What an AI Micro Drama Costs: Per-Minute Breakdown by Model (2026)

Plan on $2.80 to $28.38 per finished minute of vertical video if you keep every clip. If you throw away two takes for every one you keep, plan on $8.40 to $85.14. The model you pick is the biggest difference, so compare the rows below.

The cheapest row is a silent model, so you'd pay extra for voices. These numbers come from the per-second prices each provider lists. We turn them into a cost per finished minute of vertical drama, then add everything else an episode needs.

All prices are in USD, as listed on official pricing pages and API hosts on October 1, 2026. They change often because of promos, new versions and regional pricing. Use the tables as a starting point and check the linked pages before you budget a whole season.

The formula

Video models charge by the second of video they make. So the core math is simple:

cost per finished minute = price per second × 60 × your retake factor

Your retake factor is how many takes you make for each one you keep. If you keep every clip, it's 1. If you make three takes for every one that makes the cut, it's 3.

Nobody can look this number up for you. Dialogue close-ups with lip sync (mouths matching the words) usually need more tries than a silent wide shot. Early episodes also need more tries than episode 40, once your reference images are dialed in.

The tables show two columns: 1x (the floor, every clip kept) and 3x (the planning number we use for draft budgets). The 3x figure is our assumption, not an industry statistic. Track how many clips you actually keep for an episode or two, then use your real number.

Video cost per finished minute at 9:16 720p

These rows assume vertical 720p video (768p for MiniMax), with native audio on wherever the model has it. Native audio means the model makes the sound too. They also assume no video inputs, which cost extra on several platforms (more on that below).

The takeaway: MiniMax H3, Veo 3.1 Lite and Seedance 2.0 Mini are cheap. Veo 3.1 and Seedance 2.5 on fal cost the most.

How to read the table:

• Seedance on BytePlus charges per token, not per second. A token here is a unit of video size: seconds × width × height × 24 ÷ 1024. So a 720×1280 frame is 21,600 tokens per second, which turns $10.70 per million into about $0.23 a second.
• Silent models aren't cheaper if your scene has dialogue. Hailuo 2.3 and Wan 2.2 don't make sound. You'll pay separately for voices and a lip-sync pass (see below).
• Veo 3.1 only makes 8-second clips at 1080p or 4K in the Gemini API. Reference images also force 8 seconds, so plan your shots around that length.
• Failed or blocked clips aren't charged on the Gemini API, BytePlus (moderation failures) or Alibaba Model Studio, according to their docs. But takes that render fine and just look wrong still cost you. Those are the expensive ones.

If you need 1080p

You probably don't need native 1080p. Most vertical drama is watched on phones. Plenty of creators make clips at 720p, then upscale (enlarge with software) only the ones they keep.

If you do want native 1080p:

The takeaway: 1080p costs more on most models, and on Seedance 2.5 it's more than double the 720p price.

Subscriptions vs API

If you're making one series by hand, an app subscription can beat paying through the API. The API is where you pay per clip from code instead of using the app. Here's what each listed plan buys, using the app's own credit costs:

• Kling Standard lists $10 a month for 660 credits, with a $6.99 first month when we checked. Commercial use and watermark removal are listed as subscriber perks.
• What Kling credits buy. Kling 3.0 with native audio costs 9 credits per second at 720p and 12 at 1080p. So 660 credits is about 73 seconds of 720p video a month.
• Hailuo Standard lists $14.99 a month for 1,000 credits. MiniMax H3 costs 7 credits per second at 768p in the app, so that's roughly 142 seconds a month.
• Google AI Pro is $19.99 a month. It adds 1,000 Flow credits on top of the 50 daily credits everyone gets.
• What Flow credits buy. Google's Flow help page says a Veo 3.1 Fast clip costs 20 credits for non-Ultra users. An 8-second Veo 3.1 Quality clip costs 100, so 1,000 credits is 50 Fast clips or 10 Quality clips.

Subscriptions run out fast at series scale. A 90-second episode at 3x needs about 270 seconds of video, which is more than a month of Kling Standard.

That's when paying through the API, or a bigger plan, starts to make sense. Promos also change monthly, so check the plan pages before you commit.

The rest of the bill

Video is the biggest cost, but it isn't the only one.

The takeaway: after video, these items add only a few dollars per episode, plus a couple of monthly subscriptions.

Text-to-speech (turning written lines into a voice) is cheap at drama scale. Writers often aim for about ten lines of dialogue per episode, which is well under a thousand characters. Even at $100 per million characters, that's about ten cents an episode.

Worked example: one 90-second episode

Say episode 1 runs 90 seconds. You make three takes for every second you keep, so 270 seconds of video. You also make 30 keyframe images (the stills you animate) at $0.08 each.

The takeaway: the same episode costs about six times more on Veo 3.1 than on MiniMax H3.

Add a music subscription and an editor if you don't have them already. Now scale up to an 80-episode season of 90-second episodes, which is 120 minutes total.

Multiply the per-minute 3x numbers by 120. That's about $1,296 on MiniMax H3 via fal, $2,722 on Kling 3.0 via the official API, and $8,640 on Veo 3.1 Standard.

Six ways to spend less

1. Draft cheap, finish expensive. Rough out the episode at low resolution with Veo 3.1 Lite, Seedance 2.0 Mini or MiniMax H3, then remake the keepers on your main model. Seedance 2.5 has this built in: per BytePlus, you can make a 480p draft and finish it at 1080p within seven days.
2. Lock the still before you animate. A rejected keyframe image costs a few cents. A rejected 10-second clip with audio costs dollars.
3. Match the model to the shot. Pay top rates only for dialogue close-ups. Wide shots and inserts rarely need sound, and silent video costs less on Kling, and on Veo through Google Cloud and fal.
4. Make clips at 720p. Upscale only the clips that make the final cut, instead of making every throwaway take at 1080p.
5. Watch reference costs. Seedance and Wan 3.0 charge for video references as input seconds. MiniMax charges for reference images past the first few, so use stills where a still does the job.
6. Mix models carefully. A cheap model for wide shots and a better one for faces can cut the bill. But the look changes between models, so leave time for color matching, and keep each character's close-ups on one model.

You can also self-host the open-weight Wan 2.2 models. Open weights means you download the model and run it yourself. Then you pay for GPU time instead of per-second fees.

We don't quote GPU rental prices because they vary too much. The self-hosting guide lists the hardware you need, from the official READMEs.

Keep going

• Choosing a video model compares the five model families on features as well as price.
• Each model page has full price tables per host: Seedance, Kling, Veo, Hailuo, Wan.
• How to make an AI micro drama walks through every step, and lip sync and dubbing covers voices.
• The tools directory lists plan prices for the apps mentioned here.