just a remark if you go to load model after starting up and only get the NCNN load model which will not load the .pth file, it means you need to click on the dependency manager and install the pytorch one and then ope nthe load model from that category
Hello. chaiNNer has been updated a fair bit since this video. You now need to use the "Load Images" node to process a folder or folders. You can tick the "Recursive" option to process all the sub folders too. I have been wanting to do an updated version of this video for weeks. But, life has just been too busy. Hope this is of help.
How do you set this up to save in multiple sub directories? If I point to a root with multiple sub directories, upscale the images inside of the separate directories and save.
I'd get the latest version of chaiNNer (v0.21.1) and use the "Load Images" node. Be sure to enable the "recursive" toggle to process all images in all subdirectories. I'm in the process of making a very quick demo to show the basics. I still haven't played with the other options yet. Hope this is of help.
A few weeks ago, @heshwelgolle had a similar issue with n Apple M1. It was suggested to use the NCMM modules rather than the PYTorch ones that I use. I haven't heard if that solution worked. I'm on an Intel/Windows system.
Thank you for this great tutorial. This is an amazing Tool. I have one Question: I installed packages from the dependencies manager, but i cannot find the models anywhere. Do i have to install model for upscaling separatly. Thank you for your help.
It has turned out to be a great tool. Has let me automate a lot of repetitive tasks. When you install the packages in the Dependency Manager, they should appear as modules in the list on the left. (PyTorch, Onnx & NCNN) I use PyTorch to make use of my GPU for processing. You will also need to download a model or more from upscale.wiki/wiki/Model_Database I have extracted & placed some in a folder. When I place the "Load Model" module, I just select the one I want to use from there. Hope this is of help.
I loaded 3 models available for loading from the program interface (by clicking the arrow in the upper right corner), but when adding the "load model" node, I click on the node and the model file search window opens. I don't know where to look for the model itself. Tell me, where are the necessary files that you managed to download, in order to indicate the path to them for the corresponding node?
The download arrow (top right) will let you choose which "dependencies" you wish to use. You can enable all of them. To get the actual GAN, you need to go to the model database ( upscale.wiki/wiki/Model_Database ) or search GIThub or the likes. Select & download one or more from there. Some are trained for specific uses. Put the downloaded file, usually with the "pth" extension (*.pth) somewhere that makes sense to you. When in chaiNNer select the model from where you put it. All of mine are in a folder called GANs. I show how I install chaiNNer etc in this video ( th-cam.com/video/qAm1X0k1pZ0/w-d-xo.html ) The chaiNNer discord is here ( discord.gg/pzvAKPKyHM ) The Dev team and lots of talented people are there, who are always willing to help.
Such a great tool, but everytime I want to use it I can't, and have to wait for it to update python for an hour or three. By ther time it's done i dont' even need it anymore. Am I doing something wrong?
When I try to install chaiNNer I get a blue windows message that it can't run the program because "it may be harmful to the computer" - I downloaded it from the GitHUB link in your description, what do I do? 😮💨
Mine was the same. I do the usual virus / malware scans. If they are good, then install and scan again. I have had no issues with any that I have installed. (Sorry, I've been a few days in replying. Have been ill :( )
@@KriemenGaming I'm going to try to download it again and give it an "exception" with the antivirus, to see if I can use the program. get well, good luck😎
Look for and run the file, "chainner.exe" I have a shortcut on my desk top. Can't remember if I created it or if the install did. In my case, the executable is here "C:\Users\UserName\AppData\Local\chainner\chainner.exe" Many GANs can be found here - upscale.wiki/wiki/Model_Database They can be place anywhere and selected in the "Load Model" node. I'll try to do an installation video as soon as I can get some spare time.
Great video, thank you. I have downloaded the app and dependencies, and run a chain to upscale using a decent model, did it all correclty, yet when I execute, the final 2 chains to Upscale and Save the image just load and load and load and never seem to finish until I stop/quit it. I am running Windows 10 and do NOT have NVDIA, so I fogured it would be slower, but did not expect that it would take over an hour. I closed all other programs, made sure I have space, I can't figure out what is making it run and never finish. Any guidance you can provide that can help me? This tool would be so valuable to me. Thanks in advance!
I think you need to use the NCNN models if using AMD. The PyTorch models require NVIDA?? It sounds like it is using the CPU rather than the GPU. If you are a Discord user, the gang here are very helpful discord.gg/pzvAKPKyHM
If the image has already been created, I don't know of any way to find out. Unless, for example, someone enters the info in the comment section of a JPG. Personally, I use 4x_NicklebackFS for most and 4x_Ultrasharp for more sharpening. Have only tried a few and the results can be very similar. Haven't tried anything with cartoon / anime styles. but, there are GANs trained for them. If you are doing faces, I have had good results with using this arc.tencent.com/en/ai-demos/faceRestoration
I'm afraid I can't offer any wise advice for use with Apple products. :( The author notes that "MacOS versions older than 10.15 are not supported at this time" and "Apple M1 laptops are mostly untested, though they should support almost everything. Although, ONNX is unable to be installed as it does not yet have an arm64 build, and NCNN sometimes does not work properly" This is in the compatibility notes here github.com/chaiNNer-org/chaiNNer
Are you able to use Gigapixel as your load model for this? I would like to use the bulk resizing feature this offers but I can't find a way to get gigapixel linked in and the others don't give me the same results with my images.
I can have at look at doing that. I'm ill with the dreaded covid at the moment. So, it might be a few weeks before I can do it. I'll put it on my "to do" list. 🙂
Have a look at upscale.wiki/wiki/Model_Database There are a lot of different GANs there, for different purposes. You may find others by searching with Google or the likes. Hope this is of help.
Sorry I have been a while replying. I am unsure of the exact usage for that. Am guessing that it may process the image in pieces?? and that is the size of each piece?? All I have heard is to just leave it set to auto. I would really have to ask the author of chaiNNer for more info.
Many models can be found here upscale.wiki/wiki/Model_Database In this demo I used 4x_UltraSharp by Kim2091. I just place then in a folder / folders of my choosing and select as needed.
As far as I know, this should run ok with older / slower hardware. New, faster GPUs should process things a whole lot faster. I have a GTX3050 Ti in my laptop and it handles things quite well. I think having more RAM & VRAM is a big plus. I have 16GB / 4GB in mine and have come close to using it all when processing. But, have never run out...
Must admit that I have never tried Gigapixel. So, I can't honestly say. There are many GANs out there that are suited for various jobs. I just have a few that I use. They suit what I need at the moment.
Yes, you can process videos. Use the "Video frame interator" to open the file. It does not process audio. So, that would have to be manually replaced afterwards. Upscaling, like I showed here, is very slow. I haven't done much with video, so can't share any more than that...
I've tested many filters, but for human face the eyes looks really weird, how to get close or better from the site letsenhance wich is very good but paid heres the image comparation the first one on left is the original, the second on right is from letsenhance, the other two bellow is using the filters 4xultrasharp, 4xultramixbalanced on to improve to get equal to letsenhance or better: i.imgur.com/zY6yDKB.png
I notice that Dream.AI often messes the eyes and the teeth. I've been running my faces through this site arc.tencent.com/en/ai-demos/faceRestoration There are some limitations with image size. It will try very hard to fix eyes, mouth areas and make the faces look a bit more realistic. And it is free to use.
just a remark if you go to load model after starting up and only get the NCNN load model which will not load the .pth file, it means you need to click on the dependency manager and install the pytorch one and then ope nthe load model from that category
Hi Kriemen, I can't find "Image file Iterator" ;; A ;; I don't know where to download it.
Hello. chaiNNer has been updated a fair bit since this video. You now need to use the "Load Images" node to process a folder or folders. You can tick the "Recursive" option to process all the sub folders too. I have been wanting to do an updated version of this video for weeks. But, life has just been too busy.
Hope this is of help.
@@KriemenGaming Oh thanks very much. Your tutorial help me a lot.
5:21 stock 5:28 upscaled
thank you so much ❤
How do you set this up to save in multiple sub directories? If I point to a root with multiple sub directories, upscale the images inside of the separate directories and save.
I'd get the latest version of chaiNNer (v0.21.1) and use the "Load Images" node. Be sure to enable the "recursive" toggle to process all images in all subdirectories.
I'm in the process of making a very quick demo to show the basics. I still haven't played with the other options yet. Hope this is of help.
@@KriemenGaming not work for me after upscale and got black image
How to use a 1x model?
Just select it and run. It won't make the image larger. But, it will do its thing on it.
great tutorial. Thank you for sharing.
Glad it was helpful!
THANK YOU FOR THIS TUTORIAL
Can't get past the LOAD MODEL bit as the pretrained model data does not exist on my Mac so it is impossible to select the same options.
A few weeks ago, @heshwelgolle had a similar issue with n Apple M1. It was suggested to use the NCMM modules rather than the PYTorch ones that I use. I haven't heard if that solution worked. I'm on an Intel/Windows system.
good.peaceful.
Thank you for this great tutorial. This is an amazing Tool. I have one Question: I installed packages from the dependencies manager, but i cannot find the models anywhere. Do i have to install model for upscaling separatly. Thank you for your help.
It has turned out to be a great tool. Has let me automate a lot of repetitive tasks.
When you install the packages in the Dependency Manager, they should appear as modules in the list on the left. (PyTorch, Onnx & NCNN) I use PyTorch to make use of my GPU for processing.
You will also need to download a model or more from upscale.wiki/wiki/Model_Database
I have extracted & placed some in a folder. When I place the "Load Model" module, I just select the one I want to use from there.
Hope this is of help.
Sorry, where i can find the models pth ?
thanks
You can find models here upscale.wiki/wiki/Model_Database
There is a newer site being built But, I cand find the link at the moment...
Found it. openmodeldb.info
I loaded 3 models available for loading from the program interface (by clicking the arrow in the upper right corner), but when adding the "load model" node, I click on the node and the model file search window opens. I don't know where to look for the model itself. Tell me, where are the necessary files that you managed to download, in order to indicate the path to them for the corresponding node?
The download arrow (top right) will let you choose which "dependencies" you wish to use. You can enable all of them. To get the actual GAN, you need to go to the model database ( upscale.wiki/wiki/Model_Database ) or search GIThub or the likes. Select & download one or more from there. Some are trained for specific uses. Put the downloaded file, usually with the "pth" extension (*.pth) somewhere that makes sense to you. When in chaiNNer select the model from where you put it. All of mine are in a folder called GANs.
I show how I install chaiNNer etc in this video ( th-cam.com/video/qAm1X0k1pZ0/w-d-xo.html )
The chaiNNer discord is here ( discord.gg/pzvAKPKyHM )
The Dev team and lots of talented people are there, who are always willing to help.
Such a great tool, but everytime I want to use it I can't, and have to wait for it to update python for an hour or three. By ther time it's done i dont' even need it anymore. Am I doing something wrong?
When I try to install chaiNNer I get a blue windows message that it can't run the program because "it may be harmful to the computer" - I downloaded it from the GitHUB link in your description, what do I do? 😮💨
Mine was the same. I do the usual virus / malware scans. If they are good, then install and scan again. I have had no issues with any that I have installed. (Sorry, I've been a few days in replying. Have been ill :( )
@@KriemenGaming I'm going to try to download it again and give it an "exception" with the antivirus, to see if I can use the program.
get well, good luck😎
Noob question: I installed chainner but i dont know how to run it, is their like a .bat file or something?
Look for and run the file, "chainner.exe" I have a shortcut on my desk top. Can't remember if I created it or if the install did.
In my case, the executable is here "C:\Users\UserName\AppData\Local\chainner\chainner.exe"
Many GANs can be found here - upscale.wiki/wiki/Model_Database
They can be place anywhere and selected in the "Load Model" node.
I'll try to do an installation video as soon as I can get some spare time.
Great video, thank you.
I have downloaded the app and dependencies, and run a chain to upscale using a decent model, did it all correclty, yet when I execute, the final 2 chains to Upscale and Save the image just load and load and load and never seem to finish until I stop/quit it. I am running Windows 10 and do NOT have NVDIA, so I fogured it would be slower, but did not expect that it would take over an hour. I closed all other programs, made sure I have space, I can't figure out what is making it run and never finish.
Any guidance you can provide that can help me? This tool would be so valuable to me.
Thanks in advance!
I think you need to use the NCNN models if using AMD. The PyTorch models require NVIDA?? It sounds like it is using the CPU rather than the GPU. If you are a Discord user, the gang here are very helpful discord.gg/pzvAKPKyHM
hi im on latest chainner version and there don't have inerator even I added "Load Images" and not work cuz I got black image after upscales
th-cam.com/video/WapMHADd76A/w-d-xo.html
Have a look at this for a more current video. Does your image have transparency / Alpha channel?
another question is there a way to discover what filters or kind of model was used in image to upscale ?
If the image has already been created, I don't know of any way to find out. Unless, for example, someone enters the info in the comment section of a JPG. Personally, I use 4x_NicklebackFS for most and 4x_Ultrasharp for more sharpening. Have only tried a few and the results can be very similar. Haven't tried anything with cartoon / anime styles. but, there are GANs trained for them.
If you are doing faces, I have had good results with using this
arc.tencent.com/en/ai-demos/faceRestoration
Sadly none of the Upscaling models (especially Pytorch models) work on an Apple M1 mac 😢
I'm afraid I can't offer any wise advice for use with Apple products. :( The author notes that "MacOS versions older than 10.15 are not supported at this time" and "Apple M1 laptops are mostly untested, though they should support almost everything. Although, ONNX is unable to be installed as it does not yet have an arm64 build, and NCNN sometimes does not work properly"
This is in the compatibility notes here github.com/chaiNNer-org/chaiNNer
If you use NCNN instead of PyTorch it should work. I’ve had success with my M1 Pro MacBook with ChaiNNer
@ArguingMeadows Thank you so much. Will have a crack at it 🙏🏻
Are you able to use Gigapixel as your load model for this? I would like to use the bulk resizing feature this offers but I can't find a way to get gigapixel linked in and the others don't give me the same results with my images.
Thank you! 🙂
You're welcome. 😊
hi there can you please create a video on how to install chainner
I can have at look at doing that. I'm ill with the dreaded covid at the moment. So, it might be a few weeks before I can do it. I'll put it on my "to do" list. 🙂
th-cam.com/video/qAm1X0k1pZ0/w-d-xo.html
Hope this is helpful. Sorry, I have been so very long getting something together.
can this bulk upscale images?
From 6:00 I show how to bulk upscale an entire directory of images. Hope this is of help.
Where to get the tools gans?
Have a look at upscale.wiki/wiki/Model_Database
There are a lot of different GANs there, for different purposes. You may find others by searching with Google or the likes.
Hope this is of help.
What does tile size mean?
Sorry I have been a while replying. I am unsure of the exact usage for that. Am guessing that it may process the image in pieces?? and that is the size of each piece?? All I have heard is to just leave it set to auto. I would really have to ask the author of chaiNNer for more info.
Is there a way to keep the same folder structure where the images are ?
Click and drag on the Subdirectory Path (use the little green dot) in "Load Image (Iterator)" into the Subdirectory Path for "Save Image."
Hi your video is very helpful, which model is the directory model in the database? Thanks!
Many models can be found here upscale.wiki/wiki/Model_Database
In this demo I used 4x_UltraSharp by Kim2091. I just place then in a folder / folders of my choosing and select as needed.
lol why this is so complicated, seems easier to use Cupscale
Does this require a good GPU?
As far as I know, this should run ok with older / slower hardware. New, faster GPUs should process things a whole lot faster. I have a GTX3050 Ti in my laptop and it handles things quite well. I think having more RAM & VRAM is a big plus. I have 16GB / 4GB in mine and have come close to using it all when processing. But, have never run out...
@KriemenGaming Thank you for your response. Well, it seems like it doesn't support windows 7. 😕
And your laptop is very powerful, lol.
Is it better than gigapixel?
Must admit that I have never tried Gigapixel. So, I can't honestly say. There are many GANs out there that are suited for various jobs. I just have a few that I use. They suit what I need at the moment.
Is it possible to use it with video files? Or i need image sequence?
Yes, you can process videos. Use the "Video frame interator" to open the file. It does not process audio. So, that would have to be manually replaced afterwards. Upscaling, like I showed here, is very slow. I haven't done much with video, so can't share any more than that...
I've tested many filters, but for human face the eyes looks really weird, how to get close or better from the site letsenhance wich is very good but paid heres the image comparation the first one on left is the original, the second on right is from letsenhance, the other two bellow is using the filters 4xultrasharp, 4xultramixbalanced on to improve to get equal to letsenhance or better: i.imgur.com/zY6yDKB.png
I notice that Dream.AI often messes the eyes and the teeth. I've been running my faces through this site
arc.tencent.com/en/ai-demos/faceRestoration
There are some limitations with image size. It will try very hard to fix eyes, mouth areas and make the faces look a bit more realistic. And it is free to use.