AI Video Cost per Second: November 2026 Prices

AI video costs about $0.01 to $0.60 per second at list price in November 2026, depending on the model, the resolution and whether it makes audio. At 720p, a finished minute costs about $1.24 to $24. The newest price is Google's: Gemini Omni 1.2 Flash came out on 2026-11-05 at about $0.07 per second at 720p, 30% below the version before it.

This page lists the per-second price of each major model, read on the vendor's own pricing page on 2026-11-12. It adds two numbers that price lists leave out: how many clips we had to make again in real production, and what 480p plus an upscale saves. We update it every month.

Key Takeaways

• Gemini Omni 1.2 Flash: about $0.07/s at 720p, clips up to 15 s. Requests to Omni 1.1 move to 1.2 prices on 2026-12-03.
• Veo 3.1 runs from $0.03/s (Lite, no audio, 720p) to $0.60/s (full model, audio, 4K). At 720p and 1080p, audio doubles the price of the full model.
• Seedance 1.0 Pro Fast is the cheapest model we found: about $0.0097/s at 480p. Seedance 2.5 is about $0.103/s at 480p.
• Kling 4.0 is about $0.154/s at 720p, and you buy it in prepaid packages from $500.
• Across five of our explainer videos we made 23 fixes for 151 AI clips, about 1 for every 7 clips. Plan for about 15% on top of list price, more for harder scenes.

The November 2026 price list

Google's launch post says: "We have dropped the price 30% from Gemini Omni 1.1 making it available at $0.07 per second for 720p video" (Google, "Gemini Omni 1.2 Flash", 2026-11-05). Every row below comes from a page like that one: the vendor's own pricing page, opened in a headless browser on 2026-11-12. "Per minute" is the per-second price times 60, for one take of every clip. The Gemini rows show four decimals because Google prices them in tokens (tokens per second x token price).

How each vendor states its price, so you can check the arithmetic:

• Gemini Omni (Google). The Gemini API pricing page (it moved from ai.google.dev in November 2026) says billing is "calculated at a rate of 5,792 tokens per second of 720p video. Under Standard pricing ($12.00 per 1M tokens), this equates to an effective price of approximately $0.07 per second (~$0.70 for a 10s clip, ~$1.04 for a 15s clip; $0.023/s for 360p ... $0.104/s for 1080p with upsampling ... and $0.209/s for 4K with upsampling)". The same page says that from December 3, 2026, "all API requests to gemini-omni-1.1-flash automatically route to gemini-omni-1.2-flash and are billed at gemini-omni-1.2-flash rates."
• Veo 3.1 (Google). The Vertex AI pricing page lists each Veo row as a price per "count", for example "Veo 3.1 Lite ... Video + Audio ... 720p $0.05 / 1 count". A count is one second of output. We checked that in our own test in October 2026, through the Vercel AI Gateway: a 4-second Veo 3.1 Lite clip without audio cost $0.12, the Vertex price of $0.03 times 4 seconds.
• Seedance (ByteDance). BytePlus ModelArk bills tokens: "Estimated token consumption: (Input video duration + output video duration) x output video width x output video height x output video frame rate / 1024". The Seedance 2.x rows are BytePlus's own examples for a 5-second 16:9 clip with no input video, for example "480p ... $0.514 per video, $0.103 per second" for Seedance 2.5. The Seedance 1.0 rows use the token prices on the same page ($2.50 per million for 1.0 Pro and $1 for 1.0 Pro Fast) and BytePlus's token counts: 48,600 tokens for 5 s at 864x480 and 102,960 for 5 s at 1248x704.
• Kling 4.0. The Kling API pricing page lists "Kling 4.0 ... No Input Video ... 1.0 Unit/s" at 720p, 1.5 units at 1080p, and 0.6 units for Kling 4.0 Flash. Units come in prepaid packages. The smallest is $500 for 3,250 units, a "discounted unit price $0.15386", valid for 180 days. Our rows use that unit price.
• Runway. The Runway API pricing page says "Credits can be purchased for $0.01 per credit" and lists gen4.5 at "12 credits per second" and gen4_turbo at "5 credits per second".

Not in the list this month: Luma's API page prices Ray3.2 in credits (for example 100 credits for a 5-second 720p clip) but does not give a dollar price per API credit, so we could not convert it. OpenAI removed the Sora 2 API in September 2026 (our write-up).

What a finished clip really costs: the retake rate

From late October to early November 2026 we made six short science explainer videos with AI clips, and we logged every fix. Across videos 2 to 6 we made 23 fixes for 151 clips, about 1 for every 7 clips (our production log). Some clips needed more than one fix. Most clips came from Seedance 1.0 Pro Fast at 480p, and a few from Veo 3.1 Lite.

The errors were physical, not cosmetic:

• A globe on a desk turned the wrong way.
• Water ran out of a sponge in a scene where the sponge soaks it up.
• Small burn marks on a surface grew into a large smear.
• A swung bucket became two buckets that spilled.

Some fixes cost nothing. A clip that moves in the wrong direction can play in reverse, if nothing else in it (rain, smoke) shows the reversal. The rest needed a new take with a more exact prompt, such as "the spots stay exactly the same size and do not grow, spread or change".

How to budget for it: if every fix were a paid retake of the same length, the cost would be 151 + 23 = 174 takes for 151 clips, or 15% over list price. So Seedance 1.0 Pro Fast at 480p goes from $0.58 to about $0.67 per finished minute. That is our own rate for this kind of footage, not a ceiling. Scenes with people, traffic or text will likely need more takes. Check every clip before it ships, at full speed and frame by frame where things move.

480p plus an upscale: half the price of 720p

Seedance 1.0 Pro Fast at 480p costs about $0.0097 per second. At 720p it costs about $0.0206, a little more than twice as much (BytePlus token counts above). In our test in October 2026, a 480p clip upscaled 3x on our own machine looked as sharp at 100% zoom as a 720p clip upscaled 2x, and the character in the scene stayed the same. All six videos used 480p clips for that reason.

This works because the price follows the pixel count. BytePlus counts tokens as width x height x frame rate x seconds, so 864x480 is 414,720 pixels per frame and 1248x704 is 878,592. An upscale costs only local compute time. It does not help with fine text or small faces, which the model must draw at the size you need.

Resellers do not always charge the vendor price

Several platforms resell the same models through one API. Their prices can differ from the vendor's. Runway's API page, read the same day, lists:

One API key for many models has real value: one bill, one SDK, no prepaid package per vendor. But compare the price per second with the vendor page before you commit volume. A gap of 2x at 480p is $5.82 for each finished minute.

A template render of the same minute

Not every second of a video needs new pixels. Titles, captions, product shots, charts and transitions can come from a template that places your own assets on a timeline. Renderly's pricing bills that at 1 credit per minute of 1080p output, by the second: $0.20 a minute on pay-as-you-go credits, about $0.145 on the $29 Creator plan (200 credits) and $0.099 on the $99 Business plan (1,000 credits).

The two are not substitutes. A template cannot film a storm, and a model cannot be trusted to spell a price or a product name the same way twice. The cheap mix is to generate only the shots you cannot get, then put them in a template with the text, the voice and the captions. We compared the two approaches in template-based vs AI-generated video. If you want both in one API call, Renderly can generate Veo 3.1 clips inside a template for 0.5 or 1.5 credits per second (how it works).

How we read these prices

• Vendor pages only. Each price comes from the vendor's own pricing page, opened in a headless browser on 2026-11-12. We do not use prices from blogs, aggregators or search snippets, because those disagree with the vendor pages and with each other.
• List prices only. No discounts, committed-use plans or free tiers, except the Kling package price, because Kling sells units only in packages.
• One take per clip in the main table. The retake section shows our own correction.
• Audio follows the vendor: "yes" means the price includes generated sound, "-" means the page does not split the price by audio.

Change log

• 2026-11-12: first edition. Gemini Omni 1.2 Flash added (released 2026-11-05). Omni 1.1 requests switch to 1.2 prices on 2026-12-03. Luma not listed: no dollar price per API credit on its page.