AI micro-drama from scratch: the mistakes that cost me $337

In short. I paid Kling $337 across three payments and, short on time and skills, generated just two one-minute episodes and two teasers of about ten seconds each. The rest of the tokens I burned in "AI slot machine" mode or simply lost. So a minute of finished video cost me $168. Below are four lessons that would have saved me that money, and the AI video cost calculator shows what your own project will cost.

Who I am#

I don't have a 100K TikTok account of AI content. But I do have:

• 30K followers on TikTok and 12K on Instagram, over 6 million views a month on TikTok — before the AI era, with live-action content;
• journalism, acting school and online courses from top American screenwriters — among them the Duffer Brothers, creators of the cult Netflix series Stranger Things;
• frontend development: I count first and generate second. I built the AI series cost calculator after my budget went down the drain faster than Kling could finish rendering the first clip.

Why I thought a season costs $30–60#

My math went like this: 30 episodes of a minute each, at $1–2 a minute, $30–60 for the season. I was used to Nano Banana in Gemini, where images are free.

Video models cost from $0.06 to $0.42 a second. A minute on Seedance 2.0 at 720p is $9. And that is if every clip comes out right the first time.

Here is what one one-minute episode really costs, video only, if 80% of each clip makes the edit and half the seconds are generated again:

The math

• Seedance 2.0
• 720p
• with sound
• ×2.5 generated per second kept

1. 60 sfinished video
2. ÷

   80%kept from each clip
3. ×

   ×2rerolls
4. =

   150 sgenerated10 clips × 15 s
5. ×

   $0.15per generated secondMaker's API

=

$22.50per episode$22.50 per minute on screen

Open in calculator(opens in a new tab)

And at the start I thought: $30 for 30 episodes. Here it's $22.50 for one. The first punch in the face from a virtual but very expensive AI camera. The nose hurts, but we keep generating. By the end of this story AI will smash our naive little face to pieces with budgets in the thousands of dollars.

Four subscription traps I fell into#

1. Buying today, "to get started". A subscription works like a gym membership for a beginner: the first month you don't go at all, or you go once. On June 30, 2026 I bought Kling Ultra (24,000 credits back then) for $157.43. I generated on the first day, then my main IT job pulled me under, and I remembered my Kling token bank three days before that mountain of credits was due to burn mercilessly. Over three evenings of 2–3 hours, 80% went on duplicates: the same character references and shots 15–20 times over, video in 1080p — just so the credits wouldn't burn. Kling Pro (now $25.99 for the first month) would have given me everything I needed.

2. "I'll take a cheaper plan and this time I'll make it." In August I moved to Premier at $79.94 and planned 15–20 teasers of 10–20 seconds. Again I put it off until five days were left in the month.

3. The cancellation that didn't cancel. At the end of August I cancelled Kling. On September 7 I was charged for another month of Premier anyway.

4. The price without tax. That time I was charged not $80.96 but $99.58: $18.62 of it was Polish VAT at 23%. Prices on plan pages and in our calculator are before VAT. In the EU, add 20–27%. The tax office is the only player in this story that doesn't need to learn to generate: it takes its 23% from the very first take.

Lesson 1. Buy a video plan once you have several scripts generated and have chosen the references for the AI actors, locations and look of your project. Otherwise you risk not using all your tokens by the end of the month or, the other way round, burning them from day one like a slot machine, in endless experiments with no scripts for specific videos.

How to start on a minimal budget: a $1.60 teaser#

From my experience. I launched a new account straight away with one-minute episodes in 1080p. But HD quality doesn't affect the TikTok or Instagram algorithm. It's like a beginner blogger buying the latest iPhone and renting a whole studio for every reel. You'd tell them to just learn to talk to a camera — any camera. Same here: if you don't want to burn $300–500 in your first month, learn to generate hooks that work, viral scripts, visuals and AI actors at basic quality on a minimal budget. Once you have results, invest in 1080p. It costs more, by the way: a third more on Kling, 2.5 times more on Seedance 2.5.

The common belief: more videos, more views. That's supposedly why mythical content farms of 30–100 accounts exist, spending $3,000–10,000 a month on generation. Do the math:

A content farm is buying 30 lottery tickets that simply don't have the winning number printed on them. A new TikTok account, in my experience, hits a ceiling of 260–300 views. $10 for that ceiling is money thrown away; $10 for 100K views is money well spent. And after a dozen assembly-line videos you can get banned, and the rest will get 0–3 views.

So I warm up the algorithm with 7–12-second teasers, 2–3 a day, with a long caption under the reel: Seedance 2.5 at 480p, Seedance 2.0 or Kling 3.0 at 720p, or MiniMax H3 at 768p. In the calculator that's the "Economy · teaser" preset, about $1.60 a teaser. On Instagram I saw a young writer who gained 50K followers on 7-second teasers with her characters and romantic scenes, at 50K, 100K and even 200K views a video.

A teaser doesn't need to run 20–30 seconds or be generated at the maximum 1080p to hold a viewer. Here are two of my videos in TikTok Studio, both 10–12 seconds long:

Three things changed between them: the first frame — not a wide shot but an action that intrigues before the first word; the first line; and the genre and language — horror in Russian versus a sci-fi drama in Polish. So it's not a clean A/B test. But at second 6 almost four times as many viewers were still there, and none of those changes was a more expensive video model or a jump from 480p to 1080p.

Lesson 2. Teasers of 7 seconds in budget formats first:

• Seedance 2.5 at 480p
• Seedance 2.0 at 720p
• MiniMax H3 at 768p
• Kling 3.0 at 720p

60+ second episodes only once several of your short videos break through a ceiling of 10K views.

Frame-by-frame storyboard: my most expensive mistake in time#

I started the old-school way: first you make a storyboard, a separate image for every shot, and then a video from each one.

1. The prompts are wrong. The AI agent gets the technical description of the shot wrong — or rather, you don't explain the picture in your director's head clearly enough. Without camera terms, and without rereading those walls of text, the prompt and the frame get regenerated 2–4 times.
2. The slot machine. Once the prompt is finally right, the rerolls for the "perfect" picture begin.
3. Extra seconds. A shot in a fast-paced micro-drama lasts 0.7–1.4 seconds, but the minimum clip is 3 seconds on Kling and 4 on Seedance. You pay for the whole clip and trim the rest.
4. The shots don't connect. Each one looks great on its own, but in sequence the transitions don't work — you come up with a different shot type and generate again.
5. The edit. Stitch all the shots together, lay the sound, do the lip-sync. Assembling a minute from a hundred one-second clips is like doing a mosaic with tweezers in boxing gloves.

The alternative is multishot mode: Kling up to 15 seconds, Seedance up to 30 (I mostly do 20). You describe every shot in one prompt, and 70–80% of the clip comes out right, sometimes all 100%. The video model understands composition, angles and transitions between shots better than Nano Banana Pro or Seedream understand the description of a single frame — at least in a beginner's hands. And in English the model voices the dialogue itself.

Editing time compared: storyboard vs multishot

Why didn't I try multishot right away? I read somewhere that it makes mistakes and the shots have to be cut anyway, so it's cheaper to generate 3-second clips from a storyboard. That information is out of date. And here is what a storyboard adds to the bill for a one-minute episode, without characters or a voice-over:

The math

• Seedance 2.0
• 720p
• with sound
• ×5 generated per second kept

1. 60 sfinished video
2. ÷

   60%kept from each clip
3. ×

   ×3rerolls
4. =

   300 sgenerated20 clips × 15 s
5. ×

   $0.15per generated secondMaker's API
6. +

   $4.02besides videoStoryboard

=

$49.02per episode$49.02 per minute on screen

Open in calculator(opens in a new tab)

Lesson 3. A beginner doesn't need a storyboard. Write the shots in one multishot prompt — the script plus character and location references — and regenerate only the 1–2 shots that didn't work. At the start, skip ElevenLabs and lip-sync too: that's $20 a month and 5–10 times more editing time. Otherwise you'll burn out and post two videos a month instead of one a day.

What I'd buy now for a 24 × 60 s season: $326 through the API instead of $408 on a subscription#

Not a subscription. Not even a specific model. I'd buy access to an API aggregator you top up as you work — for example kie.ai, where credits start at $5 and, by their terms, never expire. Aggregators' prices are often lower than the official models' public prices: Kling 3.0 on kie.ai costs $0.07 a second against $0.084 at Kling itself. But not always: Seedance 2.5 is cheapest on BytePlus, its official seller. In our calculator you can compare video platform subscriptions against API marketplaces.

Subscription discounts shatter the moment you start your series and, after two episodes, realize one of three things:

1. It's not for you. And the plan is already paid for.
2. It takes an unreal amount of time. A 24-episode micro-drama season planned for a month can in practice give you 2 episodes a month — 12 months of production for 24 episodes.
3. Rerolls inflate the budget 5–10 times over the plan. That's why "Normal" in the calculator already assumes five generated seconds for every second in the edit, and "Perfectionist" ten.

A subscription lowers the price of a second only if you spend every credit. And when you have hundreds of dollars of tokens on the account, you burn them without noticing — until it turns out that $500 of subscriptions made 10 episodes of the 60 you planned. And you need to find another $2,500 somewhere.

fal, Replicate and BytePlus also offer an API with top-ups.

Lesson 4. Until your first 3–5 episodes are done, pay only for what you generate. Take a subscription once you know your pace and your reroll multiplier. The AI video cost calculator shows the volume at which it really is cheaper: "Monthly budget" → "My plan".

The four lessons in one list#

• Lesson 1. Buy a video plan once several scripts are ready and the references for AI actors, locations and look are chosen.
• Lesson 2. Teasers of 7 seconds at 480–768p first; 60+ second episodes once several videos break 10K views.
• Lesson 3. A beginner doesn't need a storyboard: shots in one multishot prompt and the voice from the video model.
• Lesson 4. Until your first 3–5 episodes are done, pay only for what you generate: an API you top up.

And if generating a whole micro-drama season is beyond your pocket? Is AI video production really only for studios and agencies? It shouldn't be. We're building a platform where a creator posts the first episode and viewers chip in for the first season. 500 people at $1 each — and you've successfully closed a crowdfunding campaign for your first series.

FAQ#

Which subscription should a beginner buy?#

None, until you've made 3–5 episodes: pay through an API you top up. After that, the smallest plan that fits your real monthly pace. The calculator shows how many episodes a month each plan makes ("Monthly budget" → "My plan").

Do content farms pay off?#

By my numbers, no: 30 assembly-line videos on new accounts get fewer views than one well-made video, and cost 30 times more per thousand views. On top of that, platforms are limiting assembly-line AI content more and more often.

Sources#

My payments are from my Kling payment history; model prices are from the catalog of the AI video cost calculator, and the source of every price is in the calculator guide.

• kie.ai top-ups and terms: Terms of Use, Refund Policy
• Kling plans: Kling
• TopView: TopView