Making ReelShort-Style Vertical Dramas with AI (2026): format, season plan, localization, pilot budget

A ReelShort-style series is a strict format: vertical episodes of 45 seconds to two minutes, 50–75 episodes a season, each dropping the viewer into the middle of the story and ending on a cliffhanger (Reuters, 2026-08-18). ReelShort's own blog says nearly every title gives the first 5–12 episodes free, so the last free episode has to carry the season's strongest reveal. Producing it with AI means planning the whole season on paper, localizing at the parsing step (names, faces, buildings, visible text and props for the target market) so every reference image and shot inherits it, and proving the plan with a three-episode pilot before generating the rest. The money is concentrated: DramaBox and ReelShort each took about $140M of roughly $750M in category in-app purchases in Q1 2026 (Sensor Tower).

The format in numbers

Micro-drama apps sell one thing: the next episode. These are the working numbers for a series aimed at ReelShort, DramaBox and their peers, with YouTube Shorts and TikTok as traffic channels.

Two conventions shape every episode without a published platform statistic behind them: a hook inside the first three seconds and a cliffhanger at every episode end (see hook / cliffhanger and vertical micro-drama). Treat them as writing rules.

Step 1 — Plan the season around the paywall

Write the full episode list, with a one-line ending for every episode, before generating a single frame. Choose the paywall episode first: when the first 5–12 episodes are free, the last free one has to sell the unlock, so it carries the strongest reveal of the season.

Keep a 60-episode plan producible: two or three locations per episode, four to six recurring characters, no crowd scenes, and no new location unless the story returns to it. Every added character or location is another reference image to generate and another thing that can drift between episodes.

Step 2 — Write each 90-second episode to a beat template

1. 0–3 s, cold open: a line or an action inside the conflict; no establishing shot, no title card.
2. 3–20 s, want: who wants what, with no more than one line of context.
3. 20–60 s, escalation: two beats, each ending on a reaction close-up.
4. 60–80 s, turn: the reveal, betrayal or rescue.
5. 80–90 s, cliffhanger: cut on the question, not the answer.

Dialogue budget: at about 2.5 English words per second, 90 seconds holds roughly 225 words if someone speaks the whole time; plan 150–200 so reactions and the closing beat have room (see dialogue timing). Build the episode from six to eight segments of 10–15 seconds with three to five shots each, name the people in frame and nobody else, and end every line with closed lips and a small action so the model does not repeat it to fill time. The shot-level format is in the storyboard template.

Step 3 — Localize at the parsing step, once

For a US, European, Southeast Asian or Latin American audience, localization is decided where the script becomes a cast table and a scene table. Parsing runs once per series, storyboards once per episode and video once per segment, so this is where a change costs cents instead of a reshoot. Changing the target market after the reference art exists means re-parsing and regenerating every character sheet and scene plate.

• Names and forms of address native to the market.
• The first line of each appearance states apparent age, ethnicity and nationality, so the character sheet is generated for that market.
• Architecture, interiors and street furniture of the market.
• Visible text (signs, phone screens, documents, banknotes) in the target language: subtitles cannot fix text inside the picture.
• Props, food, money and institutions the audience recognizes.
• Dialogue written in the target language at parsing, not translated afterwards; translated lines usually run long and push the timing out.

Tools differ on this step. SceneMixer documents four target-region presets (Western, East Asian, Southeast Asian, Latin American) that adapt names, casting, scenery and on-screen text when the script is parsed; with a tool that has no such setting, including every video model in Step 4, rewrite the script and the character table for the market before import. Which market first: Omdia puts the US at half of micro-drama revenue outside China, while Sensor Tower counts Southeast Asia, Latin America and India at 32%, 23% and 22% of app downloads. Subtitle specs and dubbing are covered in subtitles and localization.

Step 4 — Assemble the stack and cost a three-episode pilot

Pilot budget: three 90-second episodes are 270 finished seconds; at a keep rate of two in three they bill 405 seconds of video.

API rows cover generation alone; parsing, storyboards and compositing are yours to build or buy. The SceneMixer row is the video line of a pipeline that also parses, storyboards and composites, each billed separately. Free allowances listed by the vendors: SceneMixer, 50 credits on sign-up; first project includes 15 concept sheets + first storyboard free; video is billed per second; Runway, 125 one-time credits (watermarked); Pika, 80 credits a month at 480p without commercial use; Morph Studio, 75 credits a month.

Step 5 — Package, publish and avoid the usual failures

DramaBox's Business Collaborations page lists separate contacts for rights holders who want to distribute a finished short drama, for English-language production partners and for script adaptation; ReelShort's homepage lists support and press contacts but no submission page (both checked 2026-09-12). Beyond that, titles arrive through commissions, distributors, rights agents and competitions: DramaBox ran one with Stage 32 whose grand prize was a $5,000 writer contract (final deadline 2026-02-04). Routes are compared in distribution channels. What to hand over:

• The season outline: episode list, one-line endings, the last free episode marked.
• Three finished episodes at 1080×1920 in the target language.
• A series bible: character sheets, scene plates and voice choices.
• A rights file: story rights, generator terms for the tier you used, music and voice clearances (see the copyright and labeling checklist).
• Retention numbers if the pilot ran on Shorts or TikTok, especially the drop from episode 1 to episode 3.

Episodes of 60–90 seconds clear TikTok's one-minute floor for Creator Rewards and sit inside the three-minute Shorts limit; label realistic AI content on both. The failures that sink AI pilots:

• Subtitling a series parsed for another market instead of re-parsing it.
• Opening an episode on an establishing shot.
• More than three people in one frame.
• A free-episode run that ends on a quiet episode.
• Changing the target market after the reference art is done.
• Mixing two video models inside one dialogue scene.

Key figures and sources

Sources

• https://sensortower.com/blog/state-of-short-drama-apps-2026-report
• https://www.investing.com/news/stock-market-news/microdramas-boom-in-a-shrinking-hollywood-as-studios-chase-a-tiktok-audience-4864722
• https://www.reelshort.com/fandom/how-much-does-reelshort-cost-01-25-1a2b555/
• https://www.c21media.net/news/microdramas-will-generate-11bn-by-2025-twice-as-much-as-fast-channels-says-omdia-report/
• https://support.google.com/youtube/answer/15424877
• https://www.tiktok.com/legal/page/global/creator-rewards-program-us/en
• https://www.stage32.com/happy-writers/contests/Stage-32-%20-Drama-Box-Screenwriting-Competition
• https://www.dramaboxdb.com/business
• https://scenemixer.com/guides/make-reelshort-style-dramas/
• https://platform.minimax.io/docs/guides/pricing-paygo
• https://www.alibabacloud.com/blog/603505
• https://ai.google.dev/gemini-api/docs/pricing
• https://docs.dev.runwayml.com/guides/pricing/
• https://help.runwayml.com/hc/en-us/articles/50404627334547-Free-plan-details
• https://pika.art/pricing
• https://www.morphstudio.com/pricing