這些是我們在開發 EGAKU 過程中的實戰紀錄:嘗試了什麼、花了多少錢、哪裡失敗、怎麼解決,並以我們自己生成的圖片和影片為例。每篇單獨購買,Pro 以上可免費閱讀全部指南。
Pro 以上可免費閱讀所有指南。
Training a style LoRA on public-domain art only — choosing images, settings, failures and fixes
How we built four anime style models (LoRAs) from images with a clear rights status: how we chose the images, every training setting, where we got stuck, and the checkpoint and strength we shipped.
Earning from what you make — the real numbers, what creators have always done, and what PR and marketing research says
No income is promised. Using public surveys, papers and company figures, this guide shows the reality, how creators earned in the past, how marketing and PR research thinks, and the rules for selling AI-made work in several countries, all with sources.
Keeping the same character — what research knows, what still breaks, and how to actually do it
Why does the second image become a different person, and why does extending a video change the face? Based on 25 papers, this guide explains each method's strengths and limits in plain terms, adds practical notes for creators, and checks them with a worked example made on EGAKU.