使用OpenClaw和Gemini 3 Flash构建更复杂模型的实验

I've been experimenting with OpenClaw and Gemini 3 Flash to build harder things. For a long time I've wanted a model to write its own prompts and make good images that interest me, prompts that aren't keyword slop. I had some agents work on this problem, including making a vector search of 500+ prompts I've previously shared, to use as inspiration.