AI Resonance · 阅读最新日报 · AI 入门推荐

2026-10-09 · America/Los_Angeles · 社区动态 · #20

how i stopped our video agent from confidently picking models that can't do the job

i work on the backend of an ai video agent (commercial product, not linking it, just want to talk about the routing part) we have 1000+ video models. if you just let the llm pick, it grabs ones that can't even take the user's reference image or do 9:16 haha so i filter in code first. every model that can't handle the inputs (reference images, length, aspect ratio, audio) is gone before the llm sees anything. then we score whats left (release date, description, user memory) and the llm only picks from a short list. the user still gets a draft with the model and cost before anything runs video…

how i stopped our video agent from confidently picking models that can't do the job

热度 43.9 / 100;排名与评分保留该期记录。