I'm going round in circles with the training of this model on CivitAI, so here's a variation trained with different steps (1000, 2000 and 3000) on OpenArt. I'm going to play with it a bit to test the rendering and I'll let you do the same.

Otherwise, the main lines are in the title, Surreal Anime, a style that's more or less ‘Anime’ but that can go off the rails from time to time to try and get off the beaten track.

Have fun and share your experiments!!!

After a wide range of tests on the 3 versions, here's a short report/guide. Feel free to test it yourself and share your images.

** 3k steps version: the easiest to use, good results with a strength of 1 and almost everything you suggest will take the style of the lora. On the other hand, it's sometimes a little cartoonish and lacks finesse and subtlety. In short, it's easy but not always extremely creative.

** 2k steps version: Sometimes you have to search a bit and vary the strength between 1 and 1.50 depending on the prompts. The results are finer and more delicate than the 3k version, while remaining technically fairly simple. In short, a well-balanced version.

** 1k steps version: This is technically the most complicated to manage, you sometimes have to go very high to influence the prompt in the style of the lora. Some of the images in the gallery go up to 1.7-1.8 depending on the prompts. However, if you find the right settings, this is certainly the most expressive and delicate version. Among the most beautiful images I've been able to generate so far. In short, complicated, but give it time and freedom and you'll be positively surprised.