blu8k

Through AI illustration, I pursue the beauty of female silhouettes, features, and lines, exploring the true essence of my own aesthetic. Embracing the harmony between the intent of fine-tuned prompts and the serendipity of AI, I am on a journey to find the one piece that truly resonates with the soul. My idea regarding tsubaki.3 : [Mastering Tsubaki.3: Stop Trying to Micro-Adjust with Prompts] AI image generation is fundamentally a gacha. Tsubaki.3 has such a massive latent space that tweaking just a single word can completely overhaul the composition and facial features. In my view, narrowing down your ideal image using words alone is nearly impossible. So, how do you work around it? In my opinion, the real solution is to create "Save Points" using Reference Images or a LoRA to progress. ・Chained Reference Iteration (No training needed): Start with a short prompt to pull a ~60% "decent roll" ➔ Save that point in @image1 and prompt: "Keep face and hairstyle unchanged, only change the outfit" to step-by-step build toward your goal. ・LoRA (When you need zero drift): If you're mass-generating or want fine details locked down with 100% precision, don't wrestle with words—mold the character with a LoRA. Stop trying to roll a 100-point masterpiece in a single shot—using your best outputs as "save points" is the fastest way to avoid getting stuck!