วีดีโอ

where should i put that pth file in comfy ui?

Wow... Outstanding work, congratulations! And thank you very much!

I learned so much from using your models and new watching your videos. Thank you! Please keep making them!

Is it weird that is taking so long to upscale a pic?

Hello Sir, Could you please guide me how can i use early stopping during the training of Real-ESRGAN?

couldn't get the model to load. just says this node is currently running indefinitely, but thank you though. because i am upscaling a game texture, I just wasted a whole heap of time thinking it needed the in game model the texture it was attached to.

This has been great for me for upscaling videos. However I don't think you showed the audio part. After you upscale the frames (Real ESRGANx4 for me) and use LOAD IMAGES for the new frames, you ALSO can use LOAD video and dump both the frames AND the audio track to the SAVE VIDEO node. (Assuming you pulled all frames from the source, saving them out, the audio is perfectly in sync.) For large video upscaling jobs, extract the source frames, start TEN instances of chaiNNer and for the LOAD IMAGES, use WCMatchGlob expression "*0.jpg" for the first instance of chaiNNer, "*1.jpg" for the second instance and so on... a clunky way of parallel threads without file access conflicts.

Hey Philip, my validation from the config file is not working and not showing up on my logger but it does not give me any error as well but doesn't work no matter what I tried with training the Real-ESRGAN model. Any advice on how to resolve it?

It's been a long time since I've seen such a beautiful linux ui interface.

Hi sir, can you please share with me you kaggle notebook to train my own model and validate it into kaggle ??

Hi! I just found you during a deep dive into this entire topic. I enjoy restoring older images in my spare time and up until this point I've exclusively used Adobe products. Your content is incredibly helpful for a 'newbie' like me. Thanks!

Thanks for the tutorial.

how i can find you discord?

Great video !

Hey, bullshit stupid guy, why you think that we must watching your nonsence video in order to understand how to upscale with chainner? Yes, western morons that dont know nothing more than youtube and tiktok may be it will be interesting, but there are still outside western morons normal people that hate you and such stupid peoples that make money on things that other smart peoples (chainner) done. And you use the worst models that are only for moronic digitally born photos and not for real world photos. But that is maximum what morons can do.

what do you think of github Image-Super-Resolution-via-Iterative-Refinement ?

how much ram do u installed here?!

Keeping that CFG scale high and denoise level low ;) This is pretty neat. Anything where generative AI and game development/modification intersect is pretty interesting.

I've been looking for these upscale models " 100k [epoch 507, psnr 30.517, ssim 0.83536, lpips 0.031511] or 150k [epoch 760, psnr 31.331, ssim 0.84292, lpips 0.029] " and I can't find them, do you have a link to them?

But what about a sequence of images (.exr)? Won't it create artifacts or difference in frames? Or would I have to transform the Sequence into a video and then use ChaiNNer?

For now (April 2024) Gigapixel AI is still much better in upscaling and face restoring...

Thanks for this amazing tutorial.Sir,could you please make a tutorial for the custom training of HAT model for image restoration.(HAT: Hybrid Attention Transformer for Image Restoration) I hope u will consider my request

Fabelhaft!! I can't believe this video (and the information that I found with it) is at least 6 months old... feels so NEW! Well, new to me. You've created some beautiful models (which I use in ComfyUI) - thank you for sharing your process. I'm currently using your 4xFaceUpLDAT model to great effect with UltimateSDUpscale. The setting need work but wow! how wonderfully this works!

I know your website since a while, but I didn't know you had a youtube channel. That's the most in-depth upscale video I've seen. Thank you sir 👌

Thanks for the video. Could you please tell me your GPU specifications? As i want to train my SR models on my local GPU.

thanks a lot for all the work and explanations

very good work, not only about work, also detailed about sources. Could u share your page link too.

züri.jpg :DDDD

what about how to run it in python or C++?

The update came out today, the program is great!

Do you have .param or .bin versions of these models? I ask because Upscayl software allows you to upload custom models, but they must be one of these two. Thanks!

My use case is upscaling skin textures from relatively smooth sdxl portraits where the face is a about 400px in height. I’ve been getting done interesting results either latent space upscaling, but if there’s a case to be made for this kind of upscaling model in comfyui using something like Ultimate SD Upscaling with spread cycles to actually generate skin texture? Does your model generate per se, or only extrapolate? That’s what in don’t understand…

This could work

amazing!!!

Hi, what is memory consumption for training? Let say for batch 2 & 4?

hey, i was wondering if you could lend a helping hand. I did the installation step by step as you did, and everything seem to be working just fine. However, when it starts to upscale it never really saves in the chosen directory. It keeps on trying to upscale forever. Do you have any tips on how to fix it? Thanks!

I have error when i try upscale face as in your video.: An error occurred in a Upscale Face node: Failed to run Face Upscale. Input values: • Image: RGBA Image 4096x4096 • Model: Value of type 'nodes.impl.pytorch.architecture.face.codeformer.CodeFormer' • Upscaled Background: None • Output Scale: 1x • Weight: 0.7

Good video, thanks. I wanted to test your 4xFaceUpDAT model, and compare the results with the other ERSGAN based Stable Diffusion upscale models I've used (for ex. swinIR, 4x-UltraSharp, ERSGAN etc.). Would you happen to have any example .py code which I can reuse, to test this model please? I tried to use this model similar to the other SD models, but that obviously doesn't work because this is a DAT based model. I guess I can install ComfyUI etc., but would prefer doing this directly from .py scripts.

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?

nice video

any chance you could add the training yml file for 4xNomos8kDAT ? i would like to train a model with 2x and 3x on my data. the model is really good and thank you for releasing it.

Good video. Thanks.

Colab-trainner only gives me errors. Someone knows how to use it properly?

As someone not experienced with node programs I really appreciate the detailed directions and pacing. Thank you!!

does this work on live action footage? im only just started to learn using ESRGAN

Thanks so much for all the work you did in putting the model comparison tools together too. That is pure gold!

This stuff is amazing! Thanks!

Thanks for the video. Can we merge two different realesrgan models and combine their capabilities. Like combining anime and other comic art style ? If not what could be the possibilities ?

Looks like a nice detailed explanations...Is there a English version of this process. Can this be done through Chainner ?