So I decided to make a FLUX LORA based on existing works that I made, just "port" it by using existing datasets but refined to make it work with FLUX image generation that emphasizes natural language prompting rather than using usual keyword tags.This LORA will works on most FLUX-based models if found working and using FLUX image generation requires to have at least Nvidia GPU with 12GB VRAM installed. Since FLUX uses 8-bit precision method so you may get some inconsistencies unless any additional image-fixing LORA had been applied.Prompt guide and configurations:NOTE: Requires SD WebUi Forge (not reForge) or ComfyUI for FLUX image generationPrompt text examples (first one are trigger words):<pre><code>chines3luka. a woman in red chinese dress and a very long pink hair standing with alluring pose.</code></pre>Example 2:<pre><code>chines3luka. a woman in red chinese dress and a very long pink hair standing with alluring pose. She has a fairer skin, beautiful blue eyes, large breasts with visible cleavage and wearing a garter belt

(poor photography:-2), (extra legs:-2), (extra fingers:-2)</code></pre>Sampler (tested working): Euler (non-ancestral), DPM2, DPM++ 2M, LMS, Heun, HeunPP, IPNDM, IPNDM_V, DEISSchedule Type: Simple ONLYDistilled CFG: 3.5Steps: 30CFG: 1

Megurine Luka Chinese Dress FLUX

First version of LORA. Trained on 512x512 images. May get inconsistencies when trying to make image beyond 1024x1024.

Files with SHA256: 9caa3688665f3fd1add2e42af2624d3e6b5907adbbeeaca69dfbb46a24d436de