Configuration

Configuration

All settings live in the Configuration block at the top of epubify.sh.

Paths

Variable	Description
INPUT_DIR	Folder holding the PDFs to convert. Must exist — the script stops if it doesn't.
STAGING_DIR	Staging folder for PDFs awaiting conversion. Created if missing.
OUTPUT_DIR	Root for converted documents. Created if missing.
MODEL_CACHE	Where pdf2epub's downloaded models live between runs. Created if missing.

Container

Variable	Description
IMAGE	Image to run. Default ghcr.io/overcuriousity/pdf2epub:latest.
BUILD_CONTEXT	Optional path to a pdf2epub checkout. Used to build the image locally if the pull fails; leave empty to skip that fallback.
INPUT_READ_ONLY	Mount the staging tree read-only. Default true. Set false if a conversion fails with a read-only filesystem error.
EXTRA_ARGS	Extra arguments appended to every pdf2epub invocation, e.g. EXTRA_ARGS=(--start-page 10 --max-pages 50).
CONTAINER_MEM_MAX	Hard memory ceiling per conversion. Default 8G; empty disables.
CONTAINER_TIMEOUT	Wall-clock limit per conversion. Default 2h; empty disables.

The two ceilings exist so a malformed PDF that makes the model pipeline balloon gets killed locally instead of taking the desktop session with it.

GPU

Variable	Description
USE_GPU	Use an NVIDIA GPU where the host supports it. Default true; falls back to CPU with a notice if the container runtime is missing.
GPU_IMAGE	Tag for the locally built GPU image. Default epubify/pdf2epub:gpu.
GPU_DOCKERFILE	Dockerfile used to build it, resolved beside the script. Default Dockerfile.gpu.
APP_STATIC_DIR	Path inside the container handed a writable tmpfs, so pdf2epub can create it while running as your UID. Update the Python version if a future image ships a different one; empty disables the mount.

See Installation for what GPU support needs on the host.

Behaviour

Variable	Description
PROCESS_DELAY	Seconds to pause between conversions. Default 2.
EXCLUDED_BASENAMES	Filenames (without extension) to skip entirely, e.g. sample, preview.
TIDY_NAMES	Normalise separators and apply title case to output names. Default true. See Name Tidying.

Error log

ERROR_LOG defaults to error_log.txt inside OUTPUT_DIR. It's created lazily — a clean run leaves no file at all.