Patreon
Get early access to build and test build, be able to try all epochs and test them by yourself on Patreon or contact me for support on Discord
Recommended settings
Make sure you are aware on the usage instructions of LORA
Sampler: DPM++ SDE Karras (Recommended for best quality, you may try other samplers)
CFG Scale : 5 to 10
LORA weight for txt2img: reccomended 0.2-0.5.
Insutrctions
My advice is to start with posted images prompts then change them, in any case you will find different keywords for each model (wich should be used with a weight of 1.2-1.3).
Also in version description on the right you will find more details about each hairstyle model.
Enjoy!
If you'd like to support my efforts and have access to early and bleeding edge builds, please think about joining my Patreon: https://www.patreon.com/betterCuteAsians
Description
Very simple to use, just use the keyword with a proper weight, like 1.3.
works with both females and males.
FAQ
Comments (22)
great work as always, my only issue with your lora models is size. lora of this nature should be around 9-36mb imo or a single file of 144mb. just imaging downloading 10 files of 144mb only for hairstyles or costumes.
Actually, the proper size for an "Anime LoRA" its 8-32mb, for a realistic one is much more. consider that realistic images contains much more details compared to anime ones. If you want a good realistic LoRA thats the size you should expect, and twice that for a LyCORIS
@antonio_riolo2610 I often use the resize tool included in kohya on Loras over 100Mb to gain space on my hard drive, and end up with files under 40Mo with the basic settings and a high Lora rank
Even with realistic ones, I don't notice that much of a quality loss with the resized Lora.
Did you try to use this kind of tool? Did you had adverse effect when doing so?
Edit : With your hairstyle Loras, all of them end up under 20Mb.
I just posted a review on the half buzzcut with an example of the original size Lora and a resized one, if your curious of the result you can get.
@Iuot i never tryied to use this tool, i'm curious, ill give to it a look next week, thanks for the advice :)
@antonio_riolo2610 You're welcome ^^
I didn't made a lot of test with the tool, the only setting I had changed was "Desired LoRA rank".
Too low, you might loose in quality (default is rank 2 on Kohya).
But it seem there is no "too high" : after a certain rank number, the size of the Lora and the quality doesn't seem to change at all between the resized versions. (I'm not sure what a "Lora rank" is, but it might depend of the trained Lora, so don't be afraid to input a rank "too high" to not have a rank lower than the original.)
Ex : I had resized a Lora with rank 2, 4, 16, 23, 32 and rank 64.
The ranks 2, 4 and 16 weighted something like 2mb, 5mb and 10mb and gave sort of similar result as the original Lora, but acted differently in some ways.
The ranks 23, 32 and 64 were like 18mb, all had the same size, and they acted the same way too. (Close to original).
Awesome hairstyles, i hope could have asian man hairstyles,ancient chinese hairstyle would be cool too.
Thank you so much for sharing these LORAs. I really needed hairstyles. I was using prompt-made hairs :/ Also poses are really nice ^.^
The emo one is great! I'd like an update to it that works with different camera shots, though. It seems to want to snap them to the exact same shot from the previews. It takes a lot of fiddling to get others, and is still unreliable.
So cool, can you do a mohawk? It's really hard to produce with text prompts only
Thank you for these. Any chance of getting a soft afro in the future?
Would adore an obscenely long braid, really struggling to get one that lays on the floor.
If you ever find the time i would really like a pageboy haircut for women. I tried to train one myself but I'm on an amd gpu :/
I love this, but the issue I've been having is it keeps changing the clothing of my character, specifically with the emo hairstyle.
Hey, I haven't tried any of these LoRAs yet (I just discovered them) but this could be solved by altering your prompt. If the hair color is consistently changing something like a jacket then adding to the negative prompt could be beneficial. For example, if you have (red hair) or similar in your positive prompt and the jacket you want as black turns red put (red jacket:2) in the negative prompt. If that doesn't help, trying changing the order of things in your prompt—the more important things go first—or change the weight of the prompt concepts so see if that helps at all.
The training dataset tags say "half_buzzcut_heairstyle" instead, should that be used as the trigger phrase instead?
I would love one massive Lora that combines all these into one
You might be able to get something working with lycoris, but it honesly has a very low chance of working, they'll end up bleeding into eachother.
I like the very long hair, but I wish there was more hairstyles with long side bangs that drape a over the chest just above the navel.
Any chance for some men hair syles as well?
Does this middle parting model work for men?
Nice. How come all the people look Asian?
Can I use it on a fixed photo to achieve the effect of changing the hairstyle?
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.