This LoRA was trained on Anima Preview 3 using carefully selected color images from the Kimagure Orange Road manga.
When using Anima Preview 3 as the base model, character likeness is more faithful to the original manga style. However, image quality can be somewhat difficult to control. For a good balance between acceptable image quality and accurate character reproduction, it is recommended to keep the LoRA weight between 0.7 and 0.9. You are welcome to experiment with different settings and share your feedback.
When using waiAnima as the base model, the overall image quality tends to be better, but character fidelity is slightly lower compared to Anima Preview 3. The characters appear a bit cuter. In this case, it is recommended to set the LoRA weight to 1.1 or higher.
When generating images, please include the following positive prompts: mdk_manga, ayukawa madoka, 1girl, 1man, solo, long hair, black hair, brown eyes, medium breasts, masterpiece, best quality, amazing quality, score_9, score_8, score_7
And the following negative prompts: worst quality, low quality, score_1, score_2, score_3, blurry, jpeg artifacts, sepia, bad hands
It is recommended to combine tags with natural language in the prompt, as this generally produces better results.
Description
FAQ
Comments (5)
Instead of LoRA, Checkpoint Training was mistakenly selected.
Well, I must admit I don’t fully understand it. I would appreciate some guidance
@liyangpkbbs
"Checkpoint" refers to things like anima or z-image,flux etc.What you created is a LoRA that runs on a checkpoint called anima.
@liyangpkbbs Simply select LoRA again in the Type section under Details.
@fantaseed Oh, this is an obvious mistake. Thank you.