Spent the morning in Rooster's Good News Playground. This is an admin tool I built for generating a good news story plus illustration for the daily newsletter.
I pick a candidate story and the LLM does a neutral extraction, normalizing the copy. Then it generates an image prompt. From there, an editorial illustration and a Southern-voice rewrite render side-by-side, so I can compare providers.
Currently testing OpenAI, Google Gemini ("Nano Banana"), and Hugging Face. Some results are great. Others, not so much. I probably need to tailor my prompt approach to each model.
Today's story: a woman who's spent 14 years collecting donated bicycles and tricycles, then giving them free to families who can't afford them. Pretty cool!