Author Topic: Torrent Descriptions: An Advanced Guide  (Read 14249 times)

Offline kureshii

  • Former Staff
  • Member
  • Posts: 4509
  • May typeset edited light novels if asked nicely.
Torrent Descriptions: An Advanced Guide
« on: May 03, 2009, 06:40:38 am »
This thread is meant as a walkthrough-of-sorts for uploaders who are concerned with the finer points of making descriptions that are simple, effective, load quickly and create a good impression on readers. I'll be starting this off with a series if guides, to be updated over the course of the week. Contributors are welcome as well. Please do provide feedback and corrections on the guides if you have any.

I am not familiar enough with HTML or CSS (Javascript neither) so I will not be giving in-depth descriptions or code examples of them. I would suggest you look up more well-written HTML tutorials or references; I personally always have CSS Reference from W3 Schools open in a tab when making a description, and find it an immensely helpful reference page. For HTML parameters I just do a quick Google.

Here, I list some of the standardisations I use in these guides:
  • Image dimensions are always listed in width-x-height format. 500x800-pixel image means an image 500 pixels wide and 800 pixels high.
  • The suggestions made here can be carried out using Paint.Net. This is my image editor of choice for this set of of guides, so if you prefer a guide for a more advanced image editor you'll have to do some Googling. If Paint.Net can't do any of the below-mentioned things, I'll mention it in some way (or if I don't, let me know).

Those who are interested in doing more reading, might want to take a look at this image optimisation guide from the YUI Blog.
« Last Edit: July 07, 2009, 09:06:26 am by kureshii »

Offline kureshii

  • Former Staff
  • Member
  • Posts: 4509
  • May typeset edited light novels if asked nicely.
Re: Torrent Descriptions: An Advanced Guide
« Reply #1 on: May 03, 2009, 06:41:32 am »
Part 1: Image-resizing


What is image-resizing
Simply put, image-resizing is changing the size of an image from its native ("normal") dimensions, either by shrinking it or enlarging it.


Is this image resized?
How do you tell if an image has been resized? Most browsers will tell you in some way, in the object's properties window. For instance, right-clicking on an image in Firefox and selecting 'Properties' displays this window:



That is what you will see when you pull up Image Properties for a resized image (not being displayed at its native resolution), in Firefox.


Image-resizing in descriptions
For description-making purposes, image-resizing in the browser is not ideal. The image-resizing algorithms (formulas/equations) used by most browsers are optimised for speed over image quality (I'm sure you have all seen some horribly-resized images at some point in time), and result in less-than-ideal resized images.

Worse still, despite these optimisations, re-rendering these images as you scroll seems to put quite a load on some web browsers (particularly Firefox), causing what many would know as 'laggy scrolling'. I don't know what causes this, and it would be lovely if someone could shed some light on the matter.


Alternatives
As far as we're concerned for the purposes of description-making, resizing images in the web browser is not something we want to do. Instead, always display images at their native resolution (images inserted with <img src="[...]" /> tags always display at their native resolution).

For preview purposes one might resize images in the sandbox (by setting width or height parameters in the <img> tags) to test out different image sizes in a description. However, for the actual description, proper image resizing should be carried out in an image-manipulation program, saved, and then uploaded to the image-host for display.


Image-shrinking in screenshot thumbnails
The more sharp-eyed among you might have noticed that some of the screenshot thumbnails are, in fact, resized. This is because the screenshot feature produces 150px-wide thumbnails by default, but not all image-hosts create their thumbnails at the same width (in particular, Photobucket makes 160px-wide thumbnails).

This can be fixed by using [screenshot=160] in the opening screenshot tag (replacing '160' with an appropriate number representing width in pixels where appropriate). Doing so will not make those thumbnails any prettier, but it helps a lot in eliminating 'scroll lag'.


Shrinking large images
In most cases, resizing large images without any additional image-processing should produce a smaller image of good quality, but in some cases this might produce sub-optimal images. As an example, take this image from Konachan, a lovely render of Horo on a detailed background of vegetation. Let's resize this in Paint.NET without any special effects.

Here is the image resized to 50% (in both dimensions).
And here is the image resized to 25% (in both dimensions):

Notice how 50% resizing creates aliasing effects ("jaggies") around Horo's outlines, while 25% resizing pretty much kills it outright (look around the thinner edges of her hair).

Thin lines and tiny details are easily lost in resizing large images. Just imagine: in a 25% resize, details in a 4x4 box of pixels are resized to just 1 pixel! That is a lot of detail lost. It is obvious that image information will be lost in shrinking; bigger features shrink quite well since they have more information, but what about fine features (such as Horo's thin wiry outline) that don't preserve well in an image-shrink?

Meet your good friend, the Gaussian Blur. Just about any image manipulation program will have at least a Guassian Blur effect. Those who want the details on how it works can read the excellent Wikipedia entry, otherwise all you need to know is that it blurs each pixel by spreading it out. What this means is that our 2-pixel Horo outline is now wider, although more blurry. But we don't need to worry about a little blurriness, since we're shrinking the image; hardly anyone would notice.

Here is the image with a Gaussian Blur of 2 pixels applied, and subsequently resized to 50% (in both dimensions):
And here is the image with a Gaussian Blur of 4 pixels applied, and subsequently resized to 25% (in both dimensions):

If you compare them to the previous images you can see that Horo's outline is now less aliased ('jaggy'), at the cost of a little blurriness, but it looks much easier on the eyes.

In this case the Gaussian Blur also cost us much detail in the foliage; ideally we can preserve foliage detail by selecting only Horo for the Gaussian Blur effect, but I leave that as a writeup for another contributor.


Enlarging small images
Just don't do it. The results do not look that good (unless you are hardcore enough to find a fractal-resizing algorithm), and... you're better off just not doing it. Try to find higher-quality images, or work with whatever quality you can get.


On Aspect Ratio
Unless you have really good reason to do so (for intentional visual effect, or to consciously fix a bad aspect ratio), when resizing an image you should always keep a constant aspect ratio. This means that when you change the height or width of an image, you should change the other dimension in proportion as well. If you shrink the width by 40%, shrink the height by 40% as well. Do not shrink the width and not the height, just to make the image fit into a rectangle of fixed size! If you have to fit an image to a certain size, the best way is to resize as small as possible with constant aspect ratio and then crop.

One small exception: Tiny un-noticeable changes in aspect ratio can be overlooked. Say, you have a 103x100-pixel image and you want to fit it into a 100x100 rectangle. You could crop 3 pixels from the width, or you could just resize the image to 100x100 and change the aspect ratio by about 3%. Does it matter which one you choose?

Try both; You'll see that you wouldn't really notice the difference. I'll leave the decision to you, just resizing is faster, while resizing+cropping usually gives the best result. Just remember that the process doesn't matter, it is the end result that is most important. People who look at your description won't care how you resized it, as long as it doesn't look bad :) Personally, I've resized 825x819 images to 100x100 without anyone noticing (I think).


Summary
Screenshot thumbnails can be made to display at their native resolution by editing the opening screenshot tag. A large image that produces aliasing on shrinking can be made to produce a softer effect by applying a Gaussian Blur prior to shrinking. Images in descriptions should always be displayed at native size.
« Last Edit: April 04, 2012, 12:22:00 pm by kureshii »

Offline kureshii

  • Former Staff
  • Member
  • Posts: 4509
  • May typeset edited light novels if asked nicely.
Re: Torrent Descriptions: An Advanced Guide
« Reply #2 on: May 03, 2009, 06:41:56 am »
Part 2: JPEG compression and usage


This part is a quick introduction to the salient features of JPEG compression, which will help you decide what compression settings to use, and also when not to use JPEG compression. Those seeking a more detailed explanation can refer to the well-written Wiki article.


Lossy Compression
Lossy compression simply means that as you compress the file, you throw the less important parts of it away, keeping only the important information. In some cases the human senses might not be able to notice these differences.


How does JPEG achieve such small filesizes?
By throwing away information, and lots of it. This is usually a bad thing, but bandwidth on the web is limited, so if all our lossy-image formats were to be replaced by lossless-image formats, the amount of data we have to download will be increased by almost 3 - 10 times (depending on image size; Bigger images compress better than smaller ones). So a compromise has to be made somewhere. If you wisely choose your compression settings, you can achieve lossy compression with minimal loss of quality.

JPEG achieves its compression mainly through 2 features in its compression pathway:
Downsampling
Quantisation


Downsampling
Typically, JPEGs are first converted from RGB colour space (each pixel colour represented by a red, a green and a blue value) to a colour space called YCbCr (colours represented by 1 brightness value and 2 colour values). Y is the brightness value, while Cb and Cr represent blue and red colour components.

This is done because the human eye is better at perceiving changes in brightness, and worse at perceiving differences in colour. Effectively, this means you can reduce information in the colour channels and the human eye is less likely to notice. And this is what happens in a normal JPEG image.

Normally, one Cb and one Cr value is specified for each pixel (no downsampling, a.k.a. 4:4:4 downsampling). To reduce the amount of information, the number of Cb and Cr values may be reduced, such that every 2 pixels share one Cb and one Cr value (4:2:2 downsampling). If desired, this can be further reduced, to letting 4 pixels share 1 Cb&Cr value (4:2:0 subsampling). This effectively approximately halves the amount of information, in a manner that is nearly imperceptible at 100% zoom. [Y channel untouched, Cb channel reduced to 1/4, Cr channel reduced to 1/4, remaining information is 6/12 or ~1/2]
(More information on J:a:b chroma downsampling can be found on Wikipedia)


Quantisation
This is based on a mathematical theory (see: Fourier Analysis) that irregular patterns can be broken down into a sum of regular wave patterns. Not just any wave patterns, but special ones (see the Wiki article for mathematical details).

At this point, each channel (Y,Cb,Cr) of the image is broken up into 8x8 blocks of numbers, each number representing the channel value of the pixel at that point. If we treat this 8x8 block as an irregular pattern, we can break it down into a sum of patterns. In the ideal world, we'd represent this pattern as a sum of many many regular wave patterns. The more patterns we use, the more accurately the pattern can be represented, although the change in accuracy becomes more and more unnoticeable as we use more and more patterns. Since we are filesize-limited, JPEG uses 64 wave patterns, represented by 64 squares as shown below:


Image Licensing: Public Domain

This magic is called the Discrete Cosine Transform. I won't go into detail about it, but essentially what you need to know is that most image information happens to be concentrated in the upper-left corner (i.e. those patterns make up most of the image information).

The theory goes that the human eye is better at discerning changes in colour and brightness in coarse patterns than in fine patterns. This means we can throw away most of the patterns in the bottom-right corner, and pretend they play no part in the image. In most cases, the effects are unnoticeable at a glance.


When to use/not to use JPEG compression
For photographs
JPEG works well on images with messy patterns, or a broad range of patterns. It works amazingly well for photographs (most people won't even notice 50% compression).

Patterns
JPEG carries out block-splitting (into 8x8 blocks). This makes it unsuitable for patterned images, unless they repeat in an 8-pixel pattern. To demonstrate, take a look at the following 24x24px patterns:

     

The same patterns after compression to 0% quality:
     

While 0% quality is an exaggeration that isn't likely to occur in real life, the banding patterns you see when images are not 8-pixel aligned can occur at as high as 50-70% compression, and may or may not be more noticeable depending on the pattern contrast. In short, patterned images should either be minimally compressed, or compressed losslessly (using a lossless format, e.g. PNG)

Line art and vector art
Images with fine lines, or highly contrasting regions (e.g. black circle on white background, or a block of blue on a yellow background) create clear lines that do not compress well in JPEG. Since it is difficult to break down a line as a sum of patterned blocks, going lower than 90% compression creates very visible artifacting ("messy/noisy pixels") around those lines. If you have such an image, try compressing it at a high quality setting (>95%), or with a lossless compression format.

Text
For the same reasons outlined above, text does not compress well in JPEG. However, from personal experience, text still looks all right if compressed at ~95% quality or higher. Go any lower and you will start seeing artifacts in the text background.

Because of the way block-splitting is done (in 8x8 blocks) and the way the human eye perceives images, artifacting from JPEG compression affects large-sized text less obviously than it does small-sized text.

Most other images
JPEG at 88-95% quality setting works well for most images. Some compress better, and can be saved at as low as 70% compression. Others may need 96% or higher quality.

Small images
Small images (~100x100px) can be compressed as low as 80% without artifacting being noticed, since image details are small to begin with and hide the artifacts well. They can be further compressed to 70% or lower if you don't mind minor artifacting.

This is assuming there is no text in the image, otherwise see the caveats mentioned above for text, and apply plenty of judgement-by-eye.

Saving images for further editing
If doing further editing on an image, NEVER save it as JPEG. Each time you do a JPEG save, you throw away more information on the image. While this repetitive information loss is not too drastic if saving at high quality settings (most of the information is already lost when making the first JPEG, so further saving as high-quality JPEG throws a little information away each time), you want to retain as much information as possible in an image prior to the final saving.

Of course, balance this with your need for convenience. If you are only going to make minor image edits, and the image does not need to be high-quality, a high-quality JPEG can be used as an intermediate file format.


Controlling JPEG compression
To control your JPEG-saving settings, you first need a program that supports it :)

Most image editors will at least give a quality setting for you to choose how much information to discard at the quantisation step. Much of the compression occurs in the 90-100% quality range; dropping the compression from 100% to 90% can reduce the image filesize by up to 50%. After that you get diminishing returns, so strike a balance between image quality and filesize.

The more advanced image editors give you more options. I'm not a Photoshop user (someone else has to do those screen-captures), so here's a sample of what the GIMP JPEG-saving dialog box looks like (with Advanced Options expanded):



A quick description of the options:
Show preview in image window:
This let's you see the effects of different compression settings on the image (and also calculates the resulting filesize). Leave it checked.

Optimize:
As far as I know, this carries out further lossless compression on the remaining image data. It usually gives you a smaller filesize, so you can leave it checked. It is largely unnecessary when using the Progressive option.

Progressive:
This specifies a different method of storing the numbers generated in the quantisation step (see the JPEG Wiki for details). It also usually gives you a smaller filesize, so you can leave it checked.

Save EXIF data: Save extra information about the image. This adds about 0.1kB to the image, so it's not an issue.

Save thumbnail: Make smaller versions of the image and attach them. These are used for previewing in your OS, or on the web.

Smoothing: Apply smoothing to the image. I don't use this.

Subsampling: Select the type of downsampling to use on the colour channels. 4:2:0 downsampling (smallest file) is recommended unless you really have use for the extra information retained by 4:4:4 downsampling.

DCT method: Integer is good enough for most purposes, but you can set to Floating-Point if you are a quality freak. The difference is minimal at most, and imperceptible for the most part.
(For those interested, this just describes the method used to perform the Discrete Cosine Transform magic. See the JPEG Wiki for more details)


Filesize comparison
PNG
2.06MB

JPEG (90% quality)
792.4kB (Optimise off, Progressive off)
732.8kB (Optimise on)
690.6kB (Progressive on)
690.6kB (Optimise on, Progressive on)
930.2kB (Using 4:4:4 instead of 4:2:0 downsampling)
Thumbnail adds 2-3kB

JPEG (100% quality)
1709.2kB (Optimise off, Progressive off)
1262.9kB (Progressive on)

JPEG filesize-vs-quality chart

via softwarelivre.org

Summary
In general, JPEG works for most images, and works especially well for photographs and small images. It does not work too well for text, line art and vector art. Do not save as JPEG if the image is undergoing further editing.

This concludes the JPEG compression guide.
« Last Edit: June 18, 2013, 01:08:20 pm by kureshii »

Offline kureshii

  • Former Staff
  • Member
  • Posts: 4509
  • May typeset edited light novels if asked nicely.
Re: Torrent Descriptions: An Advanced Guide
« Reply #3 on: May 03, 2009, 06:42:47 am »
Part 3: Designing a Description


In Part 3 of this series, we discuss the design of a description. I'm not going to try to preach about what makes a good description (apart from adherence to description requirements), because frankly I don't know that myself. But I would like to raise some considerations that I hope pass through the minds of would-be templaters and description-makers.


Design Philosophy
Different people do different things with their designs, and hence they tend to end up with different approaches to doing things. Here at Boxtorrents, the torrent description system is designed to give users more flexibility over their description page designs. This gives more opportunity for creative expression, but also results in many different approaches to description design that are sometimes inconsistent.

If there is one design philosophy that all torrent description templates should follow, it should be Readability above all else. A plain, readable description is better than a pretty image-studded description with difficult-to-read torrent information. At a glance or quick scroll, the reader should be able to see what the torrent contains (maybe not in full detail, but at least a quick overview). If you are squinting to read the album title or fansub group information in your description preview, maybe you should rethink that font colour or background you are using.

This also means that the user should not have to click too many times to get to important information in the torrent. All information should be easily accessible with one click. Links to such information (if using javascript to create pop-up information boxes, for instance) should be easily seen and clearly labelled, and should be big enough to be easy to click for the average user.

If using a listing of some sort, each list item should be clearly labelled. If images are used as links, users should be able to tell what information that link contains, so such image-links should either be accompanied by text-labels, or be self-descriptive.


Layout
A good layout should not be too cluttered, but should not have huge tracts of empty space either. For instance, do not cluster all torrent information to one side of the screen if the other side is otherwise empty.

Space out your content, and lay it out in a manner that is easy to read and does not require the reader's eyes to jump about the page too much.

The torrent title is usually placed at the top of the page, but you can try out more interesting positions for it, although this should not compromise on readability.

Borders are useful for separating your description into individual blocks that make it easier to focus the reader's attention. However, over-use of borders can create a divisive effect that makes information seem to be scattered all over the place.

To explain this more clearly, borders give the reader the sense of organisation; if the reader sees 4 bordered boxes, he gets the mental impression of there being 4 parts to the description. For a small number of bordered elements this is helpful, but if you have more than 12-or-so bordered boxes, it creates a disorienting effect, especially if the bordered elements all small-sized. Which box should the reader focus on first?


Contrast
The human eye is better at perceiving contrasts in lighting than at perceiving differences in colour (see JPEG Compression in Part 2). Therefore, a good description should make use of contrast rather than colours to separate parts of the description. However, some colours contrast well and can be used to do this too.

A good example of this would be putting background images in descriptions. While this gives users a pretty background to look at, it also (in most cases) spoils the advantage of contrast afforded by a plain white background. Images placed on a background are difficult to differentiate without sufficient contrast with their surroundings, and text is harder to read as well. To alleviate this, the use of background images is discouraged unless there is good reason to do so.

If background images must be used, they should be carefully picked to retain good contrast with the description content. An easy way to do this is to apply a fade/watermark filter to the image to lighten it. Text can be made readable by placing them in boxes with an appropriate background colour.


Focal Point
In most descriptions, the focus is on the torrent information. However, aspiring description-makers might want to bring the audience's attention to certain aspects of the anime that aren't described in the provided information. To do this without compromising on readability, a focal point should be created. For instance, an image from the series/movie could be placed in the middle or slightly off-center, with torrent information placed around the image in an easy-to-read manner.


Design Blocks
If possible, try to create your description based on design blocks. For anime torrents, these design blocks are typically Anime Information, File Information, Plot Summary, and an image. Grouping torrent content this way makes it easier to rearrange elements and try out alternative layouts. For music compilations consisting of many albums, try grouping the albums into common themes instead of putting them all in a list, or combining them into a grid layout.


Fonts
After making a few descriptions, description-makers may be tempted to try playing around with fonts. This is strongly discouraged, since one cannot guarantee that all PCs will have the same font (if using an uncommon font). Stick to "web-safe" fonts such as the ones listed below, or others that you may know of. Test these fonts at various sizes (up to 5 points larger/smaller) to make sure they display properly and are easy to read (not everyone uses the same font size).

If you absolutely must use a particular font you have that isn't widely available on most computers, creating it as an image is the best way to ensure it displays the way you want it to.

Recommended sans-serif fonts:
Arial (standard sans-serif font)
Verdana (easy to read at small font-sizes)
Helvetica
Lucida sans

Recommended serif fonts:
Times New Roman (standard serif font)
Georgia (easy to read at small font-sizes)
Lucida


Formatting and paragraphing
A chunk of text without any formatting or paragraphing is a huge pain to read. Do not do this.

Bold text draws attention (especially when coupled with a slightly larger font-size) and should be used for headings, or first-column labels.

Italicised text is usually used in paragraphing for emphasis, but is acceptable formatting for subheaders as well. Try not to abuse it too much.

Increasing the font size also has the effect of drawing the reader's attention, but also makes the text easier to read. Definitely use this for headers, and if possible also for anime/file information.

CSS 2.0 has a few more formatting properties that enterprising description-makers might want to take a look at. In particular, the line-height property allows one to control the line-spacing, which makes long plot summaries easier to read. The letter-spacing and word-spacing properties are also useful for spacing out letters/words at smaller font-sizes, where they tend to clump together.


Theming
To create a sense of consistency in the description, the description-maker might want to coordinate the colours of page elements, such as hyperlinks, image/division borders and text colours. Here, colour-picking software is of immense aid (kureshii uses ColorArchiver, although it is by no means the best or the only colour-picking software). Try picking colours from images used in the torrent, or colours that contrast well (remember the Design Philosophy).


Works in progress
While we understand that certain special torrent descriptions might take more time to tweak and code, the reader should not be forced to suffer a template-in-progress when he/she visits the torrent page. If you have to, create a separate, plain description for readers until your prettier template is tested in the sandbox and ready for use.


Summary
All descriptions should follow a "Readability above all else" design philosophy, and page elements should be arranged to make information as easy to read as possible.
« Last Edit: April 04, 2012, 12:26:40 pm by kureshii »

Offline kureshii

  • Former Staff
  • Member
  • Posts: 4509
  • May typeset edited light novels if asked nicely.
Re: Torrent Descriptions: An Advanced Guide
« Reply #4 on: May 03, 2009, 06:43:14 am »
Part 4: Preparing Images for Your Descriptions


In Part 4, we discuss how to prepare images for use in torrent descriptions. While it's easy to just pull a random image off Google Image Search, a little discernment can help improve images a lot, and spending a bit more time can make your torrent description look simply amazing with the right images.

Note: All JPGs were saved at 95% quality. Filesizes of images should be further reduced when being used in images, as per Part 2: JPEG compression.


Hosting Images
Before talking about preparing images, you will first need to find a place to host them. Photobucket, Imageshack and xs.to are just some of the places where your images can be hosted (most Boxtorrents users host their images on one of these 3 providers). When signing up for an account, take note of how much storage space and bandwidth you are given. If you are planning on making many uploads, this might become a deciding factor.


Getting Source Images
We begin this guide with a short note on finding good source images. The best sources are, of course, high-quality scans. There are many sources for scans, they should be easily found on sites like AnimePaper, or on imageboards such as danbooru. Try to find a good image that:

1) is at least 1000 pixels along one of the dimensions
2) does not have visible artifacting ("messy" pixels along edges, and other imperfections)
3) is not terribly scanned (i.e. not too washed out or too dark, does not have scan-lines, etc)

If you're making a description for an OST collection, check if it comes with album scans. If it does, those should be a good source for images.

If you can't find scans to work with, wallpapers or user-made vectors are another source to check. These are usually found in the same place as scans, and the same criteria applies to them as well. Try to use images that have not been processed too much.

As a last resort, if you really cannot find a scan or wallpaper to your liking, you can try image search engines (such as Google Image Search). Find the cleanest image you can, and make sure it is of a decent size (~300x300 or larger). You do not want to be caught in a situation where you have to upsize your images.

Note: Always check your sources. If they say you can use the image only by giving due credit, then do so. It's best if the image is public domain or use-with-due-credit.


Note 1a: Preparing Images from low-quality sources
For low-quality JPGs that you find off image search engines, the general rule of thumb is to do as little processing as possible. These images have had a lot of visual information thrown away, and the more processing you do to it, the more information you discard. Pretty much the only processing you want to do for these is some tone adjustment (see below) and cropping.


Note 1b: Preparing Images from high-quality sources
For high-quality scans or vectors, we have plenty of information to work with. We want to highlight pertinent features in the image, and hide or de-emphasise other parts of it. We begin by resizing the image to a size that's easier to work with.


Preparing Images for Processing
Needless to say, the first thing to do is to crop away parts of the image you don't need. This is easily done by selecting the part of the image you want to keep, and then selecting Image -> Crop to Selection (GIMP or Paint.Net).

Most scans will come at around 2000x2000px; this is a little too large for us to work with, so we begin by resizing to about 1000x1000px (see Part 1: Image Resizing).

A general rule of thumb is to resize to 2 or 3 times the size of the final image. If your final image is 200x300px, then it is useful to resize to 400x600px to work with. This makes it easier to see roughly what the final image would look like, and also gives you some room for hiding artefacts and other imperfections that come with image processing.

Some examples done using Paint.Net:
Example 1 - Kanon Original Soundtrack: Before -> After
  Gaussian Blur (Radius 2), resized to 1000x1000px
Example 2 - Clannad Afterstory OP/ED Single: Before -> After
  Noise Removal (Radius 3, Strength 0.3), Gaussian Blur (Radius 2), resized to 990x1000px
Example 3 - Air Movie Soundtrack: Before -> After
  Gaussian blur (Radius 2), resized to 500x378px

Once the image is at a good size, we can begin with the fun stuff. In the rest of this guide, I will be using Paint.Net as my main tool, using GIMP occasionally to illustrate some points.


Adjusting for colour
Have you ever seen descriptions with multiple images, where some of the images are darker while others are lighter, some are more faded while others are richer in colour? We'll try to eliminate that using the 3 methods outlined below, so that the images in the description all look consistent.


Part 1: Levels
At this point, I'd like to introduce you to the Levels tool. you can find him under Colors -> Levels (GIMP) or Adjustments -> Levels (Paint.Net). He's a less advanced, but easier to use version of his brother, Curves. I'll be working with the Kanon OST image. Nice scan, but we can do some stuff with it to make it look nicer in a description.

This is what the Levels tool interface looks like in Paint.Net:


It looks like a scary curve, but it's pretty easy to interpret. Each pixel in the image is represented as a mix of 3 colours: Red, Green and Blue. Each colour is represented by a value from 0-255.

The Input Histogram is a count of the number of pixels for each given value. The top of the curve represents pixels with value 255, and the bottom represents pixels with value 0. The 3 coloured curves, of course, represent values for each colour channel (R, G or B).

We can easily see that the peak occurs at about value 238. This means that we have a lot of pixels with red, green or blue values near 238. This is not surprising, as most of the image is near-white, thus having RGB values around value 238.

We can improve this image by making the background full-white so it looks cleaner. Notice that there are 2 number-input boxes at the top and bottom of the curve, and corresponding indicators (the small black triangles). Let's say we drag the top slider down, or give it a value of 226 (use the up/down arrows in the input box, or type '226' and press 'Tab').

What you've effectively done is tell Paint.Net that you want all colour values that are 226 or higher, to be changed to 255 (the largest value) instead. Colour values from 0 to 266 are adjusted (stretched) accordingly so that they now fill the scale from 0 to 255. The result can be seen in the main image window, and in the histogram to the right.

Because we increased the colour values of all the pixels, the image now looks a little bright-ish. We can darken the dark parts of the image by moving the bottom indicator slightly upwards, or typing '10' in the bottom box.

This is what the Levels interface will look like after inputting the values:


Click Ok and your changes will be saved. the image now looks much brighter, although also a little lighter.

Paint.Net's levels display doesn't tell the full story. Let's take a look at GIMP's Levels tool (before, followed by after):
 

Notice the white gaps in-between the black curves? Those come about because we're scaling 213 (226 - 10 = 213) colour values to fit 256 values. Since colour values can't be stored as decimal numbers, the numbers are rounded off, resulting in gaps. This is where we lose colour information; at those white gaps, we will never know how many pixels were supposed to have that value. This is the tradeoff we always have to make when processing images.

This is the end result of level adjustment (followed by more Gaussian blur, noise removal and 50% resizing):



Part 2: Saturation
I'll use the image from the Air Movie OST to demonstrate what saturation can do for you. It's another lovely image that doesn't really need any work. but let's say we want to bring out the golden colour of Misuzu's hair; In this case, Saturation is the tool you're looking for.

This is what it looks like:


You can find Saturation under Adjustments -> Hue / Saturation. By increasing Saturation by 10%, we see that Misuzu's hair now looks richer and more golden. But as a result, her scarf/tie looks really red now, and some parts of the clouds in the background have now taken on a somewhat-odd tinge of blue. We could, of course, saturate only Misuzu's hair by selecting only that part of the image for filtering, but I will not cover selection and more advanced techniques here (look to Google and more professional sites for that).

This is the end result of bumping up Saturation by 10%:

The picture was intentionally blurred slightly, to hide some imperfections. It would look just fine with better focus too actually.


Part 3: Hue
Nagisa looks lovely in this scan from the Season 2 OP/ED single, but her skin looks kind of pale. Let's give her a healthier skin tone using Hue.

Hue is the companion tool to Saturation, and they can be found in the same place, Adjustment -> Hue / Saturation. This is what Hue tool interface looks like:


By moving the Hue point left by 5 points (value of -5), we make the image redder, thus giving Nagisa a rosier skin tone. Shifting it left gives it a yellower colour, making her appear jaundiced. Try the full Hue range and see what colours you get.

This is the end result of reddening the Hue by 5 points (followed by more Gaussian blur, and a 50% resize):



Summary
Once you've started making more than a few descriptions, you will start to have ideas of what you want your description to look like. You might want to adjust image colours and properties to better suit coloured backgrounds, or to make a group of images look more consistent in tone. This part of the description-making guide covers very basic techniques that you can use to do so.
« Last Edit: April 04, 2012, 12:32:56 pm by kureshii »

Offline kureshii

  • Former Staff
  • Member
  • Posts: 4509
  • May typeset edited light novels if asked nicely.
Re: Torrent Descriptions: An Advanced Guide
« Reply #5 on: May 09, 2009, 07:53:44 am »
Part 5: Optimise Page-loading
« Last Edit: August 04, 2009, 11:21:44 am by kureshii »

Offline kureshii

  • Former Staff
  • Member
  • Posts: 4509
  • May typeset edited light novels if asked nicely.
Re: Torrent Descriptions: An Advanced Guide
« Reply #6 on: May 09, 2009, 04:25:03 pm »
Part 6: Placeholder

Offline kureshii

  • Former Staff
  • Member
  • Posts: 4509
  • May typeset edited light novels if asked nicely.
Re: Torrent Descriptions: An Advanced Guide
« Reply #7 on: May 09, 2009, 04:25:11 pm »
Part 7: Placeholder

Offline kureshii

  • Former Staff
  • Member
  • Posts: 4509
  • May typeset edited light novels if asked nicely.
Re: Torrent Descriptions: An Advanced Guide
« Reply #8 on: May 09, 2009, 04:25:20 pm »
Part 8: Placeholder