Wan2.7, Alibaba's latest video generation model, is now fully integrated into ComfyUI via Partner Nodes — update to v0.18.5 or later to unlock it. The integration supports five distinct task types: Image-to-Video in three modes including audio-driven generation, Text-to-Video with multi-shot narration, Video Continuation, Reference-to-Video with up to five real-person inputs and vocal timbre matching, and text/style-transfer Video Edit. Ready-made workflow templates are live in the Template Library under Video — search "Wan2.7."
Kuaishou released Kling 3.0 Turbo on June 17, building a dedicated fast-iteration layer into the Kling 3.0 family — 3–15-second clips at 720p or 1080p, multi-shot prompting up to six shots per generation, and native multilingual lip-sync in English, Mandarin, Japanese, Korean, and Spanish. Pricing runs approximately $0.11/sec at 720p and $0.14/sec at 1080p with audio included. The workflow play: draft in Turbo, then promote winners to Kling 3.0 Pro.
YouTube's newest Shorts algorithm update includes an AI content-repetition filter that suppresses channels posting too-similar formats, hooks, or visual styles — and it has been incorrectly flagging consistent human creators who post on schedule with consistent thumbnails. The algorithm's performance testing window has shrunk to 30–60 minutes post-upload, and channels need 70%+ average retention to maintain full distribution. Separately, the July 8 YouTube Studio update replaced Shorts like buttons with heart icons and added a consolidated "Notices" column in the Content tab.
ByteDance's Seedance 2.5 entered closed enterprise beta this week; the public API goes live on BytePlus ModelArk July 16. The flagship feature is genuine 30-second native generation with no clip-stitching required, paired with support for up to 50 full-modal reference assets simultaneously. The model adds 3D white-model previz for locking spatial composition and motion before materials/lighting are applied, plus controllable editing that preserves motion, camera position, and lighting while swapping subjects — a post-production workflow tool baked into generation.
YouTube's inauthentic-content monetization policy — which made pure AI-automated video ineligible for AdSense since July 2025 — is now being applied aggressively. Channels running fully automated pipelines without narration, editorial commentary, or original framing are receiving demonetization notices this week. Eligible AI-assisted content still earns $3–$9 RPM; the enforcement line is whether a human made a real creative contribution. YouTube requires manual disclosure labels for AI-generated content and penalizes undisclosed synthetic media with reduced recommendations.
The MPA's cease-and-desist against ByteDance — the first issued against a major AI company, over Seedance 2.0 training data — remains unresolved as Seedance 2.5 launches. Disney, Warner Bros., Paramount Skydance, Sony, and Netflix all sent their own legal letters in February. A July 7 report revealed those same studios are reportedly deploying Seedance on internal productions. ByteDance added C2PA watermarking and face/character filters by March 2026, but with no court resolution in sight, commercial pipeline risk remains real.
Google opened Veo 3.1 to every Google account through Google Vids in April, delivering 10 free video generations per month with no subscription required. The model outputs native 9:16 vertical video for Shorts, upscales to 1080p and 4K, and generates synchronized native audio. Google AI Pro/Ultra subscribers additionally get Lyria 3 custom music generation, Lyria 3 Pro, and customizable AI avatars. Google AI Ultra accounts get up to 1,000 Veo video generations per month.
Runway pushed two updates in the first week of July: Agent Skills (July 2, all plans) enables users to build complete ad campaigns, produce commercials, and localize creative from a single natural-language command — the clearest move yet toward agentic production. Nano Banana 2 Lite (July 1, paid plans) is a fast image generation model for high-speed reference and concept work. Together they position Runway as a production-task agent rather than a clip-generation tool.
Midjourney v8.1 became the platform default on June 10, running 4–5x faster than v7 with improved prompt adherence on fine details and small elements. As of early July, v8.2 has appeared in the preview command slot on the website, signaling a near-term wider release. Image-to-video extension remains active: 5-second generations extendable to 21 seconds. No headline capability jump yet, but the iteration pace is picking up.
