Skip to main content
Featured image for AI Video Statistics 2026

AI Video Statistics 2026: 130+ Sourced Stats

Updated November 10, 2026 · 29 min readBot Memo

By: Editorial Staff

The AI video statistics on this page start with a shutdown. OpenAI announced Sora in February 2024 as a model that could generate a minute of video, shipped it that December capped at 20 seconds, relaunched it as Sora 2 in September 2025, and closed the consumer app on 26 April 2026.

Its Help Center set the API shutdown for 24 September 2026, and Sora no longer appears on OpenAI’s API price list.

Kuaishou’s Kling AI went the other way, with more than 850 million yuan of revenue in the second quarter of 2026 and a funding round of about 19.8 billion yuan completed on 30 October.

And since 2 August 2026, Article 50 of the EU AI Act has applied to AI-generated video, with machine-readable marking required from launch for new systems and from 2 December 2026 for those already on the market.

The figures below are grouped by what they measure, because forecasts, run rates, downloads and label views get mixed up constantly. Each stat carries its source and date.

On this page

AI Video Statistics: Market Size, Funding And Valuations

Three Forecasts Size Two Different Markets

The most common error in AI video coverage is treating these forecasts as versions of one number. They cover different things.

1. Grand View Research values the broad AI video market at $4.6 billion in 2025, growing from $5.5 billion in 2026 to $42.3 billion by 2033, a 33.7% compound annual rate from 2026 to 2033. That scope covers AI video generators plus enhancement, editing and analysis tools. (Grand View Research, 2026)

2. The same firm sizes the narrower AI video generator market at $788.5 million in 2025, reaching $3,441.6 million by 2033 at a 20.3% rate from 2026 to 2033. (Grand View Research, 2026)

3. Fortune Business Insights puts the generator market at $847 million in 2026, reaching $3,350 million by 2034 at 18.8%. (Fortune Business Insights, 2026)

4. This page previously carried Grand View Research’s earlier edition of the generator report, which projected a 19.9% compound rate from 2024 to 2030. The current edition runs 2026 to 2033 at 20.3%. (Grand View Research, 2024; Grand View Research, 2026)

5. Grand View Research’s 2024 generator report also described significant merger and acquisition activity in the AI video generator industry. The firm gave no deal count for that claim. (Grand View Research, 2024)

The two generator forecasts land within about 3% of each other, a year apart. The gap between them and the broad AI video figure is a difference of definition. The revenue disclosed by operating companies further down is firmer than any of these forecasts.

Three blocks showing three terminal-year AI video forecasts. Grand View Research forecasts 42.3 billion dollars for the broad AI video market by 2033, which covers generators plus enhancement, editing and analysis tools. The same firm forecasts 3.44 billion dollars for the narrower AI video generator market by 2033. Fortune Business Insights forecasts 3.35 billion dollars for the AI video generator market by 2034.

Deepfake Market Forecasts Disagree By An Order Of Magnitude

AI video and deepfakes overlap but are separate markets, and the two published deepfake estimates cannot both be right.

6. Spherical Insights & Consulting sized the global deepfake AI market at $6.83 billion in 2023 and forecast $119.34 billion by 2033, a 33.12% compound rate from 2023 to 2033. (Spherical Insights, March 2024)

7. In the same report, video deepfakes are expected to hold the largest segment share over the forecast period. (Spherical Insights, March 2024)

8. Asia Pacific is expected to hold the largest regional share, while North America is expected to grow fastest. (Spherical Insights, March 2024)

9. MarketsandMarkets sizes the same market at $0.53 billion in 2024 and $0.85 billion in 2025, reaching $7.27 billion by 2031 at a 42.8% rate. Its starting figure is roughly a thirteenth of the Spherical Insights base, and the terminal figures differ about sixteenfold. (MarketsandMarkets, July 2025 edition)

Both reports sit behind paywalls, so neither forecasting method can be inspected.

Side-by-side comparison of two deepfake AI market forecasts. Spherical Insights: 6.83 billion dollars in 2023, 119.34 billion dollars by 2033, 33.12 percent compound annual growth. MarketsandMarkets: 0.53 billion dollars in 2024, 7.27 billion dollars by 2031, 42.8 percent compound annual growth.

2025 Funding Nearly Doubled To $3.08 Billion

10. Global funding for AI-related video companies reached $3.08 billion in 2025, up a reported 94.6% from $1.58 billion in 2024. (Press report, February 2026)

That is the credible growth benchmark for the category. The “840% AI video growth” figure that circulates on vendor blogs has no traceable primary source and is not used here. For where generative video sits inside the wider picture, see our AI startup funding statistics.

Kling AI’s Round Is The Largest On This Page

11. Kling AI secured about 19 billion yuan ($2.8 billion) of a roughly $3 billion target round in early July 2026. Kuaishou said on 30 October 2026 that the round had completed at about 19.8 billion yuan, published as $2.98 billion and worth about $2.9 billion at July’s implied exchange rate. Either conversion is more than three times Luma’s $900 million Series C. (Press report, July 2026; Press report, October 2026)

12. Published accounts of the July deal put the pre-money valuation at $15 billion and the post-round valuation at $18 billion, down from a $20 billion target reported in May 2026. (Press report and Press report, July 2026; Press report, May 2026)

13. The July terms gave investors a 16.7% stake. At completion on 30 October 2026, Kuaishou held two-thirds of the equity and 52.45% of the voting rights. Alibaba and Tencent invested in July, and units of Kingdee International and MetaX, along with Gongqingcheng Junke, came in as new investors in the final round, partly taking over other investors’ commitments. (Press report, July 2026; Press report, October 2026)

14. Kling has reportedly picked CICC, Goldman Sachs and UBS for a Hong Kong listing of at least $1 billion, as early as 2027. No filing has been reported. (Press report, October 2026)

The completed round came within about $100 million to $180 million of the entire prior year of global funding for AI-related video companies, depending on the yuan conversion.

Runway, Luma And Synthesia Absorbed The Western Capital

15. Runway has announced three rounds on its own site: a $141 million Series C extension in June 2023, “over $300M” in Series D led by General Atlantic in April 2025, and a $315 million Series E in February 2026. Runway states no valuation for any of them. (Runway, 2023; Runway, 2025; Runway, 2026)

16. The Series E was led by General Atlantic with participation from NVIDIA, Adobe Ventures, AllianceBernstein, AMD Ventures, Fidelity Management & Research Company, Mirae Asset, Emphatic Capital, Felicis and Premji Invest. (Runway, February 2026)

17. The Series E was reported at a $5.3 billion valuation, a figure Runway’s own announcement does not state. (Press report, February 2026)

18. Runway’s total funding stands at roughly $1.05 billion including the Series E. (Research estimate, 2026)

19. Luma AI announced a “900M” Series C on 19 November 2025, led by HUMAIN with significant participation from AMD, alongside Andreessen Horowitz, Omniva, Amplify Partners and Matrix Partners. The round came with a plan to build a 2 gigawatt compute supercluster, Project Halo, with HUMAIN. (Luma AI, November 2025)

20. Luma did not state a valuation. Press coverage put it at roughly $4 billion. Luma’s announcement also gives a headcount of over 130. (Luma AI, November 2025)

21. Synthesia raised a $200 million Series E at a $4 billion valuation, led by GV, on 26 January 2026. (Synthesia, January 2026)

22. On 31 March 2026 Runway launched a $10 million fund for pre-seed and seed startups in AI, media and world simulation, alongside a Builders programme open to startups from seed through Series C. (Press report, March 2026)

Ranked bars of AI video funding rounds as stated by the companies: Luma Series C 900 million in November 2025; Runway Series E 315 million dollars in February 2026; Runway Series D over 300 million dollars in April 2025; Synthesia Series E 200 million dollars in January 2026; Runway Series C extension 141 million dollars in June 2023.

For how these valuations compare across generative AI, see our generative AI market stats, and for the companies behind the models, our list of top AI video generation startups.

AI Video Platform Usage And Adoption

Sora Shipped Shorter Than It Was Announced

23. OpenAI announced Sora on 15 February 2024, describing its largest model as “capable of generating a minute of high fidelity video”. (OpenAI, February 2024)

24. At announcement Sora could sample widescreen 1920×1080 video, vertical 1080×1920 video and everything in between. (OpenAI, February 2024)

25. The product that shipped on 9 December 2024 generated video up to 1080p and up to 20 seconds long, in widescreen, vertical or square formats. (OpenAI, December 2024)

26. ChatGPT Plus subscribers got Sora at no extra cost, with up to 50 videos a month at 480p or fewer at 720p. The Pro plan offered 10 times more usage, higher resolutions and longer durations. (OpenAI, December 2024)

A minute at announcement became 20 seconds at launch. That gap between announced and shipped capability recurs across this category.

Sora 2 Launched Faster Than ChatGPT, Then OpenAI Closed It

27. OpenAI released Sora 2 on 30 September 2025 with an invite-only iOS app in the US and Canada. OpenAI called the original Sora “the GPT-1 moment for video” and said Sora 2 “may be the GPT-3.5 moment”, which is the company’s own framing. (OpenAI, September 2025)

28. The app passed 1 million downloads in under five days, faster than ChatGPT, despite the invite gate. In their first seven days Sora took 627,000 iOS downloads against ChatGPT’s 606,000, on Appfigures estimates. (Press report, October 2025)

29. The Android app followed on 4 November 2025, 35 days after iOS. (Press report, November 2025)

30. Cumulative installs reached 9.6 million by January 2026, but consumer spending across that base was $1.4 million, about 15 cents per install. (Press report, January 2026)

31. Downloads fell 32% month over month in December 2025, then 45% to 1.2 million installs in January 2026. (Press report, January 2026)

32. Sora slipped from first on the App Store free chart in October 2025 to 126th. (Press report, March 2026)

33. Users peaked near 1 million and fell to around 500,000, while the service cost around $1 million a day to run, the Wall Street Journal reported. Appfigures estimated about $2.1 million of in-app purchase revenue across its whole life. (Press report and Press report, March 2026)

34. OpenAI’s Help Center states that the Sora web and app experiences were discontinued on 26 April 2026 and dates the end of the Sora API to 24 September 2026. Sora no longer appears on OpenAI’s API pricing page, and OpenAI’s primary pages give no reason for the decision. (OpenAI Help Center and OpenAI, 2026)

Timeline of OpenAI's Sora: announced in February 2024 as able to generate a minute of video; launched in December 2024 at up to 20 seconds and 1080p; Sora 2 released with an invite-only iOS app in September 2025; the app and web experience discontinued on 26 April 2026; the API discontinued on 24 September 2026.

Sora is the sharpest single data point yet on how fast consumer retention can collapse in this category. It is one product, so it does not settle the question on its own. The revenue in the next section comes from paid plans and business contracts, and Kling’s mix has leaned toward consumers.

Veo And Flow Passed 500 Million Videos

35. Google announced Veo 2 on 16 December 2024, stating it could deliver resolutions up to 4K and extend to minutes in length. That was a stated capability at announcement, not what shipped that day. (Google, December 2024)

36. Veo 3, announced on 20 May 2025, was the first Veo model to generate video with audio. (Google, May 2025)

37. Over 40 million Veo 3 videos were generated across the Gemini app and Flow in the seven weeks to 10 July 2025. That count covers two surfaces combined. (Google, July 2025)

38. By late July 2025 Veo 3 had generated over 70 million videos globally. (Google Cloud, July 2025)

39. Enterprise customers generated more than 6 million videos on Vertex AI between the June 2025 preview and general availability at the end of July. (Google Cloud, July 2025)

40. Google’s Flow filmmaking tool passed 275 million videos generated within five months of launch, and Google put the count at over 500 million on 25 November 2025. (Google, October 2025; Google, November 2025)

41. Veo 3.1’s Extend feature lets users build videos “lasting for a minute or more” by continuing an original clip. That is stitching, not a single generation. (Google, October 2025)

42. By February 2026 users had created over 1.5 billion images and videos in Flow. That figure counts images and videos together, so it is a different unit from the video counts above. (Google, February 2026)

43. Veo 3.1 Lite, released on 31 March 2026, was priced at launch at less than 50% of Veo 3.1 Fast at the same speed and generates clips of 4, 6 or 8 seconds at 720p or 1080p. The Fast tier’s price has since been cut, so the current gap is 50% at 720p. (Google, March 2026)

44. At I/O in May 2026 Google said Flow had launched in over 140 countries. It published no new cumulative Flow video count at that event, so November 2025’s figure of over 500 million remains the newest video-only count. The same event introduced Gemini Omni, and Google’s developer documentation lists 22 October 2026 as the earliest shutdown date for its Veo 3.1 preview models, with Omni named as the replacement.

Veo 3.1 remains available through Google Cloud. (Google, May 2026; Google AI for Developers, 2026; Google Cloud, 2026)

Three blocks of Google usage figures that are not comparable. Over 275 million videos generated in Flow in its first five months, to October 2025, videos only. Over 40 million Veo 3 videos in seven weeks to July 2025, across the Gemini app and Flow combined. Over 1.5 billion images and videos created in Flow by February 2026, images and videos counted together.

For how generative video is landing in production workflows, see our AI filmmaking statistics.

Kling AI Reports 60 Million Creators And 600 Million Videos

45. As of December 2025 Kling served more than 60 million creators, had generated over 600 million videos, and counted more than 30,000 enterprise users. (Kuaishou, January 2026)

46. Kuaishou launched the Kling AI 3.0 series in February 2026. (Kuaishou, May 2026)

47. Kuaishou calls Kling AI 3.0 “the industry’s first video generation model supporting native 4K output”, a vendor claim with no independent benchmark attached. Kling 4.0, announced on 29 September 2026, doubled the maximum clip length to 30 seconds. (Kuaishou, August 2026; Press report, September 2026)

48. The 412.3 million daily and 797.3 million monthly active users in Kuaishou’s second-quarter release belong to the Kuaishou app, not to Kling. Kuaishou publishes no separate Kling subscriber count. (Kuaishou, August 2026)

Runway And Luma Compete On Model Claims

49. Runway launched Gen-4 on 31 March 2025 and Gen-4.5 on 1 December 2025. (Runway, December 2025)

50. Gen-4.5 took first place on the Artificial Analysis text-to-video leaderboard at 1,247 Elo, ahead of Veo 3 at 1,226, Kling 2.5 at 1,225 and Sora 2 Pro at 1,206, per Runway’s launch post. Newer models have since entered that leaderboard, so the ranking is a snapshot. (Runway, December 2025)

51. Runway’s GWM Worlds 2 research model, published on 3 September 2026, generates continuous 720p video at 24 frames per second with audio at 48,000 Hz. (Runway, September 2026)

52. Luma describes Ray3, released on 18 September 2025, as “the world’s first reasoning video model” and the first to generate 16-bit HDR video. The same page gives two different Draft Mode speedups, “up to 20x” and “up to 10 times” faster. (Luma AI, September 2025)

53. Luma’s current API model is Ray3.2. Luma publishes no Dream Machine user count on its own site. (Luma AI, 2026)

Runway calls Gen-4.5 the world’s top-rated AI video generation model, Luma calls Ray3 the world’s first reasoning video model, and Kuaishou calls Kling 3.0 the first with native 4K. All three are vendor assertions.

AI Video Revenue, ARR And Cost Per Second

Runway Says Its Run Rate Doubled In Five Months

54. Annual recurring revenue reached $200 million in September 2026, roughly double the $100 million the company was at in April 2026. Co-chief executive Anastasis Germanidis stated the figure on 8 September 2026. (Press report, September 2026)

55. Runway says its net revenue retention runs above 300%, meaning existing customers roughly triple their spending. (Press report, September 2026)

The April 2026 figure is inferred from Runway’s statement that recurring revenue doubled between April and September.

Kling Grew Revenue Fast And Lost Money Doing It

56. Kling’s annualized run rate was $100 million in March 2025 and $240 million in December 2025. Kuaishou defines that run rate as monthly operating revenue multiplied by twelve. (Kuaishou, January 2026)

57. It reached approximately $500 million in March 2026, disclosed with Kuaishou’s first-quarter results. (Kuaishou, May 2026)

58. Full-year 2025 revenue was 1.04 billion yuan, about $150 million, which was 0.7% of Kuaishou’s total revenue. (Press report, March 2026)

59. First-quarter 2026 revenue exceeded 650 million yuan, about $90 million, up more than 300% year on year. (Kuaishou, May 2026)

60. In the second quarter of 2026 Kling AI generated revenue of over 850 million yuan, up more than 200% year on year. The figure is quarterly revenue, unaudited, and not an ARR. (Kuaishou, August 2026)

61. Nearly 70% of Kling’s revenue in the first quarter of 2025 came from consumer subscriptions. That split is a year and a half old, so read it as a snapshot. (Press report, May 2025)

62. Kling posted net losses of 500 million yuan in 2024 and 1.9 billion yuan in 2025, Caixin reported. (Press report, July 2026)

Runway’s $200 million is annual recurring revenue. Kling’s $500 million is a run rate taken from the single month of March 2026. Second-quarter revenue of at least 850 million yuan annualizes to at least 3.4 billion yuan, roughly level with the March run rate at mid-2026 exchange rates.

MiniMax And Synthesia Disclose On Different Bases

63. MiniMax, listed in Hong Kong, reported revenue from its AI-native products of $42.6 million for the six months to 30 June 2026, up 100.9% from $21.2 million a year earlier. That bucket includes Hailuo AI and other products; MiniMax does not break out Hailuo video revenue. (MiniMax, August 2026)

64. MiniMax’s total first-half revenue rose 283.1% to $116.6 million, more than its full-year 2025 revenue of $79.0 million. The figures are unaudited. (MiniMax, August 2026)

65. Synthesia says it scaled from $40 million to $140 million of annual recurring revenue over three years. The figure is self-reported and unaudited, published in an engineering blog post. (Synthesia, February 2026)

Four blocks of AI video revenue disclosures on different bases. Runway: 200 million dollars of annual recurring revenue in September 2026. Kling AI: more than 850 million yuan of revenue in the second quarter of 2026. Synthesia: 140 million dollars of annual recurring revenue, self-reported in February 2026. MiniMax: 42.6 million dollars of revenue from AI-native products in the first half of 2026, a bucket that includes Hailuo.

What A Second Of AI Video Costs

Published API rates are the one cost figure in this category that can be checked directly.

66. Runway charges $0.05 per second for gen4_turbo and $0.12 per second for gen4.5, billed in credits at one cent each. The same page lists third-party models from 2 credits to 150 credits per second, or $0.02 to $1.50. (Runway, 2026)

67. Google prices Veo by model and resolution, with these rates covering video with synchronized audio. Veo 3.1 and Veo 3 both run $0.40 per second at 720p and 1080p, the Fast tier of each runs $0.10 at 720p, and Veo 3.1 Lite runs $0.05 at 720p. Veo 3.1 with audio runs $0.60 per second at 4K, and Veo 2, priced for video without audio, lists at $0.50 at 720p. (Google Cloud, 2026)

68. Luma prices Ray3.2 per clip, which works out to $0.03 per second at 540p and $0.24 at 1080p on a five-second clip, and $0.045 and $0.36 on a ten-second clip. HDR output is listed at double the standard rate. (Luma AI, 2026)

69. OpenAI listed Sora 2 at $0.10 per second at 720p and Sora 2 Pro from $0.30 to $0.70 depending on resolution, until the API closed. Sora no longer appears on that pricing page. (OpenAI, rate card as listed to September 2026)

Ranked bars of selected published per-second AI video API prices at 1080p and below in November 2026. Google's Veo 3.1 and Veo 3 cost 40 cents per second with audio at 720p or 1080p, Luma Ray3.2 at 1080p costs 24 cents on a five-second clip, Runway gen4.5 costs 12 cents, the Fast tier of Veo 3.1 and Veo 3 costs 10 cents at 720p, OpenAI's retired Sora 2 API was listed at 10 cents at 720p, Runway gen4 turbo costs 5 cents, Veo 3.1 Lite costs 5 cents at 720p, and Luma Ray3.2 at 540p costs 3 cents on a five-second clip. Veo 3.1 with audio costs 60 cents at 4K, outside the range shown.

A minute of generated video, before retries and editing, costs $1.80 at the lowest rate in that chart, Luma’s 540p price on five-second clips. At Veo 3.1’s 1080p rate it costs $24, and at its 4K rate $36. The percentage savings quoted around the web come from video tool vendors’ own marketing, so this page does not repeat them.

AI Video In Marketing, Advertising And Training

Generative AI Creative Is Heading From A Fifth To A Third

Ad buyers give the cleanest adoption signal in this category, because they report on budgets.

70. IAB’s 2025 report put generative AI at 22% of video ad creative in 2024. Buyers surveyed that year expected about 30% in 2025 and 39% to 40% by 2026. (Press report and Press report, July 2025; IAB, July 2025)

71. The same report found 86% of digital video ad buyers either used or planned to use generative AI for video ad creative. (IAB, July 2025)

72. IAB’s 2026 edition, published on 14 July 2026, reports that one third of buyers’ ad assets will use generative AI this year, up from one fourth in 2025, with 43% projected by 2027. (IAB, July 2026)

73. Nearly two in three buyers now use generative AI for digital video creative, up from half in 2025. (IAB, July 2026)

74. 40% of buyers want humans in the loop, rising to 50% among small and medium spenders. (IAB, July 2026)

75. Among smaller buyers, 96% are not satisfied with their current level of generative AI use for creative ad production. (IAB, July 2026)

76. US digital video ad spend is projected to pass $80 billion in 2026 on 11% year-on-year growth, nearly 20% faster than the total ad market. (IAB, 2026)

Ranked bars showing the share of digital video ad assets built or enhanced with generative AI rising from 22 percent in 2024 to a quarter in 2025, with buyers expecting a third in 2026 and IAB projecting 43 percent by 2027.

Buyers now expect a third of ad assets to use generative AI in 2026, down from the 39% to 40% they projected a year earlier.

What Video Marketers Report Using

Wyzowl is an explainer-video production agency, so it sells into the market it surveys. Its 2026 edition drew on 266 respondents split into marketers and consumers, fielded in late 2025, and it overwrites one URL each year, so earlier editions cannot be read at the link.

77. 93% of video marketers see video as an important part of their overall strategy, against the 88% this page cited from the 2024 edition. (Wyzowl, 2024 and 2026)

78. 92% plan to spend around the same or more on video marketing in 2026, against 85% planning to maintain or increase spend in the 2024 edition. (Wyzowl, 2024 and 2026)

79. 84% of consumers want to see more video from brands in 2026, against 89% in the 2024 edition. Wyzowl’s own series has stayed between 83% and 91% since 2018, so it is not a trend. (Wyzowl, 2024 and 2026)

80. 67% of marketers who do not use video plan to start in 2026, against 68% in the 2024 edition. Non-users are a small minority of respondents, so this is directional only. (Wyzowl, 2024 and 2026)

81. 63% of video marketers say they have used AI video tools to create or edit marketing videos, up from 51% in the 2025 edition.

This page previously cited 75% from the 2024 edition; secondary write-ups of the 2025 edition quote the same 75% as the prior-year figure, which would make Wyzowl’s series 75%, then 51%, then 63%. (Wyzowl, 2024 to 2026)

82. 82% of video marketers say video has given them a good return, and 91% of businesses use video. (Wyzowl, 2026)

An Animation Studio’s Survey On AI Versus Handcrafted Video

Broadcast2World, a studio that sells handcrafted animation, ran an ungated 2024 marketing survey. It publishes no sample size, field dates or sampling frame.

83. 90% of businesses felt handcrafted videos had an edge over AI video productions. (Broadcast2World, 2024)

Training Video Is Where Vendors Report Clear Outcomes

The three figures below come from a November 2023 survey of 102 of Synthesia’s own customers working in learning and development, most at companies with 1,000 or more employees. Synthesia sells AI video, so this is a vendor asking buyers whether its product works.

84. 68% of respondents saw a positive impact on learning satisfaction scores from using AI video in training courses. (Synthesia, November 2023)

85. 57% saw a positive impact on course completion rates. (Synthesia, November 2023)

86. 60% saw a positive impact on average time to completion. (Synthesia, November 2023)

87. Synthesia lists 240+ AI avatars and one-click translation into 140+ languages, as of November 2026. This page previously gave 160+ as an avatar count; Synthesia now lists 240+ avatars. (Synthesia and Synthesia, 2026)

AI Video Provenance, Watermarks And Labelling

What AI Video Is Marked With Today

88. OpenAI’s Sora 2 system card committed to C2PA metadata on all assets, a visible moving watermark on videos downloaded from sora.com or the Sora app, and internal detection tools. (OpenAI, September 2025)

89. OpenAI’s current provenance page lists video in its content-type table only with the note “Sora is discontinued”, and says of metadata that “platforms, editing tools, and file conversions may remove it.” (OpenAI Help Center, 2026)

90. Over 20 billion pieces of AI-generated content had been watermarked with Google’s SynthID by November 2025. That count covers all AI-generated content SynthID has marked, not AI video alone. (Google, November 2025)

91. Every Veo 2 output carries an invisible SynthID watermark, Google said at launch. (Google, December 2024)

92. SynthID verification in the Gemini app had been used 50 million times globally by May 2026. That counts checks, a different measure from the 20 billion watermarked items. (Google, May 2026)

93. Runway shipped Gen-3 Alpha with C2PA provenance standards. (Runway, June 2024)

The C2PA Standard Reached Live Video

94. More than 6,000 members and affiliates have live applications of Content Credentials, the open standard published by the C2PA, per the standards body’s February 2026 release. (C2PA, February 2026)

95. Content Credentials version 2.3 enabled the standard for live video, covering broadcast and streaming content. (C2PA, February 2026)

96. TikTok moved up to the C2PA Steering Committee on 27 July 2026. (C2PA, July 2026)

Platforms Label AI Video, And Meta Publishes The Counts

97. YouTube requires creators to disclose when they use AI to meaningfully alter or generate photorealistic content. Creators who consistently fail to disclose may get a label applied manually, content removed or suspension from the YouTube Partner Program. (YouTube Help, 2026)

98. YouTube says disclosing AI content won’t limit a video’s audience or its eligibility to earn money. (YouTube Help, 2026)

99. On 10 March 2026 YouTube expanded likeness detection to a pilot group of government officials, journalists and political candidates, with the caveat that detection does not guarantee removal. (YouTube, March 2026)

100. From 1 to 29 October 2024, Facebook recorded over 380 billion user views of its “AI info” label on organic content and over 26 million user label actions. (Meta Transparency Center, February 2025)

101. Instagram recorded over 1 trillion label views and over 60 million label actions in the same window, and Threads over 150 million label views. These are label views, not counts of AI content or removals, and Meta says its labelling method is still evolving. (Meta Transparency Center, February 2025)

102. TikTok added a slider in November 2025 letting users tilt their For You feed away from AI content. It reduces AI content rather than removing it. TikTok also said it uses C2PA content credentials to identify AI content and would start adding invisible watermarks to AI content made with its own tools and to content uploaded with those credentials. (Press report, November 2025)

EU AI Act Article 50 Applies From 2 August 2026

103. The EU AI Act, Regulation (EU) 2024/1689, was published in the Official Journal on 12 July 2024 and entered into force on 1 August 2024. Article 50 applies from 2 August 2026. (EUR-Lex, 2024)

104. Article 50(2) requires providers of AI systems generating synthetic audio, image, video or text to ensure outputs are “marked in a machine-readable format and detectable as artificially generated or manipulated.” (EUR-Lex, 2024)

105. Article 50(4) requires deployers of a system that generates or manipulates deep fake image, audio or video content to disclose that it has been artificially generated or manipulated, with a lighter disclosure duty for evidently artistic or satirical work. (EUR-Lex, 2024)

106. Systems placed on the market before 2 August 2026 have until 2 December 2026 to meet the marking obligation, and the grace period covers that obligation only. The Digital Omnibus on AI, Regulation (EU) 2026/1744, wrote the date into the AI Act. (Cuatrecasas, July 2026; European Commission, July 2026)

107. By the end of July 2026, about 190 organisations had signed the Commission’s voluntary Code of Practice on transparency of AI-generated content. (European Commission, July 2026)

Two Major Labels Settled, Sony Escalated

108. Universal Music Group settled its AI copyright suit against Udio on 29 October 2025 and signed licensing agreements covering recorded music and publishing. (Press report, October 2025)

109. Warner Music Group settled on 19 November 2025 and signed its own licensing deal. (Press report, November 2025)

110. Sony Music filed a second suit against Udio in July 2026 covering 30,117 sound recordings, with potential damages reported at over $4.5 billion. (Press report and Press report, July 2026)

Disney Signed A $1 Billion Sora Deal, Then Walked Away

111. On 11 December 2025 Disney and OpenAI announced a $1 billion equity investment with warrants and a three-year licence giving Sora access to more than 200 animated, masked and creature characters across Disney, Marvel, Pixar and Star Wars. Talent likenesses and voices were excluded. (OpenAI, December 2025)

112. That investment never closed. It was announced subject to definitive agreements, and when OpenAI shut Sora down Disney dropped both the stake and the licence. (Press report, March 2026)

Hollywood Sued MiniMax, And US Takedown Rules Took Effect

113. On 16 September 2025 Disney, Warner Bros. Discovery and NBCUniversal sued MiniMax in the US District Court for the Central District of California over its Hailuo AI video and image generation. (Press report, September 2025)

114. Judge Stanley Blumenfeld denied MiniMax’s motion to dismiss on 22 May 2026. (Loeb & Loeb, June 2026)

115. Under the TAKE IT DOWN Act, covered platforms had until 19 May 2026 to provide a removal path for non-consensual intimate imagery and to remove reported content within 48 hours.

The FTC sent compliance letters to 15 platforms on 11 May and warning letters to 12 companies offering nudify tools on 20 May, and civil penalties run to $53,088 per violation. (Federal Trade Commission and Federal Trade Commission, May 2026)

AI Video Detection And Video Analysis

A Fifth Of A New YouTube Shorts Feed Is AI-Generated

116. Of the first 500 Shorts shown to a brand-new YouTube account, 104, or 21%, were AI-generated in a sample run by a video editing company. The data was collected in October 2025 and published on 28 November 2025. (Kapwing, November 2025)

117. In the same sample, 165 videos, or 33%, were classified as low-quality “brain rot” content. (Kapwing, November 2025)

People And Detectors Cannot Reliably Spot AI Video

118. Across 1,276 participants tested on synthetic images, video, audio and audiovisual stimuli, accuracy on video alone was 50.7%, against 50% for a coin toss. The mean across all modalities was 51.2%. (Communications of the ACM, 2025)

119. The DF26 benchmark, a September 2026 preprint, found that both human performance and state-of-the-art deepfake detectors are close to random chance on AI-generated video. Its dataset holds 271 real and 2,420 synthetic public-speaking videos from seven modern video models. (arXiv, September 2026)

120. A meta-analysis pooling 56 papers and 86,155 participants put overall human accuracy at 55.54%, and 57.31% for video. The 95% confidence intervals crossed 50%, so the result cannot be separated from chance. (Computers in Human Behavior Reports, 2024)

121. A University of Florida study found the opposite split, with participants correctly identifying real and fake videos about two thirds of the time while performing at chance on still pictures of faces. (University of Florida, 2026)

122. The RobustSora preprint found detectors partly key off visible watermarks: manipulating the watermark shifted accuracy by -9.4 to +1.6 percentage points, a mean change of 6.6 points. (arXiv, 2025)

Human detection ranges from chance to about two thirds correct, depending on the study and what is shown. And the two safeguards each have a documented weak point: provenance metadata can be lost in ordinary handling, by OpenAI’s own account, while detectors on modern video perform near chance and partly read visible marks rather than generated pixels.

Video Analysis Reads Footage As Well As Making It

123. You Only Look Once, published at CVPR 2016, is a real-time object detector whose base network runs at 45 frames per second, with a fast version above 150. (Redmon et al., CVPR, 2016)

124. The YOLO paper shows the model generalises to new domains, degrading less than DPM and R-CNN when tested on artwork. It was built to mimic how people glance at an image and instantly know what objects are in it. (Redmon et al., CVPR, 2016)

125. Viso.ai, which sells a computer vision platform, describes object classification in security cameras as a way to detect hazardous or suspicious objects. (Viso.ai, October 2023)

126. Google used its own AI model to analyse 8,000 campaigns from top advertisers on YouTube, finding connections between videos from Brazil, France, Germany, India, Japan, Korea, the UK and the US. This is Google analysing ads on its own platform, so it is first-party research without a control group. (Think with Google, June 2024)

127. The same analysis found brands using storytelling to convey shared experiences and human connection. (Think with Google, June 2024)

128. People are more likely to trust a brand when a familiar creator uses it. In one Lululemon campaign featuring the YouTuber Casey Neistat, the comments indicate Neistat improved viewers’ perception of the product; the video drew more than 200 comments. (Think with Google, June 2024)

The Research That Made Generated Video Possible

129. MAV3D (Make-A-Video3D), from the Meta AI Make-A-Video team, generates three-dimensional dynamic scenes from text descriptions. (arXiv, January 2023)

130. VideoPoet, from Google and Carnegie Mellon University, is a large language model for zero-shot video generation that can extend a one-second clip to ten seconds. (arXiv, December 2023)

131. AnimateAnything animates static images with fine-grained motion area and motion strength guidance. (arXiv, November 2023)

132. Dreamix, from Google Research and The Hebrew University of Jerusalem, showed that video diffusion models work as general video editors. (arXiv, February 2023)

133. OpenAI’s Bill Peebles said in February 2024 that there would be “a new class of entertainment experiences” and that “the line between video game and movie might be more blurred.” That is an OpenAI executive’s view, in a piece that is otherwise a teardown of Sora’s flaws. (Press report, February 2024)

Key Takeaways

Sora’s consumer app did not last. Sora 2 passed 1 million downloads faster than ChatGPT, and the app closed about seven months later after installs fell 45% month over month in January 2026.

Paid subscriptions and business customers carry the category. Runway reached $200 million of annual recurring revenue in September 2026, and Kling reported over 850 million yuan of second-quarter revenue, with consumer subscriptions making up nearly 70% of its revenue when it last published that split, for the first quarter of 2025.

The largest round on this page went to China. Kling completed a round of about 19.8 billion yuan on 30 October 2026, published as $2.98 billion, close to the $3.08 billion raised by all AI-related video companies in 2025.

Growth and profit are separate questions. Kling lost 1.9 billion yuan in 2025 while its revenue climbed.

Labelling now has law behind it. Article 50 of the EU AI Act has applied since 2 August 2026, systems already on the market have until 2 December 2026 to add marking, and C2PA reports more than 6,000 members and affiliates with live Content Credentials applications.

Detection has not caught up. Humans and state-of-the-art detectors perform close to chance on modern AI video in the DF26 benchmark.

Advertisers have lowered their adoption expectations. Buyers now expect a third of ad assets to use generative AI in 2026, down from the 39% to 40% projected a year earlier.

Frequently Asked Questions

How Big Is The AI Video Market In 2026?

It depends which market. Grand View Research sizes the broad AI video market at $5.5 billion in 2026, growing to $42.3 billion by 2033. The narrower AI video generator market is much smaller: $788.5 million in 2025 per the same firm, or $847 million in 2026 per Fortune Business Insights.

Company revenue is firmer. Kling reported over 850 million yuan in the second quarter of 2026, and Runway reached $200 million of annual recurring revenue in September 2026.

Is Sora Still Available?

No. OpenAI’s Help Center states that the Sora app and web experience were discontinued on 26 April 2026 and dates the end of the Sora API to 24 September 2026. Sora no longer appears on OpenAI’s API pricing page, and OpenAI’s own pages give no reason for the shutdown.

How Many People Use AI Video Generators?

Kling reported more than 60 million creators and over 600 million videos generated as of December 2025. Google’s Flow passed 500 million videos by November 2025. Sora reached 9.6 million cumulative app installs by January 2026. Luma publishes no Dream Machine user count.

What Is Kling AI’s Revenue?

Kling’s annualized run rate reached approximately $500 million in March 2026, up from $240 million in December 2025. Quarterly revenue exceeded 650 million yuan in the first quarter of 2026 and 850 million yuan in the second. Full-year 2025 revenue was 1.04 billion yuan, about $150 million.

How Much Does AI Video Generation Cost?

Selected published API rates run from $0.03 per second for Luma’s Ray3.2 at 540p on a five-second clip and $0.05 for Runway’s gen4_turbo and Veo 3.1 Lite, up to $0.40 per second for Veo 3.1 and Veo 3 at 1080p and $0.60 for Veo 3.1 with audio at 4K. A generated minute, before retries and editing, costs $1.80 at the lowest of those rates, $24 at the 1080p rate and $36 at the 4K rate. Third-party models on Runway’s price list run from $0.02 to $1.50 per second.

Does AI-Generated Video Have To Be Labelled?

In the EU, yes. Since 2 August 2026, Article 50 of the AI Act has required deployers of deep fakes to disclose them and providers to mark AI-generated video in a machine-readable format, with systems already on the market given until 2 December 2026 to add that marking.

YouTube requires creators to disclose realistic AI-generated or altered content, and Meta applies an “AI info” label.

Can People Tell AI Video From Real Video?

Barely. A study of 1,276 participants found 50.7% accuracy on video, against 50% for chance. A meta-analysis of 56 papers put video accuracy at 57.31%, with confidence intervals that cross chance. The DF26 benchmark found both humans and state-of-the-art detectors close to random chance on modern generated video, while one University of Florida study found participants correctly identified real and fake videos about two thirds of the time.

What Share Of Video Ads Use AI?

Buyers expect one third of digital video ad assets to use generative AI in 2026, up from one fourth in 2025 and 22% in 2024, with 43% projected for 2027.

How Much Has Been Invested In AI Video Startups?

Global funding for AI-related video companies reached $3.08 billion in 2025, up a reported 94.6% from $1.58 billion in 2024. In 2026 Kling completed a round of about 19.8 billion yuan, published as $2.98 billion, Runway raised $315 million in February, and Synthesia raised $200 million at a $4 billion valuation in January. For the largest rounds of early 2026, see our top AI funding rounds of Q1 2026.

Sources And Methodology

Each figure carries its source and date, and paywalled or press-only figures are marked as such. Market forecasts, annualized run rates, recognised revenue, downloads, users and label views are kept separate throughout. Vendor surveys are identified as such, with the vendor’s commercial interest stated next to the figure.

Currency: yuan figures are given in yuan, with the US dollar conversion each publisher used where one exists. Luma printed its Series C as “900M” with no currency symbol; press coverage reports it in US dollars.

Retired from earlier versions of this page, each with a reason:

“Sora is still unreleased to the public as of August 2024.” Sora launched in December 2024, was succeeded by Sora 2 in September 2025, and its app closed on 26 April 2026. The full arc replaces the 2024 snapshot.

“Video content was forecast to account for 82% of all internet traffic by 2022,” credited to Cisco’s Visual Networking Index. The Cisco source said 82% of consumer internet traffic, a smaller denominator, and it was a 2018 forecast for 2022 from a programme Cisco has since discontinued. No successor figure exists, so none is offered.

A MarketsandMarkets claim that deepfakes enable virtual and augmented reality history reenactment for education. The cited MarketsandMarkets page carries no such claim; it lists AR and VR only as adjacent technologies.

A closing claim that AI will create hyper-realistic virtual environments. The cited article is a teardown of Sora’s flaws and does not make that claim. The only forward-looking view in it is OpenAI’s Bill Peebles, attributed to him above.

Muse.ai as an AI video platform. The domain now serves an unrelated Meta personal AI agent, so the product this page described no longer exists.

Lumen5’s “branded templates and images” description. Lumen5’s homepage no longer describes image generation, and custom templates now sit in an enterprise services tier.

Five claims about AI replacing voice actors, credited to a voiceover coaching blog. The post is undated, affiliate-monetised, cites no data, and contains a false example (a posthumous Robin Williams voice clone in “Call Me By Your Name”, a 2017 film without him).

Broadcast2World’s 71% planning to use AI-generated video a lot more, and its 50% spending under $500 per video. Neither figure remains on the studio’s published benchmark page.

Broadcast2World’s “24% of businesses are already working on AI-generated videos”. The same survey page also says 33% have started using AI for video, with no reconciliation, so neither figure is used.

A “49% time efficiency, 40% cost efficiency, 27% expanded content capacity” breakdown credited to McKinsey’s State of AI. That breakdown does not appear in the report.

Coca-Cola’s 2023 partnership with OpenAI and a BESTSELLER case study with Synthesia, credited to Grand View Research. Company anecdotes from 2023 carry no measurement.

A reported 12 million Kling monthly active users and a 25 to 30 million Dream Machine user range. Neither company publishes these figures.

Corrected rather than dropped:

The CVPR 2016 paper is You Only Look Once, not “You Look Only Once”.

Dreamix is from Google Research and The Hebrew University of Jerusalem, not “The University of Jerusalem”.

VideoPoet is from Google and Carnegie Mellon University.

MAV3D is attributed to the Meta AI Make-A-Video team; the arXiv page lists authors, not affiliations.

Synthesia’s learning and development survey dates from November 2023, not 2024, and surveyed 102 of Synthesia’s own customers.

This page previously gave 160+ as a Synthesia avatar count; Synthesia now lists 240+ avatars.

The deepfake market forecast is now credited to Spherical Insights & Consulting, the firm that produced it, rather than to the press release that announced it.

Every Wyzowl figure is from the 2026 edition, set against the 2024 figures this page cited earlier.

The Think with Google analysis is first-party research by Google on its own platform, and the page has moved to business.google.com.

YouTube’s disclosure policy is cited to YouTube’s own help page.

Limitations. Private company revenue depends on voluntary disclosure. Download and user counts come from third-party app analytics that can differ from operator figures. Cumulative generation counts are operator self-reported.

Market forecasts are vendor estimates built on differing definitions and are treated as context, not measurement. The press figures on Sora’s economics come from subscription coverage, and the IAB figures are taken from the association’s published newsroom summaries.

Bot Memo

About the author

Editorial Staff

The Editorial Staff at Bot Memo is a team of writers, analysts, and AI agents dedicated to mapping the global AI startup ecosystem. Led by Chintan Zalani, the team tracks thousands of funding rounds, classifies companies across verticals, and distills it all into actionable intelligence for investors and founders.

Subscribe to AI Funding Memo

Weekly pre-seed and seed AI deal intelligence with early trends.