Cover Design Master

Problem

Book cover PDFs already contain vector elements such as titles, author names, ISSN, barcodes, and solid fills. Dropping a generated background underneath them can easily hurt legibility. Image models may also emit letters, numbers, watermarks, or flatten the page into a non-editable raster. This skill treats “adding a background to a cover PDF” as a verifiable pipeline: the background must be text-free, sit under the source PDF, and preserve editable text, barcodes, and fills.

How It Works

• analyze_layout.py reads page size, orientation, text density, barcodes, and solid-fill blocks, then emits a structured layout brief.
• The layout prompt requests one seamless background with no front/back divider; when dense text is detected, it asks for a light, low-contrast overall tone.
• The drawing prompt expands topic keywords into visible scenes, lighting, material, and composition, such as an offshore CO2 storage scene with a platform, seawater, strata, and injection well.
• The merged prompt keeps the NO TEXT constraint and selects 1536x1024, 1024x1536, or 1024x1024 based on PDF aspect ratio.
• prepare_background.py uses oversize plus bottom cropping to remove model watermarks and export a precisely proportioned cover-background.png.
• composite_pdf.py performs layered composition with merge_page, so source PDF text remains selectable and barcode white blocks are not recolored.
• Optional add_caption.py inserts editable SimSun captions when 封面论文： or 专题： is present.

Boundaries

This workflow is best for adding a unified background under an existing PDF, not for rebuilding the cover layout or moving text. The main subject is constrained to the middle-lower right half, while the left half remains low-contrast and text-friendly; people default to Chinese facial features. Results depend on the drawing model, so the prompt must specify lighting, color, and focal elements to avoid flat single-color output. If the PDF lacks topic clues, confirm the book title and subject first.