Reduce processed image display/downloading overhead

[smikhai1]

• I have searched to see if a similar issue already exists.

Is your feature request related to a problem? Please describe.
In the Gradio library, processed images are saved in the PNG format by default, as observed in the save_img_array_to_cache and save_pil_to_cache functions within gradio.processing_utils module. This behavior results in significantly larger file sizes compared to the JPEG format: specifically, for 4K images, the PNG files are approximately 10 times larger than their JPEG counterparts. This leads to increased download and displaying times for the front end, adding unnecessary overhead to the user experience.

Describe the solution you'd like
To enhance flexibility and efficiency, it would be beneficial to allow users to select the file format for saving images in the Gradio cache. This could be implemented by introducing an additional parameter, such as format, to the constructors of all image-related components.

As an example, in my application I manually modified the default behavior of the aforementioned functions to save images in JPEG format, which significantly improved the results' delivery time and saved extra 10 seconds

Additional context
Add any other context or screenshots about the feature request here.

[freddyaboulton]

Do you want to open a PR for this @smikhai1 ? I think we can add an img_format argument to gr.Image that controls the format the image is saved in when it's a numpy array or PIL image. cc @abidlabs

[abidlabs]

Agreed. The name of parameter could be just format since we're already working with the gr.Image component and to mirror the similar format parameter in gr.Video

[smikhai1]

@freddyaboulton yeah, sure. I will make a PR

[freddyaboulton]

We should preserve the file format used during upload if possible see #7657

[dfl]

apologies I wasn't able to run the test suite locally 🙇‍♂️
had a bunch of dependency issues on my mac

hoping maybe someone can take this and run with it?

[freddyaboulton]

Hi @dfl ! Yea someone can get it over the finish line. Thanks for the PR again.