AI Video
Plan an LTX Video Product Demo from a Storyboard
Break a product demo into short LTX Video shots, lock a reference still, and edit the clips under one voiceover.
- LTX Video
- Storyboard
- Product
- Editing
On this page
LTX Video is a short-clip model. It is good at one motion on a picture you already like, and it is a poor place to invent a two-minute commercial in a single prompt. A product demo that feels finished is an edit. You storyboard six to ten shots, generate a still for each, animate the shots that need motion, and let the voiceover cover the seconds between them. The model never has to remember the whole script.
This workflow is for a physical product on a table: a bottle, a device, a book. Faces are harder, and a spinning logo with tiny type will smear. If the product name must be readable, composite the real logo in an editor after the generation. Asking the model to render the wordmark is how you get a label that almost says the brand.
Storyboard before you generate
Write the demo as shots a camera could actually make. Each shot gets a duration, a single move, and the line of voiceover it sits under. If you cannot describe the move in one sentence, split the shot. LTX will invent extra motion when the prompt is a list of events.
| Shot | Seconds | Move | Voiceover job |
|---|---|---|---|
| Wide table | 4 | Slow push in | Name the product |
| Hero still | 5 | Hold, tiny drift | Say the one benefit |
| Hand enters | 4 | Hand places the object | Show the action |
| Detail | 4 | Lateral slide | Point at the part that matters |
| Wide again | 4 | Pull back | Say the next step |
Lock one lighting setup
Generate the hero still first and reuse its lighting words in every later prompt: soft window light from the left, gray table, 50mm, shallow depth of field. Change the camera, not the room. A demo falls apart when shot three is noon daylight and shot four is a neon studio. The product should be the same object. Put it in the still. Do not ask the video model to redesign it.
4 second product shot, 24 fps, 16:9. Use the reference still of a matte black device on a gray table, soft window light from the left. One move only: a slow push in. The device stays sharp, the background falls off, no new objects, no text, no logos.
Generate, then throw most of it away
Render two or three takes of each shot. Keep the take where the product shape holds for the whole clip. Warp on the last frames is normal. Cut before it. A 2 second usable piece is better than a 6 second clip you have to apologize for. Do not upscale a broken shape and hope the brand will not notice. The silhouette is the product.
- Export stills at 16:9 so the video model is not inventing the crop.
- Name files shot-03-take-2.mp4. You will re-open the edit next week.
- Drop takes on a timeline and mute them. The voiceover decides the cut.
- Add the real logo and captions in the editor, not in the model.
Edit for the sentence
Record the voiceover before you fall in love with a clip. A 45 second demo is about 110 to 130 words at a calm pace. If the script is 300 words, you do not have a 45 second demo. Lay the voice down, then place shots under the nouns. When the narrator says "the dial," the detail shot should already be on screen. When the narrator is between nouns, hold the hero still.
Music under a product demo should be quiet enough that the click of the object, if you recorded one, still reads. Generated clips often have no useful sound. Plan on silence plus voice plus a restrained bed. Label the description as generated footage of a product, and do not present a synthetic hand as a customer's hand.
Limits to plan around
- Pros: a locked still keeps the product recognizable across shots.
- Cons: small type, screens, and transparent glass still melt.
- Pros: short takes are cheap to retry compared with a full commercial.
- Cons: a single 30 second generation will not follow a five-beat script.
Screens and labels are a composite
If the product is a phone or a device with a UI, do not ask LTX to invent the interface. Record or design the real screen, and track it onto the generated device in the editor, or shoot the screen as a separate still and cut to it. Generated UI almost spells a real word and then does not, and viewers notice text before they notice lighting. The same rule covers serial numbers and regulatory marks. Leave a clean blank on the model and add the real mark in post if the law or the brand requires it. A demo that must be accurate is a demo that uses the real pixels for anything a customer would read. The model is there for the camera move and the material, not for the typography.