license: cc
A Textual Inversion Embedding for Stable Diffusion, trained on SD2.1
This is an embedding to provide a very loose painterly style consistently to varied subject matter (suprisingly hard to achieve in SD2 without embeddings - subject matter shifts aesthetic far too strongly)
Prompts for sample images are very straightforward - subject matter by laxpeint. Or painted by laxpeint.You can put in 'by laxpeint' early or late in the prompt, it seems to be effective no matter what.
For some imagery, it may be important to include 'photography' in your negative prompt, as SD has a strong leand toward it. For instance, the African businessman below would have a fairly photographic suit before adding 'photography' to the negative prompt.
The style is pretty heavy-handed - a loose painterly alla prima kind of look, but responds well to minimizing it. So in Automatic1111 if you ask for (by laxpeint:0.5) you will get a nice mix of the embedding style and whatever other style you might prompt for, or Stable Diffusion's default, bland appearance for 'painting' prompts.
You can rename the .pt file to whatever you like to use as your prompt phrase. Just make it unique.
Also note - all the below sample images were generated using the sampling method DPM++ SDE, but results vary quite strongly with other samplers, such as Euler. Still very painterly, just a bit more soft and atmospheric.
