Follow me on X/Twitter for exclusive content: https://x.com/Citron_Legacy 😁
Please checkout all my Dragon Ball Loras: https://civarchive.com/collections/268
Please support the artist by checking out their page🔞: https://www.pixiv.net/en/users/16220317
Description
There was a bounty (https://civitai.com/bounties/463) for this art style and also I'm pretty sure someone commented asking for it.
Yamamoto's signature is very prominent so I had to do a lot of photoshopping to get rid of it in the dataset.
Training Data
Trained on: 120 images
Training Model: AnyLoRA_noVae_fp16-pruned
flip_aug: True
Num of Repeats: 10
Unit is Epochs or Steps: Epochs
Number of Epochs or Steps: 10
Training Batch Size: 2
Network Dim: Dim 64 and Alpha 32
Lora Creation Process took: 1:06:02
FAQ
Comments (10)
Thanks bro, I was just going to train a LoRA with that style, you beat me
Glad you like it!
Please share if you make any cool images. I'm always eager to see what my Loras can do.
really nice, share yours as well please
I'm trying to create an original character, but every time I try to generate it, it changes in appearance even though I put the same description. Do you know a way to make the character's specifications constant?
I'm planning on making a guide on creating an original character.
Here is the informal start of a guide:
Regardless if you use this lora or not checkpoints in general do not consistently create the exact same character every time. (unless the character was already trained)
Now that I've made a few OC I can say that the process in general takes a lot of image generations and then only a small percentage will achieve the look you want. You would use this small quality dataset for the next step.
For example, look at this Memira post https://civitai.com/posts/812240. The images are not very good but they capture a rough draft of the look I want. Hair, skin color, outfit is all good but the eyes, horns, and wings are a mess.
The next step would be to take these images and in the img2img tab refine them until they looked perfect.
At minimum we need 8 images (and different art styles would be ideal) so I have 10 images with 2 art styles.
I'll explain this process better in the real "Creating an OC Lora Guide" so this is just the start to give you an idea of what I mean.
Edit: after reviewing the 10 images I have some doubts about this design. Its a little too similar to Shemira. There is nothing unique about it. So I'm thinking making giving her short white hair and a body suit instead of a dress would change it up.
I'll either try to generate images like this or use img2img to inpaint the things I want.
All that to say, I dont know how to achieve "constant character specifications" without putting in manual work.
If making an OC was easy there would be a lot more of them on Civitai.
@CitronLegacy Thank you, I will wait for this guide
it would be great if you make an SDXL version of it
Noted.
I added it to my list of SDXL Loras to make.
@CitronLegacy great
doesnt work with ADetailer u_u
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.