Duck River | Free Demo & Real Money Betting on Live Ducks (2026)

Duck River Technology: How AI Vision Counts Every Duck

Most online games generate results with a random number generator. Duck River does something else entirely.

The system points a camera at a real river, canal or pond. An AI model trained on waterfowl recognition processes each video frame. When a duck enters a predefined detection zone, the model tags it, assigns it a tracking ID, and increments the count. That count is the result of the round.

What makes this technically difficult is not just the detection part. Ducks move unpredictably: they swim in circles, dive, reappear, overlap with each other. The model needs to handle occlusion (when one duck passes behind another), re-identification (when a duck leaves the frame and comes back), and motion blur from water reflections. 155.io trained their vision engine on thousands of hours of waterway footage across seasons and lighting conditions to handle these edge cases.

The detection grid is visible on the player's screen. Every duck inside the zone gets a bounding box in real time. You can count along with the AI. If a duck enters the zone at second 48 and leaves at second 53, you see it get tagged and added to the tally. There is no hidden logic.

So instead of a hash-based seed determining your outcome, actual animals on an actual body of water determine it. The AI is just the referee.

155.io published a technical whitepaper covering the model architecture: it runs a modified YOLO-based detector optimized for low-latency inference on edge hardware co-located near each camera site. Frame processing happens locally, not in a centralized cloud, which keeps the detection time under 40 milliseconds per frame.

Dolby Millicast in Duck River: Sub-Second Streaming Explained

Streaming a live camera feed is simple. Streaming it with under 500 milliseconds of delay is not.

Duck River uses Dolby Millicast (formerly Millicast, acquired by Dolby in 2021) for real-time video delivery. Standard live streaming platforms (Twitch, YouTube Live) operate with 3-10 seconds of latency. That gap is fine for entertainment but terrible for a game where the timer counts down from 55 and every second matters. Dolby Millicast uses WebRTC-based delivery, which pushes the video from the camera to your browser in under half a second.

Why does this matter for the player?

When you watch a duck enter the detection zone on your screen, that event happened less than 500 milliseconds ago on the actual waterway. The feed is as close to real-time as current streaming technology allows. Two players watching the same camera from different continents see the same duck enter the zone within milliseconds of each other. That simultaneity is what makes Duck River work as a multiplayer prediction game. Everyone watches the same reality at the same time.

Standard RTMP-based streaming would add a 4-8 second buffer. In that window, a camera could capture 3 or 4 additional ducks, which means the player's view would be out of sync with the actual count. Dolby Millicast eliminates that problem. The player sees what the AI sees, when the AI sees it.

The adaptive bitrate is worth mentioning. If your connection dips (moving between Wi-Fi and mobile data, for example), Millicast drops the video resolution before it drops frames. You might see a slightly grainier picture for a few seconds, but the feed never freezes and the latency stays consistent. That trade-off keeps the game playable even on inconsistent networks.

Provably Fair: How Duck River Records Every Result On-Chain

"Provably fair" gets thrown around a lot in online gambling. Most of the time, it means a game uses a hash-based seed system: the server generates a seed before the round, hashes it, and reveals the original seed after the round so players can verify. It works, but it still relies on the operator's server generating the seed honestly.

Duck River takes a different path. The result is not generated by a server at all. It comes from the real world (actual ducks counted by AI on a live feed). The blockchain layer sits on top of this as a record-keeping mechanism.

Here is how the verification chain works:

1. The round starts. The camera feed goes live. The AI begins counting.
2. At the 55-second mark, the timer expires. The AI locks the final count.
3. That count, along with the camera ID, timestamp and a snapshot hash, gets written to the blockchain.
4. The transaction is immutable. Nobody (not 155.io, not the casino, not anyone) can alter the recorded result after it posts.

Players who want to verify a specific round can look up the transaction using the round ID provided in the game interface. The blockchain entry shows the final duck count, the exact timestamp, and the camera location.

The system does not use a custom chain. 155.io opted for an established public blockchain for transparency. Running verification on a private chain would defeat the purpose, because the operator could still theoretically control the chain itself.

One detail that often goes unnoticed: the snapshot hash. At the moment the round ends, the system captures a frame from the video feed and hashes it. That hash gets recorded alongside the count. If a dispute ever arises, the original frame can be compared against its hash to prove the AI's count matched what the camera actually showed at that instant.

10 Live Cameras Across 3 Continents: Duck River's Global Network

Duck River operates cameras across Europe, North America, Asia and Australia. The feeds rotate to maintain 24/7 coverage across time zones.

Each camera is paired with a local edge compute unit that runs the AI inference model. The detection parameters are tuned per site: the Amsterdam canal camera processes wider frames because ducks tend to spread out across the water, while the Kyoto stream camera uses a narrower detection zone since the waterway itself is smaller. Sydney's unit accounts for different waterfowl species (black swans appear occasionally, and the model correctly excludes them from the duck count).

The rotation schedule ensures that at least 2-3 cameras are active at any given hour. During European daytime (when Amsterdam, Paris, London, Bruges, Copenhagen and Prague are in good light), the system has the widest selection. Overnight in Europe, the feeds shift to Sydney, Vancouver and New York.

Duck River RTP Breakdown: Payout Rates from 91.5% to 93.5%

RTP stands for Return to Player. It is the percentage of total wagers that a game returns to players over a large number of rounds. The key phrase is "large number." Over 50 rounds, your actual return can be anywhere from 0% to 300%. Over 50,000 rounds, it converges toward the published RTP.

Duck River has four RTPs because it has four bet types:

The house edge is the inverse of RTP. On Range bets, the casino keeps 6.5 cents per dollar wagered on average. On Exact bets, it keeps 8.5 cents. The difference reflects the risk profile: Exact bets require a larger margin because the x18 multiplier concentrates payout variance.

How does this compare to other formats? Crash games like Aviator or Spaceman run at 96-97% RTP. Blackjack with correct strategy sits above 99%. Standard video slots hover between 93% and 96%. Duck River's range of 91.5-93.5% falls slightly below software-based crash games.

That gap has a technical explanation. Running 10 live cameras around the world with sub-second streaming, processing video through AI models at each site, and recording every result to a blockchain is more expensive than generating a random number from a server seed. The infrastructure cost gets built into the house edge.

For the player, the trade-off is concrete: you accept a slightly higher house edge in exchange for results that come from the physical world rather than from software. Whether that trade-off is worth it depends on how much you value transparency and provability over raw RTP percentage.

One thing to keep in mind: the RTP is identical across all casinos that carry Duck River. 155.io sets the parameters at the game level. Stake, Shuffle, Roobet (and others via Hub88) all run the same backend, so the odds do not change depending on where you play.

155.io Studio: The Developer Behind Duck River's Live Game Engine

155.io is a Swedish studio that started in 2024 with a simple question: what if game results came from the real world instead of software?

Their first release was CCTV Rush Hour, a game that used live traffic cameras to generate betting outcomes based on vehicle counts. The concept proved that real-time computer vision could replace traditional RNG in a production gambling environment. It also surfaced the hard engineering problems: latency synchronization, edge AI inference, camera reliability during weather events, and on-chain verification at scale.

Duck River is their second title. Same core architecture, applied to waterways and waterfowl instead of roads and cars. The AI model is different (trained specifically on duck morphology and movement patterns), but the streaming pipeline and blockchain layer carry over from CCTV Rush Hour.

The studio runs a lean team. Most of their engineering investment goes into three areas:

• Computer vision model training and edge deployment
• Streaming infrastructure (Dolby Millicast integration, camera management)
• On-chain result recording and verification tooling

Distribution goes through Hub88, which connects 155.io's games to licensed casino platforms globally. The casinos handle the player accounts, payments and compliance. 155.io handles the game engine, the cameras and the AI.

Upcoming titles include Ducks.io (plastic duck racing on real water channels), Marbles and Stairpong. All follow the same principle: physical events, filmed live, verified on-chain. No RNG anywhere in the pipeline.

Duck River launched on March 24, 2026. It went live simultaneously on Stake, Shuffle and Roobet. It is also available on 1win and PIN-UP.

Duck River FAQ: Technical & Gameplay Questions Answered