If you want to achieve the look shown in Figure 2, please use this checkpoint; otherwise, the portraits may turn out in a realistic style (though that looks pretty good, too).
https://civarchive.com/models/2357432/tulpainterly-xl?modelVersionId=2651420
Illustrious-style oil painting LoRAs are relatively scarce, yet I happen to be quite fond of this aesthetic. Consequently, I decided to try training one myself. However, the results seem a bit inconsistent—sometimes the output doesn't quite capture the look of a true oil painting. If there is enough interest, I will work on expanding the training dataset to improve it. Using the model is very simple: just include the trigger words. Below, I've included one of my favorite generated images.
It turns out that what I actually needed to train was a checkpoint, not a LoRA. However, I'm not quite ready to train a checkpoint myself just yet—but fortunately, there are some excellent ones already trained by experts available on Civitai! I've actually used them before.
V1↓
V1.5+tulpainterly-xl
Haha, it seems it unintentionally achieved an unexpected effect. Before using Lora...
After using LoRA
Description
Actually, I've already finished training, but I'm too lazy to upload it. Since it's an early training session, if feel not enough for oil painting. try increasing the intensity... zzZZ...
FAQ
Comments (2)
Truly interesting, and a very fun LoRA in my opinion. Thank you for informing us, I find this all fascinating and relevant to our work with these tools!
Thank you! I’m glad you find it interesting.I trained it as a LoRA instead of a checkpoint because it can combine with different base models in interesting ways. With stronger-style checkpoints it works as a decent optimizer, and with milder ones you can get different oil painting brush effects by adjusting the strength. I also often use it with another style LoRA of mine (though I don’t dare share that one — the company’s legal team claims to be invincible…).