AI Video Generation Cost in 2026: Faster Than Real Time, Cheaper Than You Think

In late August 2026, the honest answer to "how fast is AI video generation" stopped being "a few minutes per clip" and became "faster than you can watch it." The proof was a live stream called Infinite Slop. Anyone in the chat could type a prompt; the most-upvoted prompt got generated next in an endless sequence. It ran on a fal-fine-tuned version of the MiniMax H3 model that fal says is 50x faster than the base model, and it produced 15 seconds of video in about 9 seconds. levelsio, who built it, reported 37,000 day-one viewers and more than 2,000 concurrent watchers within 24 hours.

That weekend is the clearest real-time AI video generation inflection point yet for agencies. It resets the AI video generation cost math, changes what you can promise a client, and shifts which format a video deliverable should ship in. Two days before the stream went live, Google shipped Gemini Omni 1.1 Flash to general availability and undercut most of the market on per-second price, giving agencies a credible second anchor for the cost math. Here are the dated numbers, the new model comparison, the mobile-first format shift, the $2M AI film festival, and what it all means for agencies selling video.

How Fast Is AI Video Generation in 2026?

Infinite Slop is an interactive AI-generated TV channel where the chat decides what airs next. Viewers type prompts, the community upvotes a queue, and the winner gets generated into a 15-second clip that flows from the last one. It launched August 29, 2026, with a hard cap of 4 videos per minute, because that is roughly how much real-time generation the model could keep up with.

The model behind it is MiniMax H3 Max, a post-trained variant of MiniMax's open-weight H3 built by fal, the serverless inference platform, which also sponsors the stream's compute. fal's own numbers say H3 Max is "nearly 50x faster than the base H3 model." The claim is not just vendor marketing: Design Arena's independent benchmarks measured 6.4-second image-to-video generation (18x faster than its arena average) and 4.7-second text-to-video generation (24x faster). levelsio measured 15 seconds of video generated in 9 seconds, and Wharton professor Ethan Mollick ran his own tests and concluded H3 Max "can now create reasonably high quality AI video in less time than it takes you to watch it."

Caveat worth keeping: the viewer counts are levelsio's own counters, not independently audited. The speed numbers, by contrast, have independent corroboration from Design Arena and Mollick.

What Does AI Video Generation Cost in 2026?

fal's public pricing for MiniMax H3 Max: $0.08 per second of 768p video, or $4.80 per minute, at list price. The 50%-off launch window priced it at $0.04 per second for the first 14 days (through roughly mid-September 2026), and the first five generations a day are free; that intro window is why early coverage, including the first version of this post, quoted the cheaper number. The base MiniMax H3 endpoint costs $0.06 per second, or $3.60 per minute, at the same resolution.

What that works out to per deliverable at list price:

Raw generation is the floor, not the full bill. Voiceover, editing, on-brand direction, and client revisions still cost real money, and a 90-second spot with multiple takes and reshoots will bill well above $7.20. What changed is the cost of iteration. A rejected take used to mean waiting 2–5 minutes per render and eating human review time on top; now a rejected take costs about a dollar and roughly 9 seconds. That changes how agencies can price revisions, A/B tests, and concept rounds. For a fuller cost model, see our AI agency pricing 2026 work on the AI agency pricing calculator for the broader pricing picture.

Gemini Omni 1.1 Flash: A New Per-Second Pricing Floor

On August 27, 2026, Google made Gemini Omni 1.1 Flash generally available through the Gemini API in Google AI Studio, the Gemini Enterprise Agent Platform, Google Flow, and as a scene-extension tool in the Gemini app. It is the first major vendor model to publish a per-second rate below a dime: $0.03 per second for 360p draft output, $0.10 per second at 720p, $0.15 per second at 1080p, and $0.30 per second at 4K, with native synchronized audio billed in the same per-second rate.

The capability set is what makes the price interesting for agencies: scenes can be extended in 10-second increments up to a 40-second cumulative length, the model reads up to 10 seconds of prior context, first- and last-frame keyframes give you shot control, and you can reference up to 3 seconds of video input for character and style consistency. One caveat on the top of the ladder: 1080p and 4K are upscaled from the 720p render, not generated natively. The 360p draft tier is built for storyboarding and iteration, runs up to 60% faster than 720p, and costs about a third as much, which makes it the cheapest major-vendor per-second rate in the market.

The short version for decision-stage queries: on "how much does AI video cost per second in 2026", expect $0.03 to $0.30 per second on Omni 1.1 Flash depending on resolution, with the top competitors at $0.05 to $0.70 per second. On "best cheap AI video generator" in 2026, the answer is the $0.03/s 360p draft tier, with Runway Gen-4 Turbo at $0.05/s and fal's MiniMax H3 Max at $0.08/s list (intro $0.04/s) close behind, plus a watch-out that Sora's API shuts down on September 24, 2026.

Build your own quotes from these rates with the AI video cost per second calculator, and see our Gemini Omni 1.1 Flash for agencies guide for the full capability breakdown.

Why 9:16 Is Winning: The Mobile-First Shift

One day after launch, levelsio switched the stream from landscape 16:9 to portrait 9:16 output. His explanation was one line: "Switched from landscape 16:9 to portrait 9:16 generation now because most traffic is mobile!"

The same force is reshaping agency video deliverables. Most AI video is consumed on phones, in TikTok, Reels, and Shorts, all of which are native 9:16. An agency that still defaults every deliverable to 16:9 is producing for the desktop share of the audience. The practical read: design for vertical first, adapt to horizontal when a client's channel actually demands it, and let the placement data decide. If your client's traffic is mobile, the format decision was already made for you.

Can AI Make Films? The $2M Answer

The same weekend, the answer to "can AI make films" got institutional backing. On August 30, 2026, Balaji Srinivasan announced the Network School Astana AI Film Festival, in partnership with the Republic of Kazakhstan, with the tagline "It's time to decentralize Hollywood with AI." Entries are accepted worldwide through aaiff.ai, and the festival runs in Astana.

The prize structure is worth reading closely. The official site lists a $2,000,000 total fund: $1,000,000 in competition prizes across two sections plus a $1,000,000 production fund invested into new AI productions. Balaji's announcement tweet said "$2M in prizes"; the official breakdown is $1M awarded in prizes and $1M reserved for production funding. Grand Prize in the thematic competition, themed "The Future Worth Living In," is $450,000.

For agencies, the festival is less about the prize money and more about what it signals. A national government and a well-funded school are now running a worldwide competition that assumes AI film is a legitimate production medium. That is a demand-side signal for the AI video tools for agencies market: clients will begin asking for AI film and video work, and the agencies that can already produce it get to set the price.

What This Means for Agencies Selling Video

Three concrete changes follow from this weekend. The AI video agency 2026 playbook is pricing and packaging, not technology bets.

First, reprice your video deliverables around per-second reality. When raw generation costs as little as $0.03 per second on Gemini Omni 1.1 Flash's draft tier, or $0.08 per second list on fal's fastest H3 Max model, you are selling speed, direction, and client management, not render time. Per-finished-second pricing, or project pricing built from it, is defensible in a way hourly render billing no longer is.

Second, sell iteration. Real-time generation turns "here are two concepts" into "here are five concepts, pick a direction, and we will refine live." Revision rounds become a differentiator instead of a cost center, and you can price them accordingly.

Third, default to 9:16 for social deliverables. The format shift is already visible in the data that drove the Infinite Slop switch, and it is where the volume is. Vertical-first is now the safe default, with horizontal as the adaptation.

If you are building out the capability, start with FLUX 3 video for agencies, our breakdown of pricing and client use cases for FLUX-based video work, and compare the tool landscape in AI video tools for agencies before committing to a stack.

AI Video Questions Agencies Ask

How fast is AI video generation in 2026?
Fast enough to beat playback. fal's H3 Max, a post-trained MiniMax H3 variant, generates 15 seconds of video in about 9 seconds, and independent benchmarks measured 4.7-second text-to-video and 6.4-second image-to-video generation in late August 2026. "Faster than you can watch" is now a literal description.

How much does AI video cost per second in 2026?
Gemini Omni 1.1 Flash bills $0.03 per second at 360p draft, $0.10 at 720p, $0.15 at 1080p, and $0.30 at 4K with audio included. Competitors: Runway Gen-4 Turbo $0.05/s and Gen-4.5 $0.12/s; Sora 2 $0.10/s (Pro $0.30–$0.70/s, API sunsetting Sept 24, 2026); Kling $0.084–$0.42/s by resolution; fal's MiniMax H3 Max $0.08/s list at 768p with a $0.04/s 14-day intro. A 10-second 1080p Omni clip runs about $1.50 in raw generation.

What is the cheapest AI video generator in 2026?
By per-second API price, Gemini Omni 1.1 Flash's 360p draft tier at $0.03/s is the cheapest major-vendor rate. Close behind: Runway Gen-4 Turbo at $0.05/s and fal's MiniMax H3 Max at $0.08/s list ($0.04/s intro for the first 14 days, free 5/day). Watch-out: OpenAI's Sora API shuts down on September 24, 2026.

Is Gemini Omni 1.1 Flash cheaper than Sora or Runway?
At 720p, Omni 1.1 Flash ($0.10/s) matches Sora 2 and undercuts Sora 2 Pro ($0.30/s) and Runway Gen-4.5 ($0.12/s). Its 360p draft tier ($0.03/s) is the cheapest major-vendor rate overall; Runway Gen-4 Turbo ($0.05/s) undercuts it at 720p, but Omni adds 40-second scenes, keyframe control, and 4K upscale.

Can AI make films now?
Yes, and the market is organizing around it. The Network School Astana AI Film Festival, launched with the Republic of Kazakhstan in August 2026, carries a $2M total fund, $1M in competition prizes plus a $1M production fund, with worldwide entries at aaiff.ai under the banner "decentralize Hollywood with AI."

Should my agency offer AI video services in 2026?
The economics support it. Real-time generation collapsed the cost of iteration, mobile-first 9:16 is the dominant consumption format for short-form video, and per-second API rates now start at $0.03/s. Price per finished second or per project, verify current model pricing before quoting, and treat the film festival as the demand signal it is.