I present my Jonas Balkar (SWTOR) LoRA.
I initially wasn't happy with V1 so I removed it completely and retrained other versions to come up with a more consistent model. I've since learned quite a bit more about the training tags so I have made alterations and added a handful more training images.
V4 (clip skip 1) I refined a bit more, both in terms of the training tags and I got more selective/specific on the training images. This version generates very nicely and is more versatile with various style LoRAs. I recommend using an adetailer to clean it up if the shot is rather zoomed out.
V4 basic prompt tags are as follows:
jonasbalkar, 1boy, black hair, blue eyesV4 outfit tags:
shoulder straps, brown jacket, white shirt, leather bracers, brown belt, striped blue pants, bootsV4 optional style tags (that I trained):
realistic (use this if you wanna increase detail a bit)photorealistic (you can add this for realism shots)My generation testing for V4 was done using Forge on about 3-4 CFG (for NSFW) and 4-5 CFG (for SFW) at 20 steps (DPM++ 2M SDE - Karras). I also used an adetailer for my testing with this version.
V3 was trained on clip skip 1 and had several NSFW training images added to the set to give it more overall range. This version generally renders nice faces but I recommend using an adetailer to clean it up.
V3 basic prompt tags are as follows:
jonasbalkar, 1boy, black hair, blue eyesV3 outfit tags:
shoulder straps, brown jacket, white shirt, leather bracers, brown belt, blue pants, bootsMy generation testing for V3 was done using Forge on about 5-8 CFG (the sweet spot seemed to be about 6-7 CFG but sometimes 5.5-6 when doing NSFW with some style LoRAs) on 30+ steps (DPM++ 2M SDE - Karras). I also used an adetailer for my testing with this version.
V2 was trained on clip skip 1 using only SFW training images (but still makes excellent NSFW renders).
V2 basic prompt tags are as follows:
jonasbalkar, 1boy, blue eyesV2 outfit tags:
shoulder straps, brown jacket, white shirt, belt, jonaswrist, pants, bootsMy generation testing for V2 was done using Forge on about 6-8 CFG with 30+ steps (DPM++ 2M SDE - Karras).
Description
V3 was trained on clip skip 1 and had several NSFW training images added to the set to give it more overall range. This version generally renders nice faces but I recommend using an adetailer to clean it up.
FAQ
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.





