Reality vs AI generated content becoming indistinguishable - 410 second workflow using wan 2.2 text-to-video

The prompt I used was pretty strange which explains these odd outputs:

Fashionista, s4msung, Modern Loft Design Aesthetic, Modern Loft Design represents a perfect fusion of practicality and visual appeal. This approach, true to its name, highlights structural building elements that traditional interior design typically conceals. Prime examples include exposed concrete walls and visible ductwork. The emphasis centers on practicality and function while aesthetics adopt a contemporary viewpoint. Components borrowed from commercial spaces, manufacturing facilities and storage buildings dominate loft-styled homes. These unfinished industrial components create bold visual impact. A loft design styled residence typically features open concept layouts with different areas flowing together, separated only by strategically placed furniture pieces. Using this approach, interior designers don’t need decorative accessories to achieve sophistication or trendiness. The loft design aesthetic provides homes with metropolitan character, enhanced by raw materials and visible elements like steel hardware and finishes borrowed from classic warehouse aesthetics. This design philosophy might not suit every homeowner’s taste, though it’s hardly divisive. Loft-styled residences can be found worldwide - places like Chicago, Germany and beyond. An urban atmosphere serves as the main distinguishing characteristic of modern loft interior design.

low-end smartphone footage, minor camera shake detected

grainy image quality, excessive digital sharpening, strong HDR artifacts, novice photography, overexposed highlights, underexposed dark areas

That prompt is absolutely wild - you threw everything at the AI and got bizarre results, which makes sense. What’s interesting is how these text-to-video models handle contradictory instructions. You’re asking for modern loft aesthetics but also want low-end smartphone footage with camera shake and HDR artifacts. The AI tries to reconcile these opposing demands, which creates those strange outputs you mentioned. When I’ve experimented with WAN 2.2, the more contradictory elements I include, the more unpredictable it gets. Sometimes this creates genuinely creative accidents, but usually it just confuses the model about what you actually want.