Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

One step costs one minute? #167

Open
conanwsz opened this issue Dec 20, 2024 · 4 comments
Open

One step costs one minute? #167

conanwsz opened this issue Dec 20, 2024 · 4 comments

Comments

@conanwsz
Copy link

I followed the quick start guide and ran it on my laptop (equipped with an RTX 4070 8G graphics card) via WSL Ubuntu 22.04 and the Gradio demo.
Each step takes approximately one minute. Is this normal or have I made some error?

screenshot_2024-12-20_17-40-26
screenshot_2024-12-20_17-40-06

@able2608
Copy link

Can you provide us with the generation details like the input and output size, and how many images are used for conditioning and the size of them? These info are required for us to diagnose your problem as they influence the generation time a lot.

@conanwsz
Copy link
Author

Can you provide us with the generation details like the input and output size, and how many images are used for conditioning and the size of them? These info are required for us to diagnose your problem as they influence the generation time a lot.

屏幕截图_20-12-2024_223658_127 0 0 1 I tried the second demo and set both height and width to 512.

@able2608
Copy link

Since you have ticked the use_input_image_size_as_output option, the set output size is ignored and the original reference image size 2000×3000 is used. 8G of VRAM is quite low in OmniGen's standard, and I believe that extensive offloading or system RAM fallback happens in the background when you are generating, hence the miserable speed. You might want to first downsize the image to a reasonable size (like 1024×680) and keep that option on, or untick the option and set the output size lower. To generate at 2000×3000 you might need something beefier.

@conanwsz
Copy link
Author

Since you have ticked the use_input_image_size_as_output option, the set output size is ignored and the original reference image size 2000×3000 is used. 8G of VRAM is quite low in OmniGen's standard, and I believe that extensive offloading or system RAM fallback happens in the background when you are generating, hence the miserable speed. You might want to first downsize the image to a reasonable size (like 1024×680) and keep that option on, or untick the option and set the output size lower. To generate at 2000×3000 you might need something beefier.

Setting both separate_cfg_infer and offload_model to true will significantly enhance the speed. Thanks.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants