patterns:
CROP_OVERRIDES = {
"DSC01234.jpg": {"y_ratio": 0.30},
"DSC05678.jpg": {"y_ratio": 0.00},
}
For edge-pinned images where ratio changes cannot move the frame:
TOP_ANCHORED_OVERSCALE = {
"DSC01234.jpg": 1.010,
}
Keep corrections local to filenames/cells. Avoid global crop changes unless every affected image needs the same behavior.
Safety rules for correction passes:
- Preserve existing page folders and mapping files.
- Prefer a renderer that reads existing pages and writes new spreads over a script that reorganizes files.
- Use output folder names like
10 - layouts - crop-corrections-YYYYMMDD.
- Keep a short correction log with source filenames, correction type, output folder, and verification result.
- On Windows, avoid giant one-shot patches or command lines; keep scripts and path lists small.
Revert Option
If you need to restart the process or undo the page folder classification, run the revert tool to move photos back to the root workspace directory and remove empty page folders:
python scripts/revert_organization.py --workspace-dir "path/to/workspace"
Bundled Resources
