What a 5-second vertical AI video costs in 2026 (model by model)

The short version

We price every clip before it runs in VayBe!'s video studio, so we keep a running sheet of what each model charges. Here is that sheet for the format most brands need: 5 seconds, vertical 9:16, for Reels, TikTok and Shorts.

All figures are list price on fal.ai as of October 2026, in US dollars, for generation only. No subscription, no editing time, no music licence. fal bills per second of output for most video models, so a 5-second clip is simply five times the per-second rate, unless a note says otherwise.

Four numbers to remember:

• $0.20: a MiniMax H3 Max Turbo draft at 768p (regular price; $0.13 on promo until 31 October).
• $0.40–$0.50: a solid product animation or brand reel (H3 Max at 768p, Wan 3.0 at 720p).
• $0.53: a Turkish-speaking talking head (ElevenLabs v4 voice plus OmniHuman v2).
• $2.37: premium Seedance 2.5 at 720p.

Then multiply by two to four for retries, because the first render is rarely the one you post.

Prices change often and several promotions end within weeks of this post. Treat the tables as a snapshot, and check the live price before you commit a budget.

Text-to-video and image-to-video: price per model

Regular list prices first; promo prices are in the notes. "5 s clip" is the per-second rate times five.

If you only remember one row: MiniMax H3 Max is near the top of the quality boards while costing a fraction of the premium models, which is why it is our default for product shots.

Editing, motion transfer and talking heads

Not every clip starts from a prompt. These endpoints take an existing video, a driving motion or a voice track.

Outside fal, Runway's API charges $0.28/s for Aleph 2 edits, so a 5-second edit costs $1.40.

Budget for retries, not for one render

The price tables quote a single generation. Real production does not work like that. A hand drifts, the product label warps, the camera move overshoots, the first frame is weak. You render again.

Our rule of thumb is to budget two to four times the generation cost for every clip you plan to publish:

• H3 Max at 768p: $0.40 per render, so $0.80–$1.60 per usable clip.
• Wan 3.0 at 720p: $0.50, so $1.00–$2.00.
• A Turkish talking head on OmniHuman v2: about $0.53, so $1.06–$2.12.
• Seedance 2.5 at 720p: about $2.37, so $4.74–$9.48.

Three habits keep the multiplier closer to two than four:

1. Draft cheap, finish expensive. Explore the idea on H3 Max Turbo ($0.20) or HeyGen Video ($0.10), and only send the winning prompt to the premium model.
2. Start from an image. Image-to-video from a good, correctly framed 9:16 still fails less often than text-to-video, because composition and product are already fixed.
3. Keep shots short and simple. Five seconds of one clear action beats five seconds of three.

In VayBe!'s video studio, the estimate appears before anything runs and counts against the brand's monthly budget, so the retry allowance is a conscious decision rather than a surprise on the invoice.

Turkish speech: use TTS plus lip-sync, not the video model's voice

If your clip needs someone speaking Turkish, do not rely on the video model to generate the voice. As of October 2026, Turkish speech generated inside video models is unverified for every model we checked:

• Kling 3 generates Chinese and English voices only, and auto-translates other languages to English.
• Kling 4's dialogue languages (English, Chinese, Japanese, Korean, Spanish, Portuguese, German, French) do not include Turkish.
• Seedance 2.5 reportedly handles dialogue in 10+ languages, but Turkish is not named.

The reliable path is two steps. First, generate the Turkish voice with a text-to-speech model; then drive an audio-to-avatar or lip-sync model with it.

A 5-second Turkish line is about 70–80 characters, so the voice costs under one cent. Choose the voice on quality, with an A/B test, not on price. Lip-sync models also want at least 5 seconds of audio, which fits this format exactly.

To dub real footage into Turkish instead, re-lip the existing video with sync-3 at about $0.67 per 5 seconds.

Hidden costs that change the bill

The per-second rate is only part of the story. Five details move the final number:

• Reference surcharges. H3 Max reference-to-video includes 4,096 reference tokens, then charges $0.02 per 1,000. A 5-second 9:16 reference video at 768p adds about $0.563, which is why that 5-second clip lands near $1.00 rather than $0.40.
• Input seconds count too. Kling 4's video edit endpoint bills input plus output seconds, so at least twice the clip length: about $0.94 for a 5-second edit at 720p.
• Resolution jumps are steep. Seedance 2.5 goes from about $2.37 at 720p to $5.82 at 1080p for the same 5 seconds. Reels are a phone-first format; test whether 1080p earns its price.
• Fixed durations. Veo 3.1 only renders 4, 6 or 8 seconds, Gemini Omni Flash 3–10 seconds and FLUX 3 Max at least 5. A model that cannot do exactly 5 seconds makes you pay for frames you trim.
• On-screen text. AI video models render Turkish letters (ğ, ş, ı, İ) unreliably, and Kling 4's on-screen text languages exclude Turkish. Render typography in code over the AI footage. Remotion is free for companies with up to three employees; above that, the Automators plan is $0.01 per render with a $100 monthly minimum.

And watch the calendar: H3 Max and HeyGen promotions end on 31 October, FLUX 3 Max's on 7 November, and Veo 3.1 leaves fal on 17 November 2026.

Frequently asked questions

Sources

1. fal: MiniMax H3 Max
2. fal: MiniMax H3
3. fal: What is Kling 4.0?
4. MiniMax: Video generation guide
5. Google: Veo in the Gemini API
6. Runway: API pricing
7. CostGoat: Sora pricing and shutdown
8. Spheron: Sora 2 API shutdown
9. TechNode: ByteDance launches Seedance 2.5
10. Webrazzi: OmniHuman 2.0
11. Webrazzi: ElevenLabs Eleven v4 and v4 Turbo
12. Remotion: License and pricing