Darkgrade

Darkgrade Link attacks a real and unglamorous problem. PTP has been the shared language of cameras for roughly 25 years, but the practical work of driving a body from software — property discovery, vendor-specific quirks, event plumbing, live view decoding — is usually buried in per-brand SDKs or reverse-engineered wrappers. Link replaces that with one type-safe API: copy `npm i @darkgrade/link`, connect, and you're setting ISO, shutter, and aperture, streaming live view, starting and stopping recording, subscribing to camera events, and pulling files off the body. It runs in Node.js with usb as its only major dependency, and in the browser via WebUSB, as a roughly 55kB tree-shakable bundle in pure TypeScript. Automatic camera detection with zero configuration is the feature that saves the most setup time in practice — you don't hand-write a device-matching layer before you can take a single frame. The strengths are ergonomic and structural. The bundle is small enough to drop into a browser app; the codebase is readable MIT-licensed TypeScript; and because Link is the foundation layer of a broader product, it is being exercised against real hardware rather than shipped as a demo. The vendor publishes per-brand compatibility tables covering generic, Canon, Sony, and Nikon properties, operations, events, responses, and formats, plus datasets for device info, device property descriptions, object info, storage info, and variable values. If you are an AI engineer wiring programmatic image acquisition into a vision or LLM pipeline, or a studio/robotics automation specialist who needs a camera inside a software system, that reference surface is the difference between a week of protocol archaeology and an afternoon of coding. The weaknesses are maturity and scope, and you should go in with eyes open. The docs state plainly that Darkgrade is in alpha and that APIs may change frequently without backwards compatibility — pin your dependency and budget for migration. Feature coverage is uneven by brand; the compatibility tables exist precisely because a Sony body and a Canon body do not expose identical operations. You need a PTP-capable camera plus a USB or WebUSB connection, and you must write TypeScript against the API — nothing here is a click-to-run app. The Studio stage that would deliver an automatic first edit is still in development and Create is a stated long-term goal, so do not plan a capture-to-finished-edit pipeline around roadmap items. Data and computation stay local by default, and the project states it does not train on your footage, which matters if you're shooting client work. Where it fits: teams who treat cameras as software peripherals and are comfortable with alpha-stage tooling. Where it doesn't: production shops that need backward-compatibility guarantees and a supported vendor relationship.