baoyu-image-cards

Provide your source content and invoke the skill; it analyzes the content, recommends a style, layout, and card count, then asks for confirmation before generating. You can also pass options like --preset knowledge-card or --yes to skip confirmation.

There are 12 styles (cute, notion, chalkboard, screen-print, sketch-notes, and others), 8 layouts (sparse, balanced, dense, list, comparison, flow, mindmap, quadrant), and 3 palette overrides (macaron, warm, neon) that combine freely.

It resolves backends in priority order: a user-specified backend, a saved preferred_image_backend in EXTEND.md, then auto-selection preferring runtime-native tools like Codex imagegen, falling back to installed backends such as baoyu-imagine. It never substitutes SVG or HTML rendering for raster output.

It generates the cover image first without references, then passes that image as a reference to every subsequent card, optionally combined with a shared session ID. This anchor chain prevents drift in character design and color rendering.

Yes, but only by editing the card's prompt file and regenerating with a new output path. The skill explicitly forbids patching bitmaps with ImageMagick, Pillow, or overlay scripts, preserving the flawed candidate for comparison.

Card count is capped at 1-10 per series, batch size clamps to 1-8, and output quality depends on the available raster backend. Dense layouts are incompatible with some styles like screen-print, which the compatibility matrix flags.