AI Product Photography: Packshots, Lifestyle Shots, Props

By the end of this post you'll be able to plan an AI product photography set: a clean packshot, a lifestyle scene and a prop-styled hero image. You'll also know which details to lock down so the product doesn't quietly change between shots.

One caution first. AI image models are good at atmosphere and weaker at exact fidelity. If a label, logo or shape has to be right, treat the AI output as something to check against your real product, not a replacement for it.

The three shot types, and what each is for

Most product listings and ads need the same three things.

• Packshot: the product alone, evenly lit, on a plain or lightly tinted background. It's for marketplaces, catalogues and anywhere clarity beats mood.
• Lifestyle shot: the product in use or in a believable setting. It's for ads, social posts and landing pages.
• Prop shot: the product styled with supporting objects (ingredients, fabric, surfaces) that tell a story about it. It's for hero banners and editorial-style posts.

Each one needs a different prompt, and a different tolerance for error. A packshot has almost none. A lifestyle shot can forgive a lot, as long as the product itself is right.

Start with a prompt structure

Black Forest Labs' FLUX.2 prompting guide recommends ordering a prompt as subject, action, style and context, so the most important element comes first. It also suggests 30 to 80 words for most work, with shorter prompts for quick exploration. If you want a fuller breakdown of this idea, see AI Image Prompt Structure: Subject, Setting, Light, Lens.

For product work, that translates to: what the product is (material, colour, shape), where it sits, how it's lit, and how it's framed.

Packshots: boring on purpose

A packshot prompt should describe a calm, controlled studio. Here's a starting point:

A matte black glass perfume bottle with a silver cap, centred on a seamless light grey backdrop, soft diffused studio light from the upper left, gentle shadow underneath, straight-on view, sharp focus throughout, product photography

A few notes on why it's written this way:

• "Sharp focus throughout" instead of "no blur." BFL's guide says FLUX.2 doesn't support negative prompts, so describe what you want rather than what to avoid. Other models treat negatives differently, so don't assume.
• Material words like "matte," "brushed aluminium" or "frosted glass" tell the model how light should behave on the surface.
• A plain backdrop gives you something easy to cut out or recolour later.

BFL's guide also says hex codes work best when tied to a specific object, for example the backdrop in color #E8E4DC. That's handy when a brand has a set background colour. Expect close, not exact, and check the result on screen.

Editing a real product photo

If you already have a photo of the product, editing it usually beats generating it from scratch. OpenAI's image generation prompting guide gives a good principle: state what must stay unchanged. Its examples include telling the model not to change the background, camera angle or framing, and not to add accessories, text, logos or watermarks. For product extraction it describes keeping product geometry and label legibility intact.

The same guide recommends medium or high quality settings for label-heavy work, to reduce drift over repeated edits. That's a setting in OpenAI's own tools. Elsewhere, the equivalent habit is to make one change per edit rather than five.

A prompt in that spirit:

Keep the bottle, label and cap exactly as they are. Replace only the background with a seamless light grey studio sweep. Do not change the product's shape, colour or label.

Lifestyle shots: give the product a reason to be there

A lifestyle prompt works best when it reads like a mini creative brief. OpenAI's guide suggests describing the brand positioning, the vibe, the audience and the scene together, then letting the model fill in tasteful details.

Try:

A ceramic pour-over coffee set on a wooden kitchen counter at 7am, warm sunlight through a window on the left, steam rising from the cup, shallow depth of field, shot on an 85mm lens at f/2.8, calm, slightly editorial mood

The lens line comes from the style of example in the FLUX.2 guide, which says that naming real camera gear (its example is a Hasselblad with an 80mm lens at f/2.8) gives more authentic results than a vague "professional photo." It's not magic. It nudges the look, and results still vary from run to run.

Putting a real product into a scene

Google's Gemini API image generation docs show reference-image workflows for product mockups. Their example is a logo placed on a high-end ad for a fragrance, with the instruction that the logo be integrated into the bottle. The docs also say you can mix multiple reference images, with the limit depending on the model (up to 14 on some). All generated images from that API carry a SynthID watermark, which is worth knowing if you publish them.

So the practical workflow for a lifestyle shot with a real product is:

1. Upload one or more clean photos of the product as references.
2. Describe the scene, light and camera.
3. Say explicitly what the product must keep: shape, colour, label.
4. Generate several variations and pick the one where the product is most accurate, not the one with the prettiest background.

Prop shots: fewer props than you think

Props are where AI images go wrong most often. Ask for ten objects and you'll get ten slightly odd ones. Keep it to two or three that belong to the product's story.

A tube of hand cream on a folded linen cloth, with a sprig of lavender and a small ceramic dish beside it, soft north-facing window light, pale oat-coloured background, overhead flat lay, generous empty space on the right

Tips that help:

• Choose props that explain the product: ingredients, materials, the place it's used.
• Leave negative space if you plan to add copy later. Say where it should be.
• Tie colours together. Two prop colours plus the product colour is plenty.
• Check hands, small objects and anything that should have text. These are the usual failure points.

Common mistakes

• Changing the product between shots. Different runs can alter proportions, cap shapes or label details. Use the same reference photos every time and compare outputs side by side.
• Trusting text on packaging. Small print on labels is often wrong or garbled. Re-add real logos and copy in an editor.
• Stacking adjectives. "Luxurious, premium, high-end, elegant" adds little. One concrete light and one concrete surface do more.
• Skipping a reality check. If a customer would receive something that looks different from the image, that's a problem for returns and trust, not just aesthetics. Many platforms and regions also have rules on how AI images can be used in listings, so check the ones you sell on.
• Starting expensive. Iterate on composition with a fast, cheap model, then re-run the winner on a stronger one.

How to run this in Mediafy Studio

In Mediafy Studio, pick Image, choose a model, and write your prompt. Models such as Nano Banana 2, Seedream and GPT Image are on offer, and the lineup changes over time. The Enhance button can rewrite a short prompt into a fuller one, which is a quick way to get from "perfume bottle on grey" to something with light and lens details.

Cost shows on the Generate button before you run it. A Nano Banana image is 3 credits, so a sensible routine is to check the cost of each model, draft on a cheaper one and finish on a stronger model. If you want a walkthrough of the basics, there's Your First Image in 5 Minutes.

Seeds, saved sets of reference images, can help you keep a consistent look across a series. They're for style consistency, not custom model training, so you should still verify the product itself in every image.

If you plan shoots with an assistant, Mediafy Chat can help brainstorm scenes and draft prompts. The same Studio can also animate a finished still into a short video with models such as Kling or Veo 3.1, which is a natural next step once you have a hero image. To learn the wider workflow, the AI Creator Lab course is there too.

A simple shot list to copy

1. One packshot on a plain backdrop, front view.
2. One packshot at a three-quarter angle.
3. One lifestyle shot in the main use context.
4. One prop-styled flat lay with space for copy.
5. One close-up of a material or texture detail.

Generate each at least three times, then compare against the real product before anything goes live.

Ready to try it? Sign in with Google and run your first packshot.

Sources

• Black Forest Labs, "FLUX.2 Prompting Guide": https://docs.bfl.ai/guides/prompting_guide_flux2
• OpenAI, "Image generation models prompting guide" (OpenAI Cookbook): https://developers.openai.com/cookbook/examples/multimodal/image-gen-models-prompting-guide
• Google, "Image generation with Gemini" (Gemini API docs): https://ai.google.dev/gemini-api/docs/image-generation