Alibaba's Wan 2.7 landed in ComfyUI v0.18.5 in late March and is now the most capable open-weights video model you can run locally. It generates 1080p/24fps video up to 15 seconds from text, reference images, or both — with native audio baked into the generation pipeline. Supports up to 5 real-person image inputs, vocal timbre reference, and 3x3 grid-based image gen. Open weights expected under Apache 2.0 in Q2 2026, though not yet officially confirmed. If you're still on Wan 2.2, this is the upgrade.
YouTube has completely separated the Shorts recommendation engine from long-form. Shorts performance no longer drags down or boosts long-form recs on the same channel. YouTube also rolled out a dedicated "Shorts" content type filter in search results, so Shorts now rank independently. This means you can run aggressive Shorts experiments without risking your long-form channel authority — or spin up dedicated Shorts-only channels with zero penalty.
The Kling AI dance filter is the dominant AI viral format on TikTok right now. Creators upload a single photo, pick a reference dance video, and Motion Control 3.0 generates ultra-fluid dance clips. Individual creators are reporting 1-2M+ views per clip, and AI-enhanced content is seeing 2.3x higher completion rates vs. raw single-take uploads. The "Hawak Mo Ang Beat" variant is the specific trend blowing up. Low-effort, high-reach format — worth testing immediately.
Top creators aren't loyal to a single model — they're running multi-model stacks. The emerging best practice: use Seedance 2.0 to lock in brand assets and pre-visualize complex scenes (it accepts up to 12 file inputs — 9 images, 3 videos, 3 audio clips), then hand off to Kling 3.0 for realistic human interactions and controlled motion. Seedance at $0.30/clip is the pre-production workhorse; Kling is the polish pass. This is the workflow to benchmark against.
YouTube's July 2025 policy update is now fully enforced. Templated slideshows, pitch-shifted audio, and automated narration without original commentary risk immediate demonetization. But the key nuance: AI-assisted content with face intros, personal voiceovers, or unique narrative framing is explicitly safe. YouTube's algorithm doesn't distinguish AI vs. human — it only cares about viewer response. Properly disclosed AI content receives normal algorithmic distribution. The January 2026 update added mandatory disclosure requirements for all synthetically generated media.
The 22-year-old college dropout running "Boring History" is pulling $40K-$60K/month from a network of AI faceless channels, verified by Fortune with AdSense screenshots. His 6-hour "history to sleep to" documentaries cost ~$60 each to produce using a custom pipeline (Claude for scripts, ElevenLabs for narration, proprietary TubeGen assembly tool). Operating costs run $6,500/month against that revenue. The broader space: AI-generated faceless channels have collectively hit 63 billion views, 221 million subscribers, and an estimated $117M/year in ad revenue.
Fresh RPM data confirms finance channels command $9-$21 per 1,000 views (personal finance averaging $10-$15 RPM), while horror storytelling channels hit $8-$13 RPM. Both niches reach YouTube monetization thresholds 40% faster than lifestyle or motivation content. The Mr. Nightmare model (calm narration + atmospheric music + "true" scary stories) has scaled to 1.6B+ views across 6M+ subscribers — and it's a format that maps perfectly to AI production pipelines.
YouTube Shorts now generates 200 billion daily views (up from 70B in 2023) and accounts for 75-77% of all global YouTube views. But the quality bar has risen with the volume: the algorithm's internal threshold for "good" retention on Shorts is now ~70%. If your average view duration falls below that on most uploads, your distribution ceiling drops channel-wide. The first 30-60 minutes are make-or-break — if a Short doesn't hit a performance threshold in that window, YouTube stops pushing it.
