Your Detections Are State-of-the-Art. Your Overlays Look Like OpenCV.
We render crisp, professional graphics onto your video — using the 2D engine behind Chrome, Android, and Flutter, running on the GPU, inside your pipeline. Broadcast-clean, fast enough for many streams and edge hardware, and source-available so it's never a black box in your product.
From 1 week · from $4,000 · source-available, licensed to ship in your product.
Skia Rendering, in Practice
Our renderer running live on real video — the kind of crisp, broadcast-clean overlays your models could be wearing.
Watch it full-screen on YouTube.
Great Computer Vision, Amateur-Hour Output
It's really two problems — and OpenCV, the tool everyone reaches for, leaves you stuck with both.
It looks like a debugger, not a product
OpenCV was built for debugging, not for anything a customer sees — jagged edges, blocky fonts, no real antialiasing or typography. The instant that video hits a product UI, a dashboard, a broadcast, or an investor demo, it quietly undercuts the models you worked so hard to get right.
The good-looking way won't scale
The usual fix — rendering quality graphics on the CPU, frame by frame — doesn't scale across many streams, and on edge hardware it steals the exact cycles your models need. So teams quietly settle for ugly-but-cheap, because looking good feels like it costs too much.
You shouldn't have to choose between beautiful and fast.
Browser-Grade Graphics, at Pipeline Speed
The look, the speed, and the hardware — so you never have to trade one for another.
The look — Skia
We render with Skia, the same 2D engine behind Google Chrome, Android, and Flutter. Real antialiasing, proper fonts and text shaping, vector paths, gradients, shadows, and clean alpha compositing — the quality people expect from the web and from games, on your video.
The speed — CUDA + OpenGL, on-GPU
We render directly onto the frames already living on the GPU in your pipeline — no copies back to the CPU. It scales across many streams and leaves inference headroom on edge. In practice: 1000+ FPS on an RTX 4060 for a frame with 20 boxes, each carrying 3–4 labels. Bigger cards only go faster.
The hardware — NVIDIA, data center to edge
Optimized for data-center GPUs, professional cards, and edge devices like Jetson — the same hardware your inference already runs on.
You stop choosing between beautiful and fast. You get both.
Anything You Need on the Frame — Done Properly
Whatever your models output, we render it cleanly. A few of the overlays teams ask us for:
And it doesn't stop at boxes and labels. Because we draw real vector graphics, any shape you can save as an SVG — custom markers, brand icons, arrows, zones, dials, whole illustrations — drops straight in and renders dynamically on the frame, antialiased and composited cleanly on the GPU.
If you can design it, we can render it — blazingly fast, on every frame.
Few Teams Sit at This Intersection
GPU graphics, real-time CV pipelines, and the CUDA/OpenGL interop to marry them — from the team behind Savant.
Three Hard Disciplines, One Team
High-quality GPU graphics, real-time CV pipelines, and the CUDA/OpenGL interop to marry them — that combination is rare. We ship it as source-available rendering code you license and integrate into your product, from the team that authors the open-source Savant video-analytics framework.
1000+ FPS on an RTX 4060 · source-available
GPU-Quality Graphics
Skia-grade rendering that runs on the GPU, not the CPU — the visual quality of the web and games, on your video.
Real-Time CV Pipelines
We author Savant, the open-source framework for real-time video analytics — this is our day job.
CUDA / OpenGL Interop
The hard glue that lets rendering and inference share the same frames on the GPU — no costly copies.
Proven in Production
600+ devices on four continents, from NVIDIA Jetson at the edge to data-center GPUs.
Who It's For — and Who It Isn't
This is a rendering upgrade for video people ship — not for throwaway debug views.
This is for you if
- Your video output is seen by customers, users, stakeholders, or the public — product UI, dashboards, broadcast, live events, commercial deliverables, or demos.
- Or you're hitting CPU-rendering limits on multi-stream or edge, and quality rendering is eating the budget your models need.
This probably isn't a fit if
- The overlays are internal debug views no customer ever sees. OpenCV is fine for that — we'll tell you so.
License the Code. Ship It in Your Product. We Help You Integrate.
You license our rendering code — source-available under a Business Source License — and ship it inside your product, with our integration services alongside. Source-available means it's not a black box: your team can read, audit, and adapt it, instead of betting your product on a binary you can't see into.
-
1
Call + demo
We show you the rendering, then map your stack, stream count, and target hardware.
-
2
License
You license the rendering component to ship inside your product.
-
3
Integrate, with us
Our services get it running in your pipeline — DeepStream, GStreamer, or custom. From 1 week · from $4,000.
-
4
Ship
It runs in your product, on hardware from the data center to the edge.
Not on a proper CUDA architecture yet? That's the prerequisite for fast on-GPU rendering — and it's exactly our PoC → Production work. We get the pipeline right first, then the rendering drops in.
Questions Component Buyers Ask Before the Call
The licensing and integration questions we hear most — answered before you book.
Do we get the source?
Yes. The rendering code is source-available under a Business Source License — your team can read, audit, and modify it. It is not a closed binary.
Can we ship it inside our product?
Yes — that is what the license is for. The commercial grant covers integrating and shipping it in your product. We confirm the exact grant terms with you before you commit.
Can we modify it for our pipeline?
Yes. You have the source and can adapt it, and our integration services are there to help you do exactly that.
What do the integration services cover?
Getting the renderer running in your pipeline — DeepStream, GStreamer, or custom — on your target hardware, plus support through integration. We define the exact scope and support terms with you up front.
What does it cost?
Integration starts at 1 week and $4,000, for a pipeline already built on a proper CUDA architecture — frames on the GPU, sound memory management. If your pipeline isn't there yet, we handle that first as a productization step, and the rendering drops in afterward.
See What Your Video Should Look Like
Book a call and we'll show you the demo, then scope the license and integration to your product.
Pairs naturally with a PoC → Production engagement.