It should give your subject a nice trench coat.Version 1: 25 images (resized to 1024 wide), some original image where smaller, rank 32, alpha 32, epochs 200, steps 5000, wrong captioning kinda fine tuned the concept 'trench coat', trigger word xyz_tc had litttle to no effect use 'double breasted trench coat' insteadVersion 2: 25 images (roriginal size images), some original image where smaller, some larger then 1024 wide, rank 32, alpha 32, epochs 200, steps 5000, wrong captioning kinda fine tuned the concept 'trench coat', trigger word 'xyz_tc' had litttle to no effect use 'double breasted trench coat' insteadVersion 3.2: 48 hi-res images (some extremely large), rank 128, alpha 64, epochs 400, steps 19200, new training captioning, trigger word 'xyz_tc' works. As it learned the trench coat as a new concept 'xyz_tc', it isn't entirely perfect yet. Problems with placement of buttons, multiple buckles, etc.. may occure. 'Trench coat' is what flux-dev model knows and 'xyz_tc' is what the lora knows about the trench coat... So, use xyz_tc to describe your trench coat.You can describe your trench coat some what. Like black , horn or same color as the xyz_tc buttons and buckles. belt buckled or tied Button or buckled wrist cuffs, epaulets with a smaller button, pockets with a button closure, the color of the xyz_tc like black leather, pink, or even a black and white zebra print.

Trench coat flux

<ol><li>New dataset of 48 hi-res images</li><li>New training captions (xyz_tc should work now)</li><li>Network rank 128 / Network Alpha 64 should half the learning rate</li><li>400 epochs</li></ol>

Files with SHA256: 26b333dcd34aaace1494988243ad4731ca7cc275f3e984343870ed8f0b15e308