Hello friend, the new version has a surprising performance, if you allow me to suggest next time you could add the option to remove audio from the video. For example, I remove the audio to make the conversion faster and then I synchronize it in a video editor.
Great application. it got a great potential of growing up as its going the right direction. The UI may need some tweaks like a progress bar or something. But It doesnt really matter as much as the functionality is perfect. Keep it up!
Hey just wanted let you know I am loving the update on your app :) especially with the explanation of each AI model. Thanks also for fixing the transparency issues with Windows 11.
By the way, can we no longer drag files into the program?
Hi friend, did you remove the BSRGAN x2 model or is something wrong on my end? The 2x model produced much cleaner images than the 4x for me and had less tiling issues I guess due to needing less VRAM. But now all I see are the 4x for everything.
I see, that's unfortunate but thanks for clarifying. Is there any way to download an older version with the 2x model until it is released? And the new architecture sounds cool, I just hope there's a 2x version in the future because it's necessary for smaller images where 4x looks unrealistic. Thank you!
Sadly i do not have older version prepared, and the old AI library is no more "downloadable" :(
Unfortunately, researchers on AI of this type only publish the x4 version of the trained models, because that tends to be the one that returns the best results.
hi, i am new to a lot of this and i am quite confused on how to use your scaler. I uploaded my image to it and clicked the upscale button. I waited and now it said 'all files completed'. However, where can I access the upscaled image? I don't know where to find the enhanced image anywhere.
Great tool, love it! I am already thinking about getting RealScaler and FluidFrames as well.
I seem to have an issue with the program window where it disapears into the background when I click anywhere but on the buttons or dropdowns. Even resizing / moving the window does not work. The window behind the tool gets called into the foreground.
This seems to happen while the program is processing a video.
Not a big bug but would be nice to be able to move the window during longer processing times.
Are you on Windows 11? because i noted that the "transparent effect" used for Windows 11 makes the window unclickable and not draggable, is that the case?
I cannot test this because on my machine i use Windows 10 where this transparent effect can't be applied, so any feedback is absolutely welcome :D
I came to a little problem with the scaler... If the original was taken by phone, and not "widescreen" but the "standing video", the upscaled video will be transformed to a ratio as if it was taked wide screen... Resulting in turning the people into small, fat dwarves. Can be funny, but not the result it should be. :)
Your tool here is a great tool ;) I'm using it a lot! Two-three things, I may suggest: - for the input resolution widget: I saw that you got a lot of questions about it in the comments below. An easy way to solve this would be to display the final resolution somewhere (maybe just under the input video on the top of the application) and that way we will know what resolution we will have afterward; - you put here and on your github some examples for the BSRGANx4. Have you got anywhere some examples for the other AI Model? It's a bit hard to compare, for instance, the quality of each and which model is best for what situation. ie. is RealSR_jpgx4 better for real photos or manga pictures, etc.? - is it possible to have some bench to know which AI model is the fastest (or to have clues about them)? For instance, for the exact same input video, and the exact same output resolution (ie. 480 to 1920), what would be the whole process duration for each AI model?
1. i am already thinking how to implement your suggestion
2. absolutely, as soon as possible I will upload an example with all AI on a single image
3. all implemented AIs use the same architecture, so in terms of speed they are identical. But I have in roadmap to implement other architectures both faster (but with lower quality) and slower (but with higher quality).
Really cool, thank you for your work. Some suggestions:
Sometimes I'll be running a batch of images and it will randomly error when a background process uses too much GPU or something like that, and then I lose my place. What do you think of adding a "resume" button, or even better, a way to have it retry on its own so I don't have to watch it?
Also, it doesn't clean up the resized files of finished images after it errors. Maybe it will be better to remove the resize files after each image is done, rather than all at the end, to keep disk space free.
Also, regarding the stitching, I think a more foolproof solution could be to include a small amount of "bleed" around the edges of each slice, and then to cut out the bleed when merging them back together. But, I don't know how this tech works or if that's even possible. Just pitching my idea. :) Thanks again.
maibe you clicked on Upscale button two times? because it seems that 2 upscale process are conflicting . Try to close the app and check on task manager if it is completely close, the retry :)
This happens when the input image to AI Is too big for the selected value of VRAM limiter .
To resolve this issue you can use the Input Resolution widget, so you can downscale the image before passing to AI.
Another option is to modify Vram limiter option (default is 8 gb), if you have a GPU with more VRAM you can set the correct Number. For example if you have a GPU with 24Gb, you can enter 24 :)
it always leaves a mark on the slices, I've tested it by lowering the input resolution by several %, my GPU has 2GB of dedicated memory and CPU 4, and I put these values in their proper fields, the resolution of my image is 1568x896, what do I do ?
Qualityscaler there needs to be a trial version where it is full version for 14 days or you gt process 30 to 40 images and videos then it does not work any more. If the program is up to my liking I would be willing to pay for it after I am able to test it out.
Thanks for the reply, is there a possibility to get a resize option (ex: 1920 x 1080) for images added to a future update if possible or if there is an avenue I can take to achieve this?
it is possible to do what you ask thanks to the "Input Resolution" widget.
For example:
if you have an image 960x540 (which is half of 1920x1080)
you can select Input resolution 100% (that is, the AI will use the whole 960x540 image and feed it to the AI); selecting BSRGANx2 (which enlarges the image by a factor of 2), you will have: 960x540 -> input resolution 100% = 960x540 -> BSRGANx2 = 1920x1080
or you can select Input resolution 50% (i.e., the AI will use half the 480x270 resolution and throw it to the AI); selecting BSRGANx4 (which magnifies the image by a factor of 4), you will have: 960x540 -> input resolution 100% = 480x270 -> BSRGANx4 = 1920x1080
It all depends on the original resolution of the image.
I am also thinking in the future to simplify these steps or give visual feedback of the final resolution based on the choices set :)
the "Option" button is just a header for the option list on the left bar, it will removed in next version :)
File output location is the same of the input file, for example if you load a file on Desktop, the upscaled file will be saved on Desktop but i will improve this part, thank you for the feedback :)
Hi Jangy, I saw there is a folder "AI" to store ".pth model" so I tried to import a new one (found on https://upscale.wiki/wiki/Model_Database) but unfortunately it was not enough for it to appear in the interface. Is there a way for me to import new model or not at all ?
Unfortunately it's not that simple, each .pth needs all the architecture in the app code, but if you want to try you can change the name of the "newmodel.pth" to "BSRGANx4" and see if it works. At 99% Will not work :)
Well, you've got conventional upscaling algos like Lanczos and Bicubic. There's some talk about how better results at upscaling could be produced by first upscaling 4x thru ESRGAN or BSRGAN and then downscaling 2x thru Lanczos, see: https://upscale.wiki/wiki/Model_Database
Not a video researcher so i'm not sure how if there's a possibility you could experiment with such a technique for future releases of QS
Oh def. would impose a greater processing time, I imagine you'd have to do even more with optimizations. But it would make your app very unique: I dont know of anyone else doing it.
any way to view location of upscale frames? just kind of says its working, would be nice to see where its at exactly. where to view temp files for the extracted frames? i dont want my main drive being filled up if it just goes by where the exe is installed. doing some testing on my 6900xt. also is there a way to enable multiGPU? i have a 1650 running as well for other stuff, would be nice to combine them.
Hi my friend, you can see extracted frames and generated frames in a directory called QualityScaler_temp in the installation folder of the app, at the end of the process the app Will clean automatically temp files :)
Just purchased and the app doesn't display the upscale button and won't let me adjust the window size. I adjusted my computer display setting so I could view the entire window and... the button at the bottom is indeed missing. Any help?
what resolution is your screen? Sadly the GUI used does not works properly on fullscreen, so i disabled it; and the app is made to worki with screens > 1080p.
However you can try to hidden the Windows taskbar, this can help.
I will improve this part of the app in next releases
Hello! What I could try up to now, it is a very nice software, thank you. But can it be that even just short movies take ages? I have Pentium 13900 and Nvidia 4090, so I thought letting it do its job the whole night will finish a 9 Minute video. Thank you in advance.
Hello and thank you for the fast answer. Yes, I did as you said.
AI backend I have only one, the 4090 to chose from.
With the speed now it will still take at least one, I think two days to finish.
I maybe still use something wrong, because as you say, it should run a bit faster in my machine. Also I have 64 GB RAM, if this could be a brake of the process.
What is the resolution of the video you are processing?
If you open the Windows Task Manager can you check that your gpu is being used?
In my testing lately I have found that Windows 11 has slowdowns in reading and writing files, and as you can imagine this impacts QualityScaler's speed a lot. Do you have windows 10 or 11?
Absolutely super manageable resolution from your 4090; at this point I really think it's Windows 11 which is not yet fully optimized.
If you can try there is ReviOS, an optimized version of Windows 10 or 11. free of Defender, Cortana and that kind of bloatware, updated ISO every month with patches :)
Hello to both of you!
I love this tool. Especially upscaling very old pictures or short recordings. Unfortunately I experience similar habits upscaling mp4 vids. Using multiple computer setups (win10) while Version 9 maxed out CPU and GPU the newest Version 10 takes up 14 perc max.
Version 9.0 had a bug that allowed excessive cpu usage.
I'm working on improving QualityScaler more and more on the speed issue, but basically I can't do much for now because the 3 AIs of QualityScaler offer very good quality but have a very very complex and slow architecture.
For long videos I would recommend RealScaler, another upscale app of mine that uses Real-ESRGAN as the AI. In particular "RealESR_generalv3_x4" offers much faster speed (about 4 times faster than BSRGAN) at the expense of lower upscale quality.
Hi, unfortunately, I can't get the software to work properly; the Options button does not open any new menu, and whenever I try to upscale a file I get the error message in the screenshot below. I get similar error messages with all the selectable models. Could you please assist? Thank you!
Hi my friend, the Option button does not open anything, is just a "title" of the option under :)
That error means that the image does not fit in your VRAM limiter option.
You have to select the correct number of VRAM limiter; for example if you have a gpu with 4GB of VRAM you have to select 4, if you have gpu with 6GB of VRAM you have to select 6, and so on.
What GPU do you have?
UPDATE. I saw not from the screen that you have a integrated Intel Xe, in most case for integrated graphics 2GB Vram limiter is the perfect option :)
Hello. Version 4 is the last version where embedding an image or video works for me. Even in versions 7 and 8 my image does not appear, only the clean button appears. Version 4 works without problems, just to be sure I reinstalled both DirectX and Visual-C Runetimes as you have in the instructions. I tried running QualityScaler.exe as administrator, but it doesn't let me insert images at all then. All my attempts are without any result. I have Windows 10 and the HW matches the requirements. Can you please advise me what I am doing wrong? Thank you!
There is a bug with running as Administrator, so do not use it :)
Can you tell me more informations about your files? For exampe, extension, path etc :)
Normaly when drop a single image or a single video the app create inside its directory a file called temp.png, is a temporal file to display the image on the app. Can you find this file?
Thanks, I've moved the application directory from Program Files to Desktop, and it's working now. Apparently the program couldn't write to disk, temp.png wasn't being created. It works great now. Thanks.
Hello. I can't buy the program with Paypal or a Bank card. Please sell for cryptocurrency. I am ready to transfer money to your wallet. Please help, I like your program.
All the different number selections in the UI are a bit confusing for me.
I don't really understand the idea of a percentage resize first and a multiplication factor upscale afterwards. If I have 512x512 images, which settings should I select to upscale them to 2048x2048? Would I choose 100% resize with a 4x upscale factor or would I choosea %200 resize with 2x upscale factor? - or shouldit be something different?
For example, if I want the images to be twice as big as the original - would I choose 100% bigger or 200% bigger? Then when I upscale it, what size would 4x be? Would it be 4x bigger, or 4x 200% bigger. I just don't get it.
Also, I don't understand the RAM specification - is is asking for VRAM or System RAM - or both combined?
Either way, I've tried many different settings and the upscaled results are exactly the same as the original images = 512x512 and it adds a watermark to the image saying "upscaled with github.com". It also adds all the images to the source folder so it's a compete mess after it's process the files because it keeps the original, adds a converted file and a resized file.
If you have a 512 image and want to 2048 you have to select 100% resize (that means no resize) and bsrganx4. The resize before upscalimg option goes from 1 to 100%.
VRAM/RAM is a limiter for GPU VRAM but if needed the app Will use system RAM as GPU Memory to avoid full Memory errors.
The output file is saved in the same folder of the original file, called "filename-BSRGANx2.png". The temporary files in source are deleted once upscale process in done.
Thanks for your reply - I think it's probably an error with me not meeting minimum specs. I have an Apache Pro laptopwith Nvidia1060 (3GB). I think minimum is 4GB, recomended 6/8GB+ I have since found a workaround by using a github cloud version, which works on low-spec machines (my laptop, old Lenovo tablet and my Samsung Galaxy A20 phone) using a browser...I've managed to make 1300 images in a day or two.
I'm so excited to try this app, but unfortunately it won't let me drag images into it. Whenever I attempt to, my cursor changes to a "cancel" icon as I hover over the window. Am I missing a step somewhere?
← Return to tool
Comments
Log in with itch.io to leave a comment.
How long does it take for a 3 minute video using BRSGANx4 to finish processing?
Ps.: Ryzen 7 1800x + 32 GB RAM + GTX 1660
Hi my friend,
it all depends by the resolution of the video.
for upscale video I would recommend RealESR_Gx4, which is very fast.
But if you start upscaling, the app will say the remaining time to finish the upscale.
Hello friend, the new version has a surprising performance, if you allow me to suggest next time you could add the option to remove audio from the video. For example, I remove the audio to make the conversion faster and then I synchronize it in a video editor.
Thank you my friend :D
Absolutely, I will add it as the next improvement
hi friend ,what is the best ai to use on hand written notes ?
Hi my friend,
i will say BSRGAN_x4 that is the best model in terms of image quality :)
Would it be possible to add the ability to use custom models?
Hi my friend,
do you mean other AIs?
Yup.
The problem is that "is not that simple".
To have a working AI, it is not enough to have the .pth file but it is necessary that from code you map exactly the structure of the AI model.
It is not something plug&paly that you find a model online and use it.
Ok.
but i am actually working to add new AI architectures
Great application. it got a great potential of growing up as its going the right direction. The UI may need some tweaks like a progress bar or something. But It doesnt really matter as much as the functionality is perfect. Keep it up!
Thank you so much my friend, I appreciate it. ❤️
Hey just wanted let you know I am loving the update on your app :) especially with the explanation of each AI model. Thanks also for fixing the transparency issues with Windows 11.
By the way, can we no longer drag files into the program?
hi, thank you my friend ❤️
Sadly the new GUI does not support drag&drop, only file selection by button :(
Hi friend, did you remove the BSRGAN x2 model or is something wrong on my end? The 2x model produced much cleaner images than the 4x for me and had less tiling issues I guess due to needing less VRAM. But now all I see are the 4x for everything.
Hi my friend, yes, the BSRGANx2 has been removed due to incompatibility with newer versions of pytorch :(
but I am working to integrate a new AI architecture that is comparable in quality to BSRGAN but consumes less than half of VRAM :D
I see, that's unfortunate but thanks for clarifying. Is there any way to download an older version with the 2x model until it is released? And the new architecture sounds cool, I just hope there's a 2x version in the future because it's necessary for smaller images where 4x looks unrealistic. Thank you!
Sadly i do not have older version prepared, and the old AI library is no more "downloadable" :(
Unfortunately, researchers on AI of this type only publish the x4 version of the trained models, because that tends to be the one that returns the best results.
hi, i am new to a lot of this and i am quite confused on how to use your scaler. I uploaded my image to it and clicked the upscale button. I waited and now it said 'all files completed'. However, where can I access the upscaled image? I don't know where to find the enhanced image anywhere.
Hello, my friend,
the upscaled image or video is in the same directory as the original file :)
does it replace it or does it make a new one?
because i am not seeing any change
nevermind i figured it out!
thank you my friend :)
Hi, i´m have this error with all videos:
---------------------------
Error
---------------------------
Upscale failed caused by:
'utf-8' codec can't decode byte 0xe1 in position 1774: invalid continuation byte
Please report the error on Github.com or Itch.io.
Thank you :)
---------------------------
Aceptar
---------------------------
Hi my friend,
I think it might be about the special letters contained in video names, can you try renaming a video "video" and try again?
if you join the telegram group i can help you better :)
Do you have any idea on how to fix the tiling issue?
I upscaled a video today and see noticable tiling.
Settings:
BSRGANx4, Full Precision, png, VRAM set to max of graphics card, 70% input res, 8 cores
Hardware:
RTX2060 (6GB), i5-12600k, 32GB RAM
Edit: Input Video Resolution is 640x480
Hi my friend,
you can try to use half precision.
with 6Gb VRAM in Full precision it can only support max image 360x360 without tilling
while in Half precision it can upscale image or video 600x600 without tilling
:)
I see - But would that not indicate an issue with the tiling solution?
Since you have that listed as a feature to "avoid VRAM limitation" I would think that this should be possible.
Half precision works fine but is way slower.
btw: RealScaler can upscale the same video just fine.
I am actually working with this problem.
It happens only with certain types of images and videos.
RealScaler use 2 AIs optimized to be fast and memory friendly: RealESR_Gx4 and RealSRx4_Anime.
With these 2 models and 6gb of ram you can upscale 1200x1200 images without using tilling.
Great tool, love it! I am already thinking about getting RealScaler and FluidFrames as well.
I seem to have an issue with the program window where it disapears into the background when I click anywhere but on the buttons or dropdowns. Even resizing / moving the window does not work. The window behind the tool gets called into the foreground.
This seems to happen while the program is processing a video.
Not a big bug but would be nice to be able to move the window during longer processing times.
Currently running QS 2.2.
Hi my friend, thank you ❤️
Are you on Windows 11? because i noted that the "transparent effect" used for Windows 11 makes the window unclickable and not draggable, is that the case?
I cannot test this because on my machine i use Windows 10 where this transparent effect can't be applied, so any feedback is absolutely welcome :D
Yes indeed, I am running Windows 11. Sounds plausible.
Let me know if you need any tests done to fix this - I am glad to help!
Thank you my friend, next release will have this fix :D
Mind telling us what this is!?!
Hi my friend, it is a problem of the library used to transform the python code in .exe:
https://stackoverflow.com/questions/43777106/program-made-with-pyinstaller-now-s...
Hello friend,
I came to a little problem with the scaler... If the original was taken by phone, and not "widescreen" but the "standing video", the upscaled video will be transformed to a ratio as if it was taked wide screen... Resulting in turning the people into small, fat dwarves. Can be funny, but not the result it should be. :)
Hi my friend,
I've tried several videos, even taken from cell phones, so tall and thin, and this problem never happened.
If you joined the telegram group i can help you better :)
I just bought and downloaded the app, but there there are only image output options for a video file I added. Where are the video output options?
Hi my friend,
the output for video file is now fixed to mp4 because the library used to handle video reconstruction only supports mp4 format well.
But I am already working on allowing the choice of format for videos as well :)
Hi there,
Your tool here is a great tool ;) I'm using it a lot!
Two-three things, I may suggest:
- for the input resolution widget: I saw that you got a lot of questions about it in the comments below. An easy way to solve this would be to display the final resolution somewhere (maybe just under the input video on the top of the application) and that way we will know what resolution we will have afterward;
- you put here and on your github some examples for the BSRGANx4. Have you got anywhere some examples for the other AI Model? It's a bit hard to compare, for instance, the quality of each and which model is best for what situation. ie. is RealSR_jpgx4 better for real photos or manga pictures, etc.?
- is it possible to have some bench to know which AI model is the fastest (or to have clues about them)? For instance, for the exact same input video, and the exact same output resolution (ie. 480 to 1920), what would be the whole process duration for each AI model?
As I said, it's great tool. Thanks very much!
Hello, thank you very much my friend! :D
1. i am already thinking how to implement your suggestion
2. absolutely, as soon as possible I will upload an example with all AI on a single image
3. all implemented AIs use the same architecture, so in terms of speed they are identical. But I have in roadmap to implement other architectures both faster (but with lower quality) and slower (but with higher quality).
Thank you very much to you!
Thanks for your answers! I didn't know that all the model implemented were based on the same architecture ;)
I'll wait for your next update then :))
Thank you :D!
Really cool, thank you for your work. Some suggestions:
Sometimes I'll be running a batch of images and it will randomly error when a background process uses too much GPU or something like that, and then I lose my place. What do you think of adding a "resume" button, or even better, a way to have it retry on its own so I don't have to watch it?
Also, it doesn't clean up the resized files of finished images after it errors. Maybe it will be better to remove the resize files after each image is done, rather than all at the end, to keep disk space free.
Looking forward to more from you!
Also, regarding the stitching, I think a more foolproof solution could be to include a small amount of "bleed" around the edges of each slice, and then to cut out the bleed when merging them back together. But, I don't know how this tech works or if that's even possible. Just pitching my idea. :) Thanks again.
Hi!
Thank you my friend :D
I will investigate this gpu memory problem; you are right about resized images when upscale fails, i will take this improvement :)
Thank you!
Суставы все еще там.
На самом деле, в этой программе не так много смысла, когда она оставляет такие артефакты.
Мой знакомый решил похожую проблему и выложил ее в общественное достояние.
Это класс для нарезки и сшивания контуров
https://github.com/iperov/DeepXTools/blob/master/DeepXTools/core/lib/image/Patch...
Hi my friend!
Thank you for the suggestion, i will try to integrate asap :D
Does anyone know how to fix this?
Hi my friend,
maibe you clicked on Upscale button two times? because it seems that 2 upscale process are conflicting . Try to close the app and check on task manager if it is completely close, the retry :)
Hello, I have a problem with the image after rendering, it has the mark of the cuts, what can I do to prevent this from happening?
Hi my friend,
This happens when the input image to AI Is too big for the selected value of VRAM limiter .
To resolve this issue you can use the Input Resolution widget, so you can downscale the image before passing to AI.
Another option is to modify Vram limiter option (default is 8 gb), if you have a GPU with more VRAM you can set the correct Number. For example if you have a GPU with 24Gb, you can enter 24 :)
it always leaves a mark on the slices, I've tested it by lowering the input resolution by several %, my GPU has 2GB of dedicated memory and CPU 4, and I put these values in their proper fields, the resolution of my image is 1568x896, what do I do ?
The problem is your 2GB of VRAM.
You should imagine that for each GB of vram, you can contain max 100x100px, for example:
Qualityscaler there needs to be a trial version where it is full version for 14 days or you gt process 30 to 40 images and videos then it does not work any more. If the program is up to my liking I would be willing to pay for it after I am able to test it out.
is there an existence/possibility of an android port, I could only find windows apps, sorry if this has been posted already
Hi my friend,
Unfortunately, a smartphone version is not currently possible :(
Thanks for the reply, is there a possibility to get a resize option (ex: 1920 x 1080) for images added to a future update if possible or if there is an avenue I can take to achieve this?
Hi my friend,
it is possible to do what you ask thanks to the "Input Resolution" widget.
For example:
It all depends on the original resolution of the image.
I am also thinking in the future to simplify these steps or give visual feedback of the final resolution based on the choices set :)
options button just doesn't work, and no clear indication for output location.
Hi my friend,
the "Option" button is just a header for the option list on the left bar, it will removed in next version :)
File output location is the same of the input file, for example if you load a file on Desktop, the upscaled file will be saved on Desktop but i will improve this part, thank you for the feedback :)
every time i run it, it gives bad parameter error:
Error while upscaling
The parameter is incorrect.
Is there a more detailed log of the process anywhere?
Hi my friend,
The error that is on screen is the only error report.
I have never seen this error, if possible could you please list:
What is faster for video upscaling: bsrgan or esrgan?
Thanks,
Hi my friend,
are you talking about my 2 apps?
QualityScaler has 3 heavy, high quality models, while RealScaler has 2 models:
so, if you are interested in quality you should choose QualityScaler, while if you want speed you should choose RealScaler with RealESR_generalv3_x4
:)
Hi Jangy, I saw there is a folder "AI" to store ".pth model" so I tried to import a new one (found on https://upscale.wiki/wiki/Model_Database) but unfortunately it was not enough for it to appear in the interface. Is there a way for me to import new model or not at all ?
Hi my friend,
Unfortunately it's not that simple, each .pth needs all the architecture in the app code, but if you want to try you can change the name of the "newmodel.pth" to "BSRGANx4" and see if it works. At 99% Will not work :)
can you send me the link please
Hi my friend, what link?
the old beta version of software
You can check Github, I don t have older version of the app, sorry :(
Hi Jangy, nice to see upcoming changes in QS.
what's the possibility of using a conventional upscaling algorithm (like Lanczos) and then running something like BSRGAN to fill in the gaps?
Hi my friend, and thank you :D
What do you mean, can you give me an example please ? :)
Well, you've got conventional upscaling algos like Lanczos and Bicubic. There's some talk about how better results at upscaling could be produced by first upscaling 4x thru ESRGAN or BSRGAN and then downscaling 2x thru Lanczos, see: https://upscale.wiki/wiki/Model_Database
Not a video researcher so i'm not sure how if there's a possibility you could experiment with such a technique for future releases of QS
Ohh now i get it,
Yes, this could be a great approach but the computational cost would be even greater than now :)
Oh def. would impose a greater processing time, I imagine you'd have to do even more with optimizations. But it would make your app very unique: I dont know of anyone else doing it.
Thank you my friend :D
any way to view location of upscale frames? just kind of says its working, would be nice to see where its at exactly. where to view temp files for the extracted frames? i dont want my main drive being filled up if it just goes by where the exe is installed. doing some testing on my 6900xt. also is there a way to enable multiGPU? i have a 1650 running as well for other stuff, would be nice to combine them.
Hi my friend, you can see extracted frames and generated frames in a directory called QualityScaler_temp in the installation folder of the app, at the end of the process the app Will clean automatically temp files :)
Just purchased and the app doesn't display the upscale button and won't let me adjust the window size. I adjusted my computer display setting so I could view the entire window and... the button at the bottom is indeed missing. Any help?
Hi my friend,
what resolution is your screen? Sadly the GUI used does not works properly on fullscreen, so i disabled it; and the app is made to worki with screens > 1080p.
However you can try to hidden the Windows taskbar, this can help.
I will improve this part of the app in next releases
I adjusted to 1080 and it worked! But now it gives me an error. I think it's because this computer doesn't have enough power to process.
Nice my friend!
For Intel Iris gpus or integrated gpus you should input Vram Limiter GB option 1 or 2 Gb instead of 8 :)
You are correct. Thanks. My current old computer is too weak to handle this. I'll need to get a new one at some point. Thanks for your help.
Do not worry my friend, Is not your computer.
It s the AI that Is really heavy but, even my PC struggles with my own apps haha.
But they are future prof, when gpus will return ti be affordable as they was 3 years ago :)
can you send it to me please
The parameter is incorrect.
Can you help me?
Hi my friend,
did you followed the README and installed Directx and Visual C Redistributable?
How much ram does your gpu have? You can try to change "VRAM limiter GB" value according with your gpu capability
:)
Hello!
What I could try up to now, it is a very nice software, thank you.
But can it be that even just short movies take ages?
I have Pentium 13900 and Nvidia 4090, so I thought letting it do its job the whole night will finish a 9 Minute video.
Thank you in advance.
Hi my friend, thank you! :D
The upscale phase for video is really slow but with super pc like yours should definitely be fast. Did you check this options?
For example:
* AI backend option. Did you selected the right gpu? you should have 2 gpus, one integrated in cpu and the Nvidia 4090
* VRAM limiter option. Did you select the correct amount of VRAM for your gpu? For the 4090 you should select 24 GB
* CPU number optio. Did you select the correct amount of cpus? For example your cpu has 24 cores.
Also, for the new 10.0 version that will come in few weeks there are a LOT of bugfix and improvements for video upscaling in particular :)
Hello and thank you for the fast answer. Yes, I did as you said.
AI backend I have only one, the 4090 to chose from.
With the speed now it will still take at least one, I think two days to finish.
I maybe still use something wrong, because as you say, it should run a bit faster in my machine. Also I have 64 GB RAM, if this could be a brake of the process.
Hi my friend! Thank you :D
What is the resolution of the video you are processing?
If you open the Windows Task Manager can you check that your gpu is being used?
In my testing lately I have found that Windows 11 has slowdowns in reading and writing files, and as you can imagine this impacts QualityScaler's speed a lot. Do you have windows 10 or 11?
Thank you for your feedbacks :)
The resolution is 1424x944.
Taskmanager shows my GPU is in nearly 100% use. And I use Windows 11. So maybe Win11 is the reason then. :(
Absolutely super manageable resolution from your 4090; at this point I really think it's Windows 11 which is not yet fully optimized.
If you can try there is ReviOS, an optimized version of Windows 10 or 11. free of Defender, Cortana and that kind of bloatware, updated ISO every month with patches :)
Hello to both of you! I love this tool. Especially upscaling very old pictures or short recordings. Unfortunately I experience similar habits upscaling mp4 vids. Using multiple computer setups (win10) while Version 9 maxed out CPU and GPU the newest Version 10 takes up 14 perc max.
Hi my friend,
and thank you :)
Version 9.0 had a bug that allowed excessive cpu usage.
I'm working on improving QualityScaler more and more on the speed issue, but basically I can't do much for now because the 3 AIs of QualityScaler offer very good quality but have a very very complex and slow architecture.
For long videos I would recommend RealScaler, another upscale app of mine that uses Real-ESRGAN as the AI. In particular "RealESR_generalv3_x4" offers much faster speed (about 4 times faster than BSRGAN) at the expense of lower upscale quality.
Thank you :)
Hi,
Thank you!
unfortunately, I can't get the software to work properly; the Options button does not open any new menu, and whenever I try to upscale a file I get the error message in the screenshot below.
I get similar error messages with all the selectable models.
Could you please assist?
Hi my friend, the Option button does not open anything, is just a "title" of the option under :)
That error means that the image does not fit in your VRAM limiter option.
You have to select the correct number of VRAM limiter; for example if you have a gpu with 4GB of VRAM you have to select 4, if you have gpu with 6GB of VRAM you have to select 6, and so on.
What GPU do you have?
UPDATE. I saw not from the screen that you have a integrated Intel Xe, in most case for integrated graphics 2GB Vram limiter is the perfect option :)
I hope i helped you :)
1. What is the location of the temporary png files (and other temp files) . I would like to make sure that the application purges them after a task.
2. Feature request: estimated time to finish counter.
Thank you.
Hi my friend,
1. temporary files are in the installation directory
2.thank you, I m working on It :)
Hello. Version 4 is the last version where embedding an image or video works for me. Even in versions 7 and 8 my image does not appear, only the clean button appears. Version 4 works without problems, just to be sure I reinstalled both DirectX and Visual-C Runetimes as you have in the instructions. I tried running QualityScaler.exe as administrator, but it doesn't let me insert images at all then. All my attempts are without any result. I have Windows 10 and the HW matches the requirements. Can you please advise me what I am doing wrong? Thank you!
Hi!
There is a bug with running as Administrator, so do not use it :)
Can you tell me more informations about your files? For exampe, extension, path etc :)
Normaly when drop a single image or a single video the app create inside its directory a file called temp.png, is a temporal file to display the image on the app. Can you find this file?
Thanks, I've moved the application directory from Program Files to Desktop, and it's working now. Apparently the program couldn't write to disk, temp.png wasn't being created. It works great now. Thanks.
Nice!
Thank you for your feedback :D
This is one of those things, where it's so cool and impressive. I had to buy it.
Thank you my friend 😆
Hello. I can't buy the program with Paypal or a Bank card. Please sell for cryptocurrency. I am ready to transfer money to your wallet. Please help, I like your program.
Is it possible to request multi video support? I want to drag and drop an entire folder for upscaling. Tyvm!
Hi, this feature is already in development :)
thank you!
cant download the new version 7.0 even though I paid for this product?
Hi, why you can t download?
it took a day for 7.0 paid link to come up.
Sorry, i don t know what are you talking about. I Just uploaded the new version as Always 🤔
i think the issue was itch.io. i was only able to download the version with watermarks. the paid version showed up a day later.
Ok, thank you :D
i go to open it and it says it comes with a trojan
Hi, is a known bug of the library used to pack the python files in .exe :)
https://stackoverflow.com/questions/43777106/program-made-with-pyinstaller-now-s...
Really cool package. Is there a feature to preserve audio from a converted (upscaled) video file?
Hi, thank you :D.
yes, i'm working on this feature, it will be available on 5.0 :)
All the different number selections in the UI are a bit confusing for me.
I don't really understand the idea of a percentage resize first and a multiplication factor upscale afterwards. If I have 512x512 images, which settings should I select to upscale them to 2048x2048? Would I choose 100% resize with a 4x upscale factor or would I choosea %200 resize with 2x upscale factor? - or shouldit be something different?
For example, if I want the images to be twice as big as the original - would I choose 100% bigger or 200% bigger? Then when I upscale it, what size would 4x be? Would it be 4x bigger, or 4x 200% bigger. I just don't get it.Also, I don't understand the RAM specification - is is asking for VRAM or System RAM - or both combined?
Either way, I've tried many different settings and the upscaled results are exactly the same as the original images = 512x512 and it adds a watermark to the image saying "upscaled with github.com". It also adds all the images to the source folder so it's a compete mess after it's process the files because it keeps the original, adds a converted file and a resized file.
Thanks.
hi,
If you have a 512 image and want to 2048 you have to select 100% resize (that means no resize) and bsrganx4. The resize before upscalimg option goes from 1 to 100%.
VRAM/RAM is a limiter for GPU VRAM but if needed the app Will use system RAM as GPU Memory to avoid full Memory errors.
The output file is saved in the same folder of the original file, called "filename-BSRGANx2.png". The temporary files in source are deleted once upscale process in done.
Strange you are finding this problem:
what is your PC specs ?
what is your image extensions ?
Thanks for your reply - I think it's probably an error with me not meeting minimum specs. I have an Apache Pro laptopwith Nvidia1060 (3GB). I think minimum is 4GB, recomended 6/8GB+ I have since found a workaround by using a github cloud version, which works on low-spec machines (my laptop, old Lenovo tablet and my Samsung Galaxy A20 phone) using a browser...I've managed to make 1300 images in a day or two.
https://myalbum.com/album/gH5srPyFm6Z4qW/
ok, thank you for your feedback :)
Did you tried with VRAM limiter?
Because for 3gb VRAM you can choose 2gb and see if works :)
Ah OK - thanks! I didn't know about the limiter. I'll try it. Thank you
Thank you ! :D
I'm so excited to try this app, but unfortunately it won't let me drag images into it. Whenever I attempt to, my cursor changes to a "cancel" icon as I hover over the window. Am I missing a step somewhere?
Hi, thank you. Which extension has your file? :D
I've tried both .jpg and .png
Strange, it should work with theese files.
Try to rename one file calling simple like "image.jpg", there is a known bug with files that have latin symbol in the filename.