AI Image Prompt
Chinese description → structured English prompt. On most models, English prompts are more consistent than Chinese ones.
Paste the generated English prompt into Midjourney / Stable Diffusion / Flux and similar models. The more specific the subject description, the better (material, action, quantity) — style and lighting set the tone of the image. The text in square brackets is a blank I left for you; replace it with your own details.
When to use it
- You can picture the image clearly — subject, material, action — but do not know the English keywords, and want a starting prompt you can paste straight into Midjourney.
- You want a fixed style-plus-lighting combination for a consistent series of images, or need the --ar parameter added for a specific placement.
When not to
- The model has to understand a complex Chinese narrative, or accurate Chinese typography inside the image: this tool does not translate, Chinese text is concatenated as-is, and most models handle Chinese poorly.
- You need Stable Diffusion negative prompts, weight syntax, LoRA or ControlNet control — this page emits a positive keyword string only.
Assembled locally: your description is never uploaded and no model is called. Limits: keyword concatenation only — no Chinese translation, no negative prompt or weights; --ar works in Midjourney only, other platforms need their own size settings; output quality depends on the model version.
Frequently asked
Can I use a Chinese prompt for image generation?
You can submit one, but results are unreliable. Midjourney's Chinese understanding is limited and often ignores or misreads it; Stable Diffusion's usual CLIP text encoder is trained mostly on English, so Chinese prompts barely register (SD3 and Flux with T5 do better, but English is still safest). That is why this tool outputs English keywords. It does not translate for you, though — whatever you type in the subject field is concatenated as-is, so write the English yourself, e.g. orange tabby cat rather than 橘猫.
Can I paste the result directly, and why does the image differ from what I pictured?
You can paste it, but it is a comma-separated keyword string, not a full model instruction: no weight syntax (::2), no negative prompt, no --stylize / --chaos / --seed, no LoRA trigger words. The same prompt produces very different images across models, versions and seeds, and Midjourney version changes are especially visible. Treat it as a starting point and add or remove keywords based on what comes back — that beats trying to write the perfect prompt in one pass.
Does --ar do anything in Stable Diffusion?
No — it is ignored or read as ordinary text. --ar 16:9 is Midjourney syntax; in Stable Diffusion WebUI you set width and height, or use the platform's aspect-ratio option. SDXL is also sensitive to ratios it was not trained on, so common ones like 1:1, 3:4 and 16:9 are safer, while extreme ratios tend to produce repeated compositions.