The only prompt in this collection written in Vietnamese, and the only one where the difference between two runs is a single missing object.
Holds across both: the café, in detail — raw concrete ceiling slab, exposed spiral ducting, minimal black pendant lamps, pale wooden tables, black chairs, and a coffee grinder and bar softly out of focus at the back. The outfit is exact twice: grey ribbed halter crop with the deep U neckline, black washed high-waisted jeans. Long straight dark hair with a soft side part, thin eyeliner, calm and slightly cool expression, soft frontal light from the left. The section headings clearly did their job.
Varies, and it is the whole point: một tay cầm ly nước đá màu đen với ống hút đỏ đưa lên ngang vai — one hand holding a black iced drink with a red straw, raised to shoulder height. The second run does it exactly: a clear glass of cold brew, ice visible, red straw, held up beside her. The first run has nothing in either hand at all. Same prompt, same everything else, and the prop simply did not appear.
Both runs also ignore the framing. The camera section asks for a medium shot từ hông lên, from the hips up. Both come back tighter than that, cropping around the waist. That is the same failure as everywhere else here: wardrobe and setting are objects and they arrive; where the camera stands is an instruction and it drifts.
Writing the prompt in Vietnamese cost nothing. Worth recording, since it is the obvious question: the model handled the section headings, the fabric vocabulary (vải rib knit, denim mềm) and the negative list without any sign of struggling. The one instruction it dropped was dropped in a way that has nothing to do with language.
