Z-Image Chinese Text Test: 30-Prompt Accuracy Protocol
A reproducible 30-prompt protocol for measuring Chinese text accuracy without inventing scores: prompt set, scoring rules, review fields, and publication thresholds.
By Z-Image Photo Editorial Team
This article defines a 30-prompt Chinese text suite and its scoring method. We intentionally do not publish a fabricated “accuracy percentage” before all outputs have been transcribed and independently checked. The public bilingual example above is a completed Z-Image run; the wider scorecard remains a reproducible protocol.
The 30-prompt matrix
Use six groups of five prompts: 2–4 character headlines, 5–8 character headlines, mixed Chinese and digits, Chinese plus English, two-line layouts, and punctuation-heavy copy. Keep wording, seed policy, resolution, and review conditions fixed.
Score character by character
- Exact character accuracy: correct characters divided by requested characters.
- Sequence accuracy: whether the full requested string appears in order.
- Extra-text rate: invented characters or words not requested.
- Layout pass: line breaks and hierarchy match the brief.
Prompt template
Editorial poster with the exact Chinese headline “城市灵感”, one line only, large centered sans-serif lettering, no other words, high contrast, clean background.
For production work, add verified copy after generation. See the Chinese and English text workflow.
Sources and test record
The cover image was generated with Z-Image using seed 930203. AI-generated images and lettering can contain errors; inspect outputs before publishing.
Show the cover prompt
Minimal black cobalt and violet design poster with the exact Chinese headline and a short English subheading, crisp grid typography, no extra words.
Try these ideas in the generator
Open the AI image generator and test the prompts, styles, and text-to-image workflows from this article.
Related Articles
Keep exploring more prompt guides, model comparisons, and AI image generation tutorials.
50 Z-Image Turbo Prompts Tested: What Actually Works
Patterns from 50 documented Z-Image generations across portraits, products, architecture, fantasy, illustration, nature, and text-bearing scenes.
Best Z-Image Prompt Structure: Subject, Light, Camera and Composition
A practical Z-Image prompt order that turns vague style requests into visible direction for subjects, lighting, camera, composition, materials, and constraints.
Common Z-Image Prompt Mistakes and How to Fix Them
Fix vague quality words, conflicting lights, overloaded scenes, ambiguous counts, plastic skin, impossible reflections, and unreliable generated text.