AI models don’t really have a live-updated “database” of celebrities. It’s more about what they were trained on, so coverage can vary quite a bit.

If you want more control, you usually need to:

• fine-tune a model
• or use external data (RAG / embeddings)

For prompt-based workflows, I’ve been using a small tool to organize and test prompts, which helps get more consistent results. It’s here if you’re interested: https://promptgenerators.cc/

Not perfect, but it makes experimenting a lot easier.