Vercila is a real-time, on-device engine for live facial de-aging. It's the same core that powers the FaceReady desktop app, offered for licensing as an embeddable SDK. Here's how it works, where it runs, and where it sits in the market.
Vercila takes a video frame containing a face and returns the same frame with the signs of age eased back — under-eye shadow, the lines a lens over-sharpens, the flatness that reads as tired — while leaving the person's features, expression and real skin texture intact.
The engine uses a residual technique: rather than generating a new, smoothed face and compositing it over the original, it computes a small correction and applies only that. The subject's own pores and micro-detail remain in the image, and the result tracks expression and motion exactly, because it's the original frame minus the camera's harshness — not a synthetic replacement. That's the core difference from regeneration-based beauty filters, which tend to look waxy.
De-age is the lead capability. The same engine exposes complementary real-time adjustments — relight and warmth — as independent parameters, so a licensee can use de-age alone or combine it with subtle relighting in one pass over the frame.
Effect strength is adjustable from subtle to stronger via a single normalised parameter — the host application decides how much, and exposes it however suits its own UI.
The correction is produced by a neural network trained for this single task, executed on the device's own machine-learning hardware so it keeps pace with live video. It runs natively on both Windows and macOS, from one core behind a common interface — a single codebase, not two — so behaviour is identical across platforms.
The engine is compiled ahead-of-time to native code (NativeAOT). There's no just-in-time compilation and no managed runtime to ship or warm up: the host loads a native library and calls it directly. That keeps startup immediate, memory predictable, and the integration footprint small — which matters when the engine sits inside another vendor's application or a system virtual-camera component.
The engine integrates through a native C-ABI — initialise, process a frame, release — callable from C, C++, or any language with native interop. It's been validated end-to-end inside a desktop virtual-camera pipeline, taking frames from a live webcam and returning de-aged frames for the system camera feed, including the colour-format handling such pipelines require.
Detailed per-frame timings, the throughput approach, and tuning guidance are shared under evaluation agreement rather than published — they're in the technical overview.
The face-effects SDK market runs on-device and prices by monthly active users. It has two poles: an opaque, sales-led premium end, and a transparent, self-serve commodity end built for AR decoration. Vercila is positioned deliberately between them — transparently priced, but specialised.
| Dimension | Broad AR-filter SDKs | Vercila |
|---|---|---|
| Primary job | Masks, makeup, novelty AR | Live, natural de-aging |
| Strength | Breadth of effects | One thing, done convincingly |
| Desktop | Mobile-led; Windows often absent | Windows + macOS today |
| Delivery | Per-app SDK embed | Embed, or virtual camera across any app |
| Pricing | Opaque (premium) or commodity | Transparent, specialised |
For a video, conferencing, streaming or camera-software vendor, the practical edge is twofold: Vercila leads on a capability the decoration-focused SDKs treat as a side feature, and it delivers on desktop — including through a virtual camera that works in any conferencing app with no per-app integration.
Because inference runs locally, there's no server in the loop: nothing is uploaded, nothing is transmitted, nothing is retained. For any vendor shipping a face-processing capability into regulated, enterprise or privacy-sensitive environments, this removes an entire class of data-handling and compliance exposure — the video simply never leaves the user's machine. Even the broad face-AR leaders run their core engines on-device; "cloud," where it appears in this market, is effect delivery, not processing. Vercila has no cloud-inference path by design.
The technical overview covers integration, performance and per-MAU licensing tiers. Get it, or start an evaluation against your own footage.