In CS2, Karambit Gamma Doppler Phase 1s have a mix of green, purple, and blue throughout the blade. Diamond Gems are Phase 1s that show a predominantly blue color throughout the blade. They stand out as highly unique with their strong crystal blue color.

Diamond Gems were initially popularised by Slaid, a youtuber who made a fantastic guide about them. I highly encourage you check out his work, he really set a high standard for guides in CS2.

CSFloat screenshots comparing my rank 1 (p547) to one of the worst patterns.

In this guide, I've taken Slaid's idea of a diamond gem, and tried to recreate it through color data analysis. Slaid's work on the diamond gems was incredibly well done, so I wouldn't want to just call this an 'improvement' - that's up to the community to decide.

Every pattern was ranked by it's blue percent - the amount of the blade that can be considered blue. I managed to find every pattern's blue percent with extreme accuracy.

To be clear, this guide ranks the patterns in terms of the 'flawlessness' of their blue colour. It does not factor overall colour, wear, dark spots, positioning etc.

There's a detailed section describing my method at the end of the guide.

I've decided to present the ranking purely as a ranking, without cutting it into tiers.

Doppler patterns are so similar to each other that it's very difficult to draw any line separating them. There will never be a significant difference between the best T1 and the worst T2. My program created a ranking of all patterns, entirely from data. I think reducing all of that information into a simple Tier 1 or Tier 2 is only removing detail. Tiers are supposed to be a useful simplification, but instead for me they were the only measurement of how good a pattern is, undermining my approach because it reduces all the valuable detail down to simple groupings.

Another huge reason is that diamond gems already have an existing guide and demand. When I first released my guide, I made tiers and got a lot of backlash (mainly because I made Tier 1 unnecessarily exclusive). The tier boundaries had far more impact than the actual data, even though they are completely arbitrary. While this is partially because of my mistake, I feel that any tier system I make will take away from my actual work.

I'm only speaking about how tiers affect this guide, not other people's guides. This is all because of my data-based approach.

If you really want tiers, I would suggest the following:

Tier 1 - patterns with more than 97% blue
Tier 2 - from 97% to 94% blue
Tier 3 - from 94% to 90% blue
BTA/Tier 4 - from 90% to 85% blue

This is purely a suggestion and I would prefer if someone talked about a pattern by it's rank or it's blue % when referencing my guide.

The full ranking is here: https://pastebin.com/xnGpNkrW

You can also use this youtube video to see where the program identified flaws in each pattern. Change the speed to view individual patterns easier.

• I don't want to simply call this guide an 'improvement' on Slaid's ranking. There are definitely going to be many areas where his is better. I would like to keep working on this, to create something that measures many factors, to create a really perfect ranking. I know it's possible, but eventually I have to move on. At the end of the day, people will always pick their favourites themselves.
  
  
• This program is not a perfect list, unlike the Karambit Max Pinks, there is lots of subjectivity around dark spots, wear, severity of flaws, and shade of blue that it doesn't account for. However, I believe this is a very strong basis for consistency of blue, the primary factor.
  
  
• I will include the full ranking of every pattern at the end of the guide. If you want to get a diamond gem without going broke, I am sure there are some banger patterns beyond the top 200.

Huge thanks to:

- Slaid for his amazing work and help with writing this guide (https://www.youtube.com/@TheSlaid)
- David from CSBlueGem for the screenshot generation method (@CSBlueGem on Twitter)
- @VqltUK for the cover image and help with the guide
- @r4vendel, @Vancho666_CS for help with the rankings/guide

This guide required me to completely rebuild every part of the process, from the screenshot generation to the data analysis, because diamond gems have so many factors to consider.
Throughout this process I went through more than 15 different versions of the program.

Screenshots were generated from in-game inspects, using a method shared with me by David from csbluegem (Check out his twitter @CSBlueGem)

These screenshots were then cropped to pictures that only contain the pixels from the blade of the karambits.



After some testing, I realised my previous method of measuring average RGB value wouldn't work on the diamond gems, because they are much more about consistency of blue color instead of the average color. I had to do actual color value analysis. So I moved to HSV colorspace, which can represent the Hue of a pixel in a single value.

In HSV colorspace, I created filters that would determine which pixels were blue. These had to be extremely accurate because the line between light blue and green is almost imperceivable in the screenshots. This took a lot of fine-tuning to get extremely accurate. In-game, certain colors can only be seen under the correct lighting conditions, but these filters can always detect what colors are present.


*There is no point in float values of hue, it just happened because I initially designed the filters in opencv, where hue scaled from 0-179, then impemented in PIL where hue is 0-255.

The line between 'crystal blue' and green is extremely fine, but an important distinction in the guide. To deal with this, a different filter was applied to that area, that removed any pixels with a Value below 187 (vMin=187). This basically separated that range by their brightness, which removed all of the muddier green but kept the good crystal blue.


Many patterns show a kind of 'lavender' color between blue and purple. I classified this as blue, and it was allowed through the filter.



Patterns 773 and 497 before and after the filter.

I used these filters to measure what percent of each blade can be considered blue, and then ranked each image by it's blue%.

Note - while some of these colors might not seem 'blue' in the screenshots, when evaluated in-game under different lighting conditions, I think the filter will prove itself accurate.

I feel it's best to try to 'perfect' the single most important measurement rather than having multiple conflicting ideas.