Trained for 5000 steps (slow cooking) on about 40 images consist of half official illustrations and half synthetic data by Nano Banana Pro.<h3 id="important-usage-info!!!">※Important Usage Info!!!</h3>If you're generating his classic look: Red suit, white ruffled collar, black vest → ONLY USE AT 0.6 STRENGTH. This LoRA is very cwispy when it comes to the default look. Also remember to mention that the collar has three layers, or it might just generate four.However, if you're generating other outfit, feel free to use 0.8-1 strength, it's cooked medium-well on those thanks to the powerful NBP :')The trigger sentence is "an illustration of miledg man, anime-style drawing of a man wearing__"Recommended Aspect Ratio: 3:4, 2:3, 9:16, 1:1※Extra info about the "unnecessary feelings" pose...This pose is very hard to reproduce, unless you're either using ControlNet, or you're extremely lucky (like me lol). And it's actually not my LoRA's issue, I've tried the prompt directly with vanilla Z-Image too, it was just really hard to understand for AI, apparently. <pre><code>But in case you guys are curious, here's the prompt: "an illustration of miledg man, anime-style drawing of a man wearing a red suit with a white ruffled collar and a black vest. The ruffled collar has only three layers. He is standing with his left hand holding his right arm, right arm hanging straight downwards, he is looking down to the left side with an annoyed expression and visible blush, his head turned to the left side fully, teeth clenched slightly. The image is a medium close-up, showing his upper body only."</code></pre><h3 id="enjoy!">Enjoy!</h3>

Files with SHA256: 037506d077872267f14d962874dedc1b19d9bace6a6e7f86feb8bf5640f6af29