Shadows in AI Photos: Why Your Subject Floats

Your subject looks like it is floating when the contact shadow is missing: the dark seam exactly where a foot, chair or bag touches the surface. It is only a few pixels wide, yet it is the first thing that makes an image read as fake. AI models struggle with it, especially when you place a person into a new scene or swap a background. Below you will find the three types of shadow, what makes an edge hard or soft, and the words that steer it in your prompt.

The three shadows in every image

Every lit scene contains three kinds of shadow, and each does something different:

• Contact shadow: the narrow, dark seam where your subject meets the surface. This is what plants your subject on the ground. Without it, everything floats.
• Cast shadow: the shadow your subject throws onto the environment, away from the light. This tells the viewer where the lamp or sun is.
• Form shadow: the dark side of the subject itself, the side turned away from the light. This gives shape and volume.

Models usually render the form shadow well, because it is baked into the subject. The cast shadow often comes along halfway. The contact shadow is the hardest one, and it happens to carry the most weight.

Why a shadow edge is hard or soft

Edge hardness depends on the size of the light source relative to your subject, not on how bright the light is. A small source gives a crisp edge, a large source a soft one.

Physically, a shadow has two parts. The umbra is the darkest core, where the light source is fully blocked. The penumbra is the rim around it, where the source is still partly visible (Wikipedia). The wider that penumbra, the softer the shadow looks.

That is why even a shadow in harsh sunlight never has a perfectly sharp edge: from Earth, the sun spans roughly half a degree, about 32 arcminutes (Wikipedia, Angular diameter). That transition zone widens the further your subject sits from the surface. A foot on the pavement gets a tight, dark shadow; an outstretched arm high above the ground gets a vague one.

Translated into your prompt:

• Hard shadow: direct sunlight, a bare flash, a spotlight. Terms: hard direct sunlight, sharp defined shadows, high contrast.
• Soft shadow: overcast sky, large window, softbox. Terms: soft diffused light, overcast, soft gradual shadows.

What viewers actually notice

Viewers are poor at spotting a wrong shadow direction, but a missing shadow registers immediately. In a 2019 study by Nightingale, Wade, Farid and Watson, participants scored an average of 61% correct when judging whether shadows in a scene were consistent. On the scenes that actually contained an error they scored 46%, below chance (Attention, Perception, & Psychophysics, 2019).

The practical takeaway: do not burn renders chasing a physically perfect shadow angle. First make sure a shadow exists at all, and that your subject makes contact with the ground. That buys far more believability than an angle accurate to within ten degrees.

How to describe shadow in your prompt

Name three things: light direction, hardness, and contact with the ground. For example:

"woman sitting on a wooden bench in a park, late afternoon sunlight from the left, long soft cast shadow falling to the right, dark contact shadow where her shoes meet the gravel"

Building blocks that work:

• Direction: light from the left, backlit, side lighting, sunlight from behind the camera.
• Grounding: contact shadow under the feet, subject firmly grounded, shadow anchoring the subject to the floor.
• Character: long shadows, soft shadow falloff, subtle ambient occlusion.

Keep the light direction consistent with the rest of the scene. If you describe a window on the left, let the shadows fall to the right. In the photo generator you can render the same prompt twice, once with and once without the grounding line. The difference is usually bigger than you expect.

Watch out with composite images

With a replaced background or a person taken from a reference photo, the shadow almost never works out on its own. Your subject was lit for the old scene and is being pasted into a new one.

Two things help:

1. Pick a background with similar light. A subject shot in soft indoor light rarely lands well in harsh midday sun. Choose a scene with the same hardness and direction.
2. Ask for the shadow explicitly. State in your prompt that the subject casts a shadow on the new surface, and name the direction. Without that instruction, models often drop your subject in without any shadow at all.

If you work with background removal or an AI editor, remember that a cut-out subject literally loses its shadow. You have to ask for it back in the next step.

Common mistakes

• No contact shadow. The subject hovers a few centimetres above the ground. Ask explicitly for a dark zone at the point of contact.
• Two light sources contradicting each other. A face lit from the left with a cast shadow also falling left. Name one main direction.
• Shadows that are too dark. A real shadow is rarely black; ambient light always fills it. Ask for soft shadow with visible detail instead of pitch black.
• Light direction drifting across a series. If you are building a series with the same persona, lock the light direction the way you lock face and clothing.

Frequently asked questions

Why does my AI subject look like it is floating?

Almost always because the contact shadow is missing: the dark seam where the subject touches the surface. Put it in your prompt explicitly, for example "dark contact shadow where the feet meet the ground".

What is the difference between a contact shadow and a cast shadow?

The contact shadow is the narrow, dark zone at the point of contact with the surface, and it handles grounding. The cast shadow is the shape your subject throws further onto the environment, showing where the light comes from.

How do I get softer shadows?

Describe a larger light source: overcast sky, large window or softbox. A bigger source relative to your subject creates a wider transition zone and therefore a softer edge.

Does the shadow angle have to be exactly right?

No. Research shows viewers barely notice a shadow pointing the wrong way. Presence and grounding matter more than a precisely correct angle.

Shadow is one of the cheapest ways to make an AI image read as real: a single line about light direction and ground contact often does more than an extra render at higher resolution. Create an account and test it on your own scene in the photo generator. Render the same prompt with and without grounding, and see which version sticks.