Support Makina69 on Ko-fi! ❤️ - Ko-fi ❤️ Where creators get support from fans through donations, memberships, shop sales and more! The original 'Buy Me a Coffee' Page.
Experimental lora to generate gundam-like mecha. Use at your own risk.
Recommended base model: Mechamix
Primary trigger word: gundam
Models keywords:
astray
barbatos
sinanju
gundam_freedom
strike_freedom
gundam_god
exia
crossbone
rx78-2
gn00
sazabi
sentinel
gundam_justice
strike_freedom
aerial_wing
gundam_diver
hi-nu
sword_strike
full_armor
unicorn_destiny
astraea
bael
marchosias
palm tree
gundam_testament
gundam_strike
shenlong
qant
lindwurm
gundam_wing
cloudy sky
Description
FAQ
Comments (5)
Hey, mind sharing your training settings, for dataset repeats & pics numbers ? Thank you, awesome stuff
51 pics 60 reps
"learning_rate": 0.0001,
"unet_lr": 1e-04,
"text_encoder_lr": 5e-05,
"warmup_lr_ratio": null,
"unet_only": false,
"net_dim": 128,
"alpha": 64.0,
"train_resolution": 768,
"height_resolution": 960,
"num_epochs": 3,
@Makina69 Thank you
Might also wanna add more pics of the gundam face closeups and add tag ro ot face or something. I wanna do this but too busy. Do share your result with me
@Makina69 I am training on multiple subjects right now. Testing dataset repeats vs epochs. 30 epochs and 3 epochs had not good results for me. For now i think dataset repeats is good at 10, and to increase epochs instead. Maybe 150 to 300 steps per image whatever the number of pics (maybe up to 500 i don't know yet…). I set Batch size 6 but might test 1 or 2.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.