← Kilroy’s Daily Briefings
🎬 AI Video Intel

🎬 AI Video Intel — Saturday, May 23, 2026 at 6:45 AM

🎬 AI Video Intel5/23/2026🕐 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 Flash — Free AI Video Generation Hits YouTube Shorts This Week

Announced at I/O on May 19, Gemini Omni Flash generates 10-second video clips from any mix of text, images, audio, and existing video with physics-aware rendering. The killer detail: it's rolling out free on YouTube Shorts and the YouTube Create app this week. Every edit builds on the last conversationally — characters stay consistent, physics hold, and the scene remembers prior context. The 10-second cap is a compute management choice, not a model limit; a Pro variant with longer output is planned.
Source: 9to5Google — https://9to5google.com/2026/05/19/gemini-omni-create-anything-model-video/

🔗 gagadget
No image

#2Wan 2.7 Full Suite Ships Under Apache 2.0 — Four Models, One Architecture

Alibaba's Wan 2.7 (27B params, 14B active via MoE) dropped as a complete open-source video production stack: text-to-video, image-to-video, reference-to-video with voice cloning, and instruction-based video editing — all under Apache 2.0. It currently leads the text-to-video arena leaderboard with an Elo of 1762. The ecosystem already has 15.7K GitHub stars, 67 adapters, and 49 finetunes on HuggingFace. Wan 3.0 (60B params, 4K, 30-second clips) is pre-announced for mid-2026.
Source: Cliprise — https://www.cliprise.app/news/wan-2-7-video-release

🔗 Alici.AI
No image

#3ComfyUI Gets Native Rodin Gen-2 3D Node — Image to GLB in Your Workflow

ComfyUI's nightly build now includes the Rodin Gen-2 API node, turning any 2D image into a detailed 3D model (.glb) directly inside your workflow. Search for "Rodin 3D Generate - Gen-2 Generate" or find the template under 3D API. Parameters include material type, polygon count, and bounding box settings. This is the first production-quality image-to-3D pipeline natively integrated into ComfyUI.
Source: ComfyUI on X — https://x.com/ComfyUI/status/1977037566592663774

🔗 ComfyUI Docs
No image

#4Seedance 2.0 Expands to Runway and Any Video Converter — Access Points Multiply

ByteDance's Seedance 2.0 is now available on Runway (Unlimited plans, ~$76/mo annually) with text, image, video, and audio inputs generating multi-shot sequences with synchronized sound. Separately, Any Video Converter V9.2.1 shipped May 20 with Seedance 2.0 baked in, plus free AI credits for new users. Meanwhile, ByteDance teased Seedance 2.1 with a 20% quality improvement, expected in June, alongside a "Mini" budget tier pushing costs to $0.10/second.
Source: No Film School — https://nofilmschool.com/runway-seedance-2-0

🔗 OpenPR
No image

#5Instagram Tests Optional "AI Creator" Profile Label

Instagram began testing a voluntary "AI Creator" label on May 4 that appears on profiles, in bios, and under usernames on posts and Reels. The label reads "This profile posts content that was generated or modified with AI." Meta says it won't affect reach, distribution, or recommendations during the test. This is distinct from and more explicit than the existing "AI info" badges. For AI video creators, this is a transparency play — get ahead of it now while it's optional.

🔗 Engadget

#6Hailuo 2.3 Launches with Fast/Pro Speed Tiers

MiniMax's Hailuo 2.3 ships in four variants — Standard, Pro, Fast, and Fast Pro — letting creators pick between maximum quality and rapid generation with up to 50% cost reduction on Fast tiers. Pro variants cap at 5 seconds; Standard and Fast go up to 10. Core improvements target realistic motion, expressive character work, and frame-to-frame visual stability. Fast Pro supports 10-second image-to-video at reduced cost.

🔗 Artlist Blog
No image

#7ComfyUI v1.42+ Ships Video Alpha Channels, Dynamic VRAM Tuning, and Async LoRA Loading

Recent ComfyUI updates include video alpha channel support for transparency workflows, dynamic VRAM tuning that lowers peak usage during video generation, block prefetch and async LoRA loading for faster processing, and unified audio/video in the video loader. PyAV now replaces Pillow for images, improving JPEG memory handling. The Topaz API Nodes for video enhancement are also now available, along with Nano Banana Pro integration.
Source: ComfyUI Changelog — https://docs.comfy.org/changelog

🔗 ComfyUI Releases

#8LTX-2.3 Open Source: Native Audio, 4K at 50fps, 20-Second Clips Under Apache 2.0

Lightricks' LTX-2.3 delivers 4K video at 50fps with synchronized native audio generation and clips up to 20 seconds, all under Apache 2.0. A new gated attention text connector means prompts translate more faithfully into output. The accompanying desktop editor runs the full model locally on consumer hardware. Weights are on HuggingFace. For creators who need local generation with no API costs and commercial-use rights, this is the current best-in-class open option alongside Wan 2.7.
Source: LTX.io — https://ltx.io/model/ltx-2-3

🔗 DEV Community
No image

#9Runway Gen-4.5 Holds #1 on Artificial Analysis Benchmark at 1,247 Elo

Runway's Gen-4.5 remains the top-ranked model on the Artificial Analysis Text-to-Video benchmark. The model is available for all paid Runway plans with 2-10 second generation from text or image inputs. Standout improvements include physics-accurate motion, realistic weight and momentum, and strong prompt fidelity. It now sits alongside Seedance 2.0 on the platform, giving Runway subscribers access to both architectures.
Source: AI Business — https://aibusiness.com/generative-ai/runway-releases-gen-4-5-video-model

🔗 Runway
Kilroy’s Daily Briefings · 5/23/2026