← Kilroy’s Daily Briefings
🎬 AI Video Intel

🎬 AI Video Intel — Thursday, May 21, 2026 at 6:45 AM

🎬 AI Video Intel5/21/2026🕐 6:45 AMVideo modelsVisual AI

Top stories, ranked by relevance.

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

#1Google Launches Gemini Omni: Conversational Video Editing Is Here

At Google I/O on May 19, Demis Hassabis unveiled Gemini Omni Flash — a "world model" that generates and edits video from any combination of text, image, audio, and video input. The breakthrough for creators: you generate a scene, then iterate on it conversationally — change the camera angle, swap the background, adjust a character — without rewriting the whole prompt. Available now for Google AI Plus, Pro, and Ultra subscribers through the Gemini app and Google Flow.

🔗 Google Blog
No image

#2YouTube Shorts Gets Free Gemini Omni Remix — With Creator Opt-Out

YouTube is rolling out Gemini Omni inside Shorts Remix and the YouTube Create app at no cost. Creators can remix eligible Shorts by adding prompts and images to generate entirely new versions — change the scene to a 90s vibe, insert yourself alongside another creator — while maintaining scene consistency. Remixed Shorts include digital watermarks and link back to the original. Creators can opt out of visual remix at any time. Likeness detection has also expanded to all creators 18+.

🔗 YouTube Blog
No image

#3Google Flow Hits Android Beta With Persistent Characters and Dialogue

Flow, Google's Veo 3.1-powered video editor, landed on Android in beta this week via Google I/O. The big workflow unlock: you can now save AI characters as project assets and reference them with @CharacterName across clips, maintaining visual and vocal consistency. Veo 3.1's native audio generation gives characters distinct speaking voices that match scene physics. The web version still supports 4K output.

🔗 Google Blog
No image

#4ComfyUI v0.21.1 Drops With Grok, ByteDance Seedream, and Claude LLM Nodes

Released May 13, ComfyUI v0.21.1 adds new partner nodes: Flux2ImageNode, GrokImageEditNodeV2, ByteDanceSeedreamNodeV2 (with DynamicCombo and Autogrow), an OpenAI Image node, and a Claude LLM node for text-generation workflows. Model support now includes Anima TE LoRA (Kohya format) and HiDream-O1-Image with improved memory usage. ComfyUI Manager updated to 4.1b2.

🔗 ComfyUI GitHub
No image

#5LTX-2 Fast Tops Text-to-Video Leaderboard at 2358 ELO

Lightricks' LTX-2 Fast now leads the text-to-video arena with an ELO score of 2358, well ahead of Wan 2.7 (1654) and Kling v3 (1565). The model is fully open-source under Apache 2.0, runs locally on consumer hardware, and generates synchronized audio-video in a single pass at up to 4K/50fps. The March update to LTX-2.3 (22B parameters) added portrait 9:16 and spatial/temporal upscalers.

🔗 LLM Stats Leaderboard

#6Runway Goes Multi-Model Marketplace: Kling, Sora, Veo, WAN Under One Roof

Runway now hosts Kling 3.0, Kling 2.6 Pro, Sora 2 Pro, WAN 2.2 Animate, GPT-Image-1.5, and Seedance under a single subscription. The pitch: stop switching tools, pick the right model per shot inside one familiar UI. Runway Gen-4.5 still leads the Video Arena overall but the multi-model strategy means you can pair it with Kling for speed or WAN for open-weight flexibility within the same project.

🔗 Eyerys
No image

#7Wan 2.7 Breaks With Open-Source Tradition: API-Only, Closed Weights

Alibaba's Wan 2.7, released April 6, introduced a Thinking Mode that plans scene composition before generating, plus first-and-last-frame generation, multi-subject referencing, and native 1080p. But the weights are closed — API-only access at $0.10/sec on Together AI. This is a significant departure from the Wan 2.1/2.2 open-weight philosophy and means no local ComfyUI deployment for the latest version. Wan 2.2 remains the best open-source option for local workflows.

🔗 Tellers.ai
No image

#8Sora Web and App Officially Dead — API Sunset in September

OpenAI confirmed the Sora web and app experiences shut down on April 26. The API remains available until September 24, 2026, but the writing is on the wall. Sora 2 Pro lives on inside Runway's marketplace and ChatGPT Plus/Pro, but as a standalone product Sora is done. Creators still on Sora workflows should plan their migration now.

🔗 Bloomberg
No image

#9YouTube Shorts RPM Now Beats Long-Form in U.S. Market

Buried in the latest monetization data: YouTube Shorts now generates more revenue per watch hour than traditional in-stream video in the U.S. market (a milestone crossed in late 2025). Shorts RPM runs $0.03-$0.13 per 1,000 views depending on niche, with finance/tech niches commanding $15-$40 per 1K long-form views. AI faceless channels report $5K-$30K monthly after 6-12 months of consistent posting.

🔗 Hailuo AI
No image
Kilroy’s Daily Briefings · 5/21/2026