Video for Reels, Shorts and TikTok with a neural network

Short vertical video is the cheapest way to reach a new audience in 2026. But shooting it by hand every day is hard: you need an idea, a frame, light, editing. A neural network covers the most expensive part — the picture in motion. You describe a scene in words or animate a ready photo, and then it's just a matter of adding captions, sound and posting. This chapter is about how to make clips for the feed so that people watch them to the end.

Which platform and which format

Almost all short-form is vertical 9:16. But platforms have their own habits for length and delivery:

• TikTok. 9:16, 15–30 seconds for recognition, 60–90 for a breakdown or tutorial. The algorithm likes clips set to a trending sound and with no slow "warm-up" at the start.
• Instagram Reels. 9:16 for reels, 4:5 and 1:1 for the feed. 15–60 seconds. Visual recognizability matters: keep one style from clip to clip.
• YouTube Shorts. 9:16, up to 60 seconds. Here retention and usefulness are valued — Shorts works well as a "cut" and B-roll to a long video.

The first three seconds decide everything

The feed is scrolled by a thumb in fractions of a second. If there's no movement or intrigue in the first frame, the clip is scrolled past before your story begins. So generate the first frame separately and pickily: bright movement, an unusual angle, "what is that even?". Slow intros and a 3-second logo kill watch-through.

Describe a hook in one sentence — see what the model produces. For your first time, try: "a sharp camera dolly through a stream of coffee into a cup, slow motion, droplets in the air, warm light". Video costs more than pictures: a clip is available after signing up and onboarding — which grant starter tokens.

What to assemble a whole clip from

A finished reel isn't just the picture. In Twelver the whole pipeline is assembled in one chat, with no separate apps:

1. Scenes. text to video for frames from scratch or animating a photo if you already have a shot of the product or hero.
2. Voice-over. voicing text — a live narrator without recording yourself on a mic.
3. Music. a background track for the mood, with no disputes over the copyright of someone else's hits.
4. Subtitles. automatic subtitles — most feed clips are watched without sound, and captions hold attention.

Common mistakes

• Too much action in one frame. "A person runs, turns, waves and jumps" in 5 seconds falls apart. One clear movement per scene.
• Text inside the generation. Letters still "drift" with models. Overlay headings and captions on the finished clip, don't ask the network to draw them.
• Every clip in a new style. Social is about recognition. Keep one visual handwriting (light, colour, rhythm), and the feed starts reading like your brand.
• Zero sound. Even if people watch without sound, a track and subtitles affect watch-through and reshares.

What's next

Feed content often advertises a product. The next chapter — how to assemble a selling ad video from an ordinary product photo, one you won't be ashamed to put into promotion.

In the Twelver chat the clip, voice-over, music and subtitles are assembled in one conversation — no separate apps needed. Starter tokens for video are granted after signing up and onboarding.