AI Product Photography

Most e-commerce catalogues contain two kinds of images: the handful that were properly shot, and the rest: supplier photos, phone snaps, whatever came with the product. The gap between them is visible on every category page.

Closing that gap by shooting everything is a budget problem. Closing it with AI is a fidelity problem: general image generators invent products, and an invented product cannot go on a product page.

The distinction that matters

There are two things people mean by "AI product photography":

Generating a product. You describe a bottle and the model draws one. Useful for concepts and moodboards. Useless for a catalogue, because the bottle it draws is not the bottle you sell.

Re-photographing a product. You supply the actual product image and the model builds everything around it: the set, the light, the surface, the shadow, the composition. The product stays yours.

Magnifiq does the second. Your input image is the product; the generation is the photo shoot around it.

What you can compose

The Photo workspace takes up to 14 input images per generation and works in five modes:

A vision model reads your product images and drafts the prompt for you, so the starting point is a description of what you actually uploaded rather than a blank field.

Starting from a style instead of a blank page

Twenty-one curated blueprints each bake in a full setup: prompt, composition mode, aspect ratio and quality tier. Levitation against a pastel gradient. Overhead flat lay. Wet beauty with rim light. Exploded view. Banner with copy space reserved for a headline. You drop the product in and generate.

Browse them in the style library.

Output that fits every channel

• 11 aspect ratios, from 1:1 through 21:9 cinematic and 9:16 vertical
• Smart crop: set the focal point once, export every ratio in one zip
• 4x upscaling for print and hero placements
• JPEG, PNG and WebP, with background removal to transparent PNG when a channel needs it

Fixing a shot without regenerating it

The Editor picks up where generation stops. Paint a mask over a region and describe the change, and only that region is regenerated. Outpaint to a wider frame without touching the original pixels. Remove the background, correct exposure and white balance, or push resolution 4x.

That matters more than it sounds: the usual failure mode of AI imagery is that one detail is wrong and the whole generation gets thrown away. Masked editing turns that into a thirty-second fix.

Input quality, honestly

The product needs to be clearly visible and in focus. Lighting, background, surface and styling are what gets replaced. Sharpness is not. A well-lit supplier photo on a white background is an excellent input. A blurry photo on a dark carpet is not, and no amount of prompting will make it one.

Disclosure and content credentials

Every export carries a signed C2PA manifest and IPTC labels declaring the image AI-generated. Any later edit breaks the cryptographic seal, so a viewer can distinguish an untouched file from a tampered one. The credentials are written into the file without re-encoding a single pixel, and can be stripped per download if you need a clean export.

This is the machine-readable disclosure the EU AI Act asks for, and it ships by default rather than as something you have to remember.