I built an RNN to generate new non-existent Minecraft nicknames. The model was trained on over 200,000 real Minecraft player nicknames on the Google Colab platform (as they provide free GPU power and the training took only 15 minutes)

You give the model a starting string and it will generate new nicknames with it that are not in the training dataset.
For example, I gave the model the starting string "cool" and it generated:


cool4uk
cool_YD
coolia
cool_jean994
cool133328
coolride
cool_vilofulson
coolpvp
cool_play3r
cool_maceo
cool_truch
cool207
cool_psycho
coolmen
cool3919
cooldive935
coolete
cool063
coolmancara
coolMS
cooleman998
cool666
cooladoww2012
cool_puggywhhsy
coolcool


If I give it the starting word "star":

stari9903
star24
starvinqx3
stariow169
starboy4
starki
starow
starin
staro_YTA
starlish6
starusza
starvodcraft
starkager
stariost
starios
starAishae
starcool
starinka
starios
star95
star4craft
starXD
starcog
starwFox
starking


Let's try with the single letter "a":

alex52
azyt
ahaven
aPRAYSAY
aPVP
azweartraz14
ainfixelo
asha2024
aPlayer420
agTee352
alteals
a_Skot
aCZerobrine
avel28
akimBrado
avid19
aPROYAND
agrochmilk
aPadryXD
apusta2282
ahwolf
ammurfrost
axXgamer696
aPBAYYBEAST
ador_NotchDay


The interesting thing about the model is that instead of randomly putting together characters, it takes words and patterns from the training dataset of nicknames and combines them together to produce new somewhat meaningful names.