Sora’s consumer app is officially gone, and its API is now on a countdown clock. NVIDIA opened up a new open-weights world model at Computex. Runway is leaning harder into “world models” with new revenue numbers to back it up. TikTok’s AI labels have now touched well over a billion videos, and YouTube’s slop crackdown keeps expanding. Here is what actually happened in AI video through mid-June 2026, and what creators, marketers, and builders should be doing about it.<\/p>\n\n\n\n
The rest of this post walks through each story in more detail.<\/p>\n\n\n\n
The biggest structural shift in June 2026 is less about a single new model and more about an old one finally exiting the stage.<\/p>\n\n\n\n
Sora’s wind-down is now in motion.<\/strong> OpenAI confirmed that the Sora web and app experiences were discontinued on April 26, 2026, with the API set to follow on September 24, 2026. The shutdown is happening in phases: new accounts and API keys stopped being issued earlier this year, a refund window for long-term Pro subscribers opens this June, read-only mode for video libraries arrives in August with generation capacity cut by roughly 80%, and the servers go fully offline in September. Developers still building on sora-2 or sora-2-pro are being pointed toward alternatives like Seedance 2.0 and Wan, since OpenAI has not announced a successor video API.<\/p>\n\n\n\n Kling 3.0 now leads the leaderboards.<\/strong> On the Artificial Analysis-style arena rankings, Kling v3 sits at the top with a score of 2031, followed by LTX-2 Fast at 1930 and HappyHorse-1.0, an Alibaba-backed entrant that first appeared on leaderboards back in April, at 1885. Kling’s combination of native 4K output, a storyboard tool with per-shot camera control, and lip-synced audio continues to make it a favorite for prosumer advertising work.<\/p>\n\n\n\n Veo 3.1 remains the safe default for audio-heavy work.<\/strong> Rankings published this month still put Veo 3.1 at the top for creators who need strong realism, good motion, and synchronized audio in one package, with Kling 3.0 as the best value alternative for high iteration counts. Runway Gen-4.5 continues to be the pick when camera control and structured editing matter more than raw benchmark scores, while Pika remains popular for fast, short-form iteration and Luma’s Ray3 and Ray3.14 models are the go-to for atmospheric, environment-heavy shots.<\/p>\n\n\n\n NVIDIA enters the field with Cosmos 3.<\/strong> At Computex on June 1, 2026, NVIDIA released Cosmos 3, an open-weights “omnimodel” for physical AI that handles text, image, video, ambient sound, and action generation in one architecture. It ships as Cosmos 3 Super and Cosmos 3 Nano on Hugging Face under an open license, backed by a large academic paper. NVIDIA claims top open-source rankings on several benchmarks, though those claims haven’t been independently verified yet. Cosmos 3 is positioned more as infrastructure for robotics and world-model research than a direct creator tool, but it adds another serious open-source option to the stack.<\/p>\n\n\n\n The open-source side keeps moving too.<\/strong> Helios, a 14B-parameter autoregressive diffusion model from a Peking University, ByteDance, and Canva collaboration, shipped under an Apache 2.0 license in March and is notable for real-time generation at close to 20fps on a single H100. Lightricks’ LTX-2 remains a strong open-weights option for audio-video generation, with NVIDIA shipping optimized versions for RTX hardware.<\/p>\n\n\n\n The overall picture: the gap between the leading commercial models is narrow and shifts month to month, while the open-source tier has gotten meaningfully stronger since the start of the year.<\/p>\n\n\n\n After a blockbuster February for fundraising, June has been more about execution than new rounds, but the numbers are notable.<\/p>\n\n\n\n Runway says its world-model bet is working.<\/strong> Following its $315M Series E at a $5.3B valuation in February, Runway reportedly added $40M in annual recurring revenue in the second quarter of 2026 alone. The company has expanded beyond pure video generation since releasing its first world model in December 2025, with a second one planned for later this year. Runway is now framing itself as a long-term competitor to Google in world models, a category it shares with Luma AI (valued near $4B after its $900M Series C) and World Labs (which has raised roughly $1.29B).<\/p>\n\n\n\n The funding pattern from earlier in the year is holding.<\/strong> Global funding for AI video companies hit $3.08B in 2025, up nearly 95% from 2024, and the early-2026 mega-rounds, Runway’s $315M, Luma’s $900M, and Synthesia’s $200M Series E at a $4B valuation, set the tone for a year where capital keeps concentrating around a handful of foundation-model labs while a longer tail of workflow and editing tools builds on top of them.<\/p>\n\n\n\n Runway’s startup fund is now active.<\/strong> The $10M Runway Fund and Builders program, announced in late March, formally launched in May as an investment vehicle for early-stage startups working across AI, media, and world simulation, continuing a strategy Runway says it had already been pursuing quietly for years.<\/p>\n\n\n\n Distribution rules continue to be where AI video policy gets real, and June brought incremental but meaningful tightening across the board.<\/p>\n\n\n\n TikTok’s labeling has now touched over 1.3 billion videos.<\/strong> The platform’s AI content policy requires visible labels on AI-generated visuals and audio depicting realistic people or scenes, enforced through a combination of C2PA Content Credentials, invisible watermarking, and creator self-disclosure. TikTok’s automatic detection now catches an estimated 35 to 45% of AI content without any self-disclosure, up sharply from just 18% in early 2024. The enforcement mechanism remains a three-strike system: after three unlabeled AI uploads, an account’s reach drops by roughly 60% for 30 days and Creator Fund earnings are paused. TikTok has also added a requirement that advertisers using voice clones or digital likenesses upload signed consent documentation directly into TikTok Ads Manager, reviewed within five business days.<\/p>\n\n\n\n YouTube’s “inauthentic content” policy is still reshaping the faceless creator economy.<\/strong> The platform’s updated guidance continues to disqualify mass-produced, templated, or machine-made uploads with no human creative input from the Partner Program, while AI-assisted content where a creator’s voice, editing, and editorial judgment are central remains monetizable. The practical effect has been a wave of demonetizations among faceless channels, alongside a parallel trend of operators hiring on-camera hosts to stay compliant. Independent reporting earlier this year found that a significant share of videos recommended to children on YouTube and YouTube Kids appeared to be AI slop, which is part of why this remains a top priority for the platform heading into the second half of 2026.<\/p>\n\n\n\n Meta’s unified policy across Instagram and Facebook is now the baseline.<\/strong> Since February, both platforms have required a separate in-post AI disclosure tag, distinct from caption text, putting Meta roughly in line with TikTok’s labeling approach even if its detection relies more heavily on self-declaration and third-party metadata partnerships.<\/p>\n\n\n\n The compliance patchwork is now a real operational problem.<\/strong> A single piece of AI video content distributed across platforms may need a TikTok label, a Meta disclosure tag, a YouTube flag, and soon an EU-mandated watermark, each with slightly different rules about what counts as “AI-generated” and how it must be disclosed.<\/p>\n\n\n\n The industry tension that defined the spring is still very much live in June.<\/p>\n\n\n\n Seedance 2.0 remains a flashpoint.<\/strong> ByteDance’s video model, which generated viral unauthorized videos depicting SAG-AFTRA President Sean Astin, Tom Cruise, and Brad Pitt earlier this year, drew direct condemnation from both SAG-AFTRA and the Human Artistry Campaign, who called the videos “blatant infringement” and an attack on creators’ rights to their own likeness and work.<\/p>\n\n\n\n SAG-AFTRA’s contract negotiations are entering a critical window.<\/strong> The union’s agreement with the studios is up for renewal this summer, and AI protections, including the proposed “Tilly tax” that would price the use of synthetic performers at roughly human-equivalent rates, remain a central bargaining issue. This builds on the union’s broader pattern over the past two years: AI protections were a defining issue in the 2023 film and TV strike, the centerpiece of the 2024 to 2025 video game performer strike that concluded with a ratified deal in mid-2025, and a continuing theme in state-level legislation the union has supported, including AI protection bills signed in New York late last year.<\/p>\n\n\n\n The studio-AI relationship remains pragmatic but cautious.<\/strong> Disney’s $1B investment in OpenAI and its role as a content licensing partner on Sora continue even as the consumer Sora app shuts down, since the underlying Sora 2 model and API persist until September. Meanwhile, Netflix and Disney have both been reported to use Runway’s tools internally for storyboarding and pre-visualization, reinforcing the pattern that AI video remains a production-pipeline tool inside studios rather than a replacement for the pipeline itself.<\/p>\n\n\n\n A practical checklist for June 2026:<\/p>\n\n\n\n If you’re still using Sora or building on the Sora API:<\/strong> Start migrating now. The refund window for Pro subscribers opens this month, and read-only mode with an 80% capacity cut arrives in August, well before the September 24 shutdown. Seedance 2.0 and Wan are the most commonly cited drop-in alternatives, though both carry their own rights and policy complications.<\/p>\n\n\n\n If you run AI-assisted or faceless channels:<\/strong> Audit your output against YouTube’s inauthentic-content criteria and TikTok’s three-strike labeling rules. The safest path on every major platform is the same: keep a visible human creative role (scripting, editing, narration, original format design), self-disclose AI elements at upload, and treat AI as part of your workflow rather than the entire product.<\/p>\n\n\n\n If you’re running paid AI video creative:<\/strong> TikTok’s consent documentation requirement for voice clones and digital likenesses is a five-business-day review process, so build that into campaign timelines now rather than at launch. Keep tracking the EU AI Act’s Article 50 deadline in August, since any creative shipped into the EU after that date will need machine-readable provenance metadata and visible labeling.<\/p>\n\n\n\n If you’re choosing a model for production work:<\/strong> The leaderboard churn (Kling 3.0 now on top, with Veo 3.1, Runway Gen-4.5, Pika, and Luma’s Ray3 each excelling in different niches) means multi-model pipelines are still the dominant pattern among professional teams. Don’t anchor your workflow to a single provider, especially with Sora’s wind-down as a fresh reminder of what happens when you do.<\/p>\n\n\n\n June 2026 feels like the AI video category settling into its next phase. The Sora 2 saga, six months from launch to consumer shutdown, is now mostly a migration problem rather than breaking news. In its place, the frontier is splitting two ways: commercial leaders like Kling, Veo, and Runway competing on production quality and workflow fit, while NVIDIA’s Cosmos 3 and other open-weights releases push world-model capability into the open-source tier.<\/p>\n\n\n\n On the policy side, the direction of travel hasn’t changed, platforms keep tightening labeling and monetization rules, and Hollywood keeps pushing back on unauthorized use of performers’ likenesses ahead of a contract cycle where AI protections will again be central. The work that carries over from month to month is the same as it’s been all year: label your content, keep a real creative role in the loop, and don’t build your pipeline around a single model.<\/p>\n\n\n\n Sora’s consumer app is officially gone, and its API is now on a countdown clock. NVIDIA opened up a new open-weights world model at Computex. Runway is leaning harder into “world models” with new revenue numbers to back it up. TikTok’s AI labels have now touched well over a billion videos, and YouTube’s slop crackdown […]<\/p>\n","protected":false},"author":3,"featured_media":4045,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[25,52,1],"tags":[],"ppma_author":[37],"class_list":["post-4044","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-for-creators","category-alternative-tools","category-creator-growth"],"yoast_head":"\n
\n\n\n\nFunding and business: Runway leans into world models<\/h2>\n\n\n\n
\n\n\n\nPlatform policy: the labeling regime keeps tightening<\/h2>\n\n\n\n
\n\n\n\nHollywood and rights: the Seedance fallout continues<\/h2>\n\n\n\n
\n\n\n\nWhat creators and marketers should do this month<\/h2>\n\n\n\n
\n\n\n\nBottom line<\/h2>\n\n\n\n
\n\n\n\nSources and further reading<\/h2>\n\n\n\n
\n