Mia Cha of the Manhwa "Secret Class"
Feel free to buy me a coffee ☕
--------------------------------------------------------------------------------
V1: This is the unchanged file.
V2: Version with new methods that I have been testing.
Let me know if you would like me to implement these methods in future LoRAs.
--------------------------------------------------------------------------------
Trained: 140 images
--------------------------------------------------------------------------------
Recommendations:
Checkpoint: Below are the models that I personally recommend.
Weights: Values between 0.5 and 0.9 are considered decent, with 0.7 being the most suitable in my opinion, although the choice depends on the individual and the type of image he/she wishes to generate.
Vae: Anything-V4.0 (Opcional)
Sampler: DPM++ SDE Karras, 20 to 30 steps.
Sampler: DPM++ 2M Karras, 20 to 40 steps.
Hires.fix: R-ESRGAN 4x Anime6B , 0.5 denoising.
Negative prompt: This is an example of a general negative prompt that was used to generate the images.
" badhandv4, negative_hand-neg, verybadimagenegative_v1.3, ng_deepnegative_v1_75t (worst quality), (low quality), (normal quality), lowres,normal quality, ((monochrome)), ((grayscale)), ((watermark)),bad anatomy,bad hands, bad mouth, bad tongue, bad arms, extra arms, bad eyes, extra limbs "
Description
FAQ
Comments (11)
nice, much appreciated
Thank you! I would like to know which of the versions you prefer, so I can decide which one to implement in the next LoRAs.
@Mornix I think the v1 was somewhat better, but similar. It seemed like v1 could make the character feel a bit more "original"
May I ask how you train these? They are amazing!
Sure, I'll explain it more concisely. First, I choose the character images and set them to 768 x 768 pixels. Then, I clean the images by removing unwanted text or effects. Next, I assign descriptions and filter out problematic tags. Finally, I start training so that the AI learns and improves over time. I would like to know which version you like best to implement it in future LoRAs.
@Mornix I agree this one in particular I think works really well! I'm not sure what the difference is but for what it's worth on my end, your Yu Jeongwon lora was really hard to work with so it's probably best to not use that one as a reference.. but that's just my experience!
@cleopatracleo I thank you for the comment. I will try to publish a more versatile version of Yu Jeongwon LoRA. The main difference is that in version 2 I have worked with scaled images using R-ESRGAN 4x, while in the first version, the model was trained with unscaled images.
@Mornix all good! no pressure, thanks for all the loras you've done and looking forward to the next one! :D
Nice. It really looks exactly like her and with the actual style.
when you gonna add Eun-Ae June) the main character
I am finishing the character dataset and will upload it on Friday.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.