v2.0 was baked with the SAI offset LoRA - this made the model interact poorly with fine-tuned checkpoints also utilizing the same. But even with the base SDXL model, lended a warm color cast to outputs. All-in-all, 2.0 was flawed. 2.1 is all that 2.0 was supposed to be, with the SAI offset LoRA stripped.
Version 2.0 after working on a full checkpoint model, Painter's Checkpoint, I felt I had really completed everything I sought to capture in this particular painting style. I got the balance of detail and gestural mark making and atmosphere that represents a personal ideal aesthetic for the medium of oil painting.
But I missed the versatility of a smaller LoRA file, which can be reduced in strength, combined with other LoRAs more effectively, and run with models other than the base SDXL model. So - I ran a LoRA extraction from my own checkpoint model. That thing was nearly as large as an SD1.5 model, but could be resized down to this 200k file, maybe my last ClassiPeintXL LoRA. Nearly exactly capturing the style of the Painter's Checkpoint, but allowing so much greater versatility.
This is by far the most consistent and versatile LoRA for generating images with SDXL that capture the feel of genuine oil paintings for whatever subject matter you want to prompt.
Recommended settings for use:
You can go here (pastebin) to download a ComfyUI workflow like what I used, but without custom nodes that are embedded in my image uploads on CivitAI.
Start with a full 1.0 LoRA strength and adjust down to 0.7 or 0.8 for a subtler painterly effect. You can adjust upward (to 1.2 or maybe a little more) to maximize the painterly appearance, but it can start to introduce some quirks
Use the LoRA with your preferred SDXL model with no refiner. I have so far just stuck with base SDXL1.0 but other finetunes work great as well.
I recommend the DPM samplers, but use your favorite. Some may produce softer painting styles that don't suit my taste as much but whatever you prefer is great.
Don't do anything special for your prompt - just describe what you want to see. I usually start my prompt with 'oil painting of ...' If you skip this, sometimes an image can tend toward more photographic, with blurring backgrounds and a kind of odd mesh between painterly and photographic bokeh. It can be a nice look and you can deliberately get it some by using particular photographer's names. But prompting 'oil painting' keeps everything on target with the LoRAs intended style.
Feel free to add artist names (Klimt, Mucha, etc) and the LoRA will merge its style with theirs with usually fantastic results.
I'm happy to share my work for free, just as the base model SDXL came free to us all. But if you felt compelled to offer some material form of thanks, you can buy me a coffee, or you can buy my (darn near unreadable, I warn you) short story 'Willem & Ellene'.