v1.1
I cut out the background during training and remade it so that no impurities are visible.
I changed the Dimension value to make the file size a little smaller.
Use training model is Animagazine XL V3.
This character is [Sokushi cheat pr] Tomochika Dannoura.
I trained based on anime pictures.
This is the trigger word.[Tomochika_Dannoura]
Distortion appears in the eyes and face. Please use ADtailer or Hires.fix to fix it.
I think everyone has their own interpretations of facial expressions and drawings, so please try adjusting them using the prompts.
If there's anything extraneous in the background, cancel it out with a negative prompt. I think it will disappear in most cases.
This is made by a beginner, so if you have any improvements, please leave a note.
Description
FAQ
Comments (6)
Can I ask your lora training settings and data assets?
I am creating it using Kohya_ss GUI.
It was created with 20 images, 20 repetitions, Network Rank (Dimension) 72, Alpha 36, and 15epochs.
I uploaded this because around 11Epoch, there was less disorder in the clothes.
I'm glad if you can use it as a reference.
@supercubaa01950 oh, which way did you use? images with subject selected or images involved backgrounds with character? also wondering which model used as a base model
@difficult133133
Only one has a white background. The rest was just cropping the anime capture and resizing it. The base model also uses Animagazine XL V3. Did something go wrong? If there are any improvements, I would like to fix them when updating the version.
@supercubaa01950 I just came back to ai drawing so I just wondered because it seemed to be different from SD 1.5. this lora's good but eyelids seem to be drawn weird without adetailer (think it can't help) and I don't know this is correct that clothes seems overtrained. It's just a personal opinion from used-one newbie haha
@difficult133133
XL looks terrible when the face is far away. I think this is an unavoidable specification.
In training, if you modify Network Rank etc., the effect on the pattern may be alleviated.
I'm also a beginner. I'll try it out next time I update.
Thank you for your feedback.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.