How AI product videos are made.
AI product videos are made in stages: a brief, a locked reference of the real product, direction and boards, generated stills, image to motion, human review and fixes, then edit, sound and cutdowns. At We2 that runs as a film sprint of about two weeks. The AI generates frames. People decide what the film is and what ships.
The short version
Brief, lock the real product, direct and board, generate stills, turn them into motion, review every shot, edit with sound, then cut for every placement. The AI makes frames. A person makes every decision that matters.
This is the workflow We2 uses for AI product video and AI fashion films. It is a general process, not a recipe tied to one tool, and it holds whatever the current models are.
Brief and references
Every film starts with what it is for: the product, the audience, the placements and the date. A product page loop, a launch film and a batch of ad hooks are different films, even on the same product.
We collect references for the look: the brand system, past campaigns, films the team likes and, just as useful, films it does not.
Lock the product reference
This is the step that decides whether the film is any good. We build a reference of the real product from every side, with close-ups of the logo, print, label, hardware and true colourway.
Everything generated later is checked against it. The model, light and styling can change. The product does not.
Direction: look, characters, boards
A person directs the film. We fix the look, cast consistent characters, choose settings and board the film shot by shot, keeping hard moments for the model, like hands on small parts, short or out of frame.
You approve the board before anything is generated, so the creative call is made by people, on paper, where it is quick to change.
Stills first
We generate key stills for each shot before any motion: the product in the setting, the character, the light. A still is quick to judge and quick to redo, and errors in shape or print show up clearly.
Stills that pass become the starting frames for motion, and some become key visuals in their own right.
Image to motion
Approved stills are turned into motion with image to video models, shot by shot. Several takes are made for each shot, because motion is where a product is most likely to bend, melt or change.
Review and fix
People review every take against the product reference: logo shape and position, text, geometry, colour, hands, motion artefacts and continuity between shots.
Takes that fail are remade. Where text must read exactly, the real artwork is placed in the edit instead of trusting the model.
Edit and sound
The film is cut for rhythm and for the hook, then sound is added: music and effects that sit on the cut. The first seconds get the most attention, because that is where an ad is won or skipped.
Cutdowns and delivery
Each ratio is framed for its placement rather than cropped from one master: 9:16 for Reels, Shorts and Stories, 4:5 for the feed, 1:1 for grids, 16:9 for YouTube and banners. Files are exported and handed over.
What fails most often
On real products, the same few things go wrong again and again. Knowing them is most of the review.
- Text and small logos. Generated lettering warps or turns into lookalike nonsense.
- Hands. Fingers on zips, caps, straps and buttons are still the most visible failure.
- Fabric and material. Heavy fabrics swing like light ones, rib knits flatten, metal goes soft.
- Construction. Seams, stitches, eyelets and hardware multiply, merge or vanish.
- Continuity. The product changes shade, length or print position between shots.
Each of these is a reason to keep a person in the loop, not a reason to avoid AI. We weigh it against a live shoot in AI video vs a traditional shoot.
Do it yourself or hire
If you need simple motion on a plain product for a product page, a photo to video app may be all you need. Check every output against the real product before it goes live.
Hire it out when the film carries a launch, when the product has prints, logos or text that must stay exact, or when you need a set of hooks and cuts for paid social. That is the work We2 does, from the brief to the files, alongside the store or site the films will live on. You can see finished films on AI films and the rest of our work on our work page.
Questions about the process.
Want this run on your product?
Send the product and where it will run. Selling on Shopify? See AI video for Shopify and D2C brands.