Memory usage problem

[CCJetWing]

When I use the gradio_generate webui, my computer can run smoothly, generating a picture in 10 seconds. However when I use API to run the process, it would take up large amount of memories, slowing down the process to nearly an hour, and sometimes causes the 'CUDA out of memory' error.
How can I fix the problem?

[Vimukthi Randika Senarath Yapa]

What is the GPU used for the experiments?

[Weifeng Chen]

which API？

[CCJetWing]

I am using single RTX3090 GPU, and the /process API of gradio_generate.py

[CCJetWing]

which API？

Still no clues why I got this problem.