AI product photography without a studio

Studio product photography has a fixed cost: lights, a backdrop, a photographer, and a day of reshoots when the packaging changes. AI collapses most of that cost — but only if you treat it as a production pipeline rather than a slot machine. This guide walks through the workflow we use for ecommerce hero shots, lifestyle scenes, and ad variants.

Start from the real product, not a description

The single biggest failure mode is asking a model to invent your product. It will produce something that looks like a generic version of the category: a plausible bottle, a believable sneaker, a watch with the wrong bezel. Those images are useless for a product page because they misrepresent what the customer will receive.

Use the actual product photo as the source. Image-to-image keeps the silhouette, label position, and material cues while the prompt changes the scene. If you only have a phone photo on a desk, that is enough — a clean, evenly lit source beats a dramatic one.

• Shoot or source one flat, evenly lit photo per SKU — front-on, no heavy shadows.
• Crop tight; leave a little space around the product for the model to extend.
• Avoid reflections and busy backgrounds that get baked into the edit.

Write the scene like an art director

A product prompt has four jobs: keep the product, describe the surface, set the light, and name the camera. Vague prompts produce stock-photo mush. Specific prompts produce images you can ship.

• Product: "the same matte black bottle, label unchanged, centered".
• Surface: "brushed concrete plinth, faint water droplets".
• Light: "soft window light from the left, gentle falloff, no harsh specular highlights".
• Camera: "85mm product shot, shallow depth of field, eye level".

If the product has text on the label, say so and check the output letter by letter. Image models still garble small type; a real label crop composited later is often faster than re-rolling ten times.

Build a scene library, not one-off images

Most stores need the same product in a handful of recurring contexts: a clean hero, a lifestyle scene, a seasonal variant, and a bundle shot. Define those scenes once and reuse the prompt skeleton across every SKU. The catalog stays visually consistent, and new products take minutes instead of a shoot.

Keep the skeleton in a doc with the exact wording that worked. When a model or a season changes, you update one scene rather than rediscovering the prompt.

Consistency is a production advantage: the same light, the same surface, the same crop across every product page.

Hero shot, lifestyle shot, ad variant

These are three different jobs, and they need different levels of intervention.

• Hero shot — product only, neutral background, no props. Optimize for a clean crop and accurate colour. This is the image that goes first on the product page.
• Lifestyle shot — the product in use, with hands, context, and a little imperfection. Optimize for plausibility: scale, shadows, and contact with surfaces.
• Ad variant — a tighter crop with room for copy. Generate three or four directions and pick one; do not over-polish a variant you will A/B test.

For the ad variant, generate the background first, then composite the product on top with free.edit. That two-step approach keeps the product pixel-accurate while the scene stays creative.

Fix the last ten percent by hand

AI gets you to a convincing image; the last ten percent is craft. Check the contact shadow where the product meets the surface, the direction of every highlight, and the label text. A product that floats a millimetre above the plinth reads as fake even when everything else is right.

• Add or soften a contact shadow under the product.
• Match highlight direction to the scene's light source.
• Straighten the product; AI scenes often introduce a slight tilt.
• Recompose the label if the text is wrong — do not ship a garbled brand name.

Keep the pipeline honest

Two rules keep AI product photography defensible. First, the product must be the real product — no invented features, no altered packaging. Second, disclose AI assistance where your market or platform requires it. The image is a rendering of a real item, not a substitute for one.

Check the model's output at full resolution before publishing. Colours drift between models, and a red that looks right in a thumbnail can be orange in a hero image. Compare against the source photo, not against the previous generation.

A workflow you can repeat

Here is the loop, end to end:

• Source a flat, evenly lit photo of the real product.
• Run it through text-to-image or image-to-image with one scene prompt.
• Generate three variations; keep the one with the most accurate product.
• Composite and clean up in free.edit.
• Export at the sizes your storefront needs, then reuse the prompt skeleton for the next SKU.

Every generation on FreeInd shows its credit cost before you run it, so you can budget a scene library instead of guessing. Start with one product and one scene, then expand the library as the workflow settles. Create a free account and run the first SKU today.