Image-2 Prompt Generator

The problem

Short Chinese scene descriptions often produce inconsistent image prompts: missing lighting, weak composition, or generic mood terms. image-prompt-skill targets that gap by turning everyday descriptions such as “a rainy café window” or “running in a park at night” into structured English prompts suitable for GPT Image-2. It is not an image-generation API and does not call external services; it focuses on prompt engineering: adding concrete visual details, selecting style vocabulary, controlling length, and producing comparable variants.

How it works

The skill decomposes the input along several dimensions:
- Light and time: maps morning, overcast, rainy night, and timeless dream states to terms like golden hour, diffused overcast light, and neon lights.
- Scene and composition: converts interiors, streets, landscapes, portraits, food, and fantasy spaces into cues such as cozy interior, street photography, flat lay, and surreal composition.
- Mood and style: detects calm, melancholic, romantic, dreamcore, or magical tones, then applies presets such as photorealism, film, cinematic, illustration, minimal Chinese aesthetic, and dream/fantasy.
The output usually includes 2-3 style variants, a Negative Prompt, and an --ar aspect-ratio suggestion, such as 9:16 for portraits, 16:9 for desktop wallpapers, and 21:9 for widescreen cinematic scenes.

Limits

It is useful when you want to quickly turn Chinese scene descriptions into English image prompts, especially when comparing styles, negative prompts, and aspect ratios. It does not generate images or call APIs; precise real people, brand assets, or exact text layout need additional constraints. For very abstract requests, the skill will infer reasonable details while staying close to the original intent and avoiding unmentioned people or objects.