Image Generation Assistant

Problem

AI image generation often breaks down not because a model cannot render, but because requirements are scattered. Different use cases need different styles, aspect ratios, and quality levels. When users provide only keywords or an article, key details such as subject, mood, and scene are missing. After a poor first image, prompt edits, tool switches, and old results can easily get tangled. This skill structures the process into confirmable steps and avoids asking users to type A/B/C.

How It Works

The core approach is structured guidance with clickable options:
- Step 0 defines the use case first: social media, blog posts, posters, avatars, etc., because use case drives style and ratio recommendations.
- Input understanding: restates the brief and asks for subject, scene, or mood when needed.
- Style matching: recommends 3–4 styles per batch, with refresh batch and custom input fallbacks.
- Prompt engineering: builds a detailed Chinese prompt covering subject, scene, lighting, color, atmosphere, and image quality.
- Tool selection: supports Tongyi Wanxiang, Wenxin Yige, Tencent Hunyuan, Jimeng, plus Nano Banana 2 and GPT Image 2, with API key setup guidance when needed.
- Ratio and quality: suggests aspect ratios and quality tiers based on use case.
- Version control: regeneration updates parameters and saves results as _v1, _v2, and later versions.

Boundaries

It fits cases where a vague brief needs to become actionable image parameters, especially for social graphics, articles, posters, and avatars. It does not guarantee brand compliance; commercial use still requires checking copyright, portrait rights, and platform policies. For article-to-image work, it is better to extract core scenes and mood than to illustrate every sentence.