Test 19
Trigger word: m1er0sk
A LoRA model trained on my own pencil sketches.
( but color version is digital painting )
Just like this:
Notice: Ver. color (both for Anima and Illustrious) is generally not suitable for use alone, it's only suitable for use with the sketching LoRa. ( And the prompt "traditional media" doesn't work for this LoRa ). Color LoRa was trained using my colored practice drawings. Because it's for specific practice, these drawings don't use any lines and don't focus on the character's form.
Anima is indeed a very different model from Illustrious, and undoubtedly excellent. However, training and using LoRa with it has caused me considerable anxiety.
It's better suited for digital painting, seemingly automatically masking the texture of traditional pen and paper, making it unsuitable for my casual sketches.
Incidentally, the June LoRa training for Anima failed, so I'm using the May version as a substitute.
Description
FAQ
Comments (12)
Amazing. This will be a great successor to the Watercolor Sketch LoRA on Pony.
Thank you!
But I'm not sure, because the training data only has a few dozen images, and it reflects my strong personal artistic style.
My Illustrious images that don't specify which checkpoint was used are using the oneObsession v2.2 model.
really cool, amazing art, dude
Thank you so much for your support! Training Anima's Lora almost made my anxiety levels explode...
Big up artists who use AI.
Awesome style!
Damn❤️🔥
Thank you. Your art is excellent and I hope you get to release it one day. It's unfortunate when family causes trouble.
"Anima is indeed a very different model from Illustrious, and undoubtedly excellent. However, training and using LoRa with it has caused me considerable anxiety."
Me too. It feels much more like I'm fighting with the trainer compared to Illustrious, especially when it comes to textures and other signs of traditional media.
Thank you, I also hope that one day I’ll be able to post without any reservations.
I haven't fully mastered training Anima lora yet. It can even capture and reproduce the distortion and noise caused by image compression, yet it often fails to replicate paper texture or the uneven bleeding of watercolors, it’s quite baffling.
@Aaaher777 I just made my own lora of a style that is rough and textured. One thing I noticed is that the textures did come out for Anima Base, but all the fine-tuned models I tried overpowered it. In other words, those features might be hidden within the lora, but Anima fine-tunes are just too stubborn to show them. Whereas more "primitive" models like Base leave more up to the lora.
@uncle_krupi I completely agree, I noticed that too.
The same applies to my trained Shirow Miwa (三輪士郎) Lora, it works well on Anima base, but seems to disappear when works on other adjusted checkpoint models. It only becomes slightly noticeable when the strength >1.5.