(NOTE: The 1.5v Lora was just me retrainning and adding more images to the training data, so it might not be better than v1)
Worker Drones and Disassembly Drones from Liam Vickers and Glitch Production's show Murder drones, all credit to them. Unfortunately, the drones are extremely dissimilar from most model's perception of a person so you need to crank the lora up to 1.4-1.6 to get any decent results. Use "workerdrone" to get worker drones and "muderdrone" to get disassembly drones. I highly suggest adding "android, robot, robot joints, glowing eyes" to the prompt as well for best results.
I've tried to include as many of the characters as I can in the training data. so I've included their triggers as well. I also tried to get some like stuff the XD facial expression of the disassembly drones, the Absolute Solver Eye, and Error symbols, and drooling oil from the mouth. However, I haven't had any luck with that so give it your best shot if you want.
Description
FAQ
Comments (6)
You should definitely update this with all of the new training data available from the new episodes, I would be willing to donate hardware to the project if necessary.
If I get some free time I probably will.
@MrOreos Nice! Thanks!
When I tried using this LoRA I got a bad lora layer name error, is that a problem with my setup, or the LoRA?
AssertionError: Bad lora layer name: lora_unet_down_blocks_0_downsamplers_0_conv.lora_mid.weight - must end in lora_up.weight, lora_down.weight or alpha
Strange Im 90% sure its your setup. I have never had that error
try to use it as a loha insted of lora
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.