lucchetto / superimage

Sharpen your low-resolution pictures with the power of AI upscaling

SuperImage

Sharpen your low-resolution pictures with the power of AI upscaling

SuperImage is a neural network based image upscaling application for Android built with the MNN deep learning framework and Real-ESRGAN.

The input image is processed in tiles on the device GPU, using a pre-trained Real-ESRGAN model. The tiles are then merged into the final high-resolution image. This application requires Vulkan or OpenCL support and Android 7 or above

Or get the latest APK from the Releases Section.

🖼 Samples

📊 Benchmarks

Results on Qualcomm Snapdragon 855 (Vulkan)

Mode	Input resolution	Output resolution	Execution time
4x (generic)	1920x1080	3840x2160	3 minutes
16x (generic)	1920x1080	7680x4320	11 minutes
16x (drawing)	1920x1080	7680x4320	3 mins 42 seconds

📱 Screenshots

      

💬 Community

You can join the Telegram group for support, discussions about AI image processing, and off-topic stuff

協 Contribute

You can submit feedbacks or bug reports by opening an issue. Pull requests are welcome !

📚 TODO

• Support images with transparency
• Batch processing
• Web and desktop versions

📝 Credits

• Pre-trained models and original implementation from Real-ESRGAN
• Pictures by Satoshi Hirayama, Skitterphoto, 天江ひなた and Ryutaro Tsukata

⚖️ License

SuperImage is licensed under the GNU General Public License v3.0

superimage's Issues

Downscaling after processing (to original/fraction size)

It'd be beneficial (in many cases) to have option to downscale the image as well

(Since the upscaling removes noise, improves quality, but many times at resolution that's not needed (to store, share)

Notification info request

Is it possible to add to the notification the remaining time of the upscale process? That would be very handy.

Feature Request - Ability to colorize

Windows app?

This app is sooo great. Any chance to having Windows version?

Rename the 4x and 8x scale factors to 2x and 4x, respectively

Usually scale factors describe the scale of only one side of the image, not the whole image. Also, even if you want to describe the size of the whole image, it should be 16x, not 8x

Publish on F-Droid

Since this app is FOSS, can it be published on F-Droid?

App crashes while upscaling

The app crashes while upscaling. I tried to reinstall it from different sources i.e.play store, f-droid, GitHub but the issue persists.

Device Info -
Device: RED8BEL1
Model: realme RMX3561
Android: 13

Logs - logcat_03-01-2023_16-35-57.txt

Screenshots of Logs -

does not work on moto e7 (64gb 4Ram) - mediatek g25

when you press to make the image it starts to load and takes a while, when "0" appears the application closes

Add ultra sharp model

Yes add ultra sharp model, it's on stable Diffusion but it also works using Vulcan, Do it

Ultrasharp is the best in many terms

Feature Request: Please support Face Restoration.

Please support Face Restoration, it will be more useful.
https://github.com/TencentARC/GFPGAN

App font issue

Would it be possible to use a more "universal" font for the app? Because the currently used one has an issue in the Greek translation that's been recently added. The Greek letter "μ" appears like of another font. Check below.

Remove the extra space before the exclamation mark in the "upscaled successfully!" message

Batch processing

Would be nice if the user could select multiple images for batch processing (the model type should probably stay the same for all the images)

Edit: there's batch processing in the TODO, I'm blind 🤦‍♀️

Romanian translation

Hello, i want to translate this app in romanian. (If possible)

The images have a white background

I am updating png images with transparent backgrounds, specifically anime images, but when you get them you have a white and non-transparent background

error when building with cmake

[CXX1405] error when building with cmake using E:\SuperImage-master\realesrgan\core\CMakeLists.txt: -- Android: Targeting API '24' with architecture 'arm64', ABI 'arm64-v8a', and processor 'aarch64'
-- Android: Selected unified Clang toolchain

Optional: Can I know minimal which file is working for generating upscaled images from the mnn model? or for simple integration to new project which files i need?

Need to know about upscaling file

Can I know minimal which file is working for generating upscaled images from the mnn model? or for simple integration for upscaling image with mnn to new project , which files i need?

Support for loading custom models

Would be nice if the user could load a custom upscaling/enhancing model, like one from here https://upscale.wiki/wiki/Model_Database

Not sure if all of these are supported by MNN, however

Upscaled image orientation bug

When upscaling portrait photos, the upscaled image is turned counterclockwise to landscape. The orientation data of the original image should be preserved.

Still old version in fdroid

According to aurora Droid theirs a 1.3.5 version out but this has not been uploaded to fdroid.

Can't find resource for bundle java.util.PropertyResourceBundIe, kеу hour.few

When I try to launch in any form, I get an error

windows 11, amd r5 5600g, 64x

Stored app data issue

I've noticed that, everytime an image is opened in the app, the app's size is increased by the data size of the image. Thus, the app's size is increased more and more by time! The bad thing is that these data are stored in the app's data, not its cache. So:

1. Is it necessary that these data are stored? I think after an upscale is over or cancelled, the data should be deleted after exiting the app.
2. If the data are necessary, they should be stored in the app's cache.

P.S. Can't wait for the desktop (Linux) version! 😄

Upscaling adds background to the image

I tried upscaling the image (IMAGE1, PNG), with all of the upscaling options (x4, x16) but it returns the upscaled image (IMAGE2) with a black background. IMAGE1 has transparent background. Also, tried with Real-ESRGAN vulcan (model realesr-animevideov3) which returns the image (IMAGE3) without the background. Is this expected or a bug?

IMAGE1

IMAGE2

IMAGE3

x86 x64 processor support

please add support not only for mobile processors, but also for desktop x86 x64

multithreaded computing

now the program uses only one core of my processor on the phone, so the calculations are very slow. multithreaded computing will be able to improve the speed. please add the parameter how many cores to use in the processor.

Will there be new OpenSource versions in Github? Or only Play Store versions from now on?

Hi, I noticed that Play Store version has been updated to 1.5.0 meanwhile.

Here in github it's still 1.3.4. New versions available one day or not?

What's the idea? Maybe a one-time-purchase in Play Store without ads and tracking?

Thanks.

2.0.0 Release on GitHub

In Google Play is a 2.0.0 release oft this app but not here on GitHub. Neither the source of the gplay 2.0.0 as it seems.

• Is a OSS release of 2.0.0 planed?
• Is the 1.x.y serie still maintained?

eigen directory 404

eigen directory 404，Does anyone know the new address 。

Add monochrome icon

Please add the themed icon.

Feature request: Tensor chip support

While GPU acceleration is nice, wouldn't so-called "AI" hardware, like Google's Tensor Chip, be even faster on a mobile device?

Internet permission

I see in https://f-droid.org/en/packages/com.zhenxiang.superimage/ that version 1.3.4 added permissions:

• have full network access
• view network connections
• view Wi-Fi connections

What does it need these permissions for?

Desktop version out?

I've been looking recently for the desktop version, and I came across a website that does have it, but I'm not sure if it is the official one (seeing that there's no package of it in the repository here)

https://superimage.io/

That's all. App works good on phones, one pf the best utility apps I got installed. Keep up the good work!

What is the license in case I want to use only the image-scaling "engine", and how do I do it?

Hello,

I've searched for a nice library that uses neural-network to enhance/enlarge images of low resolution for quite some time, to be able to use it on an Android app.
I've only found one of Google, but it was very restricted and was more of a POC than a library:
https://ai.googleblog.com/2021/07/high-fidelity-image-generation-using.html?m=1
https://youtu.be/WCAF3PNEc_c
https://github.com/Janspiry/Image-Super-Resolution-via-Iterative-Refinement

Is it ok that I use what's available here to do it?
What is the license of just using the "engine" that does it? Is it ok that I make a closed sourced app that uses it without any issue? I don't intend to change the code of the original "engine" except for maybe if I'm forced to, to adjust it to make it work, if that's an issue.
And, how can I do it? Which files/folders should I check out?

Parallel image processing

Can you add an option to upscale multiple images at once. Because i think my device is capble of handling multiple heavy processes.

Also if you can add option to select multiple images once and then process them in serial manner.

I appreciate your work in this project.

Windows 11 – Consistent Crashing Upon 'Upscale'

%Appreciation for an incredible app on Android. Having some difficulties with the Windows version, though. I was wondering if someone could be of assistance, or maybe this has to do with my hardware and is not known yet.

Every time I try to upscale an image the APP CRASHES on Windows 11. Sometimes 16x (generic) will work but rarely does 16x (drawing).

Update: 4x (generic) is not working either. I actually can barely get the app to function at all on Windows 11.

---

System Information

  Time of this report: 10/13/2023, 09:36:04
         Machine name: CB0C2BCFDAAB9C9
           Machine Id: {E52D7407-3212-4294-8B31-1ADD11E3EDD6}
     Operating System: Windows 11 Home 64-bit (10.0, Build 22621) (22621.ni_release.220506-1250)
             Language: English (Regional Setting: English)
  System Manufacturer: ASUSTeK COMPUTER INC.
         System Model: ROG Zephyrus G14 GA402RJ_GA402RJ
                 BIOS: GA402RJ.319 (type: UEFI)
            Processor: AMD Ryzen 9 6900HS with Radeon Graphics         (16 CPUs), ~3.3GHz
               Memory: 16384MB RAM
  Available OS Memory: 15596MB RAM
            Page File: 16656MB used, 5339MB available
          Windows Dir: C:\WINDOWS
      DirectX Version: DirectX 12

Batch processing

I've just noticed that in your todo's list you've added "Batch processing". Since the app's process is really heavy enough for the vast majority of mobile phones, batch processing would probably takes ages to be completed, drain the battery, and most likely freeze the phone. I think this is a feature suitable only for the upcoming desktop version. Better not include it in the mobile version. It'd be only extra, useless code, that would increase the app's size.

more settings to improve

there are no settings for changing the image enhancement right now. the default settings are used. Because of this, artifacts appear in places, I am sure that they can be removed if you add the ability to change the power of image enhancement.

Material You design, icons and system font

This way the app will look more professional and native to the modern Android ecosystem

Closes #9?

RU translation

Can I translate this app to the Russian language?

The images have a white background

I am updating png images with transparent backgrounds, specifically anime images, but when you get them you have a white and non-transparent background

Basic UI

The app is great, but how did you decide on the UI? The app is simple so it doesn't require a lot of things, but wouldn't it have been easier to conform to the Material 3 standard?

Image

Request for chinese language.

Dear developer,

Hi, i love your app superimage. If possible, could you please add chinese language?

Cmake Error while builing

CMake Error at CMakeLists.txt:43 (add_subdirectory):
The source directory

E:/Android Project/SuperImage-master/realesrgan/core/MNN

does not contain a CMakeLists.txt file.

Import image bug

Tapping the "Select image" button opens the default Android image picker. If you choose the option it has to select an image through one of the available apps on the top (see the red arrow in the attached screenshot), doesn't work. The respective app opens, but selecting an image doesn't import it.
Using a Nokia 6.2 with Android 11.

App Crashes When Upscaling and "Selected mode" is "x8 (generic)"

Using "x4 (generic)" and "x8 (drawing)" both work fine but when using "x8 (generic)" the app crashes during "Initializing...".

SuperImage: F-Droid 1.2.3
Phone: G7 ThinQ LM-G710EM
Android: 10

Feature request : x4 (drawing) model

Would it be possible to integrate a lighter and faster "realesrgan-x2plus-anime" model (which would be labeled as "x4 (drawing)" ?

Add this app to fdroid, izzyondroid

Please! It would be very cool to have this great way avaible to install from a foss app store.

add algorithm GFPGAN

I have studied the issue of image enhancement. this algorithm is the best way to deal with bad images. Can you add to this algorithm?
https://github.com/TencentARC/GFPGAN

the current algorithms that they are worse in the program, I did a comparison. if you strive, then you need to add the best algorithm to the program

improvements without increasing the image size

Please add the ability to improve the picture without increasing the size of this picture. I have large pictures after the scanner and I only need to improve them without increasing the size