AI Now Swaps Brands in Video in Real-Time Based on Who's Watching

Decentralized GPU networks are demonstrating significant cost advantages over traditional cloud providers for AI video processing. **Key Performance Metrics:** - Processing over 1 million minutes of real-time AI video daily - 50-80% cost reduction compared to centralized cloud pricing - Pay-per-use model with automatic scaling The approach leverages globally distributed GPUs that would otherwise remain idle, creating a more efficient resource allocation model. This infrastructure has already proven successful with video transcoding applications. The model represents a practical alternative to services like AWS, offering both cost efficiency and scalability for compute-intensive AI workloads.

Livepeer's ecosystem documentation is now publicly accessible and open for community contributions. Builders can directly edit and improve how the protocol is explained, discovered, and integrated. This follows a documentation restructure initiative launched in September 2025, which aimed to modernize the docs with a stakeholder-focused, AI-first approach. The restructure targeted four key groups: developers, delegators, gateway operators, and orchestrators. The move enables the community to participate in shaping how Livepeer is presented to new users and developers. [View the forum discussion](https://forum.livepeer.org/t/rfp-documentation-restructure/3071/13?u=mehrdad)

Livepeer has distributed $19,000 from its treasury to support four critical open source projects: FFmpeg, OpenCV, ComfyUI, and oapi-codegen. The initiative goes beyond financial support - the ComfyUI maintainer now collaborates directly with Livepeer's AI SPE team on updates and improvements. This move demonstrates how web3 protocols can sustainably fund the open source infrastructure they rely on, creating tighter integration between projects and their dependencies.

The **AI x Open Media Forum** in Buenos Aires brought together 50+ participants for discussions on the future of open media infrastructure. **Key outcomes:** - Active conversations with [Coinbase](https://coinbase.com), [Story Protocol](https://storyprotocol.xyz), [Arweave](https://arweave.org), and XMTP - These discussions have evolved into **ongoing integration talks** led by the Livepeer Foundation - Forum featured symposium-style format with lightning talks and parallel creative/technical tracks **Major themes explored:** - Authenticity and provenance in media - Real-time video workflows - Trust across open media layers - Creative accessibility in high GPU cost regions A notable moment came from creator Franco, who highlighted how affordable compute enables creation in regions where GPU costs are prohibitive - underscoring the importance of open, accessible video AI infrastructure. The event concluded with a 500+ person celebration featuring Refraction x Livepeer, streamed live to Farcaster and Base App. [Read more details](https://forum.livepeer.org/t/rfp-devconnect-assembly/3101/6?u=mehrdad)

Livepeer's AI SPE has shipped production-ready custom workloads, allowing developers to bring their own containers to run custom AI jobs on the network. **Key developments:** - Embody has been running custom workloads for 6 months - Streamplace is preparing to launch soon - PyTrickle enables Python applications to join the network with a single package import This milestone follows the approval of the AI Video SPE Stage 3 proposal, which secured 35,000 LPT in funding. The initiative aims to unlock decentralized media-related AI compute jobs with real-time AI pipelines, batch and generative AI optimizations, and enhanced on-chain metrics. Developers can now leverage Livepeer's infrastructure for custom AI workloads, expanding the platform's capabilities beyond standard video processing. [Read the full retrospective](<https://forum.livepeer.org/t/ai-spe-phase-4-retrospective/3208>)