Three runs. The prompt was not kept — nothing to paste, and I would rather leave it empty than invent one from the pictures.
What all three agree on: the costume, in detail. Sleeveless red web-print crop with the black spider centred, blue side panels, matching leggings with the red web running down the shins, the same seam placement and the same textured weave every time. Whatever the prompt said about the outfit, it said it precisely — that level of repetition does not happen by accident.
What drifts: everything else. Two runs are in a dark wood-panelled lounge with louvred shutters and warm pendant lamps; the third is a high-rise apartment with a city skyline through floor-to-ceiling glass, a black leather sofa and a standing lamp. Hair goes long and loose, then a low bun, then cropped and slicked back. Read backwards, that suggests the prompt was specific about wardrobe and vague about the setting — “dark upmarket interior at night” would produce exactly this spread.
The mirrors are where it comes apart. The second run has a mirrored column down the left side, and the reflection in it does not match her — the figure reflected is sitting upright with a different silhouette while she is reclining. The first run does something similar further back. Reflections are the classic failure of every image model: the scene and its mirror image are generated as if they were two separate pictures that happen to sit next to each other. Worth checking any frame with glass in it, because it is exactly the sort of wrongness the eye skates over.
