AI / Design Communication & Image Prompting
Image Prompt
Chinese description → structured English prompt
Turn the Chinese description below into an English image-generation prompt. Structure: subject and action → materials and environment → style → lighting → composition and camera → aspect-ratio parameter (--ar). Requirements: be specific enough that it can actually be drawn; do not use words like 'beautiful' or 'premium feel'. [Chinese description]
How to use
Lock the subject line first; if two people would draw different things from it, no amount of style wording will save the prompt.
Good for
- You only have one Chinese sentence in your head ('a convenience store entrance on a rainy night') but need usable images out of Midjourney or Stable Diffusion — splitting that sentence into subject, environment, lighting, and camera sections keeps the result from being only half right.
- One subject needs a batch of style-controlled images (a series of illustrations, one character in different scenes) — freeze the subject and environment sections, swap only style and camera, and the batch stays comparable.
Not for
- You are still at the 'show me a few directions' stage — nailing down lighting and lens parameters this early narrows the options; four to six loose direction sketches are a better spend.
- The image must contain exact text, a logo, or a layout you can ship — image models cannot spell reliably, so that work belongs in a typesetting or vector tool.
Example
Input
A convenience store entrance on a rainy night; a woman just off work stops in front of the automatic door, holding a plastic bag, water streaks on the glass
What you get
A six-section English prompt: the subject section keeps only 'woman, stopping, looking up'; environment covers wet asphalt reflections and cold white fluorescent light inside; style locks Japanese realistic photography with light grain. Lighting is overhead fluorescent as key with sodium streetlight as rim, and the composition is a 35mm medium shot from slightly below, subject at the right third, with --ar 3:4 plus a negative prompt suppressing garbled signage.
Common mistakes
- Stacking only style words (cinematic, 8k, masterpiece, trending on artstation) while the subject gets one clause — the mood lands, but the picture is not the thing you asked for.
- Putting 'premium feel' or 'good atmosphere' straight into the prompt — the model can only return the average taste of its training data. Translate the feeling into physical conditions instead: low saturation, a single light source, generous negative space.
Try tweaking it like this
- Replace the fixed style words with 'reference the visual language of a specific photographer or film and name its concrete features' — one subject then yields one consistent visual system instead of a different flavour each time.
- Append 'output only the prompt, no explanation' — the reply collapses into one pasteable line, which matters when you generate in batches.