ChatGPT
ClaudeThis page collects AI video generation statistics for 2026.
Our own numbers come from the Oakgen AI Video Index 2026.
It covers 4,069 video jobs run on Oakgen from January 10 to November 10, 2026.
Every number below names its sample size (n) and date range.
Market, ad and policy figures come from outside sources, each linked to the publisher.
Key findings
The Oakgen AI Video Index 2026 found that 88.4% of AI video jobs completed (n = 4,069, Jan 10 to Nov 10, 2026).
The Oakgen AI Video Index 2026 found that 11.5% of AI video jobs failed (n = 4,069).
The Oakgen AI Video Index 2026 found that 4.1% of all video jobs were blocked by a content or copyright filter (n = 4,069).
The Oakgen AI Video Index 2026 found that 64.5% of video jobs started from an image, not from text alone (n = 4,069).
The Oakgen AI Video Index 2026 found that Seedance v2 text-to-video ran 22.2% of all video jobs, the largest share of any model (n = 4,069).
The Oakgen AI Video Index 2026 found that 38.4% of clips were 5 seconds or shorter (n = 3,876 jobs with a set duration).
The Oakgen AI Video Index 2026 found that 9:16 vertical was the most common aspect ratio at 46.9%, just ahead of 16:9 at 45.5% (n = 2,418).
The Oakgen AI Video Index 2026 found that the median completed clip cost 394 credits, about $2.19 at the add-on rate (n = 3,128).
The Oakgen AI Video Index 2026 found that only 62.1% of trend video jobs completed, and 19.5% were blocked by a filter (n = 585, Sep 26 to Nov 10, 2026).
The Oakgen AI Video Index 2026 found that October 2026 had 1,029 video jobs, the busiest month in the sample.
How often do AI video generations fail?
The Oakgen AI Video Index 2026 found that 88.4% of 4,069 video jobs completed.
11.5% failed.
6 jobs were still processing when the data was pulled.
Failures had several causes (n = 467 failed jobs).
The content filter (OAK-4001) caused 24.4% of failures.
Internal server errors caused 23.3% of failures.
Any content or copyright filter caused 35.8% of failures.
Across all jobs, filter blocks ended 4.1% of video jobs.
Read more in how often AI video generations fail.
Does image-to-video fail more than text-to-video?
Yes, in our data.
The Oakgen AI Video Index 2026 found these completion rates by mode:
| Mode | Share of jobs | Completed | Jobs (n) |
|---|---|---|---|
| Text-to-video | 28.5% | 93.4% | 1,159 |
| Image-to-video | 64.5% | 87.3% | 2,625 |
| Video-to-video | 7.0% | 77.9% | 285 |
A job counts as image-to-video if it had any image, start frame, reference image or face input.
A job counts as video-to-video if it had a video input.
Which AI video models do people use most?
The Oakgen AI Video Index 2026 found that Seedance v2 text-to-video ran 22.2% of all video jobs (n = 4,069).
Trend videos were second at 14.5%.
Only models with at least 100 jobs are shown.
| Model | Share of jobs | Completed | Jobs (n) |
|---|---|---|---|
| Seedance v2 text-to-video | 22.2% | 92.1% | 903 |
| Trend videos (trend pipeline) | 14.5% | 62.1% | 591 |
| Seedance v2 image-to-video | 7.2% | 97.9% | 291 |
| Kling v3 Pro image-to-video | 6.1% | 96.0% | 250 |
| MiniMax H3 reference-to-video | 5.2% | 96.7% | 213 |
| Grok image-to-video | 5.1% | 97.6% | 208 |
| Kling v3 Pro text-to-video | 5.1% | 98.1% | 207 |
| Kling v4 image-to-video | 4.2% | 94.1% | 169 |
| Video upscaler | 2.9% | 72.6% | 117 |
| Grok text-to-video | 2.9% | 98.3% | 116 |
Model share reflects what Oakgen offered and promoted during the period.
It is not a market share figure.
See which AI video models people use for more.
How long are AI video clips?
The Oakgen AI Video Index 2026 found that most clips were short (n = 3,876 jobs with a set duration; 193 without one were left out).
| Duration | Share of jobs |
|---|---|
| 5 seconds or less | 38.4% |
| 6 to 8 seconds | 10.0% |
| 9 to 10 seconds | 13.3% |
| 11 to 15 seconds | 32.1% |
| Over 15 seconds | 6.2% |
What resolution and aspect ratio do AI videos use?
The Oakgen AI Video Index 2026 found that 720p was the most common resolution at 52.7% (n = 3,207 jobs with a set resolution).
1080p was 21.0%.
480p was 16.1%.
768p was 8.0%.
The Oakgen AI Video Index 2026 found that 9:16 vertical was 46.9% of jobs with a set aspect ratio (n = 2,418).
16:9 widescreen was 45.5%.
More detail is in AI video clip length, resolution and aspect ratio.
How much does an AI video clip cost?
The Oakgen AI Video Index 2026 found that the median completed video job cost 394 credits (n = 3,128).
The mean was 717.6 credits.
The mean is higher because a few long or premium jobs cost much more.
We convert credits to dollars at three rates.
The add-on rate is 180 credits per dollar.
The Basic plan is $19 for 4,000 credits.
The Creator plan is $99 for 30,000 credits.
| Measure | Credits | Add-on rate | Basic plan | Creator plan |
|---|---|---|---|---|
| Median clip | 394 | $2.19 | $1.87 | $1.30 |
| Mean clip | 717.6 | $3.99 | $3.41 | $2.37 |
Cost per completed clip varied by model (only models with at least 100 charged jobs):
| Model | Median credits | Median at add-on rate | Jobs (n) |
|---|---|---|---|
| Trend videos (trend pipeline) | 1,912 | $10.62 | 364 |
| Seedance v2 text-to-video | 787 | $4.37 | 793 |
| Seedance v2 image-to-video | 468 | $2.60 | 198 |
| Kling v3 Pro image-to-video | 437 | $2.43 | 114 |
| Kling v4 image-to-video | 266 | $1.48 | 159 |
| MiniMax H3 reference-to-video | 163 | $0.91 | 206 |
| Grok image-to-video | 119 | $0.66 | 203 |
| Grok text-to-video | 104 | $0.58 | 114 |
Failed jobs are not counted here, so the cost of a usable clip can be higher.
See AI video cost per finished clip.
How long does AI video generation take?
The Oakgen AI Video Index 2026 found that the median completed video job took at most 271.6 seconds, about 4.5 minutes (n = 3,596).
Treat every time here as an upper bound.
We measure from job creation to the last update, and later edits such as favoriting can move the last update.
| Model | Median time (upper bound) | Jobs (n) |
|---|---|---|
| MiniMax H3 reference-to-video | 38.3 s | 206 |
| Grok text-to-video | 79.8 s | 114 |
| Grok image-to-video | 80.5 s | 203 |
| Kling v3 Pro image-to-video | 220.9 s | 240 |
| Kling v3 Pro text-to-video | 279.0 s | 203 |
| Kling v4 image-to-video | 312.4 s | 159 |
| Seedance v2 image-to-video | 318.3 s | 285 |
| Seedance v2 text-to-video | 336.4 s | 832 |
| Trend videos (trend pipeline) | 676.6 s | 367 |
How many AI videos are made each month?
The Oakgen AI Video Index 2026 counted video jobs per month on Oakgen.
Months with fewer than 100 jobs are not shown.
| Month | Video jobs |
|---|---|
| April 2026 | 449 |
| May 2026 | 406 |
| June 2026 | 338 |
| July 2026 | 209 |
| August 2026 | 486 |
| September 2026 | 412 |
| October 2026 | 1,029 |
| November 2026 (1st to 10th) | 710 |
These counts show Oakgen's own growth, not the market's.
How often do people rerun the same AI video prompt?
The Oakgen AI Video Index 2026 found that 11.1% of prompts in the AI video generator were run more than once by the same person (n = 2,614 prompt groups).
The mean was 1.2 runs per prompt.
The median was 1 run, and the 90th percentile was 2.
This counts identical-prompt reruns only.
Someone who edits a prompt and tries again is not counted as a rerun.
Why do AI video trends get blocked?
Trend videos put a person's photo into a set scene, often with a song or a known character.
The Oakgen AI Video Index 2026 found that 62.1% of trend video jobs completed (n = 585, Sep 26 to Nov 10, 2026).
37.6% failed.
19.5% of trend jobs were blocked by a content or copyright filter.
Filters caused 51.8% of trend failures (n = 220 failed trend jobs).
The copyright filter alone blocked 6.8% of trend jobs.
The median completed trend video cost 1,912 credits, about $10.62 at the add-on rate (n = 364).
We do not publish numbers for single trends, because most have small samples.
Read why AI video trends get blocked.
How are AI video tools used in ads and UGC?
This section uses outside sources only.
None of these numbers come from Oakgen data.
Meta said the combined revenue run rate of its video generation ad tools hit $10 billion in Q4 2025 (Meta Platforms, Q4 2025 earnings call, 2026-01-28).
Meta said nearly 2 million advertisers used its video generation features in Q2 2025 (Meta Platforms, Q2 2025 earnings call, 2025-07-30).
Meta said nearly 10% of Reels viewed each day were created in its Edits app (Meta Platforms, Q4 2025 earnings call, 2026-01-28).
The IAB found that 86% of video ad buyers use or plan to use generative AI to build video ad creative (IAB, 2025 Digital Video Ad Spend & Strategy report, 2025-07-15).
The same IAB report says buyers project generative AI creative will reach 40% of all ads by 2026. This is a projection, not a measurement.
The IAB and Sonata Insights found that 85% of advertisers in their study use AI for social media ads (IAB / Sonata Insights, The AI Ad Gap Widens, 2026-01-15).
Wyzowl found that 63% of video marketers have used AI to make video, up from 51% a year earlier (Wyzowl, The State of Video Marketing 2026).
Do AI video ads perform better?
The evidence conflicts.
A Harvard Business School working paper found that ads with AI-generated images got higher click-through rates, but only when the images did not look like AI (Harvard Business School, AI in Disguise, 2025). It studied 16 billion+ impressions. It is not yet peer reviewed.
NielsenIQ found that AI-generated ads, even high-quality ones, caused weaker memory activation than traditional ads (NielsenIQ, 2024-12-12). This study predates 2025.
The Nuremberg Institute for Market Decisions found that identical ads were rated less natural and less useful when labeled as AI (NIM, Transparency Without Trust, 2025-06-26).
These studies measure different things: clicks, memory and stated opinion.
EU labeling rules start in August 2026, so the label effect may matter more from then on.
How big is the AI video generator market?
Research firms disagree, because each defines the market differently.
| Publisher | Estimate | Date | Definition note |
|---|---|---|---|
| Grand View Research | USD 788.5M in 2025, USD 3,441.6M by 2033 (20.3% CAGR) | 2026-01 | Proprietary market model |
| Fortune Business Insights | USD 716.8M in 2025, USD 847M in 2026, USD 3,350M by 2034 (18.8% CAGR) | 2026-10 page update | Proprietary market model |
Fortune Business Insights says North America held 41.0% of the market in 2025.
It says marketing and advertising will be the largest segment at 33.88% in 2026.
We left out figures we could not trace to a primary page.
One vendor blog cites a USD 6.2B market for 2025, far above the research firms. We did not use it.
How much are the big AI video platforms used?
Kuaishou said Kling AI served over 60 million creators and generated over 600 million videos as of December 2025 (Kuaishou Technology, 2026-01-13).
Kuaishou said Kling AI hit a USD 240 million annualized revenue run rate in December 2025 (same source).
Google said over 275 million videos were generated in Flow (Google, Introducing Veo 3.1, 2025-10-15).
What are the AI video labeling rules on YouTube, TikTok, Meta and in the EU?
| Date | Rule | Source |
|---|---|---|
| 2024-03-18 | YouTube requires creators to disclose realistic altered or synthetic content | YouTube |
| May 2024 | Meta starts adding "AI info" labels when it detects C2PA or IPTC signals or a user discloses | Meta Transparency Center, 2025-02-19 |
| 2025-11-19 | TikTok says it has labeled over 1.3 billion videos and adds invisible watermarking | TikTok |
| Live policy | TikTok requires creators to label realistic AI content | TikTok |
| Live help page | YouTube carries forward C2PA 2.1+ credentials that show a video was made with AI | YouTube Help |
| 2026-08-02 | EU AI Act Article 50 transparency rules apply; deployers must disclose deepfakes | European Commission, 2026-07-29 |
People cannot reliably spot AI video.
A study in Communications of the ACM found about 52% accuracy on video-only clips, close to the 50% of a coin toss (Cooke et al., 2025-10).
What about AI audio?
Oakgen also tracks audio jobs, such as voiceovers.
The audio sample is 1,490 jobs from December 15, 2025 to November 10, 2026.
The Oakgen AI Video Index 2026 found that 95.0% of audio jobs completed and 4.8% failed (n = 1,490).
ElevenLabs V4 ran 63.0% of audio jobs (n = 1,490).
The median completed audio job cost 6 credits, about $0.03 at the add-on rate (n = 1,394).
The median audio job took at most 14.6 seconds (n = 1,394, upper bound).
54.0% of all audio jobs were created from November 1 to 10, 2026, so this sample is young.
Methodology
Source. All Oakgen numbers come from Oakgen's production job records.
Period. Video jobs from January 10 to November 10, 2026. Trend jobs from September 26 to November 10, 2026. Audio jobs from December 15, 2025 to November 10, 2026.
Fields. We used job status, error type, model, input fields (to set the mode), duration, resolution, aspect ratio, prompt and timestamps.
Mode. A job is image-to-video if it had any image, start frame, reference image, element or face input. It is video-to-video if it had a video input. Otherwise it is text-to-video.
Credits. We used the credits recorded on the job. When that was missing, we summed the usage deductions in the credit ledger that point to the job. Only completed jobs are counted.
Dollars. We convert at 180 credits per dollar (add-on), $19 per 4,000 credits (Basic) and $99 per 30,000 credits (Creator).
Generation time. Time from job creation to the job's last update. Later edits can move the last update, so treat these as upper bounds.
Reruns. We grouped jobs by person and exact prompt text. Edited prompts count as new prompts.
Suppression. We do not publish any cut with fewer than 100 jobs.
Privacy. We report totals only. No user-level data, prompts or outputs are shown.
Limits. This is one platform. Model share reflects which models Oakgen offered. The sample is small next to the big platforms. Duration, resolution and aspect ratio were not set on every job, so those shares use smaller samples.
Outside sources. We cite each with its publisher, date and link. We did not use figures we could not trace to a primary source.
How to cite this page
You are free to use these numbers with a link back to this page.
Suggested citation:
Oakgen Research. "AI Video Generation Statistics 2026." Oakgen AI Video Index 2026, n = 4,069 video jobs, January 10 to November 10, 2026. https://oakgen.ai/research/ai-video-statistics
Frequently asked questions
How much does one AI video clip cost?
The median completed clip in the Oakgen AI Video Index 2026 cost 394 credits. That is about $2.19 at the add-on rate of 180 credits per dollar, $1.87 on the Basic plan and $1.30 on the Creator plan (n = 3,128).
How long does an AI video take to generate?
The median completed video job took at most 271.6 seconds, about 4.5 minutes (n = 3,596). This is an upper bound, because the timestamp can move after the job finishes.
Is image-to-video more popular than text-to-video?
Yes. In the Oakgen AI Video Index 2026, 64.5% of video jobs started from an image, 28.5% from text only and 7.0% from a video (n = 4,069).
How often are AI video trends blocked?
19.5% of 585 trend video jobs in the Oakgen AI Video Index 2026 were blocked by a content or copyright filter (September 26 to November 10, 2026).
When do EU AI Act labeling rules start?
The European Commission says the Article 50 transparency rules apply from 2 August 2026. Deployers must disclose deepfakes.
