How to Make an AI Video With Shots, Sound, and a Story

The main decision is what AI should create. A moving product shot, a presenter reading an explanation and a finished multi-scene video are different tasks. Starting with that distinction makes the process easier to control and prevents you from spending generations on material the final video does not need.

This tutorial follows a small vertical fragrance concept, then shows where to adapt the workflow. The product, NORI 01, is fictional. The example demonstrates making a video, not an actual product campaign or sales result.

1. Write the brief for the finished video

Describe the audience, message, format and required ending before writing prompts. A useful brief is specific enough to decide whether a shot belongs in the project.

For this example:

This is deliberately smaller than a complete brand story. A first project should prove that the subject, movement and assembly work together. More shots increase the number of continuity and timing decisions.

For an educational video, replace the visual sequence with one question and its answer. For a presenter video, write the spoken explanation first. The brief should still say what the viewer needs to understand at the end.

2. Choose text, image or presenter input

For the NORI example, product appearance matters, so a reference-led route is preferable to inventing a new bottle for each prompt. An approved image constrains the starting design; it does not guarantee that every label letter will remain stable in motion.

Runway's generative video introduction makes a useful input distinction: describe scene and motion for text-to-video, while an image-led prompt can concentrate on motion. Apply that distinction when writing the brief for whichever model you use.

If you need a speaking host, a tool such as HeyGen starts from a script or described video instead. You do not need to force that job through a cinematic scene prompt. If free access is essential, the first free text-to-video tutorial explains the account and export checks.

3. Prepare references at the intended aspect ratio

Check the image before animation. The subject should have enough space for the planned movement, and details that must survive should be clearly visible. Avoid starting with a cropped bottle if the final shot needs to show the entire product.

The initial assembly of this concept had a practical problem: a wide opening shot was cropped into the vertical canvas, cutting off the bottle's lower area. Its liquid also looked darker than the other shots. A later version replaced that opening material with a vertical source that better matched the sequence.

This is why format belongs at the input stage. An editor can crop a wide video, but it cannot recover a product edge that never fits within the chosen framing. Prepare a vertical reference or create suitable vertical material before committing to motion.

For the payoff shot, the reference was adapted to a 9:16 image with the whole bottle visible. It still contained a leaf shadow near the upper area, so it was not a completely blank lighting setup. The later video used the shadow moving away as part of the reveal.

Keep the original reference and accepted adaptations. Give each one a clear role—product identity, close-up framing or final composition—so later changes do not accidentally replace the approved bottle design.

4. Generate short shots in Dreamina

In Dreamina's Seedance 2.5 workflow, enter AI Video, choose an available model and reference mode, add the relevant inputs, then set the output and generate. The options on your account determine the available controls.

For the example's detail shot, a paid Dreamina account used Seedance 2.5, five seconds, 9:16 and 720p. The instruction asked for the bottle shoulder, silver cap and upper label to remain steady enough to read the brand, with warm side light and a subtle lateral camera move.

The useful parts of that instruction are the subject, the restrained movement and the details to preserve. You can adapt this structure to your own product:

Show the approved product in close-up. Keep its shape, cap and visible brand lettering stable. Let one soft light move gently across the surface while the camera makes a slow, subtle lateral move. One continuous shot, no hands and no new lettering.

This is an adaptable instruction, not a guarantee that the output will meet every condition. Start with one movement so the result is easier to evaluate.

A five-second Dreamina detail shot brings the bottle’s upper area into focus, adding a quiet visual beat between the opening reveal and end card.

The payoff shot used the adapted vertical reference and asked for the leaf shadow to move off the bottle during a gentle push-in. These two shots perform different jobs: one shows surface detail; the other restores a fuller view of the product.

Do not make every shot equally busy. If the opening establishes the product, the close-up can slow down and show detail. This creates a sequence with purpose, rather than a collection of unrelated moving images.

5. Review the shot against its job

Watch the whole output before deciding it is usable. Inspect the product's outline, cap, label, color, background and movement. A pleasing first frame is not enough if the bottle changes shape a second later.

Separate three kinds of problem:

• Input problem: the reference is cropped, unclear or inconsistent. Correct the source before asking for more motion.
• Generation problem: a usable reference produces a changing label, broken contact or unwanted camera move. Revise the specific instruction or regenerate the shot.
• Assembly problem: the accepted shot is too long, poorly placed or cropped by the editor. Correct the project rather than spending another generation.

Replacing the opening with a portrait-friendly shot brought its framing and color closer to the rest of the sequence. Keep the scene that works, then refine the one shot that interrupts the visual rhythm.

If only part of a good clip needs changing, inspect the available repair controls before starting again. The generated-video editing comparison separates visual transformations from ordinary timeline edits.

6. Assemble the accepted shots

Place the accepted clips into a project at the intended aspect ratio. Keep the original sources alongside the project so you can replace one shot without rebuilding everything.

For this concept, the revised Canva assembly used three generated clips followed by a static end card. The finished file is 20.167 seconds, including roughly five seconds for the end card. That final static section is edited presentation, not a fourth moving scene produced by the generator.

A simple assembly order works well:

1
1. Use the clearest opening to establish the subject.
2
2. Cut to the close-up when the viewer already knows what the detail belongs to.
3
3. Return to a fuller product view for the payoff.
4
4. Add a short ending with the intended next message.

Trim according to the useful action. A model's five-second clip does not require you to use all five seconds. Avoid holding a failed last frame merely to fill a preset duration.

Add titles as editable text layers. In this example, the sequence uses NORI 01, LIGHT. GLASS. FORM., and an end card reading EXPLORE THE CONCEPT / FICTIONAL DEMO. The demo wording makes the concept's status clear. Creating these layers in the editor makes later text changes easier than embedding new words into generated footage.

7. Control the sound separately

Choose the role of sound before assembling the final cut. Music sets the mood; narration explains the message; dialogue and effects bring a scene to life. Giving each track a clear job keeps the edit focused.

The NORI assembly muted the source audio and used an existing Canva background-music selection. That supplied one continuous audio layer across the clips instead of relying on each generated shot to carry the final soundtrack. Use material you are entitled to include, and retain its source information with the project.

If the message depends on speech, make the narration understandable before adding background music. Review pronunciation and timing in the final render. For an atmospheric product concept, leave enough space for the images instead of adding narration solely because a tool can generate it.

Model capabilities differ. Runway's sound guidance, for example, distinguishes models with native audio from Gen-4.5, which does not embed it. The native sound and dialogue guide helps when sound is central to tool choice.

8. Export, reopen and keep the project

Export the assembled video, then open the saved file outside the editing timeline. Check the beginning and ending, cuts, text, sound, dimensions and any watermark.

The finished fragrance concept is a twenty-second vertical video, exported at 1080 × 1920 with titles, an end card and a music track.

Keep both the export and the editable project. The file is ready for viewing; the project preserves the ability to change titles, music and shot order. Also retain accepted references and the source clips, so a later request does not force you to recreate the subject.

Plan the budget around accepted shots

Free access and paid upgrade: Free accounts in the US receive 120 credits per day; daily allowances vary in other regions. Start developing images and short videos with Seedance 2.0 for video and Seedream 5.0 Lite for images. For regular production, Dreamina Basic is USD 15 per month on monthly billing, with 1,575 monthly credits. Paid plans add watermark-free downloads and a faster generation queue. Credit use varies by model and settings.

Plan the budget around the shots your story needs: an opening, a detail view and a closing image can be enough for a short advertisement. Allow room for a replacement take when a new angle or composition strengthens the sequence.

Start with a small project that can reach a complete export. In Dreamina AI Video, prepare one approved subject and one short action, review it, then build the remaining shots around what works. The finished project comes from that sequence of decisions: clear input, useful motion, deliberate corrections and careful assembly.