AI Video for TikTok, Reels and Shorts - 9:16 Workflow

To make an AI video for TikTok, Instagram Reels or YouTube Shorts, pick a video model that renders natively in 9:16, split 15-30 seconds into 2-4 shots, build the first shot as a 2-second hook, and add captions and logos in your editor rather than in the prompt. On Jantaris almost every video model supports vertical output, for example Seedance 2.5 (from $0.261 per second) with single shots up to 30 seconds, or the budget Seedance 2.0 Mini (from $0.0366 per second). Below: the 9:16 models and their limits, a second-by-second structure, caption safe zones and what a 15 or 30 second video costs.

Render vertical from the start

TikTok, Reels and Shorts play full screen on a phone at 9:16, and the standard export is 1080 x 1920 px. Cropping a landscape clip to vertical throws away about two thirds of the frame and pushes your subject off the edge. Set 9:16 in the form before you hit Generate ("aspect_ratio": "9:16" in the API).

Start frames follow the same rule. In my own Z-Image Turbo test, the same prompt and seed rendered in vertical produced a different object than the landscape version, so generate start frames in 9:16 too:

Which AI video models output 9:16

These are the limits from the Jantaris catalog, which is what the form and API actually accept:

"From" prices are for the lowest resolution; Veo 3.1 Fast is billed per 8-second clip. Studio shows the exact amount for your settings under the form. The only models without a ratio setting are the Motion Control ones, which follow the input video.

How I pick:

• one long 15-30 s shot with sound: Seedance 2.5 or Wan 3.0,
• cheap drafts and daily series: Seedance 2.0 Mini at 480p or 720p, Vidu Q3 Turbo,
• polished short product shots: Kling 3.0 at 1080p,
• a character talking to camera: Kling 3.0 or Vidu Q3 with audio on.

Alibaba's official Wan 3.0 example: a vertical video built around a product passed as a reference image.

A 15-30 second structure

A short is not one prompt; it is 2-4 shots with a job each. My 20-second template:

The 2-second hook

Viewers decide almost instantly whether to keep watching, so I treat the first shot differently:

• open on motion or change, never on a static wide shot,
• put the strongest frame (effect, result, emotion) first, not last,
• write motion that starts on frame one: the camera rushes forward from the first frame, immediate, sudden,
• add a question or promise as a caption in the first second, in the edit.

Video models sometimes take a moment to get going, so I generate the hook a little longer (5 s instead of 3 s) and trim the opening frames in the edit. A cheaper model at 480p is enough to test three or four hook variants before rendering the final one.

Make it loop

Shorts replay automatically. If the last shot flows into the first, the cut is invisible. On models with a last-frame field (end_image_url on Seedance, Kling and Wan), pass the same image your video opens with.

A vendor sample from MiniMax H3 Max shows the fast-cut, high-energy style that tends to work in short formats (this clip is 3:4; the model also does 9:16):

Captions and safe zones

I always add captions in the editor (CapCut, DaVinci Resolve, or the app's own editor), not in the video prompt:

• video models distort text in motion, letters drift from frame to frame,
• a caption in the edit takes seconds to fix, a caption in the generation means a new generation,
• auto-captions from speech need a clean file without burned-in text anyway.

Where not to put text and logos:

• the bottom: caption, handle and music bar cover roughly the lower fifth,
• the right edge: like, comment and share buttons,
• the very top: tabs and status icons.

The safe band runs from about 15% to 70% of the frame height. I mirror that in the prompt: subject centered, empty space above the subject.

What a 15-30 second video costs

Seconds of video times the rate, times the number of attempts:

Higher resolutions raise the per-second rate, so I draft hooks at 480p or 720p and render only the keeper at 1080p. A soundtrack from Mureka V9 adds $0.124 per song. Failed generations cost nothing. A full breakdown with voiceover and music is in AI video cost per second.

From a photo

Got a product shot, a character or a location photo? Use it as the first frame. In Studio, paste the image URL into the start frame field or upload a file. On Kling the aspect ratio follows the image, so the photo has to be vertical already.

The full photo-to-video process is in the image-to-video guide, and camera moves that work in shorts are in AI video prompts and camera moves.

Shorts at scale: API and automation

One video a week is a Studio job. One a day is worth automating. A minimal vertical clip request:

curl https://jantaris.com/v1/generate \
  -H "Authorization: Bearer $JANTARIS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2.0-mini",
    "input": {
      "prompt": "The camera rushes forward from the first frame into a steaming bowl of ramen, chopsticks lift noodles, vertical framing, subject centered, empty space above the subject",
      "aspect_ratio": "9:16",
      "resolution": "720p",
      "duration": 6,
      "generate_audio": true
    }
  }'

The response returns an id and estimated_cost; fetch the finished file with GET /v1/tasks/{id} or a webhook. For batches, the async pattern with signed webhooks is in AI video pipelines with webhooks, and n8n, Make and Zapier setup is in the no-code docs. I still leave captions, the final edit and posting to a human.

FAQ

How do I make a TikTok video with AI?

Pick a video model with 9:16 output (Seedance 2.5, Kling 3.0, Wan 3.0), plan 15-30 seconds as 2-4 shots with a hook first, generate each shot vertically, then assemble them in CapCut with captions and music.

What size is a TikTok, Reels or Shorts video?

Vertical 9:16, usually 1080 x 1920 px. Generate in 9:16 from the start; cropping landscape footage loses most of the frame.

How much does an AI TikTok video cost?

Seconds of video times the model's rate times attempts. A 15-second video with two attempts on Seedance 2.0 Mini is 30 × from $0.0366 per second; 30 seconds on Seedance 2.5 with two attempts is 60 × from $0.261 per second.

Can the AI model add captions?

It can render text in the frame, but text in AI video distorts easily. Add captions, logos and calls to action in the edit, away from the bottom fifth and the right edge.

How long can an AI short be?

A single shot runs up to 30 seconds on Seedance 2.5 and Wan 3.0, up to 15 on Kling 3.0, and a fixed 8 on Veo 3.1. Longer videos are several shots cut together.

Can I make AI shorts for free?

Jantaris has no free plan or signup credits; the minimum top-up is $10, you only pay for successful generations and the balance never expires. Some model vendors offer limited free tiers in their own apps.