Wardeye

Alpha Free · Open source · Chrome and Edge

Place the ward.
See the table.

Wardeye is a free browser extension for people watching Riftbound on Twitch. Point at a card on the table to see its name and image, live or on replay. It works from the video alone, on your computer: no video leaves it.

Free, from the Chrome Web Store, for desktop Chrome and Edge. In Edge, allow extensions from other stores when it asks.

Wardeye on a stream, illustrated A card table seen from above, as a broadcast shows it. Every face-up card is marked at its corners in violet. The pointer rests on one card, and a hover card beside it shows that card's picture, its name and the card lying under it. A badge in the corner reads: Wardeye, 15 cards named. Under it Wardeye · 15 cards named
An illustration, not a screenshot: corner marks on every card Wardeye has named, and the outline and hover card of the one under the pointer, with the card lying under it.

01 · How it works

Three steps. No setup.

  1. 1

    Place the ward

    Add Wardeye to Chrome or Edge from the Chrome Web Store.

  2. 2

    Open a stream or replay

    Any Riftbound broadcast on Twitch, live or on replay. Theatre mode and fullscreen work too.

  3. 3

    Point at a card

    Its image and name appear beside it, with the cards lying under it, such as gear on a unit.

02 · What it does

Built for watching.

Names the table as it is played

Each face-up card on the table is outlined and named as it lands.

A hover card for every card

The card's image and name, how sure Wardeye is, and what lies under it.

Remembers what is covered

A card hidden under another is still known, and listed with the card on top.

Out of the way

Works in theatre mode and fullscreen. Turn it off with its badge, the toolbar button or Alt+R (Option+R on a Mac).

03 · Under the hood

It reads the art, not the text.

On stream a card is 70 to 140 pixels tall at 1080p, too small for its text to be read. So Wardeye recognises the art: it finds each card's corners, turns its picture into a fingerprint and matches it against every printing of every card.

  1. 01FrameThe video, as it plays
  2. 02TableOnly the table camera
  3. 03DetectThe four corners of each face-up card
  4. 04FingerprintThe card's art, as numbers
  5. 05MatchAgainst every printing
  6. 06TrackThe same card, frame after frame
  7. 07OverlayOutlines and the hover card

It runs on your graphics chip with WebGPU, or more slowly on the processor.

04 · Results

Measured, not claimed.

Alpha results, measured on real broadcasts. The reports are public.

Cards found 97.6% Of 5,465 reviewed cards from three broadcasts. The detector was trained on synthetic tables only. The detector report
Named, Barcelona 96.0% A broadcast held out of training. The identifier's results
Named, Los Angeles 99.4% The grand final, a second broadcast held out of training. The identifier's results
Same in the browser 240/240 Frames of the Los Angeles grand final: the extension's own engine read the same cards, names and plays as the reference pipeline. On GitHub

05 · Principles

Private by design. Made for viewers.

Private

  • Runs on your computer. Recognition happens in your browser. No video or frames leave your machine.
  • Public information only. It reads only the table camera: never hand cams, face-down cards or the hand lists a broadcast shows.
  • Nothing kept about you. No account, analytics, ads or history. Card names and images load from Riot's public card gallery as you watch; Wardeye ships none of them.

For viewers

  • Not a play aid. It is made for watching broadcasts: no rules enforcement, no advice, nothing for the players at the table.
  • No statistics about players. No win rates, play rates or ratings. The plays list stays in the tab and goes when it closes.
  • Free for everyone. No paid tier, no ads, no account.

Read the privacy policy

06 · Who makes it

A community project.

Wardeye is made by Federico Vietti, who also makes Gradeon, the AI card pre-grading app. It shares Gradeon's dark look; it is not a Gradeon product.

The code is open source under the AGPL-3.0, and built in the open on GitHub. The trained models ship inside the extension.

Status
Alpha
Price
Free, for everyone
Browsers
Desktop Chrome and Edge
Streams
Twitch, live and replays

07 · FAQ

Questions, answered.

Which browsers does it work in?

Desktop Chrome and Edge, version 137 or newer. A recent graphics chip (WebGPU) makes it fast; without one it runs on the processor, more slowly.

Where do I get it?

From the Chrome Web Store, for free. Edge installs it from the same page: allow extensions from other stores when Edge asks, then click Get.

How much does it cost?

Nothing. There are no ads, no paid tiers and no account.

Does it work on YouTube?

Not yet. Twitch comes first; YouTube is on the roadmap.

Is Wardeye made by Riot Games?

No. Wardeye is a fan project. Riot Games does not endorse or sponsor it: see the notice at the bottom of this page.

Is Wardeye part of Gradeon?

No. It is a community project by the person who makes Gradeon.

Can it see players' hands?

No. It reads only the table camera. Hand cams, face-down cards and the hand lists a broadcast shows are never processed.

Does any video leave my computer?

No. Recognition runs in your browser, and nothing is uploaded. The extension keeps no history.