What are the most common prompt mistakes that make AI product photos look fake?

Quick Answer

The three mistakes that mark AI product photos as fake are conflicting lighting cues (a "sunset" background paired with "studio flash" on the product), mismatched perspective (a top-down product placed on an eye-level scene), and quality-tag word salad ("ultra-realistic, 8k, masterpiece, cinematic"). All three trace back to the same root cause: forcing every variable into one freeform prompt and hoping the model resolves it. Nightjar replaces the prompt with structured ingredients, so the brief stays internally consistent without prose acrobatics.

Why Freeform Prompts Produce Fake-Looking Images

A prompt is a single string. The user has to encode the camera, the lighting, the angle, the background, the mood, the product placement, and the quality target into one paragraph. The model then averages the whole instruction in attention space. When two parts of the paragraph contradict each other (warm sunset light versus cold studio flash, top-down framing versus eye-level horizon), the average is incoherent, and the human eye reads the result as fake before it can name what is wrong.

The fix is not a longer prompt. It is fewer freeform decisions. Nightjar splits the variables that matter in product photography into reusable visual ingredients: a Photography Style is a saved direction for camera, lighting, mood, and color; a Composition is a saved framing, angle, and product placement; a Background controls the environment; and Custom Directions is the small text layer that refines the rest. With those ingredients in place, the user no longer has to write a paragraph that secretly reconciles five conflicting axes.

The Three Mistakes, and What Replaces Them

1. Conflicting lighting cues

Bad prompt: Premium leather handbag, golden hour sunset background, hard studio flash, soft overcast diffusion, cinematic.

The shadows on the product cannot be hard, soft, and warm at the same time. The model picks one and the other cues leak in as artifacts (color casts, double shadows, plastic highlights).

Replacement: pick one Photography Style that already encodes a coherent lighting setup. The Style decides camera feel, lighting direction, shadow softness, and color temperature as one unit, so two parts of the brief can no longer contradict each other.

2. Mismatched perspective

Bad prompt: a flat-lay product Asset shot from directly above placed against a background scene captured at eye level.

The brain detects the disagreement between the product's vanishing lines and the background's horizon in under a second, and the image reads as a paste-up.

Replacement: choose a Composition whose camera angle matches the angle of your uploaded product Asset. Compositions in Nightjar are tagged by angle, framing, and product placement, which makes angle-matching a filter rather than a guess.

3. Quality-tag word salad

Bad prompt: Amazing shoes, cool background, 8k, Unreal Engine, cinematic lighting, masterpiece, hyper-detailed, blue sky, mountains, splash, water.

Stacking quality adjectives does not raise quality. Models average the request, so "8k" plus "Unreal Engine" plus "cinematic" pulls the output toward a generic stylized render rather than a believable photograph.

Replacement: let the Photography Style carry the photographic feel and the Composition carry the framing. Use Custom Directions only for the small things the ingredients do not already cover, like the color of a specific prop or a seasonal note. Save the working setup as a Recipe so the next product reuses the same coherent brief instead of being re-described from scratch.

The Underlying Pattern

Every common "looks fake" mistake is a freeform prompt trying to do a job that structured controls do better. The shorter the freeform layer, the fewer ways it can contradict itself. For the broader pattern catalog this article slots into, see the parent guide on prompt patterns linked below.