What AI Cannot Do
🟥 AI Cannot DoImage Generation

AI cannot reliably generate realistic cars in Chinese county towns

Image models often mix county roads, domestic car models, license plates, and worn vehicle details into plausible but incorrect scenes.

Creator: 林启2026-06-01china, vehicles, photorealism
Challenge this
Why is this considered beyond AI?

Background

Repeated attempts to generate county-town street photos in China produced broken badges, invalid license plates, and streets that looked too urban.

Goal

Generate a set of images that local Chinese viewers judge as realistic county-town car photos, including models, plates, roads, storefronts, and wear.

Observed problems

Models drift into Western car proportions, invalid plate text, top-tier city streets, and parking layouts that do not match county-town space.

Failure analysis

The models lack a stable joint distribution for low-end domestic cars, older vehicle conditions, local roads, and Chinese plate formats.

Success criteria

At least 20 images, with over 70% of Chinese blind-test participants judging them as real county-town street photos and no obvious plate or car-structure errors.

AI tool attempt records

Midjourney · Midjourney V7

Ada Chen · 2026-06-10 · APPROVED

Model settings: photorealistic image generation

I think it failed

Prompt

Chinese county town street, realistic domestic cars, unpolished daily life, phone photography...

Tried a direct photorealistic prompt with county-town cues, then added negative prompts for luxury cars and megacity streets.

The images looked attractive but failed on license plates and car proportions.

Failed sample
Timeline

  1. 创建问题

    2026-06-01

    作者记录了图像模型在中国县城汽车细节上的稳定失败。

  2. 第一次挑战失败

    2026-06-10

    Midjourney 结果通过审核，但未满足成功标准。