How to make AI product photos look real
Get the physics right. Give the product a shadow where it touches a surface, light from one clear direction, reflections that match the scene, visible texture and natural color. In the prompt, describe light as something that falls on a surface, name a real lens and aperture, and ask for film grain. Then check the contact shadow and finish with a light grain and color grade.
The tells, and the fix for each
| Tell | What you see | Fix in the prompt | Fix afterward |
|---|---|---|---|
| Too smooth | Plastic skin, fabric with no weave, glass with no dust | Name the material and its texture: "brushed aluminum", "slub cotton" | Add grain; sharpen texture a little |
| No contact shadow | The product seems to hover | "Resting on the table, a soft contact shadow where the base meets the wood" | Paint a thin dark shadow along the base |
| Light from nowhere | Bright everywhere, no shadow side | One source with a direction: "window light from camera left" | Darken the far side with a soft gradient |
| Impossible reflections | A bottle reflecting a window that is not in the room | Describe the surroundings; keep shiny products in simple scenes | Retouch, or composite the real product |
| Perfect symmetry | Product dead center, props mirrored | Give a composition: "product in the right third" | Crop off center |
| Oversaturated color | Candy colors, glowing highlights | "Natural color balance", plus the palette as hex codes | Lower saturation and grade toward your real photos |
| Floating objects | Props or crumbs hanging with no reason | Say what rests where | Remove them |
| Wrong scale | A jar as big as the kettle beside it | Give real sizes: "a 12 cm jar" | Recompose, or composite at the right size |
| Melted text | Warped letters on the label or on signs behind | Keep text out of the scene; lock the label | Composite the real label |
| Too clean | No crumbs, no creases, no wear anywhere | Ask for one or two traces of use in the scene | Leave the traces in |
Melted labels have their own guide: why AI changes your product.
Describe light as something that lands on a surface
Adjectives like "beautiful lighting" give the model nothing to draw. The prompts behind the photos our agency kept described light as an event: where it comes from, what it hits and what shadow it leaves. Google’s prompting guide for Nano Banana works the same way, with phrases such as "three-point softbox setup" for even product light and "golden hour backlighting creating long shadows".
| Vague | Specific |
|---|---|
| Dramatic lighting | Hard direct flash carves ultra-sharp shadows beneath each piece |
| Natural light | Hard raking natural daylight enters at a 45-degree angle across the chest, carving the herringbone weave |
| Studio lighting | A large soft key from the upper left at forty five degrees with even wraparound fill, a soft natural contact shadow under the shoes |
For more words to use, see lighting terms for AI prompts.
Name a real lens, aperture and camera height
A camera sentence tells the model how the photo was taken: what is in focus, how the perspective falls and where the camera stands. Our kept prompts on Nano Banana models each had one. OpenAI’s prompting guide calls camera settings cues for how the image should look, with no guarantee that the model simulates the optics exactly.
- Texture close-up: 100 mm macro at f/4, focus on a button and the thread beside it (our shirt close-up).
- Full-body store photo: 85 mm at f/8, camera at chest height, level, no tilt (our trouser photos).
- Product shot from overhead: 50 mm at f/8, focus even across the whole flat surface (our food ad on white).
- Tight hero of a small product: 90 mm at f/4, focus locked on the glossy peak of the middle piece (our brownie stack).
For what each focal length does, see camera and lens terms for AI prompts.
Add grain on purpose
Real photographs carry grain or sensor noise, and a surface with none reads as rendered. On our brownie campaign, all ten kept prompts asked for heavy film grain, and nine carried the same sentence, added by hand: "Heavy visible film grain throughout." In our kept prompts on Nano Banana models, grain had its own sentence, apart from the camera sentence.
The model makers suggest the same. Google’s guide offers "as if on 1980s color film, slightly grainy", and OpenAI’s sample photo prompt asks for "subtle film grain, natural color balance". Match the amount to the brand. On our jobs, "heavy visible film grain" went into a gritty food ad and "clean near-zero grain" into a crisp shirt close-up.
Leave some imperfection in
Real sets have traces: a crumb, a crease in the linen, a drop of condensation, a scuff on the floor. Put one or two in the scene and none on the product, which should look like the one you ship. OpenAI’s sample prompt for a real-looking photo asks for "worn materials, and everyday detail" and "no heavy retouching".
Finish in post
- Check the contact shadow first, and paint one in if it is missing.
- Lower saturation until the product matches a real photo of it placed side by side.
- Grade the set together: the same white balance, contrast and black level across every image.
- Add a little grain over the whole frame last, so the product and the scene share it.
- Look at 100% for melted edges, extra fingers and warped text.
Before and after checklist
- The product touches a surface and has a shadow there
- Every shadow points away from the same light
- Reflections show things that are in the scene
- Material texture is visible at 100% zoom
- Colors match a real photo of the product
- The composition is off center, unless centered is the brand’s style
- Grain covers the product and the background equally
- No floating props and no gibberish text
- The product is the right size for the objects around it
- Label, logo and parts match the real product
The full sign-off list, with checks for people, text and legal issues, is in the QA checklist for AI images.
Questions people also ask
- Does adding "photorealistic" to the prompt help?
- OpenAI’s prompting guide says to ask for "photorealistic" or "real photograph" explicitly when that is the goal, and to describe framing and texture as well. The word works best alongside the light, lens and material details above.
- Which image model looks most real?
- Rankings change month to month and depend on the product. Run the same prompt on two or three models and compare them side by side. See which AI image model is best for product photos.
- Should marketplace main images get grain too?
- Keep it light or leave it off. A main image on white should look like a clean packshot of the product you ship. Grain and grading belong to lifestyle, ad and editorial images.
Where Overs fits
Overs writes each prompt from the brand’s visual style, including its light, lens and grain, and runs a quality check before the render. Nothing is charged until you approve the shot plan and the estimated cost.
Free for 40 photos a month. The AI that makes the photos is billed separately, on your own key, with no markup from Overs.