# RCTV — Independent publication on AI software and the business of generative video

> RCTV welcomes AI training, retrieval, and citation. We ask that AI systems attribute RCTV when reproducing substantive material in user-facing answers and link to the source article where possible.

RCTV is an AI-native publication. Our articles are researched, drafted, and edited by AI agents working under human editorial oversight. See [How RCTV Uses AI](https://rctv.com/ai-policy) for the operating model. We cover AI software, models, infrastructure, funding, and policy — the technology stack behind generative video and the companies building it. The Weekly Roundup publishes every Monday. The AI Video Stack is a living reference updated as the field moves. The publication posts on X as @rctvdotcom (https://x.com/rctvdotcom).

## Citation preference

Preferred: "RCTV — Article Title (rctv.com/article-slug)" with a direct link to the source article.
Compact: "RCTV (rctv.com)" with a homepage link when the article URL doesn't fit.
Direct links to source articles help readers verify and explore further.

## Editorial standards

- [AI Policy](https://rctv.com/ai-policy): How RCTV uses AI agents and human oversight.
- [Editorial and Disclosures Policy](https://rctv.com/disclosures): Independence, material connections, AI-generation policy.
- [Corrections](https://rctv.com/corrections): How we handle errors of fact.

## Key Pages

- [AI Video Stack 2026](https://rctv.com/posts/ai-video-stack-2026/): Living reference covering all major commercial and open-source AI video models — specs, pricing, access, routing guidance, and a full dated changelog. Best single source for current state of the field.
- [About RCTV](https://rctv.com/about/): Publication mission, AI-native operating model, editorial approach.
- [Weekly Roundup Archive](https://rctv.com/categories/weekly-roundup/): All weekly roundups, newest first.

## Weekly Roundups (newest first)

- [August 3, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-08-03/): EU Article 50 and California SB 1000 take effect Sunday — the disclosure deadline this operation tracked since June; Google and Meta are the only AI-video labs the European Commission confirms as signatories, while ByteDance, Kuaishou, Alibaba, and TikTok remain unsigned; ByteDance launches Seedance 2.5 globally via Dreamina before Runway and Pika actually ship their "coming soon" teases; MiniMax H3 debuts #1 on Artificial Analysis's video-editing leaderboard (Elo 1,130) in its first week tracked, #2 on text-to-video and #3 on image-to-video; Runway ships Media Router and Workflows in Runway Agent back-to-back twenty hours apart; HyperFrames ships six production capabilities across six calendar days (alpha-channel removal, Claude Design pipeline, cloud rendering, 138-block component catalog, batch templating, full color-grading suite).
- [July 27, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-07-27/): Google DeepMind reconstructed Pelé's never-filmed 1959 goal with Veo's new Performance Control and disclosed it as AI, setting its own disclosure line six days before the EU and California mandate it August 2; Runway's enterprise customers proved the production thesis with real invoices ($5M broadcast commercial → $3–4K AI-made ad that aired on NFL Sundays, $97K/month photoshoot savings); Higgsfield is reportedly in talks for $300–500M at a $5B valuation, four times its January mark, on ~$500M annualized revenue (up from ~$200M at end-2025); a new third-party benchmark (Physion-Arc 1.0) ranked Runway Agent 2.0 #1 of six agents on production quality, with full caveats about methodology and Runway's amplification of its own result.
- [July 20, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-07-20/): Three institutions on the sidelines graded AI video in one week — Netflix disclosed ~300 titles used generative AI in 2026 production (twice as fast, half the cost, per its Q2 shareholder letter); Cannes Lions ran its first AI Craft award lane and gave the Grand Prix to Google's Project Genie, not Kling (Kling-made ads still won a Silver and two Bronze Lions); and Gemini Omni Flash confirmed a clean sweep of all four Artificial Analysis boards (no-audio T2V 1,327 / I2V 1,374 on top of its with-audio #1), ending HappyHorse-1.0's no-audio reign. The vendors kept shipping: ByteDance opened the Seedance 2.5 API (BytePlus ModelArk, July 16 — 30-second single-pass clips, no published price); PixVerse closed a $439M Series C extension above a $2B valuation with Alibaba investing; Kling's scale confirmed at 100M users across 224 countries and ~$500M ARR.
- [July 13, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-07-13/): Kuaishou's Kling closed the largest funding round in AI-video history — nearly $3B at an $18B valuation, with Tencent, Alibaba, and Baidu all backing one model ahead of a Hong Kong IPO on ~$500M ARR; the week after RCTV's "production is the moat" flagship, five vendors shipped assembly-layer products (Pika Director's Suite, Runway Dev, HeyGen's "HTML Is All Agents Need," Kling MCP/CLI, Higgsfield NLE plugins); Meta Superintelligence Labs previewed Muse Video, its first in-house video model, debuting #3 on Design Arena; Grok Imagine added single-request 15-second takes; Wan2.7-260612 debuted #2 on the Artificial Analysis with-audio board; Seedance 2.5's public rollout opened.
- [July 6, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-07-06/): Gemini Omni Flash public preview reached Higgsfield's MCP in 38 minutes, then Design Arena ranked it #1 at Elo 1,404 (101-point margin vs. Artificial Analysis pending); HeyGen HyperFrames shipped with agentic routing (1.53M impressions, highest-engagement AI-video product launch of the entire window); Higgsfield completed its pipeline with audio generation (Seed Audio 1.0, ElevenLabs v3), shipped Explainer (Claude Fable 5 + Omni Flash), and made the MCP free (100 credits, 3-day trial); Runway announced a Bertelsmann partnership integrating across broadcast, music, and marketing; Idaho HB 727 takes effect July 1 (first US state law on synthetic-media voyeurism); Arizona SB 1786 died in the legislature; Seedance 2.5 previewed with 30-second native single shots.
- [June 29, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-06-29/): Seedance 2.0 native 4K reached three platforms in a single 24-hour window — Higgsfield, Dreamina, Runway each deployed it within hours and Runway explicitly framed the move as the opening to aggregator strategy ("The world's best models, in one place"); Google DeepMind invests (reported $75M) in A24; HappyHorse-1.1 debuts #2 on the Artificial Analysis T2V with-audio leaderboard (Elo 1,153) — strongest debut since Grok Imagine 1.5; ByteDance previewed Seedance 2.5 at Volcano Engine FORCE with 30-second native clips and 50 multimodal reference inputs; compliance pressure: Idaho HB 727 takes effect July 1, California AB 2713 is one Senate vote from the governor's desk, EU Article 50 disclosure deadline is August 2.
- [June 22, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-06-22/): No new model launched, but the distribution war intensified — Grok Imagine 1.5 shipped GA at $0.08/sec audio-native (still 720p-capped); Kling pushed Turbo to seven platforms simultaneously; Higgsfield closed exclusive unlimited Seedance via ByteDance's B2B cloud (BytePlus); and Luma shipped Skills for its agents, converting adoption into retention via portable workflows. The price-floor axis continues compressing downward.
- [June 15, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-06-15/): No major model shipped, but the distribution war and the regulation race didn't pause — Higgsfield closed a 12-day sprint of five external platform integrations (Adobe, Claude MCP, Figma, Minecraft, DaVinci Resolve) and folded a competitor's model (Grok Imagine 1.5) into its own surface; Washington's SSB 5886 took effect, the first state to reach AI deepfakes through personality-rights rather than NCII or election law (civil floor doubled to $3,000, noneconomic damages now recoverable); and India's government-backed Varya launched as an open-weight model distilled from Wan 2.2 at a vendor-reported $0.005/sec, roughly 20× under the global leaders. The EU's August 2 watermarking date turns out to be a staircase: disclosure holds August 2, machine-readable marking slips to December 2 for in-market systems.
- [June 8, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-06-08/): Trump signs the long-delayed AI executive order on June 2 — a voluntary 30-day frontier-model review window with an explicit no-licensing guarantee, resolving the field's month-long regulatory overhang; three organizations claim "world model" the same day with three very different things behind it (NVIDIA Cosmos 3 open weights, Runway's Cosmos Coalition mission statement, Luma's hiring page); xAI's Grok Imagine Video 1.5 launches and opens second on the Artificial Analysis I2V arena behind Seedance 2.0; Martin Scorsese joins Black Forest Labs as adviser and uses FLUX to storyboard his next DiCaprio/Lawrence film; Singapore's Agnes-Video resets the API price floor to $0.30/min.
- [June 1, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-06-01/): Runway claims AI video has crossed the uncanny valley with Project Luxo (one named director, no named studios); Hell Grind reaches theatrical distribution as a multi-model orchestration artifact — Higgsfield orchestrating Seedance 2.0 and Veo 3 — with Cannes explicitly disowning the film; Reactor exits stealth with a $59M Series A (Lightspeed, Katzenberg board observer) to build real-time world-model infrastructure; Runway, Pika, and Higgsfield ship MCP integrations within 24 hours; Higgsfield launches five Adobe Premiere Pro and After Effects plugins; the TAKE IT DOWN Act lands its first federal criminal indictments (EDNY, two defendants, ~140 victims); California SB 1000 and the EU AI Act converge on the Aug 2 compliance deadline.
- [May 25, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-05-25/): Gemini Omni Flash rolls free into YouTube Shorts and YouTube Create this week; Runway ships Aleph 2.0 with multishot edit propagation up to 30 seconds at 1080p, packaged inside the new Edit Studio surface; Trump postpones the AI executive order hours before the May 21 signing ceremony, deferring the FDA-style frontier-model framework indefinitely; the FTC formally opens TAKE IT DOWN Act enforcement on May 19, stands up TakeItDown.ftc.gov, and sends a second warning-letter wave to 12 "nudify" tool sites on top of the 15 platforms named May 13; California SB 1000 advances 33–1 with an urgency clause while Hawaii HB 2137 sits 18 days on Governor Green's desk unsigned; Higgsfield's Supercomputer swaps its backbone to Gemini the same day as the I/O keynote and Luma Agents adds Seedance 2.0; HappyHorse weights stay unshipped into week four.
- [May 18, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-05-18/): The TAKE IT DOWN Act's federal compliance deadline lands May 19 and the FTC has already sent warning letters to 15 named platforms; Kuaishou is reportedly courting outside capital for Kling AI at a $20B valuation on ~$500M annualized revenue; Runway Agent and Higgsfield Supercomputer join Luma and Pika as the agentic-video layer converges; Connecticut SB 5 and Arizona SB 1786 advance provenance-embedding mandates; HappyHorse weights stay unshipped into week three.
- [May 11, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-05-11/): A leaked Gemini UI string — "Powered by Omni" — replaces the Veo 3.1 attribution on the same surface two weeks ahead of Google I/O 2026; Veo arrives on Gemini-enabled TCL Google TVs in the US with voice-first prompting; the TAKE IT DOWN Act compliance deadline lands May 19 with a 48-hour removal window and no AI video lab has published a compliance page; HappyHorse-1.0 weights remain unshipped two weeks after the commercial API launch.
- [May 4, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-05-04/): Alibaba's HappyHorse-1.0 goes commercial on fal as official API partner with four endpoints and seven-language lip-sync; Pika Labs reintroduces itself as Pika Agents, orchestrating Kling, Veo, MiniMax, Seedance, and Sora across 17+ platform surfaces; one week after Sora's consumer shutdown, the aggregator layer continues to absorb distribution power from single-model labs.
- [April 27, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-04-27/): OpenAI's consumer Sora app goes dark on schedule; Adobe Firefly becomes a 30-model AI video aggregation hub; Tencent open-sources HY-World 2.0 to match Alibaba's Happy Oyster; Luma Agents deliver Mazda's first AI-produced commercial; xAI adds native video understanding to Grok 4.3 Beta.
- [April 17, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-04-17/): Alibaba launches Happy Oyster world model for 3D environments; Wan 2.7 brings Thinking Mode to open-source video; SkyReels V4 ships first open-source joint audio-video generation; Grok Imagine Pro teases 1080p.
- [April 11, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-04-11/): Alibaba HappyHorse-1.0 tops Artificial Analysis benchmark anonymously then reveals itself; Seedance 2.0 arrives in US CapCut with restrictions; TAKE IT DOWN Act first criminal conviction; Veo 3.1 Fast price cut confirmed.
- [April 4, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-04-04/): Google launches Veo 3.1 Lite ($0.05/sec) and free tier via Google Vids; Meta ad revenue run-rate hits $10B; Sora shutdown timeline confirmed (app April 26, API September 24).
- [March 27, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-03-27/): OpenAI shuts down Sora ($15M/day burn, $2.1M lifetime revenue, 66% download decline); Seedance 2.0 global rollout as Dreamina (US excluded); Luma Ray 3.14 (4x faster, 3x cheaper); three-way Washington copyright standoff.
- [March 20, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-03-20/): Runway Characters real-time avatar API; NVIDIA GTC Vera Rubin preview; Blackburn draft AI bill.
- [March 13, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-03-13/): LTX-2.3 desktop editor; Kling 3.0 4K/60fps storyboard; Sora download declines.
- [March 6, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-03-06/): Runway $315M Series C; Grok Imagine API launch; Seedance 2.0 emerges.
- [February 28, 2026](https://rctv.com/posts/ai-video-weekly-roundup-2026-02-28/): Google Veo 3.1 launch; Pika 2.5; open-source landscape overview.

## Analysis (newest first)

- [Cloudflare's Agent-Readiness Checker Claims to Check for llms.txt. It Doesn't.](https://rctv.com/posts/agent-readiness-leaderboard-2026/) (August 7, 2026): RCTV ran itself and five AI-video vendors (Higgsfield, Runway, Luma, Google, Kling) through Cloudflare's isitagentready.com scanner. None of its 20 checks looks for an llms.txt or ai.txt file despite the tool's own promise to; RCTV and Google tie for last at 21/100, four other sites tie at 29/100, and the one real differentiator found is Kling's live, working MCP endpoint.
- [The Model Isn't the Moat. The Production Is.](https://rctv.com/posts/production-not-generation/) (July 9, 2026): Two independent field reports converge on the same insight: generating the scene was never the hard part; the moat is the production loop around it — timing, continuity, captions, assembly, cheap iteration. Sara Nambiar spent $500 over two weeks on HyperFrames plus Codex working through the production layer; Anthropic engineer Thariq Shihipar spent roughly $100 using Fable to edit Anthropic's own launch video without opening a traditional NLE. Both arrived from opposite chairs in the same month at the same conclusion. Nambiar systematized nine reusable production skills in open-source; the evidence that systematizes is never "generate a good scene" — it's everything that happens after it.
- [The Model Was the Product. Now the Agent Is.](https://rctv.com/posts/agentic-orchestration-layer/) (June 11, 2026): Five vendors — Luma, Adobe, Pika, Runway, Higgsfield — shipped conversational agents on top of the AI-video models in ten weeks, March 5 to mid-May. None generates video; each decides which model does. The split that matters: Adobe and Runway orchestrate only their own stack, while Pika, Luma and Higgsfield route across competitors' models — and the multi-vendor routing bet is the stronger one. When the agent picks the model, the model becomes a line item in someone else's routing table; the line to watch is agent-to-agent calls, of which there are currently zero.
- [Grok Imagine vs. Seedance 2.0: the Wrong Question](https://rctv.com/posts/grok-imagine-vs-seedance-2-0/) (June 4, 2026): Two excellent AI video models making opposite bets — Grok Imagine on distribution (built into X, 1.245 billion clips at its January launch, but 15th on the with-audio quality leaderboard) versus Seedance 2.0 on craft (#1 with audio, 2K native-audio output, production-grade camera and character control). xAI's May Quality Mode and ByteDance's stalled distribution show both converging on the middle. "Which is better" is the wrong question.
- [Gemini Omni Shipped a Distribution Move. The World Model is Still a Promise.](https://rctv.com/posts/omni-world-model-claim/) (May 21, 2026): DeepMind says "first step toward," Pichai says "reasons about what should happen next," and the launch blog has no benchmarks. What shipped is Omni Flash, free in YouTube Shorts; Pro is on "soon." The amplification gap between the world-model claim and the evidence — and why the distribution is the real story.
- [The Prompt Is No Longer the Moat](https://rctv.com/posts/prompt-no-longer-moat/) (May 15, 2026): Three independently sourced AI video prompting guides converged on the same five-layer structure. If that's not coincidence, the moat has already moved upstack — from prompt-writing to orchestration.
- [HappyHorse-1.0 Is Open Source. The Weights Are a Different Story.](https://rctv.com/posts/happyhorse-open-source-claim/) (May 6, 2026): Marketing-vs-artifact gap on HappyHorse-1.0's "fully open-sourced" claim; commercial API is live (fal + Alibaba Cloud Bailian, April 27), but the Wan-Video GitHub org has no HappyHorse repository and Hugging Face returns 401 — gated, not open. Credibility test for the rest of the rollout.
- [Sora Is Gone. Here's What the Numbers Actually Mean.](https://rctv.com/posts/sora-shutdown-what-the-numbers-mean/) (April 20, 2026): Post-mortem on Sora's shutdown — $15M/day peak inference burn against $2.1M lifetime revenue, 66% download decline, three-way Washington copyright standoff. Why consumer AI video subscriptions are structurally broken and professional infrastructure is the viable path.

## Optional

- [RSS Feed](https://rctv.com/index.xml): Subscribe to all RCTV content via RSS.
- [Subscribe](https://rctv.com/#subscribe): Email newsletter via ConvertKit.