<hr />v1.0 (2024.9.12):parameter:<ul><li>Sampler: Euler a</li><li>Steps: 30</li><li>Size: 700~1200 (width and height) (Lower resolution (such as 512 * 768) can lead to a significant decline in image quality)</li><li>prompts: masterpiece, best quality</li></ul>

illustration ani

This model is fine-tuned from Novel AI 1.5. I have tried other base models before, but except for NAI 1.5, SD 1.5, and a few XL models, training LoRA or direct fine-tuning with other base models often resulted in poor outcomes, and I don't know the reason.(Anyone who knows please tell me, I want to train based on other well-trained anime models such as tponynai3)Due to the limitations of my GPU, I can only perform LoRA training with XL models. Although models like Kohaku and Animagin do not yield terrible results, the final effect is still not very satisfactory.Once I collect enough training data, I may consider renting a platform to fine-tune a large XL model.

Files with SHA256: 4ba873652edfc22000808a686459c34fed9f0eccc9ef7fb9497b29cb32f0692f