AI Video Intel — 5/25/2026

Top stories, ranked by relevance.

Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.

#1Gemini Omni Flash Rolls Out Free on YouTube Shorts This Week

Google's new multimodal video model, announced at I/O on May 19, is now rolling out at zero cost inside YouTube Shorts and the YouTube Create app — putting AI video generation in front of 2.7 billion monthly active users. Omni Flash takes text, images, audio, and video as input and generates 10-second clips with conversational editing ("make the lighting warmer," "remove the person in the background"). All output carries SynthID watermarks. This is the broadest consumer rollout of any AI video model to date.

🔗 TechCrunch

#2Runway Drops Gen-4.5 — Now Available on All Paid Plans

Released May 21, Runway's Gen-4.5 claims the #1 spot on the Artificial Analysis Text-to-Video leaderboard with 1,247 Elo. It supports text-to-video and image-to-video at 2–10 seconds with markedly improved physics, human motion, and prompt fidelity. Starting at $12/month on the Standard plan, Gen-4.5 brings all existing control modes (keyframes, video-to-video) forward while maintaining Gen-4 generation speeds.

🔗 AI Business

#3Hailuo 2.3 Ships Four Variants — Fast Cuts Cost 50%

MiniMax's Hailuo 2.3 landed May 7 with a significant update on May 16, offering Standard, Pro, Fast, and Fast Pro variants. The Fast tiers deliver up to 50% cost reduction for image-to-video, while Standard/Pro support both text-to-video and image-to-video with improved character performance and motion smoothness. Output specs range from 768p/6s (25 credits) to 1080p/6s (50 credits). Strong anime and realistic character emotion are the standout improvements.

🔗 Artlist Blog

No image

#4Runway Agent: Conversational Video Production Goes Live

Launched May 13, Runway Agent is an AI creative partner that lets you describe what you need, refine direction through conversation, and generate a finished video — all in one flow. This moves Runway from a tool-per-task model toward an end-to-end production assistant, collapsing prompt engineering, iteration, and export into a single conversational interface.

🔗 Runway Changelog

#5Google Flow Gets Android Beta and Omni Flash Integration

Also from I/O, Google Flow — the AI creative studio — now includes Gemini Omni Flash for video generation and editing, custom tools via Flow Tools, and AI agents for multi-step production. The Flow app launched in beta on Android (18+), and a separate Flow Music app is already on iOS. The web version supports 4K output with advanced scene settings.

🔗 Google Blog

#6ByteDance Preps Seedance 2.1 + Mini — 20% Quality Boost Incoming

ByteDance is executing a two-pronged strategy: Seedance 2.1 pushes the quality ceiling with a reported 20% improvement over 2.0, while Seedance Mini lowers the cost floor for high-volume production. No firm release date yet, but reporting from the past week suggests a June launch. Meanwhile, Seedance 2.0 is now available via the Runway API and fal, generating ~20-second clips with joint audio-video architecture.

🔗 Pandaily

#7HappyHorse-1.0 API Now Live — Alibaba's Dark Horse at #1 Elo

The mysterious model that topped the Artificial Analysis Video Arena on April 7 is now confirmed as Alibaba's project, led by ex-Kling architect Zhang Di. HappyHorse generates synchronized audio-video in a single forward pass using a unified 15B-parameter Transformer — no separate audio post-processing. Multilingual lip-sync across seven languages. Developer and enterprise API access launched April 27 via fal.

🔗 CNBC

#8TikTok Now Auto-Labels AI Content via C2PA — Even Without Creator Disclosure

TikTok's 2026 policy enforcement is now active with C2PA Content Credentials detection that automatically identifies and labels synthetic media, even when creators skip self-disclosure. Unlabeled AI content faces auto-labeling, distribution reduction, or removal. Repeated violations affect account standing. Notably, YouTube and Instagram have different thresholds — making cross-platform compliance a per-platform exercise now.

🔗 Storrito

🎬 AI Video Intel — Monday, May 25, 2026 at 6:45 AM

Jump to another brief

Jump to this brief on another date

Recent AI Video Intel