How to make TikTok shorts with AI in 2026 (full workflow + free tools)
Step-by-step: generate, caption, and export 9:16 TikTok shorts using AI video models. Free tier, no editor required, ready to upload in 5 minutes.
TikTok's algorithm rewards consistency: 1 short per day for 30 days beats 1 viral hit. AI video tools turn that 30-day commitment from "impossible" into "coffee break."
This is the workflow we use to produce daily TikTok shorts at the Vivix studio. Five minutes per clip, $0.50 per render, no Premiere Pro or CapCut required.
What you need
- A free Vivix account (30 credits, no card)
- An idea for a 10-second clip
- A TikTok account
That's it. No editing software, no expensive subscription.
Step 1: Pick your model
For TikTok shorts, the sweet spot is Veo 3.1 Fast (cheapest, native audio, 720p which TikTok upscales fine) or Kling 3.0 Standard (cheaper still, no audio but TikTok will let you overlay music anyway).
Open Vivix Text-to-Video and pick one. The credit cost is shown before you click generate.
Step 2: Write a prompt that works in 9:16
Vertical video has rules. The eye reads top-to-bottom, not left-to-right, so:
- One subject, centered. Multiple subjects fight for attention in a narrow frame.
- Vertical motion beats horizontal. Pan up/down or zoom, not side dolly.
- High contrast. TikTok thumbnails are tiny — bold colors and strong silhouettes win.
- First 1.5 seconds = the hook. Front-load the wow.
Example prompts that work:
- "A neon goldfish swims toward the camera in a dark aquarium, slow zoom in, cinematic, 9:16"
- "Hands kneading dough on a marble counter, top-down view, soft window light, 9:16 vertical"
- "A tiny astronaut on a blue planet looks up at three moons, slow tilt up, 9:16, cinematic"
Step 3: Generate at 9:16
Set the aspect ratio to 9:16in the Vivix model controls. Most models support it natively; for the few that don't (some older Wan versions), Vivix auto-crops with subject detection so the speaker stays centered.
Hit generate. Veo 3.1 Fast finishes in ~30 seconds; Kling in ~60.
Step 4: Add captions (the algorithm boost)
TikTok watch-time goes up ~40% with burned-in captions (Sprout Social study, 2024). Open Vivix Caption Studio:
- Drop the rendered MP4
- Whisper transcribes every word with timing in ~10 seconds
- Pick a viral caption style (Bold TikTok works best)
- Drag any line you want to fix and edit the text
- Export 9:16 with captions burned in
1 clip credit per export. Free signup includes 1 clip credit.
Step 5: Hook test before you post
Before uploading to TikTok, watch the first 1.5 seconds with sound off. If you don't feel a pull, the clip won't hold a stranger either. Re-prompt with a bolder opening shot and burn another credit.
Step 6: Upload + caption + tags
- TikTok caption: 1 sentence, 1 question. Questions drive comments. Comments drive distribution.
- Tags: 3-5 max. Mix one big (#ai), one medium (#aivideo), one niche (#aivideoartist).
- Sound: if your clip has Veo native audio, leave it. If silent, pick a trending sound — TikTok's algorithm bumps clips using trending audio.
Multi-platform: post once, distribute everywhere
Same 9:16 MP4 also works for:
- Instagram Reels(re-upload, don't cross-post — Reels demotes obvious cross-posts)
- YouTube Shorts (auto-detected as Short if vertical and under 60s)
- X / Twitter (vertical works, just not optimized)
- LinkedIn (yes, vertical AI clips perform well there now)
See the dedicated YouTube Shorts Maker page for the Shorts-specific workflow.
Cost breakdown for daily posting (30 days)
- 30 generations × $0.40 (Veo Fast) = $12
- 30 captioned exports × ~$0.12 each = $3.60
- Total: ~$16/mo for 30 daily TikTok shorts
Vs. Sora.com Pro at $200/mo or hiring an editor at $50/clip ($1,500/mo).
Common mistakes
1. Generating at 16:9 then cropping
You lose ~50% of the visual information. Always generate at 9:16 native.
2. Skipping captions
85% of TikTok is watched with sound off in public. No captions = no watch time.
3. Putting the hook at second 5
The algorithm decides at ~1.5 seconds whether to keep showing your video. The cool part needs to be first, not last.
4. Using one model for everything
Different shots need different models. Human dancing? Kling. Cinematic narrative? Sora. Dialogue or SFX? Veo. Vivix lets you swap mid-session.
Try it tonight
Sign up free — 30 credits + 1 clip credit, enough for ~3 generations and 1 captioned export. See if AI video shorts fit your channel before paying anything.
Try Vivix free — 30 credits + 30 daily
Over 100 frontier AI models in one studio. Same models on free as on paid.
Start freeBe the first to know
Subscribe to the Vivix newsletter and you'll hear it first whenever new models land or new features go live. No promo spam. Unsubscribe in one click.
We use your email only for the newsletter. Unsubscribe anytime.