Revly

Streamers work hard. They build audiences, create content every day, and pour real energy into their communities — yet when it comes to making money, they're stuck at the bottom of someone else's food chain. Platform revenue shares are razor thin. Brand deals go to the top 1%. Pre-roll ads annoy viewers and hurt retention. And through all of it, the
platforms pocket the majority.

We built Revly because creators deserve better. You built your audience — you should own how you monetize it. We
wanted to give streamers a tool that lets them run real sponsorships, on their own terms, without interrupting their
content or handing control to a platform.

What it does

Revly is a creator-first monetization platform that turns any live stream into a native ad experience — powered by computer vision, running invisibly inside OBS.

Instead of cutting to an ad break, Revly detects real objects in your stream and seamlessly transforms them into
brand placements in real time. Your coffee cup becomes a sponsor's product. The phone on your desk becomes a brand
moment. Your stream stays uninterrupted, your viewers stay engaged, and the revenue goes to you.

For streamers, it means:

Direct brand relationships — connect with sponsors and serve their ads yourself, no platform middleman
Zero content disruption — ads integrate naturally into what's already on screen, so viewers never feel the
interruption
Smart timing — Revly reads your stream in real time, holding ads back during high-energy moments and placing them during natural pauses so they never kill the vibe
Full creative control — manage your ad library, set your own rules, and trigger placements your way via a real-time dashboard or Stream Deck hotkeys
OBS native — no new software to learn, it runs right inside the tool you already use every day

How we built it

We built Revly as a full-stack real-time system that runs entirely on the streamer's machine.

Computer Vision Pipeline (Python):

YOLOv8 + FastSAM detect and segment objects at 30+ FPS on consumer hardware
Kalman filter and EMA smoothing keep bounding boxes rock-stable so ad placements never flicker
Poisson blending, perspective-aware warping, and automatic color adaptation make replacements look genuinely natural

Three deployment modes to fit any stream setup:

Overlay Mode — positions ad images directly over detected objects as OBS sources
Replace Mode — pixel-level object replacement streamed back in real time via MJPEG
AI Mode — buffers the stream, sends segments to Alibaba's Wan 2.7 generative AI for photorealistic product
replacement, then re-injects the edited footage live

Smart Timing Engine:

Audio silence detection, scene change analysis, and motion tracking work together to find the moments ads will be
least intrusive
Configurable cooldowns so you're never over-serving placements

Creator Dashboard (Node.js + Socket.IO):

Real-time ad library management with drag-and-drop image uploads
Preset positioning templates (corner, lower-third, banner, and more)
Live sync across devices — control your stream ads from your phone, tablet, or Stream Deck
HTTP trigger endpoints for full automation

OBS Integration:

Runs natively as an OBS Python script — no separate app, no extra setup
Full source control: positioning, opacity, and timing all managed from inside OBS

Challenges we ran into

Making it look real: The gap between "object detected" and "ad looks natural" is enormous. Getting perspective
warping, blending, and color matching right took the majority of our time
Temporal stability: Raw detection outputs flicker frame-to-frame. Implementing Kalman filters gave us the smooth,
stable placements that make ads feel intentional rather than glitchy
The live delay pipeline: Orchestrating the buffer → AI submission → polling → re-injection loop without dropping
frames or desyncing audio was one of the hardest engineering problems we tackled
Timing that respects the creator: Building a system smart enough to know when not to show an ad required combining
multiple real-time signal sources and careful state management

Accomplishments that we're proud of

A complete real-time ad pipeline running at 30+ FPS on everyday consumer hardware — no server required
Native OBS integration that fits directly into a streamer's existing workflow
An end-to-end AI editing pipeline that takes live footage, runs it through a generative video model, and re-injects it back into a stream in near real time
A timing engine that makes ad moments feel organic rather than forced — something that matters deeply for
creator-audience trust

What we learned

Giving creators control means thinking about every point where that control could be taken away — from platform
dependencies to ad timing to revenue routing
Real-time ML systems live or die by their threading architecture — we learned this the hard way
Generative AI video is a genuinely new creative medium, and we've barely scratched the surface of what's possible
when you apply it to live content

What's next for Revly

Creator marketplace: A direct brand-to-streamer marketplace where creators list their inventory and set their own
rates
Live bidding: Real-time auctions where brands compete for detected object slots during a stream
Impression analytics: Viewability metrics, dwell time, and audience reach data that creators can share directly with sponsors
Multi-stream management: Scale across multiple simultaneous streams from a single dashboard
Custom model training: Let streamers train detection models on their own branded objects and set pieces