AI Videos for Social Media: How to Make Short Vertical Clips That Get Watched
A Kanakaw guide · about 8 min read
Vertical 9:16 Is the Default for Social — With One Exception
On TikTok, Instagram Reels and YouTube Shorts, video is watched upright, full-screen, one clip at a time. A vertical 9:16 clip fills the whole phone screen: the subject is bigger, the first frame hits harder, and there is nowhere for the eye to wander. A horizontal 16:9 clip on the same feed gets letterboxed into a small strip in the middle of the screen, with black bars above and below — and a small clip is a clip people swipe past.
That does not mean 16:9 is dead. It still fits anywhere a video sits inside a wider page: the main YouTube feed, an embedded video on your website, a landscape ad placement, a presentation. The square 1:1 ratio works when one post has to serve the Instagram grid and Facebook at the same time, and 3:4 is a useful compromise for feed posts that are mostly viewed on phones but also appear on desktop. Kanakaw offers all five — 16:9, 9:16, 1:1, 4:3 and 3:4 — so pick the ratio for the place the clip will actually live before you generate it. You cannot change the shape afterwards without cropping the video or generating it again.
The practical rule is simple: one ratio per destination. If a clip has to work both as a Reel and as a YouTube video, generate it twice — once in 9:16 for social, once in 16:9 for YouTube. Two generations cost more than one, but a cropped, blurry compromise costs you the viewer, which is the more expensive loss.
Text-to-Video or Photo-to-Video: Pick by What You Already Have
There are two ways to make a social clip on Kanakaw, and the right one depends on what you are starting from. Text-to-video builds a scene from a written prompt. It is the choice for anything you cannot film: a product floating above a beach at golden hour, a city street in the rain at night, a slow aerial sweep over a landscape that does not exist. If the idea needs a scene that is impossible, expensive or far away, write it as a prompt and let the generator build it.
Photo-to-video starts from a picture you already have and animates it. This is the better choice when the hook depends on something real: your product, your face, a real place. A product photo works well when the motion is small and believable — the bottle rotates a few degrees, steam rises from a cup, light moves across the label. A portrait works when the movement is subtle — a slow turn of the head, a small smile. A landscape photo comes alive with drifting clouds or rippling water. One photo goes into each clip, and the photo is the anchor: the more the motion respects what is already in the picture, the more natural the result looks.
As a rule of thumb: if the idea needs a real thing, animate a photo. If the idea needs a scene that does not exist, write a prompt. Both run on the same video engines behind the scenes — Wan 2.2, MiniMax Hailuo and LTX-2 — and the generator picks the right one for the job, so there is nothing to configure there. Your decisions are the ratio, the length, the resolution and the prompt.
Write Motion Prompts That Work on Mute
Most social video is watched on mute, at least for the first few seconds. Sound is off by default in the feed, and many viewers never turn it on. Your clip has to read with the sound off: one clear subject, one movement, and something worth looking at in the very first second. If the first second is a slow fade from black or an empty room, the viewer is already gone.
A good motion prompt has five parts: the subject, the setting, one camera move, one subject action, and the mood or light. For example: "A ceramic coffee cup on a wooden table, morning light through a window, steam rising slowly, camera pushes in gently, warm and calm." Or: "A woman in a red jacket standing on a city street at dusk, she turns her head toward the camera, slow tracking shot, neon signs glowing." Each prompt describes one thing happening, not three.
Keep the motion slow. The camera moves that suit vertical social video are the gentle ones: a slow push-in toward the subject, a slow pan across a scene, a tracking shot that follows someone walking, a slow orbit around a product. Fast motion is where AI video breaks down — faces and hands can wobble, edges smear, objects morph mid-frame. Small, slow movement hides all of that. Two other things to avoid: asking for two actions at once ("she runs and the camera spins"), and putting words inside the scene (text rendered inside AI video often comes out garbled). Since Kanakaw videos are silent, put your message in captions added in your editor afterwards — CapCut or any editor will do — rather than in the video itself.
Length, Cost and Honest Limits
Five seconds is enough for a hook. On social, the decision to keep watching is made in the first one or two seconds, so a short, sharp clip often outperforms a long one. Use 5 seconds to test an idea and for daily hooks; 10 seconds for a small story with a beginning and an end; 15 seconds when the clip needs a full beat — a product reveal, a before-and-after, a joke with a setup. If you need something longer, generate several clips and stitch them together in your editor: each clip is one shot, and a sequence of shots is a video.
What a clip costs depends on where it starts, how long it is and how sharp it is. A 5-second text-to-video clip costs 238 credits at 480p, 476 at 720p and 951 at 1080p; 10 seconds costs roughly twice that, 15 seconds roughly three times. A 5-second image-to-video clip costs 357 credits at 480p, 713 at 720p and 1,426 at 1080p. The cheapest way to test an idea is a 5-second text-to-video clip at 480p. For a clip that will actually be posted, 720p is the sweet spot on phones — visibly sharper than 480p at about half the price of 1080p. The generator always shows the exact credit cost before you create, so there are no surprises.
To put that in budget terms: the $19.99 pack (1,000 credits) covers two 5-second 720p text-to-video clips. The $59.99 pack (5,000 credits) covers ten of them — a week of daily hooks with credits to spare. Video generation is members-only, and membership comes with any credit pack: 30 days with a monthly pack, 12 months with an annual pack. There is no free trial, and sign-in is with Google only.
Finally, the honest limits. Clips are 5, 10 or 15 seconds — nothing longer, so plan in shots. Videos are silent, with no audio track at all: add music, a voiceover or captions in your editor. Image-to-video takes exactly one photo per clip, so a two-photo transition needs two clips cut together. Faces and hands can wobble on fast motion, which is why slow camera moves work best. And you can use the videos you generate commercially, subject to the Terms of Service — including on a business account or in a paid ad.
Frequently Asked Questions
What size should an AI video be for TikTok or Reels?
Vertical 9:16, at 720p. It fills the phone screen on TikTok, Reels and Shorts, where a horizontal clip would be letterboxed into a small strip. Use 16:9 only for the main YouTube feed, embeds or landscape placements, and 1:1 when one post has to serve the Instagram grid and Facebook at the same time.
How much does one social video cost?
A 5-second text-to-video clip costs 238 credits at 480p, 476 at 720p and 951 at 1080p. A 5-second image-to-video clip costs 357 credits at 480p, 713 at 720p and 1,426 at 1080p. Ten seconds costs roughly twice as much, 15 seconds roughly three times. The generator shows the exact cost before you create.
Do Kanakaw videos have sound?
No. Kanakaw videos are silent — there is no audio track. That suits social feeds, where most clips are watched on mute anyway. Add music, a voiceover or on-screen captions in your editor after you download the clip.
Can I use Kanakaw videos on my business account or in ads?
Yes. Videos you generate can be used commercially, subject to the Terms of Service — including on a business account, in organic posts and in paid ads. The same applies to images you generate on Kanakaw.
Do I need a membership to make AI videos?
Yes. Video generation is members-only. Membership comes with any credit pack you buy — 30 days with a monthly pack, 12 months with an annual pack — and there is no separate membership fee. Standard image generation only needs you to be signed in.
Try it: Text to Video · Pricing · All guides