该模型是一个偏厚涂原画风格的lora模型，训练这个风格的灵感来源于之前接触过原画练习，还有玩儿过类似风格的游戏，实际应用上，人物效果要比纯场景要好一点，可能是训练集里场景较少的缘故，之后还会再优化一下该模型，就想训练出来试试看，触发词: Original painting style使用建议● 大模型推荐使用 <a rel="ugc" href="https://civarchive.com/models/7371?modelVersionId=46846">RevAnimated</a>、<a rel="ugc" href="https://civarchive.com/models/4384/dreamshaper">DreamShaper</a>● VAE建议使用 vae-ft-mse-840000-ema-pruned，Clip skip:2，CFG scale：3~5，lora weight：0.3~0.7● 采样器：DPM++ SDE Karras、DPM++ 2M Karras、Euler a；迭代步数：25~40 ● 放大算法：Euler a、R-ESRGAN 4x+、4x-UltraSharp，放大倍数：1.5~2，重绘幅度：0.4~0.5希望多出一些图，提一些小建议。This model is a lora model with a thick painting style, and the inspiration of training this style comes from previous contact with the original painting practice, as well as playing similar style games. In actual application, the character effect is better than the pure scene, perhaps because there are fewer scenes in the training set, and then I will optimize the model, just want to train it out. Original painting styleUse suggestion● RevAnimated and DreamShaper are recommended for large models● VAE - FT-MSE-84000-EMA-Pruned, Clip skip:2, CFG scale: 3~5, lora weight: 0.3~0.7 are recommended for vae● Sampler: DPM++ SDE Karras, DPM++ 2M Karras, Euler a; Number of iteration steps: 25~40● Magnification algorithm: Euler a, R-ESRGAN 4x+, 4X-ultrasharp, magnification: 1.5~2, redrawing amplitude: 0.4~0.5I hope there are more pictures and some suggestions.

Files with SHA256: a4c8b1cf5fbbbeba485226342c38d136d7f0299cc7987500d8bd44259223765a