Type | |
Stats | 2,046 3,156 |
Reviews | (209) |
Published | Feb 17, 2024 |
Base Model | |
Training | Steps: 3,000 Epochs: 100 |
Usage Tips | Clip Skip: 1 Strength: 1 |
Trigger Words | digital painting |
Hash | AutoV2 F0839026CF |
version 1.3 - brings better portraits (still very loose and gestural!) and aerial scenes.
Just write 'Digital Painting' and your straightforward prompt ideas
Fine detail? Meticulous rendering? Photorealism? Sometimes you don't want any of those - you just want a fast digital sketchy loose feeling. With this LoRA, you can crank out those quick Cintiq-drawn matte paintings and concept sketches all day long.
Probably, this is the style I should have been pursuing when training LaxPeintXL, but kept steering away toward greater detail and rendering. That model converged a lot with ClassiPeintXL. So let's take a step back, start over, and go all-in for that digital sketch aesthetic.
Why train a model like this? Can't you get this look out of the base SDXL model?
You can get much this look with the right prompts straight out of the base SDXL model using artist names like Marc Simonetti, Ismail Inceoglu, Craig Mullins etc. But these artist prompts also tend to push the subject matter toward particular time periods or genres. This LoRA will get that loose gestural digital look without needing to use any artist's name, and so retain much broader subject matter flexibility. Period piece, sci-fi, still life, landscape - prompt for anything and consistently get this same aesthetic.