SDXL version of my Melfina lora.Examples of v1.0 sdxl are generated with Foocus, log for the second picture with holding a spear: <pre><code>2023-12-01_23-35-09_9035.png

Prompt: mel, light smile, light purple hair, blue eyes, long hair, holding polearm, head wings, (purple hairband:1), armor, breastplate, pauldrons, angel wings, fingerless gloves, waterfall background

Negative Prompt:

Fooocus V2 Expansion: mel, light smile, light purple hair, blue eyes, long hair, holding polearm, head wings, (purple hairband:1), armor, breastplate, pauldrons, angel wings, fingerless gloves, waterfall background, highly detailed, cinematic, great composition, dramatic atmosphere, dynamic, thought, open friendly, warm, inviting, sincere, intricate

Styles: ['Fooocus V2', 'Fooocus Enhance', 'Fooocus Sharp', 'Dark Fantasy', 'Fooocus Negative'], Performance: Quality

Resolution: (768, 1280), Sharpness: 2

Guidance Scale: 4, ADM Guidance: (1.5, 0.8, 0.3)

Base Model: bluePencilXL_v200.safetensors, Refiner Model: None

Refiner Switch: 0.5, Sampler: dpmpp_2m_sde_gpu

Scheduler: karras, Seed: 2879922520723956751

LoRA [sd_xl_offset_example-lora_1.0.safetensors] weight: 0.1, LoRA [melfina-test-sdxl2.safetensors] weight: 0.75</code></pre>

Melfina (Black Summoner) SDXL

Version with some generated nudes added to dataset, makes outfit more flexible. And when it comes to fidelity, it doesn't seem any inferior to version trained on just screencaps, also I managed to get good results at lower rank, so it's much smaller.

Files with SHA256: 3308833880c8154947e4c3097330798a3c202eee7c839a1c03ab83b5ffbb0fdd