How to make YouTube Shorts with AI (from one product photo to a finished Short)

Of every platform I've tried for AI video, YouTube Shorts is the most forgiving. The bar for production quality is low, the format is short, and faceless content is completely normal there. If you sell anything visual — a product, a space, a craft — you can run a Shorts channel without ever filming yourself.

This is my full process, tested over about twenty published Shorts. It takes me 20-25 minutes per video, and most of that is waiting for generation.

Why Shorts works for AI video

Three practical reasons:

• The format is short. 15-30 seconds is 3-5 AI-generated clips. That's one generation session.
• Imperfection is fine. Viewers accept slightly surreal motion in AI clips far more than they would in a TV ad. The algorithm cares about whether people watch to the end.
• One photo is enough raw material. Image-to-video turns a single decent product photo into motion. No shoot, no lighting setup.

The catch: AI clips alone don't get views. The structure around them does. That's what steps 2 and 4 handle.

Step 1: Start from your strongest photo

Pick one photo of your product — clean background, good light, product clearly visible. Phone photos work. This becomes the seed for image-to-video: you upload it, then describe the motion you want.

What makes a photo "strong" for this:

• The product fills a good part of the frame
• Nothing in the shot you'd have to explain (clutter reads as chaos once it's moving)
• Decent resolution — the video can't add detail the photo never had

If you only have lifestyle shots with busy backgrounds, one of those can still work. Just expect to generate a couple of versions.

Step 2: Write the 3-beat script (5 minutes)

Every Short that has performed for me follows the same three beats:

1. Hook (0-3s): the most visually surprising frame you can generate, paired with one line of text
2. Demo (3-15s): the product doing its thing — one clear action per clip
3. Payoff (last 3s): the result, plus a nudge to follow or comment

Example, for a self-watering planter:

• Hook: the planter seen through a window, rain outside, text: "I forget to water my plants on purpose."
• Demo: close-up of the water reservoir; a cross-section shot of water rising through the soil.
• Payoff: the plant, thriving, next to a calendar marked with crossed-out days. Text: "Water it once a month."

Write these as three bullets. Each bullet becomes one generation. Don't write full sentences of narration — on Shorts, text overlay carries the story and the visuals carry the proof.

Step 3: Generate the clips

Open the generator and work through your three bullets. Two different entry points, and I use both in the same video:

• Beats that start from your photo: use Image-to-Video. Upload the photo, prompt the motion: "slow push-in on the ceramic planter, water droplets forming, shallow depth of field."
• Beats that need something you don't have a photo of (the cross-section shot, a lifestyle scene): use Text-to-Video with a full description.

Model choices that have worked for Shorts:

• Cheap and fast models for the demo beats — you'll probably regenerate one of them anyway
• Kling when the beat needs a cinematic camera move for the hook
• Seedance if a person's hands or motion appear in frame — it handles people better

Keep each clip short (2-4 seconds of footage is plenty per beat), and generate two versions of your hook beat. The hook is where Shorts are won or lost; it deserves the extra credits.

Step 4: Assemble in the YouTube app

Don't leave the platform to edit — the built-in tools are enough:

1. Clip the fat. Trim each generated clip to its best 2-4 seconds.
2. Add text overlays for the three beats. Big font, top third of the screen (the bottom gets covered by the caption and buttons).
3. Add a trending sound at low volume under the video, or no sound — Shorts watch fine muted, which is why the text matters.
4. Crop check: if any clip came out non-vertical, crop it to 9:16 here. Generation isn't always vertical by default, and a letterboxed Short looks unfinished.

Total editing time: about 5 minutes. If it's taking longer, you're polishing beats that weren't the problem.

Step 5: Package and post

• Title: one line, curiosity-shaped. "I haven't watered this plant in a month" beats "Self-watering planter demo."
• First comment: one useful sentence — price, link, or the question you want people to answer. It starts the thread the algorithm likes.
• Post time: less important than people claim. Consistency matters more; three decent Shorts a week beats one perfect one.

What 20 Shorts taught me

• The hook frame is 80% of the result. I now spend more credits on the first 3 seconds than on everything after.
• Real physics sells AI clips. Water, steam, sand, pouring, sparks — models do texture and motion beautifully, and viewers rewatch them.
• One idea per Short. The moment a Short tries to show two features, watch-through drops. The second feature is the next Short — that's how you get your three per week.
• Weak clips die quietly. When a Short underperforms, I repost the same 3-beat structure with a new hook frame before trying a new topic. It works more often than not.

Related reading

• How to use MakeClipAI: step-by-step tutorial — the basics of your first generation
• Image to video: animate your product photos — going deeper on the photo-to-video workflow
• AI video for social media: what actually works — what drives engagement per platform

Make your first Short — one photo and 20 minutes is a fair price for a channel that runs itself.