Product photography is the most misunderstood use of AI image generation. The expectation: describe the product and get its photograph. The reality: what you get is not your product.
The jar the model generates resembles yours, but the label differs, the lid has different proportions, your brand is not written on it. In e-commerce that means showing the customer the wrong product.
The correct approach is the reverse: you shoot the product, AI generates the world around it. That is the method that actually works and is widely used.
Step 1: Shoot the product properly
A phone is enough; professional equipment is not required. Put the product on a plain white or grey surface and shoot in soft, diffused light. Beside a window that does not get direct sun is ideal.
Make sure the shadow is defined and comes from one direction. That shadow direction determines the light direction of the scene you generate next.
Place the product centrally with space around it. A product cropped at the edge is hard to fit into a new scene.
Shoot several angles. Different backgrounds want different angles, and not having to reshoot saves time.
Step 2: Separate the product
Canva's background remover is fastest and works in one click. In Photoshop, the Object Selection tool gives a more precise result.
Edge cleanliness determines how convincing the scene looks in the next step. Automatic removal struggles with transparent, glossy or furry surfaces; those need manual correction.
Preserve the product's natural contact shadow if you can. A product without one floats in the new scene, and that is the most common reason these images read as fake.
Step 3: Generate the new scene
Two routes: fill in around the product with Adobe Firefly's Generative Fill inside Photoshop, or generate a separate scene and place the product into it.
Either way, state the light direction explicitly: "warm directional light from the upper left casting a soft shadow to the lower right". That single sentence is the intervention that most improves believability.
Reserve the area where the product will sit: "empty space in the centre for a product" guarantees it.
Avoid scenes containing people; add "no people" to the prompt. Hands and faces are still the details that break most often, and a broken hand ruins an otherwise finished product image.
The line you cannot cross
Consumer protection law prohibits misleading a customer about a product, and that covers imagery as much as text.
The practical rule is simple: if a customer forms an expectation about the product from this image, the image must be real. If it only conveys a feeling or a season, it can be generated.
By that rule, the autumn leaves in a campaign banner, the abstract background of a category page and the living room of a lifestyle scene can all be generated. The product itself, its label, its texture and its packaging proportions cannot.
For a corporate brand there is a further reason to prefer Adobe Firefly: trained on licensed data, it carries lower infringement risk and offers indemnification on enterprise plans.
Frequently asked questions
Can I describe my product in the prompt and generate it?
You can, but what comes out is not your product: label, packaging proportions and texture all differ. In e-commerce the image shown must represent the product sold. The right route is shooting it yourself and replacing only the background.
Why does my image look fake?
Almost always because the light does not match. If the shadow in your product photo falls right, the scene's light must come from the left. There must also be a soft contact shadow where the product meets the surface.
Which tool should I use for background replacement?
Adobe Firefly's Generative Fill inside Photoshop is the strongest and is trained on the safest data for commercial use. If you do not use Photoshop, Canva's background removal and Magic Expand reach a similar result far more simply.