SORA and Veo 3 generated videos often display subtle physics anomalies: objects moving in ways that defy gravity, water flowing unnaturally, smoke or fire behaving inconsistently, or clothing movement that doesn't match body motion. Watch for objects that appear to float, morph, or move in physically impossible ways.

AI video generators can struggle with temporal coherence. Look for objects or people that subtly change appearance between frames, backgrounds that shift unexpectedly, or details that "morph" over time. Elements may appear, disappear, or transform in ways that shouldn't happen in real footage.

While SORA handles lighting well, examine shadows carefully. Look for shadows that don't match object positions, multiple conflicting light sources, or shadows that remain static when objects move. Reflections in windows, water, or mirrors may also show inconsistencies.

Human hands remain challenging for AI. Watch for extra or missing fingers, fingers that bend unnaturally, hands that morph when moving, or hands that seem to "melt" into objects. Similarly, text in signs or screens often appears as gibberish or morphs when the camera moves.

AI-generated faces may show irregular blinking patterns: too-frequent or too-infrequent blinking, eyes that "snap" open and closed, or uncoordinated blinking between the two eyes. This is especially common in deepfake face swaps.

In AI-manipulated videos, examine where the face meets hair, neck, or background. Look for blurring, color mismatches, or unnatural edges that become more visible when the person turns their head or changes expression.

AI struggles with consistent lighting. Check if facial lighting matches the environment. Look for shadows that don't align with light sources, faces that seem too bright or dark, or lighting that stays constant despite movement through different illuminated areas.

AI-generated faces may appear unnaturally smooth (missing pores and fine lines) or overly textured. Natural skin shows subtle imperfections, blemishes, and varying texture. Watch for skin that looks like a plastic mask or has artificial-looking texture patterns.

Hair remains challenging for AI. Look for hair that doesn't move naturally with head movements, individual strands that phase through the face or shoulders, hair that appears to "float," or sudden changes in hair texture or color between frames.

The mouth interior is difficult for AI to render accurately. Watch for blurry or morphing teeth, teeth counts that change, unnatural gum appearance, or an abnormally dark or overly bright mouth cavity in AI-generated content.

Real eyes reflect light sources consistently. In AI videos, check if both eyes show matching reflections (catchlights), and whether those reflections correspond to the scene's lighting. Mismatched or missing eye reflections are red flags.

In AI-generated content with speech, watch carefully for lip-audio synchronization. Even subtle delays, misalignments with complex words, or lips that don't fully close on "B" and "P" sounds can indicate synthetic generation or deepfake manipulation.

Natural expressions precede or accompany speech and emotional context. In AI videos, expressions may lag behind the audio, appear at inappropriate times, or fail to match the emotional tone. SORA and Veo 3 generated characters may show particularly flat or generic expressions.

Look for jerky or floating movements, unnatural posture transitions, motion that seems to "skip" frames, or movements that glitch momentarily. AI-generated videos may show objects or people teleporting slightly between frames or moving in non-physical ways.

Hand gestures and body language should naturally coordinate with speech rhythm and meaning. AI videos may show disconnected gestures, hands that remain unnaturally still, or gestures that don't match cultural or personal communication patterns.

In SORA, Veo 3, and similar AI-generated videos, watch for objects that subtly change shape, size, or appearance over time. Items in the background may morph, disappear, or multiply. This "temporal instability" is a key indicator of AI generation.

AI-cloned voices may exhibit unnatural pitch changes, robotic undertones, or a "processed" quality. Listen for missing breathing sounds, absence of natural vocal fry, or overly consistent vocal tone. Real speech has subtle variations that AI sometimes fails to replicate.

AI-generated audio often has suspiciously clean backgrounds or unnatural silence. Real recordings capture ambient noise, room tone, and environmental sounds. Sudden changes in background noise, or backgrounds that don't match the visual environment, suggest synthetic media.

AI voices may struggle with natural speech rhythms, emphasis patterns, and characteristic speaking styles. Listen for unusual stress on syllables, missing or exaggerated emotional inflections, difficulty with regional accents, or failure to maintain consistent speaking pace.

Beyond lip sync, check if environmental sounds match visual actions: footsteps on different surfaces, object interactions, clothing rustling. AI-generated videos may lack these synchronized audio cues or have generic sound effects that don't perfectly match the visuals.

Play the video at slow speed (0.25x) or step through frame-by-frame using video editing software. AI-generated videos often show temporal inconsistencies invisible at normal speed: morphing objects, changing details, or subtle warping effects between frames.

Examine video metadata for manipulation signs: multiple saves, suspicious timestamps, missing camera information, or edited creation dates. AI-generated videos may lack typical camera metadata or show signs of processing through AI tools.

AI-manipulated videos often show inconsistent compression artifacts. Look for areas with mismatched quality levels, faces that appear sharper or blurrier than surroundings, or unusual compression patterns that suggest selective processing or AI upscaling.

Advanced users can analyze videos in the frequency domain. AI-generated content may show unusual patterns in spectral analysis, particularly in high-frequency details. Audio spectrograms can also reveal synthetic voice generation patterns.