The main misconception
People think an AI will make a Reel. It will not.
A video model generates one short shot: 5 to 15 seconds, one action, one camera move. A Reel is a sequence of such shots stitched together with music and text. So the work splits into three parts: come up with the sequence, generate the shots, assemble them.
Whoever skips the first part gets six seconds of beautiful meaningless motion and disappointment.
Step 1. A storyboard for vertical video
Take the idea and break it into scenes. For 30 seconds you usually need three to five.
The one-action rule. One scene is one action and one camera move. Not «a girl walks into a cafe, sits down and opens a laptop», but three scenes. On paper it looks boring, but it works.
Vertical is not cropped horizontal. Generate in 9:16 from the start. If you generate 16:9 and crop it later, you throw away half the frame along with the composition: the hero ends up past the edge and the camera drifts into emptiness.
Composition in vertical video is different. What works is anything that stretches upwards: a person at full height, a staircase, a street receding into the distance, a close-up of a face. What does not work is anything wide: panoramas, group scenes, long tables.
The first half-second decides everything. In Reels the first frame is where a viewer either stays or swipes away. So put your strongest scene first and do not waste it on titles and a logo.
Step 2. Choosing the model, and why it is a decision about money
The price spread between models is more than tenfold. This is the first decision, and it matters more than all the others.
KADR prices, 6-second clip, 1080p, sound off, October 2026:
| Model | Price, rubles | Best for |
|---|---|---|
| Google Veo 3.1 Lite | 51 | drafts, simple scenes, backgrounds |
| Seedance 1.5 Pro | 73 | volume content, lots of clips |
| Kling Video O1 (720p, 5 s) | 98 | camera movement, dynamics |
| Google Veo 3.1 Fast | 101 | good quality for little money |
| Kling v3.0 | 112 | people, complex motion |
| Seedance 2.0 | 572 | flagship, character held between shots |
This is not a public offer. Kling Video O1 tops out at 720p and only does 5 or 10 seconds, so its price is given for 5 s at 720p (10 s cost about 196 rubles). For trends where a person repeats a dance or a movement from another video there is Kling Motion Control: you upload a photo of the person and the reference video, and the model transfers the movement; at 720p about 13.4 tokens per second, 10 seconds come to about 212 rubles.
The full price table for all models and the per-minute calculation are in a separate article. The practical takeaway: run drafts on cheap models and do the final on an expensive one. Working out the logic of a scene on Veo 3.1 Lite for 51 rubles and then moving the debugged prompt to Kling is not penny-pinching, it is a several-fold difference at volume.
Step 3. Two levers almost nobody touches
Resolution. The same clip on Seedance 2.0 costs 109 rubles at 480p and 572 rubles at 1080p. More than a fivefold difference. While you are tuning the prompt and do not know what you will get, you do not need high resolution at all: you are checking whether the camera goes the right way, not examining skin pores.
Sound. On models with native audio, turning sound on doubles the price. Veo 3.1 costs 132 tokens without sound and 264 with it. And in Reels you almost certainly do not need the model’s sound: a track or voice-over goes on top anyway. Turn it off and pay half.
There is one exception: if someone speaks in the frame and lip-sync matters. Then the model’s sound is justified.
Step 4. How not to look like everyone else
The problem with AI Reels is not quality but sameness. They all look like perfume ads: gloss, slow motion, golden hour, faceless beauty. It happens because the model pulls towards an averaged pretty picture by default, and people write averaged prompts.
The cure is specifics.
- Set the lens. «Shot on 35 mm, slight distortion at the edges» gives a completely different shot than nothing at all.
- Set the character of the light. Not «beautiful light» but «hard backlight from a window, deep shadows».
- Ask for imperfection. Film grain, slight camera shake, a reflection in glass. A perfect picture reads as artificial precisely because it is perfect.
- Give a reference instead of words. If you have an image in the style you want, feed it as a sample. It is more precise than any description. If one hero repeats across shots, read about character consistency.
- Do not use words like «cinematic» and «atmospheric». They mean nothing specific, and the model reads them as «make it glossy».
Step 5. Assembly
The scenes are ready; next comes ordinary editing in any editor.
- Music. It sets the rhythm of the cuts. In practice: pick the track first, then cut the scenes to its beats, not the other way round.
- Text. Most people watch without sound, so the meaning has to be readable with the eyes. Large font, short phrases, high contrast.
- Length. Cut ruthlessly. If a scene does not move the story, it is superfluous, even if it is beautiful and you paid for it.
What a whole Reel costs
A 30-second video, four scenes, 1080p, sound off. KADR data, October 2026.
| Model | One scene | Reel (4 scenes) | With regenerations (×3) |
|---|---|---|---|
| Veo 3.1 Lite | 51 rubles | 202 rubles | about 610 rubles |
| Seedance 1.5 Pro | 73 rubles | 291 rubles | about 870 rubles |
| Kling v3.0 | 112 rubles | 449 rubles | about 1350 rubles |
| Seedance 2.0 | 572 rubles | 2289 rubles | about 6870 rubles |
This is not a public offer.
A factor of 3 is not pessimism, it is the norm. A simple scene works on the first attempt; a scene with a person can take ten. Budget from the result, not from a single generation.
For musicians: a music video with AI.
In short
Making a Reel with AI is not one button but a storyboard plus several short generations plus editing. Generate in 9:16 from the start, split into scenes with one action each, run drafts at 480p on a cheap model, and turn sound off if you are adding music anyway.
The biggest saving is hidden not in discounts but in the choice of model: the price spread is more than tenfold.
