I extracted a LoRA file from my dreambooth custom model made for hand-painted object icons generations specificly. (REPO used for this: Kohya_ss) For better understanding of how to use this model to get the best results visit the official model page. The main advantage comparing to whole custom model is of course a light weight and possibility of using this LoRA file with any 1.5 based model. (I've tested it for a while and you can get some creative results - you have a taste of it in sample generations). Outputs presented on this page are cherrypicked, but I've also did some random testing (you can check it here) and with a bit of workflow, experimenting and testing and a bit of luck you can get quiet interesting and expected results. I strongly encourage you to test it with different models and a range of parameters and prompts (prompt editing is really helpful in this case).
In sample images beside overall presentation I focused also on showing the same subject but with different models and parameters. | 
发表于 2023-6-7 20:17:53