TL;DR
Short-form video has its own grammar: 9:16 vertical, a hook in the first second, cuts every 1-2 seconds, and a loop or payoff at the end. One generic prompt does not serve Douyin, TikTok, Reels, and Shorts equally because each algorithm rewards something different. Below are four ready-to-use prompt templates (one per platform) for the same topic, the current generation limits per tool (as of June 2026), and a checklist so every clip lands native. Recommended generators for vertical work right now: Veo 3.1 (true native 9:16, up to 4K) and Kling 2.5 Turbo (fast, social-tuned, ~$0.35 per 5s clip).
What changed in 2026
Two things made native vertical generation practical:
- Veo 3.1 added native 9:16 output in its January 2026 update (previously you generated 16:9 and cropped), plus 4K upscaling and contextual audio. It generates in roughly 8-second chunks, so longer pieces are stitched. Access is through the Gemini app (Google AI Pro at $19.99/mo or AI Ultra at $99.99/mo), Flow, and the Vertex AI / Gemini API.
- Kling 2.5 Turbo outputs 1:1, 9:16, and 16:9 at 1080p/30fps in 5- or 10-second clips, priced around $0.35 for 5s and $0.70 for 10s. It is the most approachable option for batch social content.
OpenAI wound down the standalone Sora consumer app in April 2026 and folded video generation toward ChatGPT; if you rely on Sora, confirm current access on OpenAI’s pages before building a workflow around it. For the templates below, any model that accepts a text prompt and a 9:16 target will work.
Why one template doesn’t fit four platforms
All four formats are 9:16 vertical at 1080×1920, but the algorithm rewards different signals, the native style differs, and the sweet-spot length differs. A single template underperforms on every platform because it optimizes for none.
Douyin
- Format / length: 9:16, 1080×1920. 7-60s, 15s most common.
- Algorithm signals: completion rate dominates; replays, comments, and shares matter; no native “save.”
- Native style: heavy on-screen captions; fast cuts (0.5-1.5s); a first-second hook is non-negotiable; trending sounds layered under voiceover.
TikTok
- Format / length: 9:16, 1080×1920. 15s is most algorithm-friendly; up to 10 min but rarely surfaced.
- Algorithm signals: completion rate, shares, comment depth, follows from hashtags. The niche FYP dominates reach.
- Native style: subtitles on by default; “first 2s stops the scroll”; trending sounds; duet/stitch loops; monetizing creators get pulled toward 60s+.
Reels (Instagram)
- Format / length: 9:16, 1080×1920. Up to 90s; 15-30s sweet spot.
- Algorithm signals: shares and saves carry heavy weight; reach to non-followers; comment depth.
- Native style: looping clips (the end visually matches the start), aesthetic-forward color grading, a consistent brand palette, and a cover frame that reads well in the feed grid.
Shorts (YouTube)
- Format / length: 9:16, 1080×1920. Max 60s.
- Algorithm signals: average view duration, swipe-away rate, and subscribe-clicks-from-Short. YouTube optimizes Shorts to convert viewers into channel subscribers.
- Native style: more on-screen text than Instagram; “tip-format” educational clips work well; often a teaser to long-form on the same channel.
Same topic, four prompts: “remote work”
Drop these into Veo 3.1, Kling 2.5 Turbo, or any 9:16-capable generator. Each is built to the platform’s native style above.
Douyin:
9:16, 1080x1920, 15s. First frame: a cat sitting on an open laptop keyboard,
owner's hand reaches in to push it off and the cat glares.
Bold caption "Day 87 of remote work". 4 fast cuts: cat blocks screen, coffee
spills, takeout boxes on desk, person on a call from bed. Cut every 1s, last
frame loops back to the first. Trending emotional BGM under voiceover.TikTok:
9:16, 1080x1920, 15s. First 2s: a person stares into camera and says
"remote work isn't what you think", caption burned in.
Mid: three contrast cuts (suit on top + pajama pants, bed = office,
2am answering email). End: punchline "but I'm never going back". Stitch-friendly.Reels:
9:16, 1080x1920, 20s. Aesthetic-forward: natural light, beige + wood palette.
Open on a latte-art close-up, pull back to a desk with a laptop. Three location
cuts: window seat, balcony, cafe. End frame matches the start frame (seamless loop).
Cover frame: latte art. Lo-fi piano BGM.Shorts:
9:16, 1080x1920, 45s. Burned-in title "3 things nobody tells you about remote work".
Three ~12s segments, each opens with a big number caption. End with a direct
"full breakdown on my channel" pointing to a long-form video. Pace slightly faster than IG.Which tool for which platform
Pick by clip length, output ceiling, and how much you batch. Figures as of June 2026.
| Need | Veo 3.1 | Kling 2.5 Turbo |
|---|---|---|
| Native 9:16 | Yes (since Jan 2026) | Yes |
| Max clip per generation | ~8s, stitch for longer | 5s or 10s |
| Resolution ceiling | Up to 4K (3840×2160) | 1080p / 30fps |
| Built-in audio | Contextual audio | Add in editor |
| Rough cost | Gemini plan ($19.99 / $99.99 mo) | ~$0.35 / 5s, ~$0.70 / 10s |
| Best for | Polished Reels, ad hero shots | High-volume TikTok / Douyin batches |
For prompt fundamentals before you batch, see the AI video prompt basics guide and the aspect ratio guide. Tool walkthroughs: Sora / Veo for beginners and the Kling tutorial.
Step by step
- Lock 9:16 first. Generate vertical natively. Producing 16:9 then cropping wastes credits and loses the top and bottom of every frame.
- Write the first-second hook explicitly. State the opening frame:
first frame: striking visual that pays off in seconds 2-5. Do not leave the open to chance. - Keep clips 3-8 seconds. Since Veo caps near 8s and Kling at 5-10s, build to those lengths and stitch. Longer single shots lose viewers.
- Ask for medium-to-high motion. Static clips die in feeds. Name the motion: a push-in, a pan, a hand entering frame.
- Compose subject in the upper third. Phone UI and captions cover the bottom. Tell the model to keep the subject high.
- Make it loop. When the end frame connects back to the start, the clip feels native, especially on Reels.
- Generate 4-6 variants per shot and keep the one with the strongest first second.
A real batch: 5 fitness Reels
Goal: a week of fitness Reels from one session.
- Write five prompts, each:
9:16, first-frame hook, single clear motion, 5s, subject in upper third, natural light. - Generate 5 variants of each in Kling 2.5 Turbo (5s clips keep cost near $0.35 apiece).
- Pull the strongest variant of each into your editor, trim to the hook, add on-brand captions, and lay a trending audio track.
- Export each as a standalone 5-8s Reel, or stitch into one 30s compilation. Set the cover frame on the most striking single frame.
Common mistakes
- Generating before setting the aspect ratio, then cropping away the framing.
- Slow openings. On short-form the first second is the whole game.
- Subject centered, so the caption overlay covers it.
- One long clip instead of several short cuts.
- Prioritizing visual polish over a trending audio track. On TikTok and Reels, the audio often drives reach more than the visuals.
Advanced tips
- A/B test hooks for ads. Generate 10 clips with the same product but a different first-second hook each, then test which converts. See the AI video social ads workflow.
- Save platform prompt prefixes so batches stay consistent, for example:
vertical 9:16, mobile-optimized, captions-safe lower third, subject upper third. - Match the BGM to the platform. Trending sounds amplify reach; for original music see Suno for short-video BGM.
Output checklist
- 9:16 vertical, generated natively (not cropped).
- First frame is a real hook, not a generic establishing shot.
- Subject in the upper third so captions don’t cover it.
- 3-8 seconds per clip, stitched for longer pieces.
- Medium-to-high motion in every clip.
- A platform-appropriate audio track.
FAQ
- Which tool is best for short-form right now? For polished Reels and ad hero shots, Veo 3.1 (native 9:16, up to 4K, contextual audio, via a Gemini plan). For high-volume TikTok and Douyin batches, Kling 2.5 Turbo (5-10s clips, ~$0.35 per 5s). Always test on your actual platform before committing a series.
- Can I generate a full 30-second clip in one shot? Not with the leading tools as of June 2026. Veo 3.1 caps near 8 seconds per generation and Kling at 5-10 seconds, so longer pieces are stitched in an editor. This matches short-form pacing anyway, where you want cuts every 1-2 seconds.
- Will TikTok penalize AI video? There is no blanket ranking penalty as of June 2026, but TikTok, Instagram, and YouTube require disclosure of realistic AI-generated or AI-altered content. Label it in the platform’s AI-content toggle and check each platform’s policy, since requirements change.
- Do I need different prompts per platform, or can I reuse one? Reuse the core scene, but change length, pacing, and ending: a tight loop and trending sound for Douyin/TikTok, an aesthetic seamless loop for Reels, and a subscribe-oriented payoff for Shorts.
- What resolution should I export? 1080×1920 covers every platform. Generate at the highest the tool allows (Veo can hit 4K) and export down; never upscale a cropped 16:9 source.