Generative AI video, automated editing, scene detection, AI dubbing, voice cloning, and upscaling are moving from research demos into production pipelines. Get the daily rundown on which AI video capabilities are actually shipping, and which ones are still just announcements.
OpenAI announced it is discontinuing its Sora AI video app and API, stating the move is to refocus efforts and compute resources on robotics and world simulation research. According to the article, the decision follows a period of waning popularity for the app, which launched in February 2024 and had previously entered into a licensing deal with Disney. OpenAI confirmed that its ChatGPT AI image generator will not be affected.
A new report from Sportradar posits that the sports viewing experience itself has become the primary product, requiring broadcasters and platforms to integrate data and personalization. The study outlines five pillars shaping modern fan engagement, including using real-time data for narrative context, increased interactivity, and leveraging AI to accelerate live production workflows. The shift is driven by audiences moving from linear broadcasts to fragmented streaming and social media consumption.
IBM has launched the Sports Tech Startup Challenge, a year-long global program to identify AI-driven startups in the sports industry. The initiative includes regional showcases and an invite-only competition where finalists can pitch for a potential paid proof-of-concept valued at up to $100,000. The program aims to connect founders with IBM's sports partnerships and technology for use cases like performance analytics, fan engagement, and platform monetization.
Experts from Arm, Cadence, Synopsys, and other semiconductor IP firms discussed the challenge of designing edge AI hardware that can adapt to AI models evolving faster than silicon design cycles. The roundtable highlighted the trade-offs between flexibility and power/performance/area, the critical role of software compiler toolchains, and the architectural implications of emerging agentic AI workloads. Participants noted that application lifecycles, from disposable cameras to long-lifespan automotive systems, heavily influence the required adaptability of the underlying chips.
Meta's independent Oversight Board has overturned the company's decision on a viral, AI-generated fake war video, stating it should have been labeled. The Board intervened after the video amassed over 700,000 views, criticizing Meta for failing to apply a warning label. It has recommended that Meta create a dedicated policy for AI-generated content, improve automated detection tools, and be more consistent with watermarking.
Particle6, a company founded by Eline Van der Velden, has released a music video starring its AI-generated "actor," Tilly Norwood. The video, produced using AI tools including the music platform Suno, frames AI as a creative partner in response to industry backlash and debate over AI's role in entertainment. The concept continues to draw criticism from actors and unions over concerns about labor, consent, and creative ownership.
D-ID has launched 'Agentic Videos,' a new capability that transforms passive videos into interactive experiences by adding a conversational AI agent. The feature allows viewers to ask questions to a visual avatar that understands the video's script and context, responding in real-time with sub-second latency. The capability is integrated into the simpleshow platform for enterprise use cases such as interactive marketing and corporate training, and includes analytics on user interactions.
YouTube is expanding its AI-powered "likeness protection tool" to the entertainment industry, allowing actors and musicians to detect and request the removal of deepfakes. The free tool, which was previously offered to government officials and political candidates, identifies AI-generated content using a person's likeness. The move is a response to the proliferation of hyper-realistic AI videos and follows complaints about the platform's process for flagging fabricated content.
London-based AI translation startup Palabra AI has launched a new "streaming-native" text-to-speech (TTS) engine and appointed Andrey Feldman as its new CTO. The new engine is said to support 8 languages with a time-to-first-answer of approximately 100ms and includes voice cloning capabilities from a 6-second audio sample. Feldman joins the company to help scale its real-time systems for enterprise clients.
AnveVoice has launched high-accuracy Catalan speech recognition (ASR) with support for dialects and real-time streaming, targeting 4 million native speakers. The system offers production-grade voice AI for various use cases, including voice search and dictation, and is available with a free tier. It provides solutions for handling Catalan accents, background noise, and includes features like custom vocabulary support and speaker diarization.
Google Cloud introduced the Gemini Enterprise Agent Platform, a new suite for building and managing AI agents, and the Gemini Enterprise app, which enables no-code creation of AI agents for business workflows. These developments were announced alongside new eighth-generation TPUs (8t and 8i) designed for AI workloads, and the Virgo Network for high-speed data transfer. The offerings are intended to facilitate the deployment and scaling of AI applications across various industries.
Apple Music's VP of Apple Music and international content Oliver Schusser revealed that over one-third of monthly uploads to the platform are 100% AI-generated music, yet AI music accounts for less than 0.5% of total usage. Schusser also stated that Apple Music has developed its own technology for detecting AI-generated music during upload and called for industry consensus on defining AI music.
At NAB 2026, EVS unveiled AI-driven upgrades across its production portfolio, including an AI deblur feature for its Xeebra officiating platform and cinematic effects for XtraMotion super slow-motion. Its LSM-VIA system now uses AI object recognition for player tracking and automated reframing into vertical formats. The company also launched Choreon, a unified orchestration layer for its new T-Motion robotics brand.
Union Square Ventures announced Glif v2, describing it as a creative orchestration agent for AI. The tool distills learnings from millions of community-created AI workflows to manage production pipelines, including selecting models for tasks like image sequencing, video, and audio. As a user interacts with it, Glif v2 learns their specific aesthetic and processes to become a personalized creative partner.
An AWS blog post recaps a podcast with partners Thoughtworks, TwelveLabs, and LucidLink on media and entertainment trends. The discussion covered the use of AI for semantic content understanding and automated editing (TwelveLabs), the shift to iterative, cloud-based workflows enabling remote collaboration (LucidLink), and the need for platform modernization to deliver personalized experiences (Thoughtworks). The common themes identified were the need for speed, personalization, and efficient cloud-based workflows to manage rising content volumes.
Chryon has launched PRIME Translate, an AI-powered solution for simultaneous multi-language content production. The product uses NVIDIA technology to orchestrate AI audio translation, graphical text overlays, and realistic facial movement generation from a single source. According to the company, the solution is designed to help live media, news, and sports organizations transition from multi-production models to software-driven localization at scale.
Wowza has launched the Wowza Video Intelligence Framework, an AI framework designed to run alongside Wowza Streaming Engine to apply AI inference within live sports streaming workflows. The framework extracts frames from live streams, routes them to AI models, and outputs real-time metadata, clips, alerts, webhooks, and machine-readable event signals for downstream systems while streams are still live. It also supports real-time detection of quality issues (e.g., degraded image quality, obstructed lenses, misaligned feeds) and allows customers to bring their own AI models and customize detection logic.
Chyron launched PRIME Translate, an AI-assisted live production workflow aimed at enabling regional and local broadcasters to generate multiple language versions of the same live content simultaneously, including translated audio, facial movement, and graphics overlays. The product is built to work with NVIDIA technology (including Holoscan for Media and NVIDIA AI for Media’s Content Localization Blueprint) to support scalable localization and create additional localized outputs that can be monetized via expanded ad inventory.
Speechmatics and Adobe have collaborated to introduce cloud-grade speech recognition directly on-device within Adobe Premiere. This new feature enables private, offline, and accurate speech-to-text capabilities for professional creators using Premiere.
YouTube has expanded its AI-powered likeness detection technology to identify AI-generated deepfakes of celebrities. The system, which functions similarly to Content ID, allows rights owners to request either removal of the content or a share of its revenue. The company also noted plans for future audio detection capabilities and its advocacy for federal legislation like the NO FAKES Act.
Cloudflare announced a series of new product launches during its Agents Week 2026, focusing on compute, security, and tools for the 'agentic cloud' and 'agentic web.' These innovations aim to enhance its offerings in areas related to Artificial Intelligence agents and platform capabilities.
Cloudflare has built its internal AI engineering stack using its own products, processing 241 billion tokens and routing 20 million requests through AI Gateway. This internal AI infrastructure serves over 3,683 users using Workers AI for inference. The company highlights the use of its own platform for its AI development.
Music streaming platform Deezer reports that AI-generated tracks now constitute 44% of its daily uploads, equating to nearly 75,000 tracks per day. The company states its AI detection tool tags this content, which accounts for 1-3% of total streams, and that 85% of those streams are demonetized as fraudulent. Deezer is removing AI content from recommendations and is now licensing its detection technology to other companies.
iQiyi said it expects AI to create the majority of its films and shows within five years and is restructuring its core app and website toward a more social-media-like destination centered on AI-generated video. The company debuted its Nadou Pro filmmaking toolkit, announced a new social-video app, plans an international version using models including Google Veo 3.1, and said it will incentivize AI creators with an additional 20% share of advertising and membership revenue. The shift is positioned as a response to competitive pressure from short-video platforms and follows an estimated 13% year-over-year revenue decline in Q1, alongside plans for a Hong Kong listing.
Quickplay announced AI-oriented product updates and partner news for NAB 2026, including Social Signals within Quickplay AI Studio, Smart Verticalizer, and a new partnership with Visible Things. The company highlighted deployments such as consolidating 1,300 digital touchpoints for Gray Media while managing 269 live and 123 FAST channels, and completing a 12-month cloud-native transformation of TVNZ+ serving over 2 million daily viewers.
The global AI video generator market is projected to reach $847 million in 2026, growing at an 18.8% CAGR, with a significant increase in enterprise adoption across various industries. This growth is driven by substantial cost reductions and accelerated content creation, impacting traditional video production and marketing strategies. Venture capital also made significant investments into AI video companies like Synthesia, Runway, and HeyGen in the past 18 months.
Samsung Electronics showcased its 2026 AI-powered TV, display, and audio innovations at its European Tech Seminar in Frankfurt, Germany. Key features include Vision AI Companion for intuitive interaction, AI Upscaling Pro, AI Soccer Mode Pro, and AI Sound Controller Pro for optimizing viewing experiences. The company plans to expand AI capabilities across its entire 2026 TV lineup.
The article discusses the crucial metrics for speech-to-text latency in voice agents, focusing on Time-To-First-Speech (TTFS) and accuracy. It highlights how Pipecat's benchmarks are influencing the conversation around these metrics.
Sonilo has launched its AI-powered music generation service as a new Node within ComfyUI, a platform for AI model workflows. This integration allows ComfyUI users to automatically generate custom background music for videos that matches pacing, timing, and emotional tone. The service offers both video-to-music and text-to-music capabilities, aiming to streamline music creation for video creators.
Gracenote research on TV and movie search and discovery reports growing use of AI chatbots for content recommendations, with 66% of respondents saying their usage increased over the past 12–18 months and Gen Alpha at 80%. In the survey, 49% of Gen Alpha chose AI chatbots as the best source for recommendations versus 41% for streaming/cable UIs and guides, while respondents still rated traditional search higher for trustworthiness and accuracy. The report is based on an online survey of 4,003 US AI chatbot users ages 13–79 conducted from 23 January to 4 February 2026.
Fastly and LALIGA announced a joint anti-piracy innovation effort to detect and help remove illegal live streams of LALIGA matches, citing LALIGA estimates that piracy costs its clubs $700–$800 million annually. Fastly says it has built a real-time detection system using AI and proprietary content signals to identify unauthorized streams and enable more precise takedowns without broad regional blocking. The release references industry data (Grant Thornton) indicating 10.8 million unauthorized live-event retransmissions were detected in 2024, with most not suspended and only 2.7% addressed within the first 30 minutes.
INSAIT Institute and Netflix announced VOID, an AI video model designed to remove objects from video while realistically reconstructing the scene’s subsequent dynamics (e.g., how objects would move if a person were removed). The model is built on CogVideoX and uses a “quadmask” method plus simulated training data generated in Blender, and it is released as open source with code, paper, and demos publicly available.
Alibaba confirmed it is behind HappyHorse-1.0, an AI video generation model that debuted anonymously on the Artificial Analysis benchmarking platform and quickly ranked at the top of blind tests for text-to-video and image-to-video. The project is attributed to Alibaba’s ATH AI Innovation Unit and is described as still under development, amid heightened competition in generative video as rivals face setbacks such as OpenAI discontinuing Sora and ByteDance pausing Seedance 2.0 amid copyright disputes involving major studios and streaming platforms.
Synamedia announced a just-in-time AI plugin framework for its Quortex portfolio, positioning it as a way to apply AI only when specific workflow conditions in video/audio streams warrant it for processing, distribution, and delivery. The company says the system triggers AI based on detected meaningful changes or events rather than continuous, always-on processing. The capabilities are framed as applicable across Quortex Link, Quortex PowerVu, Quortex Play, and Quortex Switch.
Amagi is launching NewsPulse, an agentic AI platform designed to automatically convert live broadcast feeds and VoD news libraries into social-ready clips, vertical video, and digital news bulletins. The platform covers the workflow from ingest through real-time story segmentation, intelligent multi-aspect reframing, caption/post generation, and publishing to digital endpoints, with optional human review and policy guardrails. NewsPulse is in limited availability testing with select newsroom partners, with general availability expected in June 2026.
A NAB Show session description outlines how AI is being applied to live sports streaming workflows to improve monetization and viewer experiences. Topics include AI-based detection of natural ad breaks with real-time SCTE-35 marker generation, contextual metadata extraction for highlights and personalized replays, and localization automation (speech-to-text, translation, voice cloning/overdubbing). The session also cites AI-driven super-resolution upscaling from HD to UHD as a way to deliver higher quality while reducing costs.
Harlequin (a HarperCollins division) signed a multi-year agreement with AI shortform-video company Dashverse to co-produce 40 animated, AI-assisted "microdrama" series based on Harlequin Romance titles, starting with an adaptation of Catherine Mann’s "A Fairy-Tail Ending". The series are slated for distribution in English on global microdrama platforms including Dashverse’s DashReels, with monetization via advertising and, on some platforms, subscriptions; Harlequin says authors will receive royalties. Dashverse’s production system, Frameo, is described as a generative video studio combining text prompting with automated editing to produce short films with consistent characters and audio.
Bouygues Telecom’s InnoLab Tech, working with AWS and Devoteam, developed “Ensemble ce soir,” a generative-AI TV service designed to help households choose content from SVOD catalogs by collecting preferences via set-top box prompts and a smartphone chat initiated with a QR code. The system uses Claude 3 (Haiku) through Amazon Bedrock with RAG and publisher catalogs to generate three explained recommendations, with an architecture built around an Android app, web app, API Gateway, and AWS Lambda. The project progressed from a prototype shown at VivaTech (May 2024) to a product, followed by internal testing and a customer test phase in September 2024.
AWS outlines its NAB Show 2026 (Las Vegas, April 18–22) demonstrations and sessions focused on agentic AI for media creation, distribution, and monetization, including workflow automation and archive understanding. A featured capability is AWS Elemental Inference, described as a fully managed service that applies AI in parallel with live encoding to generate 16:9-to-vertical video in 6–10 seconds and integrate with AWS Elemental MediaLive and MediaConvert, with Fox Sports Digital and NBCUniversal cited as users. The agenda also highlights cloud/AI use cases in live sports production and fan engagement with partners and customers including Fox Corporation, NVIDIA, and the PGA TOUR.
Netflix describes a production-oriented multimodal video search architecture designed to index and retrieve moments from large volumes of raw footage by fusing outputs from multiple AI models (e.g., character, scene, dialogue) into a unified, time-aligned representation. The system persists raw annotations in an internal annotation service backed by Apache Cassandra, performs offline temporal bucketing and intersection via Kafka-triggered processing, and indexes enriched per-second records into Elasticsearch to enable low-latency hybrid text+vector search and ranking. The post also outlines query planning, result deduplication/clustering, and future work including natural-language querying, adaptive ranking from user feedback, and workflow-specific personalization.
EE TV has launched two AI-powered conversational discovery features, Smart Search and Mood Matcher, within the EE TV app to help users find content using natural-language prompts and mood-based questions. The tools aggregate results across live TV, on-demand libraries, and streaming apps available through EE TV, with plans to expand access later to EE TV Pro and EE TV Box Edge customers.
AWS outlines approaches for adding live-stream localization and accessibility—captions, multi-language subtitles, and live dubbing—using AWS Media Services and AI-based speech recognition/translation. The post compares four integration points across a typical AWS live pipeline (MediaConnect/MediaLive/MediaPackage plus CloudFront), highlighting trade-offs in latency, language support, and client changes, and recommends integration points 2 (embedded captions) and 3 (late-binding subtitle/audio tracks into HLS manifests). It also references an implementation using partner SyncWords and standards such as CEA-608/708, WebVTT, DVB Subtitle/Teletext, SRT, and HLS.
PA TV Metadata has launched EMMET, a human-in-control, AI-supported metadata enrichment tool aimed at improving content discovery, engagement, and personalisation across broadcast and streaming services. The tool adds scalable descriptive and contextual metadata (e.g., mood, themes, soundtrack details, and filming locations) and is integrated into PA TV Metadata’s Pawa management and distribution system for use on broadcaster and content-owner catalogues.
Imagine Communications has introduced AI-assisted scheduling for its Landmark Rights & Scheduling (R&S) system. This new feature aims to streamline media scheduling workflows, reduce manual effort, and improve consistency for planning decisions.
Runway is launching a $10 million venture fund to invest in early-stage startups building in AI, media, and world simulation, with plans to write checks up to $500,000. The company is also rolling out a Builders program that provides eligible startups with 500,000 API credits and access to Runway Characters, a real-time video agent API built on its general world models, aiming to expand use cases for interactive, real-time “video intelligence.”
RTL Deutschland outlined a strategy to expand “agentic AI” use across post-production, virtual production, and AI-generated commercial breaks for both streaming and linear TV workflows. Planned applications include AI-assisted analysis and structuring of reality-format footage and producing alternative narrative edits of previously broadcast episodes (with legal and editorial review). RTL also described pilots under real production conditions, including the use of virtually enhanced backgrounds in multiple series.
OpenAI shut down its AI video-generation tool, Sora, just six months after its public release, due to high operational costs and low user engagement, according to a WSJ investigation. Sora was reportedly burning through $1 million daily and its user count had declined significantly. The decision also led to the termination of a $1 billion partnership with Disney.
The article reports that OpenAI has discontinued its Sora video generation model, citing high compute costs and intense market competition. It frames the decision as a strategic shift toward prioritizing product viability and monetization over continued investment in a costly, competitive generative video effort.
This article provides an overview of the current state of AI in live streaming, detailing its applications in production, localization, quality measurement, content creation, security, and monetization. It highlights areas where AI is robustly integrated, such as automated camera operation, captioning, and ad optimization, contrasting them with latency-sensitive areas like encoding and delivery where AI's role is more diagnostic.
Wowza describes how the Mississippi Department of Transportation uses Wowza Streaming Engine as an API-driven platform to provision, monitor, and deliver over 1,200 RTSP traffic camera feeds, including automated health checks and stream management via a custom "Camera Manager" system. MDOT generates near-real-time still thumbnails from camera streams using Wowza Transcoder and sends frames to a vision model that answers incident-related prompts (e.g., crash, stalled vehicle, emergency responders), creating a prioritized operator review queue. Wowza also notes it is running a beta program with clients to add AI-powered video intelligence/scene analysis capabilities to existing Wowza-based live video workflows.
