Product photography vs CGI: when to shoot, when to render, what to generate.

What the camera does better

Photography's monopoly is context and credibility: real hands using the real product, the lab bench, the clinical bay, the texture of an environment no render fully fakes. It's also unbeatable for products that already exist in final form: a shoot captures a hundred angles of established truth quickly. When the audience needs to believe someone has actually held this thing, shoot it. As generated images get harder to tell from photographs, a real photograph of a production unit is worth more, not less, provided you can show that's what it is.

What the render does better

CGI's monopoly is access and control: the exploded view, the cutaway, the mechanism mid-motion, the finish option that doesn't exist yet, the product photographed before tooling exists at all. One CAD-built asset yields infinite angles, every colourway, print-resolution stills and animation from the same source, and stays current as the design evolves, which photography can't. For technical audiences, an accurate cutaway outsells a beauty shot, because it answers the question they were actually asking.

Where generation fits

Generated imagery is the newer third option, and it earns a place: trying out settings, moods and art direction while the design is still moving, and building a backdrop around a render that is itself accurate. What it can't do is either of the other two jobs. It has never held the product, so it can't lend the camera's credibility, and it has never read the CAD, so it can't lend the render's precision. Use it for scenes, never for the product itself, and label it wherever it appears.

Where the costs actually sit

A shoot's cost is mostly per-event (crew, location, product prep) and repeats when the product changes. CGI's cost is mostly up-front (model preparation, materials, lighting), and then each additional output is cheap. Which means the economics flip with change frequency and variant count: stable product, one hero campaign → shoot; evolving product, many variants and channels → render; both audiences at once → both, and that's not a compromise, it's the standard playbook. Our accuracy rule holds either way: renders built from production CAD, so the picture is a promise the factory keeps.

The agent-native lens: images that say what they are

When the buyer sends an assistant. Assistants researching a purchase can look at images as well as read the words around them, and they tend to lean on what the page says each image is: the alt text, the caption, the file's own metadata. A render captioned as a photograph, or a generated lifestyle scene with no label, can cost a listing its credibility once a machine notices the mismatch. Say what every image is (a photograph of a production unit, a render from released CAD, a generated scene) and use the alt text to name the model, configuration and colourway, so an assistant asked "does it come in grey with the long cable?" needn't guess from pixels. Keep the provenance metadata on generated images rather than stripping it out.

When the business runs on agents. The hard part of a mixed library is keeping it straight, and that is work agents suit. They can tag every asset with its medium, the configuration it shows, its usage rights and when those expire; find the gaps, such as a colourway with no photograph in use or a channel with no cutaway; draft the brief for the next shoot or render batch from those gaps; and list every channel still showing a discontinued variant. The choices stay with people. The art director decides what each shot must prove, and therefore its medium. Whoever owns usage rights signs off before an image of a person, a location or a customer's premises goes anywhere new.

The hybrid that wins launches

The strongest launch libraries interleave: photographed context for belief, CG product for precision, and hybrid live-action with CG where the seam disappears entirely. Plan the split at narrative stage, letting what each shot must prove decide its medium, rather than shooting first and rendering the gaps. That planning discipline is most of what separates an asset library from a folder of pictures. As more of a library is assembled by software and read by machines, it is also what keeps the library credible: every image chosen for what it proves, and labelled as what it is.