Here's your briefing, Kilroy:
---
🎬 AI Video Intel — 2026-04-29 06:45
Model & Tool Releases
• Sora is dead. OpenAI's consumer app shut down April 26. API stays live until Sep 24, then it's gone. Three senior execs left the same week (April 17) — including Sora creator Bill Peebles. OpenAI is redirecting compute to chat/coding. (OpenAI Help Center, ThePlanetTools)
• ComfyUI raised $30M at $500M valuation (April 24). Craft Ventures led. Total funding now $48M. 4M+ users, 60K+ community nodes, 150K daily downloads. Netflix, Apple, and Ubisoft are paying customers. Silverside AI used ComfyUI to produce SVEDKA's 2026 Super Bowl ad — first primarily AI-generated Super Bowl spot. (TechCrunch, GlobeNewsWire)
• Netflix open-sourced VOID (Video Object and Interaction Deletion) on April 3. Apache 2.0 licensed. Built on CogVideoX, it removes objects from video AND rewrites the physics — remove a person holding a guitar and the guitar falls naturally. Uses quadmask conditioning (primary object / overlap / affected regions / background). Beat Runway in user preference studies 64.8% vs 18.4%. (GitHub, Hugging Face, MarkTechPost)
• ComfyUI v0.19.3 (April 17): LTX text generation nodes, Quiver Arrow 1.1/1.1-max SVG models, Topaz video enhancement via Partner Nodes, Nano Banana Pro. GPT Image 2.0 also now available in ComfyUI through Partner Nodes — it reasons before generating, useful for dense text/UI/infographics. (ComfyUI Changelog, ComfyUI Blog)
• LTX updates: Desktop v1.0.3 (April 3) dropped VRAM floor to 12GB. LTXV now generates clips >60 seconds via autoregressive sequence conditioning. LTX 2.3 distilled model updated to v1.1 on April 14. Seedance 2.0 integrated into LTX Studio on April 24. Canvas collaborative workspace launched April 27. (LTX.io, LTX Studio Release Notes)
• Kling 3.0 standout: multi-shot sequences (3-15s) with subject consistency across camera angles, shared audio timeline, native dialogue in 5 languages, and the only model with native 4K output. (Pixflow, Lushbinary)
Platform Signals
• TikTok is deprioritizing AI content. Algorithm now gives AI-labeled videos ~5-8% lower reach than traditional content. Three unlabeled AI videos trigger 60% reach reduction for 30 days plus Creator Fund suspension. TikTok has labeled over 1.3 billion AI-generated videos via C2PA Content Credentials + watermark detection. Removal of unlabeled synthetic media up 340% vs 2024. (Vexub, Storrito)
• Meta unified AI labeling across Instagram/Facebook (February 2026). "Made with AI" tag now required via advanced settings — looks like the paid partnership label. Instagram Reels Originality Score penalizes recycled/duplicate content. However, Reels bonus program eligibility is engagement-only — creation method doesn't matter. (Influencer Marketing Hub, Vexub)
• YouTube: Three-strike enforcement for undisclosed AI content — warning, 90-day suspension, permanent removal. AI-generated content remains fully monetizable with proper "altered or synthetic content" disclosure in YouTube Studio. (Vexub, ReelNReel)
• TikTok followers-first rollout: New videos now go to existing followers first over several days, measuring engagement velocity before pushing to non-followers. This hits AI channels that grew on virality but have low follower loyalty. (Miraflow)
What's Working
• YouTube remains the best platform for AI video monetization. Same CPM ($3-5 per 1K views), no AI-specific rate penalty, just needs disclosure. Top AI-assisted creators pulling $500K-5M+ annually via volume + quality. Production costs down 80-95% per video vs traditional. Individual creators producing 5-10x more video than 2024 counterparts. (AI Monks/Medium, Vivideo)
• Instagram Reels bonus is creation-method agnostic — only engagement metrics matter for payout eligibility. If your AI content gets the engagement, you get paid the same. (Vexub)
• Virtual influencer brand deals: $2K-15K per sponsored post at 100K+ followers. IP licensing deals averaging $45K-120K for large campaigns. (CommuniPass)
• Paid communities are outperforming sponsorships and affiliate income for recurring revenue stability among AI creators. (CommuniPass)
Technique Watch
• VOID for clean plates. Netflix's model gives everyone access to production-quality object removal. Apache 2.0 means it slots straight into ComfyUI pipelines. Built on CogVideoX-Fun-V1.5-5b-InP with quadmask conditioning — useful for removing watermarks, fixing compositing errors, or cleaning reference footage. (GitHub)
• Seedance 2.0's unified audio-video architecture is the technique to study. The model "hears" what it's generating as it generates — a character in a large room gets natural reverb, a whisper gets proximity effect. This was previously post-production only. Multi-shot storytelling from a single prompt with phoneme-level lip-sync in 8+ languages. (Seedance)
• LTX autoregressive sequence conditioning for 60+ second clips — each segment is conditioned on the previous, maintaining motion/narrative continuity. Relevant for anyone doing long-form AI video who's been stitching segments manually. (MarTechCube)
• GPT Image 2 in ComfyUI for text-heavy assets (thumbnails, title cards, infographics). It plans composition and self-corrects, which makes it significantly better than one-shot models for anything with readable text. (ComfyUI Blog)
Worth Watching
• Post-Sora vacuum. With Sora dead and three execs gone, the video generation race is now Veo 3.1 (Google), Seedance 2.0 (ByteDance), Kling 3.0 (Kuaishou), and Runway Gen-4.5. OpenAI's retreat validates that video gen economics are brutal — watch whether Google maintains Veo investment or follows suit. (Futurum Group)
• ComfyUI's $500M valuation + enterprise customers (Netflix, Apple, Ubisoft) signals the open-source pipeline approach is winning over walled-garden APIs. The 150K daily downloads number makes it the de facto standard for production AI media. (TechCrunch)
• TikTok's followers-first algorithm + AI deprioritization is a structural headwind for AI-first accounts that grew on FYP virality. Channels with low follower-to-view ratios will feel this hardest. YouTube and Reels are now clearly the better monetization targets for AI content. (Miraflow)
• EU AI Act enforcement is ramping — content creators are now subject to transparency requirements for AI-generated media. This isn't just platform policy; it's regulation. Worth tracking for anyone distributing AI content in EU markets. (OpusClip)
• AI video generation market projected at $18.6B by end of 2026 (34% CAGR). Creator economy overall heading past $280B. The money is real and accelerating. (Grand View Research)