How to Make an AI Video: A Practical Step-by-Step Guide

Most first attempts at AI video fail for a boring reason: the prompt describes a scene, but the model has to produce a single continuous shot. Five seconds is short. The workflow below is built around that constraint, from picking an input mode to downloading an MP4 you can actually use.

Think in one shot, not one scene

Before you open any tool, decide what happens in a single take. One subject, one action, one camera move. "A woman walks through a market, then turns and smiles at the camera" is two beats, and five seconds will not hold both. "A woman walks away through a night market, camera tracking behind her" is one shot and fits.

If your idea needs several shots, generate them as separate clips and cut them together in an editor. Each generation is one shot.

Step 1: Choose text-to-video or image-to-video

There are two starting points, and they solve different problems.

Text-to-video starts from a description. Use it when you have no source image and want to control the whole frame. This site renders 16:9 and 9:16 text-to-video clips. See text-to-video for the full setup.

Image-to-video starts from a photo or illustration you already own. Upload a PNG or JPEG and describe how it should move. The composition you upload is the composition you get, which is useful when a product, layout or character must stay exactly as designed. Image-to-video currently renders 9:16 only; see image-to-video for the accepted file types and limits.

A simple rule: if the frame matters more than the motion, start from an image. If the motion matters more than the frame, start from text.

Step 2: Write a prompt in six parts

A prompt that works usually names six things, in plain language:

1. Subject — one clear thing in frame: a perfume bottle, a mountain valley, a person in a coat.
2. Action — one slow movement: steam rising, a coat moving in the wind, an object turning.
3. Camera — one move: push in, pull back, orbit, track forward, or locked off. Pick one only.
4. Lighting — direction and quality: soft morning light from the left, hard backlight, overcast daylight.
5. Style — a short visual reference: cinematic, documentary, minimal product photography.
6. Constraints — what must not appear: no text, no logos, no people, no cuts.

Here is an example you can adapt. It is a written template, not a sample render:

A ceramic coffee cup on a wooden table, steam rising slowly, the camera slowly pushing in, soft morning window light from the left, minimal product photography, no text, no people.

It is one sentence and one shot. The subject never changes, the action is slow enough to read in five seconds, and the constraints remove two common failure points.

Short prompts beat long ones. If a sentence does not change what appears on screen, delete it. A free AI video prompt generator will assemble the six parts for you if you would rather pick from options than write from scratch.

Step 3: Choose the frame before you write the prompt

Aspect ratio changes composition, so decide it first.

• 16:9 for websites, presentations and landscape video. The verified output is 1280×704 at 720p.
• 9:16 for phone-first and vertical short-form. The verified output is 704×1280 at 720p.

Neither is a crop of the other. If you compose for 16:9 and later crop to 9:16, you lose the edges and usually the subject. Compose for the frame you will publish in. Higher resolutions and square output are not offered yet.

Step 4: Generate and check the quote

Generation is a paid action, so the price should be visible before you commit. On the entry model, a five-second 720p clip costs 2 credits. New accounts receive 3 welcome credits, which is enough for one trial clip. If a job fails, the charged credits are refunded automatically.

Open the studio, paste the prompt, choose 16:9 or 9:16, confirm the credit cost, and start the render. The job is tied to your account, so the result stays in your private history.

Step 5: Review the clip like an editor

Watch the result twice. The first pass is for the idea: does the subject match what you described, and does the camera do what you asked? The second pass is for artifacts: warping edges, extra fingers, objects that appear or vanish, backgrounds that drift.

One useful habit is to change one variable per retry. If the camera moved too fast, keep everything else and change only the camera phrase. If you rewrite the whole prompt each time, you cannot tell which sentence caused the improvement.

Step 6: Download and finish in an editor

The finished file is an MP4 in your history. It is private and has no watermark. Download it and finish the work elsewhere: trim it, add music, add captions, and cut it together with other shots.

This tool generates clips; it does not edit them. There is no audio, no voiceover, no subtitle track and no timeline inside the product. Treat it as a source of footage, not a finished video.

Step 7: Keep a prompt log

Save the prompts that worked, with the frame you used. A short note like "coffee cup, push in, left window light, 16:9" is enough to rebuild a shot later. Over a few sessions this becomes your own template library, which is more useful than any generic prompt list.

What this workflow cannot do

Being clear about the limits saves credits:

• Clip length is fixed at 5 seconds. Longer durations are not offered.
• Resolution is 720p. There is no 1080p or 4K option.
• No audio, subtitles or voiceover. Clips are silent.
• No editing, extending or reference video. You cannot feed in a clip and continue it.
• No lip sync and no person replacement. Those need different models and are not available here.
• Image-to-video is 9:16 only at the moment.
• Every generation costs credits. The entry price is 2 credits per clip.

Within those limits, the workflow is reliable: one shot, one subject, one camera move, the right frame, and a prompt you can iterate on.

A five-minute checklist

Before you generate, confirm:

1. The idea fits in one continuous shot.
2. You chose text-to-video or image-to-video on purpose.
3. The prompt has a subject, action, camera, light, style and constraints.
4. The aspect ratio matches where the clip will be published.
5. You know the credit cost and have enough balance.
6. You have a plan for review and one variable to change if it fails.

If you want ready-made starting points, the companion guide has 12 annotated AI video prompt examples grouped by product, landscape, character and vertical use. When you are ready to render, open the studio or check pricing first.