In break apart by alpha just set fuziness to 0. and it keeps the sharp edges i think i just did it and the I looks as pointy as your original.

Yes i used G'MIC>Arrays & Tiles>Extract Objects.
Then I renamed each layer added a space.png layer and aligned all the layers as you see in the XCF file. I snapped to the guide to the bottom and left of the canvas edge (as Graechan did with Strawberry Fields).
The difference i believe comes from Trans "Auto Crop Resize To Selection Height" filter. This creates the exact width of the characters but allows for the entire height of the layer itself. Instead of doing a Layer>Layer Boundary Size> center and set height layer by layer which is very tedious.

Here is 2.1

Okay, here's test version of script with kerning (version 15).

make sure you all caps for text like "GIMP CHAT CUSTOM FONT IS AWESOME"... so that it's explicit and the script doesn't have to try to look for caps if it isn't there. Then this kerning works.
define a vector named for example "AT.png" and define a path start from left of A and ends at about 3/4 of the width A where the T should start.
define a vector named for example "AW.png" and define a path start from left of A and ends about 1/2 the width of A where he W should start...
and now we have kerning.
sample

that's because i still used the old .xcf that i am testing with fuzziness of 100. heheh
here's a better .xcf with vectors AT.png, AW.png, WA.png, AC.png defined



EDIT: AT could be a little wider, but you get the idea.
Notice this one has WALLACE which needs to definition kerning for WA.png
and AWESOME which needs definition of AW.png to get A and Ws closer together.

Great work on the font Wallace!
Rod and Wallace..good catch on the spacing, now we have optional kerning definition.

forgot to define vectors other cases like AV,VA, TA, AY, YE so this font has those

option to custom font script?
I don't know how it can detect which letter is which though.

It wouldn't have to. Just break them apart in order left to right. A user could make a set number of characters on one layer in a list the script knows and run the script. The script would know how many characters there were and name them accordingly. After it broke them apart from alpha. With the left most character as A and paste it to its own layer "A.png" and so on through the list. You could use a call to G'MIC for the break apart stuff. Things like ( " = i ? j ) that have two or more pieces could already be finished. That is how i do mine anyways.

And now we have kerning. Well done! Brilliant!

Cool Beans on the character spacing fix Tran!

I am not sure what you mean by = i ? j that have 2 pieces could already be finished?

this script would be very specific
i mean what if the font included smaller characters, and/or some symbols.

No Graechan, that didn't cross my mind.
If you would like to do this, be my guest.

Well when G'MIC extract objects is used it will extract every object. If you have a j you will have a "." on one layer and the rest of the character on another. So i extract those characters by hand before i run G'MIC.

What if each font file HAD to be certain characters so the script would run them every time?
You could name each small letter or symbol accordingly. Tell the script to see EVERY character as a list.
A , a , B, b, AND make the USER create the font in exactly that list from left to right and so many characters per row. Tell the script to see only that amount of characters per row.
Wouldn't that work?
This would also give you COMPLETE control over characters used in the font. No more "sorry that character is missing" error.

So basically your just
telling the script to see a list.
create every named layer... A.png, a.png ... ect ect

count the characters in the first row ( i am not sure how the script would see them by alpha maybe?)
if the first row character count is over or under ...error
if the first row is okay ...
extract first row and place in layer according to the list.

repeat until the list is finished.

OR just create all the named layers and extract EVERY object according to alpha using G'MIC and place according to row count. G'MIC extracts but doesn't name each layer. In fact sometimes layers are mixed. The Cap A could be at the top and cap Z in the middle somewhere. There is no precise order using G'MIC. Maybe David could fix this? Create a entirely NEW filter only for extracting the alphabet and naming each layer accordingly?