Note: DO NOT use "anime style" positive prompt with this model (or maybe do, I'm not the boss of you)
Images generated using Fooocus-MRE.
This is the config template I use for all samples:
{"prompt": "3d shading, 4k, bokeh, blurred background\n\n<YOUR PROMPT HERE>", "negative_prompt": "poor quality, western animation, deformed hands, 2d", "style": "sai-fantasy art", "seed": 7291318656256760812, "width": 768, "height": 1344, "p_txt": "ethereal fantasy concept art of (photorealistic:0.4), 3d shading, 4k, bokeh, blurred background\n\<YOUR PROMPT HERE>. magnificent, celestial, ethereal, painterly, epic, majestic, magical, fantasy art, cover art, dreamy", "n_txt": "photographic, realistic, realism, 35mm film, dslr, cropped, frame, text, deformed, glitch, noise, noisy, off-center, deformed, cross-eyed, closed eyes, bad anatomy, ugly, disfigured, sloppy, duplicate, mutated, black and white, poor quality, western animation, deformed hands, 2d", "sampler": "dpmpp_3m_sde_gpu", "scheduler": "exponential", "performance": "Quality", "steps": 60, "switch": 40, "sharpness": 2, "cfg": 7, "base_clip_skip": -2, "refiner_clip_skip": -3, "base_model": "moefusion.safetensors", "refiner_model": "sd_xl_refiner_1.0_0.9vae.safetensors", "l1": "None", "w1": 0.5, "l2": "None", "w2": 0.7, "l3": "None", "w3": 0.8, "l4": "None", "w4": 0.6, "l5": "None", "w5": 0.5, "img2img": false, "revision": false, "zero_out_positive": false, "zero_out_negative": false, "control_lora_canny": false, "software": "Fooocus 1.0.43 MRE"}For those using other platform here's the fantasy style for reference:
{
"name": "sai-fantasy art",
"prompt": "ethereal fantasy concept art of {prompt} . magnificent, celestial, ethereal, painterly, epic, majestic, magical, fantasy art, cover art, dreamy",
"negative_prompt": "photographic, realistic, realism, 35mm film, dslr, cropped, frame, text, deformed, glitch, noise, noisy, off-center, deformed, cross-eyed, closed eyes, bad anatomy, ugly, disfigured, sloppy, duplicate, mutated, black and white"
},At least for me using "anime style" really messes with the output images, producing unprofessional looking results.
This is my first SDXL model so take that in mind.
Model was made by training a big LoRA for the general style, and merging it together with a bunch of other LoRAs for a painterly look.