What an AI Short Drama Costs in 2026: a Line-by-Line Budget for a 10-Episode Vertical Series

SinCoSphere EditorialUpdated 2026-09-126 min readGuide

Budget = parsing + reference art + storyboards + (video seconds × price per second × takes per kept segment) + voice + compositing. For a 10-episode × 90-second series (900 finished seconds), parsing, reference art and storyboards together come to roughly $2–6 at official LLM and image list prices. One take of every video second costs $45 at the $0.05-per-second tiers (Wan 3.0 480P, MiniMax H3-Max 480P, Veo 3.1 Lite 720p) and $360 on Veo 3.1 Standard at 1080p. Video is over 90% of the bill in every scenario below, and the number of takes you need moves the total as much as the price tier does.

The formula and the inputs you need

Total = parsing + reference art + storyboards + (video seconds × price per second × takes per kept segment) + voice + compositing.

Write down your own value for each input before opening any price page; the right-hand column shows the example used throughout this guide.

Input	Example used below
Episodes × seconds per episode	10 × 90 s = 900 finished seconds
Delivery model and tier	A 1080p tier at $0.12–0.20 per second
Draft tier	A 480P/720p tier at $0.05 per second
Takes per kept segment	2 (keep one render in two); 1.5 if you keep two in three, 3 if you keep one in three
Reference sheets	20 (for example 6 characters, 8 scenes, 6 props)
LLM for parsing and storyboards	Gemini 2.5 Flash at the low end, Gemini 2.5 Pro or GPT-5 at the high end
Voice route	Native audio in the video model, or text-to-speech afterwards

All other prices below are official list prices checked on 2026-09-12; SceneMixer's come from its public pricing API when this page is built. For how credits, tokens and per-second billing convert, see the pricing-models explainer; for why published cost estimates disagree, see the cost explainer.

Line items 1–3: parsing, reference art and storyboards

Item	Assumption (replace with yours)	Low	High
Parsing one story arc	30,000 words ≈ 37,000–50,000 tokens in (Google: 100 tokens ≈ 60–80 English words), 30,000 tokens out	$0.09 on Gemini 2.5 Flash ($0.30 in / $2.50 out per million tokens)	$0.36 on Gemini 2.5 Pro or GPT-5 ($1.25 / $10)
Storyboards	10 episodes × (30,000 tokens in + 8,000 out)	$0.29	$1.18
Reference art	20 sheets × 1.5 attempts = 30 images	$1.17 at $0.039 per image (Gemini 2.5 Flash Image)	$4.02 at $0.134 (Gemini 3 Pro Image, 1K/2K)
Subtotal		≈ $1.55	≈ $5.56

Token counts are planning assumptions: the input for a storyboard call includes the instructions, the episode outline, the cast and scene sheets and the source excerpt, and reasoning models can produce more output than assumed. Gemini 2.5 Pro's rates apply up to 200,000 input tokens. Even with generous margins these three lines stay in single-digit dollars. Their real cost is indirect: a vague appearance line or a scene plate with plot debris forces a reshoot of every segment that uses it.

Line item 4: the same 900 seconds on each channel

Channel	Published rate	$ per second	900 s, one take
Wan 3.0 API (Alibaba Cloud, Singapore)	$0.05 / $0.10 / $0.20 per second at 480P / 720P / 1080P	$0.05–0.20	$45–180
MiniMax H3 API	$0.08 (768P), $0.13 (2K); H3-Max $0.05 (480P)	$0.05–0.13	$45–117
Veo 3.1 (Gemini API)	Lite $0.05 (720p), Fast $0.12 (1080p), Standard $0.40; audio included	$0.05–0.40	$45–360
Runway Gen-4.5	12 credits/s; one credit $0.024 (Standard, $15 a month for 625) down to $0.008 (Max, $76 a month billed yearly for 9,500)	$0.096–0.29	$86–259
Luma Ray3.14	20 credits/s (720p), 80 (1080p); one credit $0.003 (Plus, $30 a month for 10,000) down to $0.0017 (Ultra, $250 a month billed yearly for 150,000)	$0.033–0.24	$30–216
Kling 3.0 inside Luma	30 credits/s (720p, no audio) to 49 (1080p, with audio) at the same Luma credit prices	$0.05–0.15	$45–132
Kling 3.0 in the Kling app	6 credits/s (720p, no audio) to 12 (1080p, native audio)	Not converted (plan prices not captured)	5,400–10,800 credits
SceneMixer	Wan 3.0 · 480P 10 credits/s to Seedance 2.5 Pro 75 credits/s; one credit $0.006–0.015 depending on pack or plan	$0.062–1.12	$56–1,006

API rows are raw model prices: you still do the parsing, storyboards and bookkeeping yourself. Runway, Luma and SceneMixer are app prices that include their interfaces; SceneMixer bills parsing and storyboards separately from video. Credit-based ranges run from the lowest tier at the lowest listed credit price (largest plan or pack, yearly billing where offered) to the highest tier at the highest listed credit price. Runway's 10,800 credits exceed one month of its Max plan. Sora 2 is not listed: OpenAI's deprecations page schedules the sora-2 and sora-2-pro models and the Videos API for removal on 2026-09-24, with no replacement named.

The multiplier that decides the bill

Takes per kept segment	900 s at $0.05/s	at $0.10/s	at $0.20/s	at $0.40/s
1×	$45	$90	$180	$360
1.5×	$67.50	$135	$270	$540
2×	$90	$180	$360	$720
3×	$135	$270	$540	$1,080

Needing two takes costs as much as moving up one price tier. Two duration rules add seconds you never see on screen: Veo 3.1 requires 8-second clips when you use reference images, so a 5-second beat is billed as 8; MiniMax H3's shortest call is 4 seconds and Kling 3.0's is 3. Plan segment lengths on the model's billing grid. Before budgeting a whole series, render a three-segment pilot on your target model and count takes per kept segment; that number is yours, not a vendor's.

Line items 5–6: voice and compositing

Native audio: Google's Veo prices include audio. Kling 3.0 charges 9 credits per second with native audio against 6 without at 720p (12 against 8 at 1080p), plus 2 credits per second for voice control.
Text-to-speech: MiniMax lists HD speech at $100 per million characters. A 90-second episode of 100–150 spoken words stays under 1,000 characters, so under $0.10 an episode and under $1 for ten. ElevenLabs' Starter plan ($6 a month, 30,000 credits) is its entry paid plan and includes a commercial license.
Voice cloning: MiniMax charges $1.50 per rapid-cloned voice, billed when the voice is first used for speech. Cloning a real person's voice needs that person's consent.
Compositing: free in CapCut or ffmpeg; SceneMixer bills compositing per second in credits (rate on its pricing page).

Text-to-speech stays under $1 for a 10-episode series and compositing in an editor costs nothing; cloned voices add $1.50 each on MiniMax, and SceneMixer's compositing is its per-second rate × 900 s. The expensive voice mistake is indirect: a line given too few seconds, or a speaker left unnamed, forces a video retake, and with native audio every retake repeats the audio cost too. Voice options in depth: AI voice for drama.

Worked budgets for the same series

Scenario	Video rate	Takes	Video	Other lines	Total
Draft-grade	$0.05/s (Wan 3.0 480P, H3-Max 480P, Veo 3.1 Lite 720p)	2×	$90	$2–7	$92–97
1080p delivery	$0.12–0.20/s (Veo 3.1 Fast 1080p, Wan 3.0 1080P)	2×	$216–360	$2–7	$218–367
Draft then final	$0.05/s × 2 takes, then $0.20/s × 1.3 takes	—	$324	$2–7	$326–331
Premium model	$0.40/s (Veo 3.1 Standard, 720p or 1080p)	2×	$720	$2–7	$722–727

"Other lines" is the $1.55–5.56 subtotal above plus under $1 of text-to-speech. Two of the $0.05 tiers take no reference images (MiniMax lists H3-Max for text- and image-to-video, not reference-to-video; Google's API table shows none for Veo 3.1 Lite), so draft cast-critical segments on a tier that does. A draft pass followed by a 1080p pass is cheaper than two 1080p takes ($324 against $360) when the draft removes 1080p retakes, and the gap widens the more errors it catches. A re-render is a new sample, not an upscale, so the final still needs review. These are model costs; your review time is the other half of any production budget.

Cost-cutting checklist (and three savings that backfire)

Review and edit every storyboard before rendering; text edits cost nothing.
Draft at 480P or 720P to check blocking, speaker and dialogue timing.
Fit segment lengths to the billing grid: 8 s on Veo 3.1 with references, 4–15 s on H3, 3–15 s on Kling 3.0.
Stay inside free reference slots: MiniMax H3 charges $0.04 per reference image after the fifth.
Batch the text steps where the API offers it: Google's Batch API lists Gemini 2.5 Pro at $0.625 in and $5 out per million tokens, half the standard rate. It trims lines that are already small; no batch price is listed for Veo 3.1.
Check regional price lists: Alibaba lists Wan 3.0 at $0.041 / $0.083 / $0.165 per second in Tokyo, Frankfurt, Virginia and Hong Kong against $0.05 / $0.10 / $0.20 in Singapore.
Size the plan to the volume: a Luma credit costs $0.003 on monthly Plus and about $0.0017 on yearly Ultra; Runway's Max plan rolls up to one month of unused credits over, while Standard and Pro reset.

Three savings that backfire: cutting segments to 4 seconds (the per-second price does not change and every extra join is a continuity risk); rendering dialogue on a model whose lip-sync fails for your cast (retakes cost more than the rate saved); and skipping reference images (drift forces reshoots). Tool-by-tool prices: pricing comparison.

Key figures and sources

Figure	Value	Source	Date
Wan 3.0 API list price (Alibaba Cloud, Singapore)	$0.05 / $0.10 / $0.20 per second at 480P / 720P / 1080P	www.alibabacloud.com	2026-09-12
MiniMax H3 API list price	$0.08/s (768P), $0.13/s (2K); reference images beyond 5 cost $0.04 each	platform.minimax.io	2026-09-12
Veo 3.1 (Gemini API)	Lite $0.05/s at 720p; Fast $0.12/s at 1080p; Standard $0.40/s; audio included	ai.google.dev	2026-09-12
Luma Ray3.14 (Luma app)	20 credits/s at 720p, 80 at 1080p; Plus $30/mo for 10,000 credits, Ultra $250/mo billed yearly for 150,000	lumalabs.ai	2026-09-12
Runway Gen-4.5	12 credits per second; Standard $15/mo for 625 credits, Max $95/mo ($76/mo billed yearly) for 9,500	runway.com	2026-09-12
SceneMixer video pricing	10–75 credits per second by tier; one credit $0.006–0.015	scenemixer.com	2026-09-12

FAQ

Can a free tier finish a series?

No. Runway's 125 one-time credits cover about 10 seconds of Gen-4.5 at 12 credits per second; Morph Studio's free plan gives 75 credits a month with two models and no commercial use; Pika's free plan is 80 credits a month at 480p without commercial use; Kling gives free credits (amount shown on its site); Luma lists no free plan; SceneMixer: 50 credits on sign-up; first project includes 15 concept sheets + first storyboard free; video is billed per second. Free allowances are for testing the workflow, not for shipping 900 seconds.

Why do per-second list prices differ so much?

Resolution, speed tier, bundled audio and model size. Google's own Veo 3.1 list runs from $0.05 per second (Lite, 720p) to $0.40 (Standard). Compare the tier you will actually deliver at, and check whether audio and reference images are included or charged extra.

Subscription or credit packs?

Convert both to dollars per credit at the plan size you would really need, then to dollars per finished second. Check rollover and expiry: Runway resets Standard and Pro credits at the billing date while Max rolls over up to one month. SceneMixer: Credit packs from $1.49 (100 credits); memberships from $7.99/mo (Pro, 700 credits), 5 paid tiers, cheaper quarterly / yearly.

Should I pay for 2K or 4K?

Phones play vertical video at 1080×1920, and YouTube recommends about 8 Mbps H.264 for 1080p uploads. Higher tiers help when you plan to crop or reframe; otherwise they multiply the dominant line of the budget.

How do I estimate takes per kept segment before I start?

Render three segments of a real episode on the model and tier you plan to use, count every render, and divide by the segments you would publish. Repeat after fixing the storyboard: the second number is usually the one to budget with.

Sources

AI short-drama cost explainer Credits vs tokens vs per-second Pricing comparison Novel-to-drama workflow