[Request] For models without image recognition, use vision-capable models to assist with recognition #7551