give your reviews on how I can do better with my loras k thx.
this lora looks unstable, will see in the future.
Description
FAQ
Comments (4)
Your LoRA is well trained, but there are ways you could improve the tagging and training.
The "Blue eyes" tag appears to give Robyn a shocked expression and isn't necessary in the training. Unless you want to generate Robyn with green eyes, you can omit "blue eyes" as a tag in training.
Try tagging Robyn as "WolfWalkerRobyn". This doesn't appear to be a problem in your training, but it's possible that the base model has training for the "Robyn" tag and it might train better with a unique tag.
You could also just train a Wolf Walker LoRA that includes training data for several characters and include a style tag on every image, for example "WolfWalkerStyle." This would allow you to gen the characters in different styles and have many characters in one LoRA.
A third, but lesser improvement would be resizing the LoRA. 217.87 MB isn't awful, but I bet a single character could be trained better at a higher dim with less alpha. Say 64 dim and 8 alpha then resized to 32 dim or 16 dim.
I'm guessing the lora was trained with clips from the movie. There are a lot of action scenes where she has an "oh fuck!" expression. Hence why it's probably cooked pretty hard into the data.
Thank you for you comment, I will look forward into this.
I love wolfwalkers! Thanks for making these Loras.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.