Aimed at `npx hyperframes` users (standalone and inside monorepos), where the
native modules `sharp` and `onnxruntime-node` can't install or load.
## Native modules are now optional, and never abort the CLI
`sharp` and `onnxruntime-node` are native modules: their platform binaries ship
as optional sub-dependencies that can fail to land on end-user installs
(--omit=optional, musl/glibc, monorepo hoisting, cross-platform lockfiles,
broken npx cache). Both powered only optional commands, yet both were wired as
hard `dependencies`, so on any platform where a binary can't install the whole
CLI failed to install. Moved both to `optionalDependencies` (alongside
@google/genai) so the core CLI always installs; the native-accelerated paths
light up only when present.
Runtime handling so a missing/unloadable binary degrades instead of crashing:
- `capture` (`contentExtractor.ts`): sharp was a static top-level
`import sharp from "sharp"`, so a load failure threw on module import —
before the inner try/catch — aborting the whole command. Now a guarded lazy
`await import("sharp")` that skips SVG captioning with an actionable warning.
Marked `external` in tsup so esbuild never bundles the native module.
- `remove-background` (`inference.ts`): both `onnxruntime-node` and `sharp`
are loaded here and genuinely required. The dynamic imports are now guarded
to throw an actionable "install / reinstall with optional deps" error
(surfaced cleanly by the command's existing try/catch) instead of a raw
"Cannot find module". New tests assert createSession rejects with that
guidance — before touching the model download — when either module is
unavailable.
`contactSheet.ts` also uses sharp but is already behind a dynamic-import
boundary wrapped in try/catch, so it was never a hard-fatal path.
## inspect: content-overlap as a warning, not a blocking error
The `content_overlap` layout-audit check shipped as `severity: "error"`, and
the audit exits non-zero when `errorCount > 0`, so `inspect` failed for
compositions that intentionally layer text. Downgraded to `severity: "warning"`
so it still reports (and prints the `data-layout-allow-overlap` opt-out hint)
without breaking exit codes. Reversible.
- Extract applyMask helper from postprocess and add 5 unit tests pinning
the contract this PR is selling: fg.alpha + bg.alpha === 255 per pixel,
RGB triples byte-identical between fg and bg, and bg=null path leaves
the bg buffer untouched. Without these, a future postprocess change
(mask threshold, premultiplied alpha, gamma) could silently break the
inverse-alpha relationship and the existing plumbing tests would all
still pass.
- Add stdin 'error' listener inside spawnFfmpeg. If either encoder dies
mid-render, Node emits an unhandled error on the dead writable on the
next .write() and crashes the CLI before waitForExit's reject path
can surface the encoder's stderr tail. Doubled encoder count = doubled
failure surface, so this is worth pinning down.
- Tighten stdio param to a 3-tuple so an accidental 1-element array fails
at type-check.
- Sharpen backpressure comment: write→true means "highWaterMark not
exceeded," not "libuv flushed." Reuse-without-corruption is safe only
because session.process is slow enough that libuv drains in between.
Addresses review on PR #637.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Address Miguel's review on #612.
- Normalization std was (1, 1, 1) — that's the base u2net session, not
u2net_human_seg. Switch to ImageNet (0.229, 0.224, 0.225) to match
rembg's U2netHumanSegSession reference. Add a parity test pinning the
exact MEAN/STD values.
- waitForExit treated `code === null` as success, but per Node child_process
docs that's the signal-killed case — a SIGTERM'd ffmpeg encoder was
reporting success with a partial output. Switch to (code, signal) and
reject with the signal in the error message. Add four signal-handling
tests (clean exit, signal-killed, non-zero code, SIGKILL).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Miguel Ángel