Short answer: of these ten cinematic prompts, the ones built around one clear subject, one light source and a simple setting gave the most usable frames in the ten images I checked. The ones that asked for signage, paper, or several people produced the visible problems: invented and garbled text, a moving car where the prompt said "parked", and unrequested extra hands. The ten prompts below are the exact ones that were sent to Z-Image (recorded as `zimage`), with what each result actually shows.
What every prompt in the set has in common
Each prompt names a subject, a place, a light source, a lens length and a mood, usually in that order, and most end with "subtle film grain" or a similar phrase. That is a common way to write a shot description. Whether the lens numbers (35, 40, 50, 70, 85mm) changed the rendering is something this set cannot answer, because each prompt was generated once and no prompt was rerun with a different lens. What the set can show is how closely each image followed its text.
All ten generations are 1440x2560 vertical files, including the prompt that asked for a "wide cinematic photograph". Frame shape appears to come from the output size setting, not from words in the prompt. If you need a horizontal establishing shot, set a horizontal size; do not rely on the word "wide". (No generation settings other than the prompt and model were recorded, so I cannot say which size option was chosen.)
Where the image matched the prompt
The rainy street portrait is the cleanest match.
Cinematic street portrait of a young woman standing under a transparent umbrella at night, wet asphalt reflecting neon signs, soft blue and magenta rim light, shallow depth of field, 50mm lens, realistic skin texture, subtle film grain, composed like a quiet scene from an independent drama, natural expression, no text, no logo
The umbrella is see-through with visible droplets and metal ribs, the background breaks into blue and pink bokeh, and the gripping hand has a readable set of fingers around a curved handle. The prompt did not mention a red coat or a red handle; both appeared on their own. There is no readable text anywhere, which fits the "no text, no logo" ending, though one image cannot prove the phrase is what kept it clean.
The train platform prompt also followed its text for light and color.
Cinematic photograph of two people standing apart on a quiet train platform at dusk, one holding a small suitcase, train lights glowing in the distance, cool blue ambient light with warm highlights, 85mm lens, shallow depth of field, emotional tension, realistic clothing, subtle film grain, composed like a final scene
The teal sky, warm red and orange signal lights and wet reflective platform all read as requested, and the suitcase is clearly in the woman's hand. Two things differ from the text. The people are not "standing apart"; they stand close together and look at each other. And both figures are almost pure silhouettes, so "realistic clothing" cannot be judged: there is no fabric detail to see. The silhouette hides hands and faces, which also hides any errors there.
Where the image disagreed with the text
The desert road prompt said "a lone vintage car parked on an empty desert highway".
Wide cinematic photograph of a lone vintage car parked on an empty desert highway at golden hour, distant mountains, long shadows across the road, warm orange sunlight, 35mm lens, low camera angle, realistic dust in the air, expansive negative space, subtle film grain, quiet road movie atmosphere
The car sits on the road's center line with a dust cloud beside it, which looks like movement, not parking. "Expansive negative space" is honored: roughly the upper half of the frame is empty sky. The license plate is an unreadable smear, and the shrubs along the road have an orange-red fringe that looks more like a color artifact than desert light. If a parked car matters, this prompt would need to be reworded; I have not tested a reworded version.
Text the model invents
The clearest failure in the set is the diner.
Cinematic photo of a nearly empty roadside diner at midnight, one person seated in a red vinyl booth, neon sign glowing through the window, rain streaks on the glass, teal and red color contrast, 40mm lens, soft haze, realistic reflections, moody film still, detailed but not crowded
The mood, colors, rain streaks and booths are convincing, but the neon sign reads "SIUNCE SEIIOS", the menu board at left is scribble, and the sign's reflection on the tabletop is mirrored garble. The prompt asked for "a neon sign" without saying what it should say, and the model filled the sign with letter-like shapes. The person is a small dark shape behind a booth back, so the prompt's "one person seated" is only barely met.
The same pattern appears in the candlelit portrait.
Cinematic candlelit portrait of an older man seated at a wooden desk in a dim room, handwritten papers and an old map nearby, warm candle glow on one side of the face, deep shadows, 85mm lens, painterly but realistic photography, fine skin detail, dark brown and amber palette, quiet historical drama mood
The face is the strongest part: forehead lines, downcast eyes and candle-side lighting work together, and the hand holding the pen and the flat hand beside it show five distinct fingers each. The handwriting on the paper is pseudo-script, and the "map" is a dense web pattern with no legible labels. The pen and the act of writing were not in the prompt. Across the diner, the car plate and this paper, the pattern is consistent in this set: if an object in the scene carries writing, expect unreadable writing unless you can live with it or fix it afterward.
Extra hands and a painted look
The backstage prompt asked for a guitarist tuning before a show.
Cinematic backstage portrait of a guitarist tuning an electric guitar before a show, warm tungsten bulbs around a mirror, cables and equipment cases in the background, shallow depth of field, 50mm lens, handheld documentary feel, soft grain, realistic skin tones, focused expression, intimate music venue atmosphere
The mirror bulbs, coiled cables and flight case are all there. Two problems are visible: a second pair of hands and forearms crosses the lower foreground over a keyboard, which the prompt never asked for, and the face and hair have a slightly outlined, illustrated look compared with the umbrella portrait. The fretting hand is small in frame and partly hidden behind the guitar, so its finger structure cannot be checked.
All ten at a glance
| # | Prompt subject | What the image shows | |---|---|---| | 1 | Rainy street portrait | Clean match, readable hand, no text | | 2 | Desert road | Car in motion with dust, garbled plate, large empty sky | | 3 | Kitchen table | Figure almost entirely in silhouette against curtains; cup on a saucer; small garbled marks on the curtain | | 4 | Neon diner | Good mood; sign, menu and reflection are garbled text | | 5 | Perfume bottle | Dark bottle with glossy cap and smoke, bright light at left; the surface carries script-like scribbles | | 6 | Foggy forest | Lone dark-coated figure seen from behind in fog, centered | | 7 | Backstage | Props right; extra hands and painted look | | 8 | Train platform | Color right; silhouettes, not "apart" | | 9 | Rooftop | Profile of a person leaning on a ledge, lit window towers behind | | 10 | Candlelit portrait | Strong face and hands; pseudo-writing and pseudo-map |
For the kitchen scene, the prompt's "soft window light" shows up as backlight through the curtain, but it turns the person into a silhouette, which is a different picture from the "intimate framing" the prompt describes. The perfume result is mood-first: it reads as a dark object in smoke, so if packaging detail matters, this output would not serve.
The eleventh image is not a negative-prompt test
The last generation in the set used the original article's negative-prompt list as a normal prompt:
blurry face, extra fingers, distorted hands, unreadable text, watermark, logo, overexposed highlights, plastic skin, oversaturated colors, cluttered composition, duplicate subjectIt produced a blurred face behind two open palms with reddish creases. That is what the model drew when it was told to draw those words. It says nothing about whether a negative-prompt field improves results, so I did not use it here. Whether your workflow exposes a negative-prompt field at all is a question for your own tool.
Adapting what this set shows
The following follows from the images above, and is a suggestion, not a tested rule.
- Write one scene with one light source. The best images here (umbrella, candlelit face) each have one dominant light.
- Put "no text, no logo" on prompts whose scene contains signs, labels, or paper. The one prompt that used it with no other text-carrying objects came out clean; that is one sample.
- Do not describe a prop you need to read (sign, map, letter). Plan to add it in an editor.
- Choose the output size before worrying about lens words.
- If you want a moving subject to be still, say so in plain terms and check the result; "parked" was not enough for the car here.
To try any of these, paste them into the [image generator](/create/en), or rework a draft in the [Prompt Lab](/promptlab/en).
FAQ
Are the lens numbers doing anything? This set cannot show it, since no prompt was rerun with a different lens. Treat 35/50/85mm as shot-description language, not a verified control.
Can I use these images as final art? Only after checking every region for invented text and extra hands, as in the diner and backstage frames.
> What this article is based on. Model: Z-Image (recorded as `zimage`). Eleven images generated on 2026-06-03, all 1440x2560. I opened and inspected all eleven (six are shown). Only prompts and the model name were recorded; sampler, steps, seed and size settings are not known.