This is my first Lora. I've trained this on Google Colab using a hand captioned dataset filled with screenshots of TF2 maps (without hud or players).

It does not require a trigger word, but tags are recommended:

Try prompting like: "(name of the map), (keywords), fps game"

Adding "Team Fortress 2 map," to the beginning of your prompt may worsen quality, especially for maps that use dev textures, but could possibly improve quality for sandier maps like pl_upward.

Maps it was trained with (use one of these in the prompt): ctf_2fort, ctf_turbine, koth_sawmill, pl_upward, pl_badwater, plr_hightower, trade_mega, trade_plaza, achievement_engineer, achievement_apg, cp_5gorge, tr_walkway, cp_degrootkeep, cp_orange (x3 version)

I've captioned images in the dataset using these keywords:

|grid texture, community map| -> for cp_orange and achievement_engineer

|overcast, wooden logs| -> for koth_sawmill

|payload track/payload cart| -> for pl_upward, plr_highertower, pl_badwater

|medival, castle| -> for cp_degrootkeep, achievement_apg

You can also use these general terms: outdoors, indoors, RED/BLU intelligence room, RED/BLU control point, wooden walls, concrete walls, red brick walls, underground, cliffside, rocks, cactus