A general question #388

New Issue

claunia · 2026-01-29T21:47:27Z

claunia commented

2026-01-29 21:47:27 +00:00

Originally created by @tjasmin111 on GitHub (Aug 28, 2023).

Interesting repo. I like to know if these GAN can be used as a real-time filter in an app like what those used in Snapchat? Would it be fast enough on a good phone?

Originally created by @tjasmin111 on GitHub (Aug 28, 2023). Interesting repo. I like to know if these GAN can be used as a real-time filter in an app like what those used in Snapchat? Would it be fast enough on a good phone?

claunia commented

2026-01-29 21:47:27 +00:00

@CrazyWolf13 commented on GitHub (Sep 20, 2023):

I'm not a developer, but if you think, for a camera to be real-time; image has to be displayed nearly instant from what the camera "sees".

Gfpgan has to analyze a photo search for faces and then enhance them, which can sometimes take over 30 seconds to a minute depending on the settings, on a gaming computer with decent gpu and cpu. So i think there wouldn't be any change to get realtime, and surely not a phone, not even speaking of snapchat integration.

I guess someone would either need to rewrite the code to be lightweight enough for a phone, however next problem, how would you get all those dependencies on the phone if they are probably not even available or we need to wait for quantum computers to be built into smartphones lmao 😂

@CrazyWolf13 commented on GitHub (Sep 20, 2023): I'm not a developer, but if you think, for a camera to be real-time; image has to be displayed nearly instant from what the camera "sees". Gfpgan has to analyze a photo search for faces and then enhance them, which can sometimes take over 30 seconds to a minute depending on the settings, on a gaming computer with decent gpu and cpu. So i think there wouldn't be any change to get realtime, and surely not a phone, not even speaking of snapchat integration. I guess someone would either need to rewrite the code to be lightweight enough for a phone, however next problem, how would you get all those dependencies on the phone if they are probably not even available or we need to wait for quantum computers to be built into smartphones lmao 😂

claunia commented

2026-01-29 21:47:27 +00:00

@tjasmin111 commented on GitHub (Sep 20, 2023):

Got it. But how does snap and insa filters work? They achieve such filters in real-time nicely

@tjasmin111 commented on GitHub (Sep 20, 2023): Got it. But how does snap and insa filters work? They achieve such filters in real-time nicely

claunia commented

2026-01-29 21:47:28 +00:00

@CrazyWolf13 commented on GitHub (Sep 20, 2023):

Got it. But how does snap and insa filters work? They achieve such filters in real-time nicely

Well if you think of snapchat filters, they are built for android unlike gfpgan, and they are much less cpu and ram hungry than gfpgan.

Overall I guess it's just because they are built for such hardware, optimized for this hardware and software.

And I interpret it's like either you get speed in terms of faster processing (real-time at best) but you most surely won't get such a good result as if you would run a desktop tower pc with a dedicated graphics card for some minutes or you get the opposite, or the oppsite: slow but highly detailed

@CrazyWolf13 commented on GitHub (Sep 20, 2023): > Got it. But how does snap and insa filters work? They achieve such filters in real-time nicely Well if you think of snapchat filters, they are built for android unlike gfpgan, and they are much less cpu and ram hungry than gfpgan. Overall I guess it's just because they are built for such hardware, optimized for this hardware and software. And I interpret it's like either you get speed in terms of faster processing (real-time at best) but you most surely won't get such a good result as if you would run a desktop tower pc with a dedicated graphics card for some minutes or you get the opposite, or the oppsite: slow but highly detailed

claunia referenced this issue

2026-01-29 21:49:04 +00:00

[PR #388] Mr Anil bhai #608

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: TencentARC/GFPGAN#388