# How to Make AI Product Photos Look Real: a 7-Point Checklist (2026) > Seven reasons AI product photos look fake and the fix for each: light, contact shadows, materials, depth of field, prompt words, fidelity and editing. Published: 2026-08-22 Updated: 2026-10-14 Author: Olga Stogova Tags: ai-product-renders, product-photography-lighting, image-quality, tutorial, ai-product-photography Kive studio: Bare Factory Canonical: https://kive.ai/learn/make-ai-product-photos-look-real AI product photos look real when seven physical things agree in one frame, and they look fake the moment one of them does not. The seven are light direction, shadow edge width, the contact shadow, how highlights behave on the material, depth of field against the product's scale, the words in the prompt, and whether the product in the frame is the one you photographed or a redrawn version of it. Six of the seven are visible in five seconds without any tool. The fix for each is below, in the order you should check them, followed by the light and material physics that explain why they fail. ## Key takeaways - Check shadows against the intended lighting setup; multiple lights and perspective can explain different apparent directions. - Compare material response, contact and label geometry with the source instead of judging realism from one highlight. - Choose focus for the details the shopper needs; a sharp product is not itself evidence of a fake image. - Specific prompt instructions make results easier to evaluate, but do not guarantee physically accurate lighting or optics. - Review the whole image after a targeted edit; changes can extend beyond the requested area. --- ## The seven checks, and where each fails Use these checks against the intended image and real product reference. They are review aids, not a claim that every photo follows one lighting formula. | Check | Inspect | Action when it fails | | --- | --- | --- | | Light direction | Shadows and highlights against the intended lighting setup | Clarify the lighting and inspect the next output | | Shadow edge | Softness in context of source, object and surface | Adjust the intended light description or edit the shadow | | Contact | The product’s join with the surface | Correct a floating or missing contact region | | Material | Reflection, texture and transparency against the source | Restore or regenerate inaccurate material detail | | Focus | Whether the label and sale-critical details are readable | Ask for the needed focus or use an appropriate finishing workflow | | Prompt | Whether it describes the intended image concretely | Replace vague instructions and compare results | | Product fidelity | Label, geometry and color against the source | Reject changes or restore approved product detail | ## Light and shadow: direction, edge width, contact ### Shadow direction is the first test In a scene intended to have one dominant light, check whether the cast shadows and highlights support that setup. Perspective, object position and surface shape affect their apparent direction. Multiple lights can produce multiple shadows in a real photograph, so judge the stated lighting setup rather than rejecting every image with two shadow directions. Compare the product and a nearby prop on the same surface. Look for unexplained differences in shadow softness, contact and lighting; then inspect the source image to see whether its original shadow carried into the new scene. _Hard sun: narrow shadow edge, streaked highlight – Kive studio: Whitesand_ Reuse an approved lighting direction to reduce setup changes. In Kive, a [Studio](https://kive.ai/docs/guides/ai-studios) supplies reusable visual direction, but each generation still needs the shadow and highlight checks above. A saved preset is not a physical light rig or a guarantee that two frames agree. ### Edge width tells you the size of the source Shadow softness depends on the apparent size of the light source and the positions of the source, object and receiving surface. A nearby large softbox often gives softer transitions than a small distant light. Look for contradictions between the intended light and the result, while allowing for material and surface geometry. Highlight size alone is not a reliable measurement of the source; a curved or rough surface can spread a reflection differently from a flat polished one. The fix is to name the modifier and its distance rather than the mood: a large softbox close to the product gives the wide penumbra, a bare bulb far away gives the crisp edge, then inspect whether the generated highlight agrees. ### The contact shadow decides whether the product is really there Where an object meets a surface, light cannot reach the join. The result is a tight, dark band right at the contact point, denser and sharper than the cast shadow stretching away from it. It is small, most viewers never consciously notice it, and its absence is the single loudest signal that an image was assembled rather than photographed. This is the check that matters most for anyone swapping backgrounds. Cut a product out cleanly, drop it on a new surface, and the cast shadow may survive the move while the contact shadow does not. The product then floats a millimeter above the world it is supposed to be standing in. If you are choosing a preset rather than building a setup, the short descriptions do the work of naming the lighting state – hard direct sun, soft overhead gradient, strong directional shadow, low key spotlighting. Kive's [discover page](/discover) lists them that way, and the [writing prompts guide](https://kive.ai/docs/creating-with-ai/writing-prompts.md) covers how to keep a chosen state from drifting between generations. The fix is to generate the product standing on the scene's surface instead of compositing it over a finished background, and, when the band is still missing, to paint it in with a region edit rather than rolling the whole frame again. The full inspect-and-edit loop, defect by defect, is in [how to fix an AI image without regenerating](/learn/fix-ai-image-without-regenerating). ## Materials, depth of field and the words you use ### Specular behavior is material-specific Materials respond differently to the same light. Brushed metal can stretch reflections along its grain, polished glass can reflect and refract the scene, and fabric sheen varies with weave and finish. Use the real product reference to decide what belongs. A visible caustic is not required in every glass photograph: the light, background and geometry may make it weak or invisible. Inspect the bottle edges, transparency and reflections together rather than using one highlight as a pass/fail rule. | Material | What the highlight does | Wrong if it shows | | --- | --- | --- | | Brushed metal | Stretches into a streak along the grain direction | A round point highlight, as if polished | | Polished glass | Can bend light into caustics on the surface below | Transparency or refraction that contradicts the source | | Matte cotton | Returns almost no specular; color only | A glossy sheen, which reads as synthetic | | Leather | Broad low-contrast sheen following the curve | A hard point highlight, as if lacquered | _Soft even light: low-contrast sheen, no hard edge – Kive studio: Cobalt Minimal_ ### Depth of field has to match the product's scale Depth of field depends on aperture, focus distance, magnification and the camera setup, not product size alone. A shallow focus can make a furniture scene look miniature; an all-sharp bottle can be intentional, including in focus-stacked product photography. Decide which details the listing must show, then describe the desired focus in the prompt and inspect the result. Camera terms guide a generator; they do not enforce optical settings. ### Scale and camera height belong to the same check A product can be lit perfectly and still look pasted in because its size or viewpoint disagrees with the scene. Compare it with anything whose real size a shopper knows, such as a hand, a cup or a book spine: a serum bottle as tall as a coffee mug reads as a model, not a photograph. Then check camera height. If the tabletop is seen from above but the product is shown dead level, two cameras are in one frame. The horizon of the surface, the ellipse of a round base and the label's perspective should all agree on where the lens was. When they don't, describe the camera position in the prompt, for example eye level or a 45-degree view from above, and inspect the base ellipse first. ### Hype words versus camera words Specific light, surface and focus instructions are easier to evaluate than a string of quality adjectives. Test those descriptions on the generator you use rather than assuming any word guarantees realism. Kive's [prompt-writing guide](https://kive.ai/docs/creating-with-ai/writing-prompts) explains how to describe the intended image. This is a starting brief, not a promise that the model simulates the camera: ```text Use the attached product reference on a matte paper surface. Large softbox camera-left, soft shadow edge and visible contact beneath the base. Keep the label readable and compare the product color with the reference. Show material-specific reflections, with no distracting objects behind the product. Keep the product details in focus; allow a softer background. ``` --- ## Generate, inspect and choose a targeted correction ### The workflow that keeps all seven checks **Photograph the product properly first** Front, three-quarter and one detail, under even light, with the label sharp and the whole object in frame. Every generated frame inherits this capture, so a soft label here is a soft label everywhere. A sharp source gives you a reliable reference for the output check. **Generate the scene around the real product** Use a mode that anchors to your photograph rather than describing the product in words. In Kive that is [Product Shots](https://kive.ai/docs/guides/ai-product-shots): the source product guides the new image, and you inspect the result for unwanted changes. Then run the 200% check on label text, geometry and color against the physical item. **Fix one region instead of rolling again** A missing contact shadow, a wrong reflection or a soft corner of a label is a region problem. Select it and describe only the change with the [general edit tool](https://kive.ai/docs/editing-tools/general-edit-tool). The [product photo editor](/tools/product-photo-editor) page shows before and after examples of clutter removal, a color fix and a lighter background. Review the full result after an edit; if the composition remains wrong, regenerate and compare again. **Change one variable per frame** When you need a second frame, keep the studio, the lens and the product and change the angle or the surface. If two things change at once you cannot tell which one broke the light. This is also how a set of frames stays consistent enough to sit together on a product page; the tools that do it well are in our [rundown of AI product photography tools](/learn/best-ai-tools-for-product-photography), and the room-building side is in [placing products in lifestyle scenes](/learn/ai-lifestyle-product-photography-staging). For a product already on Shopify, start with its existing photos through the [Kive Shopify app](https://kive.ai/docs/integrations/shopify). In Shopify admin, open the product and choose **Create with Kive**, select a studio, aspect ratio and quality, then generate. Run the realism checks before adding a selected result to its product images. Use **Open advanced tools in Kive** for further editing. App Store installs can connect a free workspace you administer; generation uses your workspace credits. **What goes wrong** Three tells survive everything else. A product that floats, because the contact shadow was never generated and nobody painted it in. A highlight that disagrees with the shadow, because the scene was lit twice, once by the prompt and once by the composite. And a plastic skin on matte materials, because the prompt asked for detail instead of describing the surface. Check for these three last, on every frame, at 100% and at 200%. --- None of the seven checks needs a light meter or a rendering background. They need the habit of looking at a frame the way a photographer looks at a set: where is the light, what is it doing to the edges, is the product standing on anything, does the surface behave like the material it claims to be. The pattern in the failures is that they come from asking the model to invent physics it was never shown: a product it has not seen, a light that exists only as an adjective. Give it the photograph and the vocabulary of a real shoot, and the realism stops being a setting you chase and becomes a set of rules the frame either obeys or does not. The prompt side of that realism, subject first, objects over adjectives and light with a direction, is in [how to write AI image generation prompts](/learn/how-to-write-ai-image-generation-prompts). **Read the light in your own render** Kive studio: Whitesand Open a studio with a defined lighting state and generate one frame from your product, then run the seven checks in this article on it. Prompt: hard direct sun, single source from the left, crisp shadow edge, product on a plain surface **Studios named by their lighting state** Live from Kive Discover. The short description is the lighting, not a mood word. --- ## Sources 1. [Photoroom - Instant shadows tool]() 2. [getimg.ai - How to make AI product photos look real]() 3. [Nightjar - How to make AI product photos look real: checklist]() ## FAQ ### How do I make AI product photos look real? Treat the generation like a shoot: one light direction for every shadow, a shadow edge that matches the source, a dark contact band where the product meets the surface, highlights that behave like the material, depth of field that fits the product's scale, camera words instead of hype words, and your real product in the frame rather than a redrawn one. ### Why do AI product photos look fake? Because two or more physical rules disagree in the same frame. A shadow points one way while the highlight says the light is elsewhere, a product floats above its surface, matte fabric returns a glossy sheen, or a shampoo bottle carries the shallow focus of a macro lens. Each is visible in five seconds once you know where to look. ### How do I get a realistic contact shadow under a product? Generate the product standing on the scene's surface rather than composited over it, then check the join: a tight, dark band right at the contact, fading outward, on the side away from the light. If it is missing, add it with a region edit instead of regenerating, keeping the light direction the other shadows already agree on. ### Which camera and lens terms work in prompts? Describe the light position, modifier, surface and desired focus. Camera and lens terms can communicate a look, but a generator does not necessarily simulate those settings. Compare results on your chosen model and keep the instructions that help. The prompt example in this article is a starting brief, not a guarantee. ### How do I keep the label accurate in a generated image? Use a sharp product reference and compare the generated label with the original at 200%. Referencing an image does not guarantee unchanged pixels. If small lettering changes, restore it from approved artwork or use a workflow that preserves the original product region. Recheck the whole image after editing. ### How do I fix one flaw without regenerating the whole image? Try a targeted edit when the composition already works. Describe the local change, then compare the entire output with the previous frame: edits can affect details outside the intended area. Keep the earlier version so you can revert. Regenerate when a local correction cannot resolve the underlying composition.