A requested character LORA for Aishia from Seirei Gensouki: Spirit Chronicles! The examples are using LORA weight 1.0 and highres fix ESRGAN_4x with denoising strength 0.5.
Character trigger: 1girl aishia
Outfit trigger (less consistent for her layered dress): white dress sleeveless dress cleavage front slit frilled dress layered dress
Description
FAQ
Comments (9)
Thanks, will try her out soon~
For God's sake, please tell what model did you train it on?
animefull-final-pruned aka nai
@Akii Thank you so much bro you can't imagine how grateful I am to you. I have been trying to make a lora on my fav character but to no avail is it getting as good as yours. Can you also please provide other details like how many epochs did you use, steps, unet lr, text enc and num_repeats, etc
@AnimeKnower You can refer to existing guides such as this one https://civitai.com/articles/680/characters-clothing-poses-among-other-things-a-guide or https://civitai.com/articles/4/make-your-own-loras-easy-and-free
I'm using lower lr (1e-4 for unet and 1e-5 for text) and a typical run will have around 1000~2000 steps.
@Akii These are good but the only problem I am facing is how many repeats and epochs should I use? Like say I just have 52 images...
@AnimeKnower total_steps = number_of_epoch * repeat * count_of_image / batch_size
To end up with 1000 steps with batch_size=4 and 20 epochs (my typical setting), you need 4 repeats for 52 images
Hello mate! Nice LoRA you have there. You look pretty experienced considering the quality of your LoRAs that's why I was wondering if you could provide me with some information like:
No. of images(you used),
Repeats,
Epochs,
Unet_lr,
Text_encoder_lr,
Network_Dim,
Network_Alpha,
And lastly the Optimizer.
I know my comment might be irritating but trust me, I have no choice. I have been trying to make a LoRA from atleast last 2 months but to no avail am I being successful in it. A FREAKING 2 MONTHS ATLEAST! I would be really grateful if you could provide me with the exact information that you used to create THIS specific LoRA.
Also a side question: Did you also use NSFW images in the dataset?
And did you specifically crop onto the faces of the characters or used it as a whole?
For context, this is the LoRA maker I am currently using: https://colab.research.google.com/github/hollowstrawberry/kohya-colab/blob/main/Lora_Trainer.ipynb
Side question
Hi, I recently used kohya_ss to use my LoRa. When generating images with LoRas try to try different models and prompts to achieve desired results. If your LoRa isn't doing anything close to the intended subject, I would recommend watching a yt tutorial on different techniques
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.