// real-time generative AI cerati.org

JOE
CERATI

Senior Engineer @ Daydream. I build real-time AI video & audio on the Livepeer network — diffusion models you can play like instruments. Based in Milan.

live ring buffer

// now Lately I've been working on real-time diffusion — video and audio generated frame by frame, fast enough to respond the moment you do.

Open source · real-time video

Scope & Daydream API

Run & customize real-time, interactive generative-AI video pipelines.

Autoregressive video diffusion you can drive live — StreamDiffusionV2, LongLive, Krea Realtime, MemFlow and more — composed into one low-latency pipeline with WebRTC streaming, a timeline editor, plugins, LoRAs and VACE control. The toolkit behind Daydream's real-time AI video app.

+292,147−91,629lines of code

Open source · streaming diffusion · audio

DEMON

Diffusion Engine for Musical Orchestrated Noise — StreamDiffusion, for audio.

A ring-buffer engine for continuous music generation on ACE-Step. Several generations stay in flight at different denoise stages and advance together each tick, so shared-mutable latent curves map your control inputs into the solver the millisecond your hand moves — generative audio played like an instrument. Streaming output is bit-identical to batch.

25 Hzcontrol resolution
<100 msend-to-end latency
1 GPUsingle consumer card

+24,857−1,997lines of code

Open source · decentralized video infrastructure

Livepeer

The open network for video — and the rails under everything above.

Four-plus years here: from Livepeer Studio's APIs, SDKs and interfaces that put open video infrastructure in builders' hands, to leading Daydream — the first real-time AI video app built on the Livepeer network.

+98,734−35,121lines of code

Open source

Public activity on github.com/gioelecerati · 2022 → today

3,256contributions since 2022
458pull requests opened
376PRs merged
23public repositories
2022
2023
2024
2025
2026
less more
// stack
PythonTypeScriptRustGoJavaScriptC++SolidityShell PyTorchTensorRTCUDADiffusersFastAPINext.jsReactWebRTCFFmpegJUCEDjangoEthereum

The path here

2017 → 2022

Tangram — Co-founder & CTO

Co-founded a platform for distributed video transcoding on Ethereum. Rust transcoding clients and microservices, full-stack.

2020 → 2022

HiWay Media — Software Engineer

OTT and digital-asset-management platforms for sport live-streaming. Rust microservices, product and customer work.

2022 → now

Livepeer Studio — Full-Stack Video Engineer

Built the APIs, SDKs and interfaces empowering video builders with the Livepeer network. Contributed across Livepeer's fully open video infrastructure.

2025 → now

Daydream — Senior Engineer

Building open-source tools for real-time AI creation: Scope for video, DEMON for audio. Leading the first real-time AI video app on the network.

Studied History & Philosophy, then Piano & Composition, in Milan. The thread from mining blocks to streaming latents has been the same all along: real-time, decentralized, open.

// in case you need to reach out