https://huggingface.co/MoonShinkiro/goldcan-lora
Trained on 176 images for 5000 steps
Activate with using goldcan, gcsks
Please experiment in any way as you like.

24 messages · Page 1 of 1 (latest)
https://huggingface.co/MoonShinkiro/goldcan-lora
Trained on 176 images for 5000 steps
Activate with using goldcan, gcsks
Please experiment in any way as you like.


another comfy, wonderful model from moon!!! thank you so much for sharing it and making it, it's lovely 
ctd is just built different
What is your recommended weight?
0.8-1 has been just fine for me, it doesn't hurt to set it really low also for just a touch of style

generally don't u want to train a LoRA to work properly at 0.6 - 0.8?
that way it's not overbaked
1girl, green hair, green eyes, solo, hat, sweater, crescent moon, night, looking at viewer, outdoors, turtleneck, turtleneck sweater, beanie, upper body, parted lips, blurry, moon, short hair, black headwear, sky, bangs, blurry background, night sky, ribbed sweater, railing, building, swept bangs, tree, <lora:goldcan-moon:1.0>, goldcan, gcsks, grass, mountain, valley, flower, flower_field,
Hires upscale: 2, Hires upscaler: 4x_foolhardy_Remacri, Lora hashes: "goldcan-moon: 8831766093bb",```


Eh isn't a lora trained where it only works at that range make it.. overbaked? If mine works fine at 1.0 then I don't think it's bad is it.

correct
being able to use a lora at a weight of 1 does not mean it's overbaked
in fact, if something has to be used at a weight of .6-.8, it is overbaked where the weight had to be adjusted down to compensate
being able to use a lora right out of the gate also means less work for everyone
Just training it the way I know, I didn't know what Blob meant since I would rather use my loras at full strength normally and would hate to underclock it. Though I feel it could be overcooked anyways so I have the 3000 step and 4000 step on hand, might just upload them anyways.
o7
it also just helps that things working at 1 makes it easy to adjust the value depending on what you need