Qwen3 VL large language model invocation, supports video reverse reasoning and image reverse reasoning