← Kilroy’s Daily Briefings
🎬 AI Video Intel

🎬 AI Video Intel — Saturday, May 16, 2026 at 6:45 AM

🎬 AI Video Intel5/16/2026🕐 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.

#1ComfyUI v0.21.1 Drops with Veo3, Seedream, and Unified Audio/Video Pipeline

Released May 13, ComfyUI v0.21.1 adds Veo3 integration, ByteDance SeedreamNodeV2, Flux2ImageNode, GrokImageEditNodeV2, an OpenAI Image node, and a Claude LLM node — all with DynamicCombo and Autogrow features. The bigger under-the-hood story: a unified audio/video loader replacing Pillow with PyAV, block prefetch, LoRA async loading, and dynamic VRAM tuning with --cache-ram 2. This turns ComfyUI into a genuine multimedia creation hub.

🔗 ComfyUI Changelog

#2Wan 2.7 Live on Together AI — First-and-Last-Frame + V2V Editing

Alibaba's Wan 2.7 hit Together AI on April 3 and is now the most feature-complete open-source video model available via API. New capabilities include first-and-last-frame generation for precise narrative arcs, natural-language video-to-video editing, a 9-grid image-to-video layout for multi-angle reference, and combined visual + voice subject referencing. Outputs at 720p/1080p, 2–15 seconds, with optional audio.

🔗 Together AI Blog
No image

#3Kling 3.0 Motion Control Arrives in ComfyUI

Kling 3.0's director-level camera system (pan, tilt, zoom, dolly, rack focus) is now accessible via ComfyUI nodes. The model supports multi-shot sequences (2–6 scenes with auto transitions), 4K/30fps output, native audio sync, and physics simulation. Character consistency is locked across shots — a major workflow win for short-form creators building narratives.

🔗 ComfyUI Blog
No image

#4Seedance 2.0 Public on fal — Native Audio-Video with 8-Language Lip Sync

ByteDance's Seedance 2.0 is now generally available on fal (API access since April 9). It's the first unified audio-video generation model — sound and image are generated in the same forward pass, not post-processed. Phoneme-level lip-sync works across 8+ languages, making it immediately useful for multilingual talking-head content without a separate dubbing pipeline.

🔗 fal.ai

#5LTX-2.3: Open-Source 4K at 50 FPS with Native Vertical Format

Lightricks' LTX-2.3 (22B parameters, March 2026) generates native 1080×1920 vertical video — trained on portrait-specific data, not cropped from landscape. 4K output, 50fps, synchronized audio from a single forward pass. Free for commercial use under $10M revenue. ComfyUI nodes are optimized with NVFP8 quantization cutting model size ~30% and doubling inference speed.

🔗 Lightricks / GlobeNewsWire

#6Google Veo 3.1 Lite — Budget API Video Generation with Native Audio

Released March 31, Veo 3.1 Lite costs less than 50% of Veo 3.1 Fast with the same generation speed. Supports text-to-video and image-to-video, 720p/1080p, landscape and portrait ratios, 4–8 second durations with native audio — all through a single Gemini API call. Available on paid-tier Google AI Studio. Positioned as the volume play for creators who need high throughput at lower cost.

🔗 Google AI Blog

#7TikTok AI Detection Now at 94.7% Accuracy — Enforcement Up 340%

TikTok's automated detection identifies synthetic faces at 94.7% accuracy and AI-generated backgrounds at 87.3%, scanning content from 47+ AI generation platforms via C2PA metadata and frame-level artifact analysis. Enforcement removals for unlabeled AI content surged 340% in 2025, with 51,000+ videos removed and 8,600 accounts permanently banned. If you're posting AI video to TikTok without their built-in label, you're playing with fire.

🔗 Storrito / AuditSocials

#8Faceless AI Channels Hit 38% of New Creator Monetization — Sub-$3 Production Costs

New data shows faceless YouTube channels now represent 38% of all new creator monetization ventures (up from 12% in 2022). AI voice, automated assembly, and multilingual dubbing have collapsed production costs to under $3 per video. Top performers: finance/tech niches commanding $15–40 CPM, with channels like Fern (3D crime docs) pulling $80K+/month. Noah Morris operates ~20 channels with 2.5M+ combined subscribers, one court-case video costing $250 to produce earned $20K from 5M views.

🔗 AutoFaceless / Miraflow
No image

#9Sora Officially Dead — Market Reshuffles

OpenAI's Sora web/app shut down April 26, with the API sunsetting September 24. Active users had collapsed to under 500K while burning an estimated $15M/day in compute. The vacuum benefits Runway Gen-4.5 (currently #1 on Artificial Analysis benchmarks at 1,247 Elo), Kling 3.0, and the open-source stack. If you had Sora in your pipeline, migration paths point to Runway for quality, Kling for cost, or Wan 2.7/LTX-2.3 for local control.

🔗 Multiple / Bloomberg
No image
Kilroy’s Daily Briefings · 5/16/2026