User blog:HaiFire3344/Waifu2x.udp.jp quick intro and guide

waifu2x is a neural network-based program for upscaling and denoising images, including both art and photographs. waifu2x-caffe is a good desktop client for it, but this post will discuss the web demo, waifu2x.udp.jp.

Upscaling is typically not necessary if you're uploading something to the wiki and you have the biggest available version. Denoising simply removes noise from the image; it can't make an image less blurry if it happens to be that way.

Getting an image
I'll use Twitter as an example here, as waifu2x is often useful for denoising Twitter images, which Twitter often compresses into JPGs. Open an image from a tweet in a new tab; in the URL, the property "name" will = something like a resolution or a size; change that value to "orig" while leaving the text "name" as is. Press enter and use "Save image as" in your browser to get the best possible version of the image.

On the waifu2x website: Picking an image
You can choose from a URL or a file on your device.

Image type
Choose the correct style. For example, if your image is digital art, choose "Artwork".

Noise reduction
The "Low" option often works fine, but I would recommend using "Highest" for the best possible reduction of noise. The site has some notes on when noise reduction should be used.

Upscaling
As mentioned before, this won't always be necessary, and I've never found it necessary for FNF images. If you don't need to upscale, select "None". If you only have an exceptionally small image of something, use either 1.6x or 2x depending on what you need. Check upscaled images for quality.

Finishing up
Complete the captcha and choose "Convert" or "Download". Open the saved image in something like Paint.NET to view it losslessly and check to make sure the noise has indeed been removed in a way you are satisfied with (or that an upscaled image has good detail and is faithful to the original or whatever); with Paint.NET or a similar program, you can easily zoom in on an image to see pixel-by-pixel details and make sure noise is actually gone. If it's all good, you are done!

I may make a future post discussing removing the background from an image.