How AI Car Shadows Work — and Why Bad Ones Look Fake

You can usually tell a bad car photo composite in under a second. You may not be able to say why. But something is off, and most of the time the something is the shadow.

This article explains what the shadow is actually doing in a car photo, why it is the first thing the eye checks, how generative AI handles it differently from the old cut-and-paste tools, and what to look for in the output before a photo goes on a listing.

Why the shadow is the giveaway

A car is a heavy object sitting on the ground. Your visual system has spent your entire life confirming that heavy objects sitting on the ground block light underneath themselves. You do not consciously check for this — it is processed instantly, before you have read the price or noticed the alloys.

When a car has been cut out of one photo and pasted onto a new background without a convincing shadow, the brain registers it immediately: the car appears to float. It does not look like a car in a studio. It looks like a sticker of a car on a picture of a studio.

This is why shadows matter more than almost any other element of background replacement. A slightly imperfect wall texture goes unnoticed. A floating car never does.

What a real contact shadow actually does

Look at any car parked indoors under even lighting and you will see two things happening underneath it.

Contact occlusion. Directly under the sills, under the tyres, and in the wheel arches, the car blocks ambient light almost completely. This produces a soft, dark band that is tightest where the tyre meets the floor and fades outward. It exists even when there is no strong directional light at all — it is simply the absence of light in spaces the car covers.

Directional shadow. If the scene has a dominant light source — a window, a bank of studio lights, the sun — the car also casts a longer, softer shadow in the opposite direction. Its angle, length, and softness all depend on where that light is and how large it is.

Both have to be present, and both have to agree with the scene, for the image to read as real. A car with a crisp drop shadow pointing left in a studio lit from the left looks wrong. A car with no darkening under the tyres looks wrong even if the directional shadow is perfect.

Cut-and-paste shadows vs generated shadows

The older generation of car photo tools worked by composition: cut the car out, place it on a background image, and add a shadow as a separate layer — typically a generic dark ellipse, blurred and placed under the car. This is the "drop shadow" approach, and it is responsible for most of the floating-car listings you have seen on marketplaces.

The problem with a stamped-on shadow is that it knows nothing about the scene. It is the same shape regardless of the lighting, the same darkness regardless of the floor, and it has no relationship with the wheel arches or the underside of the specific car.

Generative AI works differently. As covered in how generative AI creates car photos, the model does not paste a background behind the car — it synthesises the scene as a whole, and the shadow is generated as part of that scene. Because the model has learned from enormous numbers of photographs what cars on floors actually look like, the shadow it produces tends to have the properties of a real one: occlusion concentrated under the tyres and sills, softness that matches the lighting, and a direction consistent with the light sources in the generated environment.

The shadow and the floor are made together, which is why the result generally grounds the car convincingly — and why the floor reflection, where there is one, agrees with the shadow rather than fighting it. That grounding work is a large part of what Motuva's scene rendering is doing in every output — how reflections and contact realism work →.

Where AI shadows still go wrong

Honesty section. Generative shadows are much better than drop shadows, but they are not infallible, and pretending otherwise would defeat the point of this article.

The cases where generated shadows are most likely to misfire:

• Unusual input lighting. If your original photo was taken in harsh, low sun with a long hard shadow, the model has to reconcile that with the lighting of the studio scene it is generating. Most of the time it does. Occasionally a trace of the original shadow direction survives into the output and conflicts with the new scene.
• Complex ground contact. Cars photographed on gravel, grass, or a steep camber give the model less clean information about where the tyres actually meet the ground. The generated contact point can end up slightly soft or slightly misplaced.
• Very dark cars on very dark floors. When the sill line and the shadow are close in tone, the boundary between them is hard for the model to define and hard for you to check.

The general rule from how AI car photography works applies here too: better input produces better output. A photo taken in flat, even daylight, on level tarmac, gives the model the easiest possible grounding job. Our guide to improving car photo quality covers how to shoot for that.

What to check before you publish

The car in the output is not a generated car: its pixels are lifted from your own photograph and composited into the studio, so nothing about the vehicle is invented. The shadow and the floor beneath it are rendered, though, and a difficult shot can separate imperfectly at the ground line — so check every output before publishing, because the listing is yours either way. For shadows specifically, the check takes about five seconds per image:

1. Is the car grounded? Look at the tyres. There should be a tight dark zone where rubber meets floor. If the car looks like it is hovering a few millimetres above the surface, try a different studio or use a different shot.
2. Does the shadow direction match the scene? Find the brightest part of the generated environment. The directional shadow should fall away from it. A shadow pointing towards the light is the classic composite error.
3. Does the shadow match the floor reflection? On gloss floors, the reflection and the shadow should sit in the same place under the car. If they disagree, the eye will catch it even at thumbnail size.
4. Is the shadow plausibly soft? Studio lighting is large and diffuse, so studio shadows are soft-edged. A razor-sharp shadow edge under a studio-lit car reads as fake.

If a shadow fails any of these, do not publish that image. Re-process it, or shoot the car again from a cleaner position. One bad floating-car photo undermines the credibility of every good photo around it.

Shadows are one of several places where AI output can look subtly wrong — reflections and surfaces are the other big one, and we cover those in why AI car photos sometimes look wrong.

Try it on your own stock

The fastest way to judge how well generated shadows hold up is to run your own cars through the tool and look at the contact points yourself. Motuva's free tier is 20 images a month, full output quality, no card required.

Start free — no card, no demo →