AI video generation · Wunmi's AI Starting Point

A clip is one piece of the finished video

Text-to-video creates a clip from a description. Image-to-video starts from an image and adds motion. A reference can help establish a subject or scene, but it does not guarantee that every frame stays consistent.

A finished commercial usually needs several clips, selection, trimming, transitions, titles, audio and a final accuracy check. Generating a clip and editing a finished video are separate jobs.

Start with one action and one camera movement. A request to walk, spin, pour, smile and change locations in a few seconds is difficult to control. Tool-supported duration, resolution and audio vary; check the settings before spending credits.

Plan a 24-second tote-bag video

Fictional brief: show a cream tote as an everyday companion. No waterproofing, capacity or durability claims. Use a real permitted reference when showing a real product.

Four six-second edit slots; trim generated clips to fit
Time	Visual and action	Camera / purpose
0–6s	Tote on a desk, curtain moves gently	Slow push-in; establish product
6–12s	A hand places a notebook beside it	Locked close-up; show daily context
12–18s	Person lifts tote from desk	Steady medium shot; show use
18–24s	Tote rests beside a chair	Locked end frame; add title in editor

These are editing slots, not a promise that a tool can generate exactly six seconds. If the tool offers five or ten seconds, generate a suitable length and adjust the edit. Keep the final total between 20 and 30 seconds.

Continuity notes: same cream bag, handle shape, warm room, soft light from the left and burgundy notebook. Keep reference images and prompt versions labelled by shot.

Make and inspect one clip first

Choose a video tool and check whether your account supports text-to-video or image-to-video. See official options.
For image-to-video, upload a permitted clean reference with the subject already framed as you want it. Avoid distorted starting images.
Choose the available duration and aspect ratio. Use vertical for a vertical social project; keep the ratio consistent across shots.
Describe the motion for one shot. Generate and preview it at normal speed, then pause on the beginning, middle and end.
Check for shape changes, flicker, unwanted camera movement and objects appearing or disappearing.
Revise one motion instruction and try again only if needed. Keep the better version.

For a concrete vendor example, Runway’s Gen-4 guide recommends a simple motion-focused prompt with a good input image. That is model-specific advice, not a universal rule for every video tool.

One image-to-video shot

The tote remains upright on the desk. The camera slowly moves closer. A curtain in the background shifts gently. Soft side lighting stays steady. One continuous shot.

Why is this easier to assess?

There is one camera movement and one small background action. You can check whether the bag stays intact. Adding a scene change and several hand movements would introduce more opportunities for errors.

Adapt the plan to four formats

Use these as shot directions. They are not fake finished advertisements or demonstrations of model output.

UGC-style advertisement plan

Plan a 24-second fictional UGC-style video for an unbranded reusable bottle. Four shots: introduce the bottle, show opening the lid, place it in a bag, end on the product. Natural phone-camera feel. Use a consenting performer or clearly labelled synthetic character. No invented customer experience or health claims. Return shot, duration, action and continuity notes.

Product commercial shot

A plain ceramic mug stands on a neutral table. The camera moves slowly from left to right while the mug stays still. Soft window light creates a stable shadow. Keep the mug shape and handle consistent throughout one shot.

Cinematic sequence plan

Plan three simple shots of a fictional person arriving at a quiet reading room: wide doorway, medium shot placing a book on a table, close-up of the closed book. Keep the same clothing, room and afternoon light. Give one action and one camera direction per shot. Avoid a scene transformation within a shot.

Short social video plan

Plan a 20-second vertical video encouraging people to read for ten minutes. Use a desk, a book and a timer as visual subjects. Give four short shots with timings totalling 20 seconds and space for captions added in an editor. Do not generate readable captions inside the clips.

Assemble, then add sound and text

Import the chosen clips into an editor such as Canva or another editor you already know. Create a project with the intended aspect ratio.
Put shots in story order. Trim weak beginnings and endings; use simple cuts first.
Add factual titles and captions as editable text. Keep them away from the edges and check on a phone.
Add music, voice or sound only when you have the rights and permission. Some generators include audio; others require it separately. Check the chosen tool rather than assuming.
Balance speech and music, check captions against spoken words, and watch once with sound off.
Export a preview, watch it from start to finish and compare the final message with the brief.

Do not clone a person’s voice or use their likeness without appropriate permission. Do not imply endorsement. Review synthetic-media disclosure rules for the platform and context where you will publish.

Your exercise: a paper edit

Plan a 20–30 second video for a fictional stationery product. Write three to five shot cards with duration, subject, action, camera and sound notes. Add the durations yourself. Sketching boxes is enough; no paid tool is required.

Check your resultThe durations total 20–30 seconds.Each shot has one main action and a clear purpose.Product, lighting and scene details remain consistent.Titles and sound have an editing plan and permission checks.No scene makes an unsupported product or customer claim.

For your own review. These ticks are not saved or graded.

Go FurtherOptional challenge. Come back whenever you’re ready.

Generate or storyboard two versions of one shot, changing only the camera movement. Choose the version that joins more naturally with the surrounding shots. Note any continuity repairs needed before export.

Fix the problem you can see

Subject changes shape: simplify motion, improve the reference or shorten the usable segment.
Flickering light: ask for stable lighting and reduce environmental movement.
Awkward hands: simplify the action or choose a framing that genuinely avoids it; do not hide a misleading product demonstration.
Clips do not match: compare reference, wardrobe, light direction and aspect ratio before regenerating everything.
Audio does not match: edit or replace permitted audio separately and verify captions.

Several short, consistent shots give you more control than expecting a complete commercial from one prompt. Editing cannot make an inaccurate product claim true.