Orelon logoOrelon
价格

Cinematic Product Videos with AI: A Van Review Workflow

2026年11月14日 · 作者:Orelon Team

探索 AI 视频模板

浏览社区创作获取灵感,打开任意模板即可在 Orelon 中继续创作。

Turn an old van review into a cinematic product video. Plan shots, write prompts, and keep the vehicle consistent with an AI video workflow.

A 2012 Ford Transit Connect is an awkward hero. It is tall, slab-sided, built to swallow cargo pallets rather than turn heads, and by now most examples carry a decade of work in their paint. That is precisely why it makes a useful test case for AI product video: if you can make a tired compact van look deliberate on screen, you can make almost any product look deliberate. What follows is a practical, repeatable workflow for turning archive material about an unglamorous vehicle into a short cinematic showcase, using an AI video generator for motion and an AI image generator for keyframes.

Why an unglamorous van is the ideal AI video test

A conventional vehicle review was written for buyers, not for cameras. It contains the raw ingredients anyway: three-quarter angles, cargo measurements, detail shots of latches and load floors, and a handful of contextual frames shot in whatever light was available. The job of generative video is not to invent a different vehicle. It is to reinterpret existing truth into a cinematic register, so that a workhorse reads as purposeful rather than plain.

The van also acts as a stress test. Flat panels and long straight body lines expose any model weakness immediately. A coupe can hide drift in its curves and reflections; a box cannot. If a sliding door track migrates, a wheel design changes between shots, or the roofline sags, you will see it in the first second of playback. Learning to hold a Transit Connect together across six shots teaches you the consistency discipline that every product video needs.

Finally, the subject forces honesty. You are not selling a fantasy car. You are presenting a used commercial vehicle that someone might actually buy, which means the visuals have to stay close enough to reality that a buyer would recognise the van in the video. That constraint makes you a better director than an unlimited budget of spectacle would.

Pre-production: build the shot list before the prompt

Most weak AI product videos begin with a prompt and end with a pile of disconnected clips. Begin with a shot list instead. Six shots is enough for a twenty-five to forty second piece, and it is enough to tell a small story: arrival, inspection, load bay, detail, departure.

Six shots that sell a utilitarian vehicle

Shot Framing Job it does
1 Front three-quarter, static or slow push Establishes identity and paint condition
2 Lateral slider along the closed flank Shows length, height, and straight body lines
3 Low front tracking at bumper height Adds weight and presence
4 Detail montage: wheel, badge, door latch, load floor Proves care and honesty about wear
5 Rear doors open, cargo volume visible Communicates the actual selling point
6 Departure shot, van leaving frame Gives the edit an ending

Write the list with durations attached, not just descriptions. A detail montage of four quick cuts at 1.5 seconds each behaves very differently from one eight-second shot, and knowing which you need changes how you prompt the model. You can browse a video template library for pacing references if you want a starting rhythm.

Build a reference board from what already exists

Before generating anything, gather stills: frames pulled from old footage, phone photos, screenshots of the same model year, and any detail shots you have. Normalise them. Crop to the same aspect ratio, correct obvious colour casts, and rename each file with its shot number so you never animate the wrong reference.

Then write a single vehicle block — a short paragraph describing only the things that must never change: body colour and finish, wheel style, trim level, ride height, roof rack or lack of one, window tint, and the amount of visible dirt. You will paste this block into every prompt. Think of it as the costume department. If the costume changes between scenes, the audience stops believing the film.

Keyframes before motion: the rule that saves hours

Text-to-video is seductive and almost always the wrong first step for product work. Generate stills first, approve them, then animate the approved frames. A bad still costs you one generation; a bad animated clip costs you ten times the time and usually forces a reshoot of the shot entirely.

Lock the hero angle first

Shot 1 is the anchor. Iterate on it until the proportions read correctly: the tall greenhouse, the short nose, the near-vertical rear. Generate a batch of twelve to twenty variations, then keep exactly one. That frame becomes the visual contract for everything else.

Keep the same van in every shot

Consistency comes from three habits rather than one trick. First, keep the vehicle block identical in every prompt and change only the camera, the light, and the action. Second, animate from a still instead of from text for every shot after the hero, so the model inherits the correct geometry. Third, derive new angles from the approved hero frame where the tool allows it — re-posing a known object is far more reliable than inventing a new one from a sentence.

If your first attempts drift, tighten the scope. A close detail of a wheel arch is easier to hold than a full vehicle in a wide environmental shot, so put the risky full-body shots late in the schedule when you have learned what the model tolerates.

Camera language that flatters a boxy vehicle

A van has no swooping surfaces to flatter it, so movement has to do the work that sheet metal cannot. The safest cinematic vocabulary is slow and mechanical: a dolly, a slider, a crane. These read as industrial and intentional, which suits a commercial vehicle better than a flourish.

Moves that work

A slow dolly-in on the front three-quarter, roughly one metre of travel over four seconds, gives the van presence without exposing geometry errors. A lateral slider along the flank emphasises length, which is exactly what a cargo buyer cares about. A low push at bumper height makes the vehicle feel heavier than it is. A gentle crane up to reveal the roof and load height works if the background stays simple.

Environmental motion is often more convincing than camera motion. Dust drifting past a locked-off frame, rain on the windscreen, or a slow shadow crawling across the flank all suggest cinema without asking the model to reconstruct a vehicle from a new angle.

Moves that fall apart

Fast orbits are the classic failure. A 180-degree arc around a rectangular object requires the model to invent two hidden faces, and it usually invents them wrongly. Whip pans, drone-style reveals with large altitude changes, and shots with two simultaneous camera moves all increase the chance of warping. Keep one motion per shot and let the edit supply the energy.

Light, grade, and the direct-flash look

Automotive footage has two dependable registers. The first is honest overcast: soft, slightly flat light that makes panel gaps and paint condition readable and never lies about the product. The second is hard, direct, on-camera flash with visible specular highlights and crisp shadow edges, which gives an editorial, almost documentary-fashion feel. The second is more striking and much harder to keep consistent across shots, because every highlight position has to agree with every other frame.

Pick one register for the whole piece and write it into your vehicle block. Then grade lightly. A small contrast lift, a modest desaturation, and careful preservation of highlight detail in the paint will do more than a heavy look. If the van is white — as many are — protect the highlights above all, because blown white panels destroy the sense of shape.

The end-to-end workflow, step by step

Step 1: ingest, clean, and catalogue

Pull every usable frame from existing material. Trim clips, stabilise what you can, and export reference stills at a generous resolution. Catalogue them by shot number in a single folder. This step feels administrative and it is the reason the rest of the project goes smoothly.

Step 2: generate and select keyframes

Start with the hero angle. Batch generously, select ruthlessly, and keep a rejects folder so you can see what you have already tried. For the remaining shots, work from the closest available reference and use the same vehicle block. Approve all six keyframes before animating any of them — a contact sheet of the whole sequence will reveal mismatches that individual frames hide.

Step 3: animate with short, specific prompts

A reliable prompt has four parts and no more: subject, action, camera, light. Something like the van's front three-quarter view, stationary, slow dolly in, hard afternoon sun. Keep it under about twenty-five words. Long prompts invite the model to add elements you did not ask for, and every added element is another thing that can drift.

Generate three to five takes per shot rather than one long one, and prefer shorter durations that you can trim. A confident four-second clip is worth more than a shaky eight-second one. If you are new to structuring these instructions, the prompt library is a good place to study the grammar before writing your own.

Step 4: assemble, add sound, and finish

Cut to music with a clear tempo, place the hero shot first and the departure last, and resist the urge to use every good clip. Then spend real time on sound: an engine idle, a door latch, a sliding door track, tyres on gravel. Product video lives or dies on sound design, and it is the step most AI-first creators skip. Finish with a light grade, a subtle grain pass, and exports in both 16:9 and 9:16.

Choosing a model and spending iterations wisely

Not every project needs the most ambitious engine. Choose on four criteria: how faithfully the model holds a specific real object, how long a usable clip it returns, whether it accepts a start frame or end frame, and how many attempts a good shot typically costs you. A tool that is mediocre at fantasy landscapes can still be excellent at slow product moves.

A practical budget rule: expect to discard seventy percent of what you generate early on and forty percent once your prompts stabilise. Plan iterations per shot rather than per project. If a shot has failed five times, change the approach — a different angle, a simpler move, a closer framing — instead of generating a sixth variation of the same idea.

Mistakes that wreck automotive AI footage

  • Animating unapproved stills. Motion hides small errors and magnifies large ones.
  • Rewriting the vehicle block. One word changed between prompts can repaint the van or swap its wheels.
  • Overloading prompts. Extra adjectives become extra objects.
  • Ignoring reflections and wheels. These are the two details viewers use to judge whether a vehicle is real.
  • Claiming things the vehicle does not have. Do not let generated footage imply condition, mileage, or features that are not true; keep the video clearly a stylised presentation.
  • Skipping sound. Silent product clips feel like tests, not films.
  • Too many shots, too fast. Six confident shots beat eighteen nervous ones.

FAQ

Can I use AI video for a real vehicle listing? Yes, provided the footage stays representative. Use it for mood, movement, and detail emphasis, keep a few genuine photographs or clips in the piece, and make clear that some shots are stylised.

How many reference images do I need? Five to ten good ones is plenty: front three-quarter, side, rear, wheel, badge, and interior. Quality and consistency matter far more than quantity.

Why does my van change shape between shots? Almost always because the shots were generated from text rather than from an approved still, or because the vehicle description changed between prompts. Animate from stills and freeze the description.

What length should the final video be? Twenty-five to forty seconds for social, sixty to ninety seconds for a dealer page or presentation. Build the short version first; it is harder and it teaches you which shots actually matter.

Do I need a video editor? Yes, at least a basic one. Generation gives you shots; editing, sound, and grade turn them into a product film.

Put Orelon to work on your next product video

Start with one vehicle, one shot list, and one approved hero frame. Generate the stills in the Orelon image generator, animate the keepers in the video generator, and keep the vehicle block taped to the top of your prompt file. When you want to compare approaches or check pacing ideas, the Orelon blog has more workflow breakdowns to borrow from. Orelon is built for cinematic ideas in motion — bring the van, and give it a better second act.