Turning product photos into video, automatically

Short answer: two approaches, and they are not comparable. Template-based rendering animates your real product photo with pans, zooms and transitions — cheap, fast, predictable, and the product stays exactly itself. AI generation (Runway, Luma, Kling, Veo) invents motion from a still — impressive, slower, roughly $0.05 to $0.50 a clip, and it will change your product. For ecommerce, template rendering is right far more often than the demos suggest.

Every product photo you own could be a video, and video outperforms stills almost everywhere it competes. Product pages, Reels, TikTok, paid social.

The gap between knowing that and having 400 product videos is the interesting part.

The two approaches, and why the difference matters

Template rendering

Take the photo you already have and animate it. Slow zoom, pan across detail, a transition between variant shots, text overlay, your brand's colours, a music bed.

Nothing is invented. The pixels are your pixels, moved around.

Built on ffmpeg, or on APIs like Shotstack, Creatomate or Remotion that wrap the same idea in something friendlier. Render time is seconds. Cost is compute, which is cents at most and often free on a machine you already have.

The result looks like a well-made slideshow, because that is what it is. That is not a criticism. For a product page, a well-made slideshow showing the actual item is usually more useful than anything generative.

AI generation

Give a model a still and a prompt and get back a few seconds of invented motion. The fabric moves, the camera orbits, the light shifts.

Runway, Luma Dream Machine, Kling, Pika, Google Veo. All API-accessible. Typically $0.05 to $0.50 for a short clip depending on model, length and resolution. Generation takes anywhere from twenty seconds to several minutes.

The output can be genuinely striking. It can also add a strap that does not exist, change your logo into something adjacent, or render a sixth finger on a hand model.

The distinction that decides it

Does the video need to show the product accurately?

Product page, catalogue, anything a customer uses to decide what they are buying: the product must be itself. Template rendering, every time. A generated video that adds detail your product does not have is a misrepresentation, and the return comes back with "not as shown".

Top-of-funnel social, mood, brand: generation earns its place, because nobody is inspecting the stitching.

What does an automated pipeline look like?

The mechanics are the same either way. Six steps, and most of the engineering is in steps 1 and 5.

1. Pick the images. Not all of them. A product with eight photos does not need an eight-shot video. Rules work well: the featured image, plus the two with the most visual difference from it. Skip size charts and packaging shots, which are usually detectable by aspect ratio or filename.

2. Prepare them. Crop to the target aspect ratio — 9:16 for Reels and TikTok, 1:1 for feed, 16:9 for a product page. Normalise resolution. This step is where most pipelines break, because product photos are wildly inconsistent, and a crop that centres on the wrong part of a 4:5 photo cuts the top off the product. Content-aware cropping, or a simple rule that biases toward the centre of mass of the non-white pixels, is worth the effort.

3. Compose. Apply the template: motion, timing, transitions, text from the product title or price, logo, music. For generation, build the prompt from product metadata and send it off.

4. Render. Seconds locally with ffmpeg, or a queued job against a hosted API. For generation, this is an async job with a webhook or a poll.

5. Review. The step everyone skips. Generative output needs a human to look at every clip. Template output needs spot-checking, mostly for bad crops. Build the queue for this or the pipeline will publish something embarrassing at scale.

6. Attach and distribute. Upload the file to Shopify Files, attach it to the product as media, or push it to the social scheduler. Shopify accepts video as product media directly, which many stores do not realise.

What does it cost, honestly?

Template rendering, 500 products. Compute is essentially free — ffmpeg on a machine you already have. Build cost is a fixed engineering project, typically 3 to 5 weeks for a full pipeline with a review queue and Shopify integration. Running cost afterwards: nothing.

AI generation, 500 products. At $0.20 a clip that is $100 for one pass. Add the failures: expect to discard and regenerate 20 to 40% of clips, so call it $130 to $140. Then add the human review time, which at 30 seconds a clip across 500 products is four hours of someone's day. The API cost is the small number.

SaaS tools that do some of this sit at $30 to $300 a month with clip quotas. Reasonable for a small catalogue, poor value once you are rendering hundreds a month, and you never own the templates.

The honest summary: template rendering has a build cost and no running cost. Generation has a running cost that never stops, plus a human in the loop that never goes away.

Where do these pipelines go wrong?

Cropping without looking. The single biggest cause of bad output. A 9:16 crop of a wide product photo cuts off the product. Automate the crop, but validate it — check the product still occupies a sensible share of the frame, and flag anything that does not.

Rendering every product identically. Five hundred videos with the same zoom and the same transition, posted in sequence, read as spam to both viewers and the algorithm. Vary the template, or render fewer.

Treating generation as fire-and-forget. It is not. Every clip needs eyes. A pipeline without a review queue is a liability generator.

Ignoring the file size. Shopify has limits on product media, and social platforms re-encode aggressively. A 200MB render gets crushed into something that looks worse than the still it came from. Target the platform's spec, encode once, correctly.

Music licensing. A track you do not have rights to will get the post muted or taken down, and on a paid ad it can cost the account. Use a licensed library and keep the receipt.

Generated video in a product listing. Beyond the return rate, it edges into misrepresentation. Several platforms are tightening rules on undisclosed AI-generated commercial imagery. Not worth the exposure on a page whose job is to describe what the customer receives.

What we would build first

Template rendering, one aspect ratio, one template, ten products. Post them. Measure.

If the videos move the numbers, extend the pipeline to the catalogue. If they do not, you have spent a week instead of a quarter, and you know something true about your store rather than something true about the demo reel.

Generation is worth adding later, for social only, once you know video works for you at all.

We build video pipelines as custom Shopify apps: your templates, your review queue, rendering on your own infrastructure, no per-clip fee. Fixed price, source code handed over.

Sitting on a catalogue of photos and no videos? Book a free 30 minute call and we will tell you whether rendering or generation fits. Or read where Shopify automation actually pays, or see our app development work.