Retrained from scratch. V6 renders legible text inside images for the first time, follows longer and more literal prompts, and over the following months ships the reference-image toolkit that matters most for narrative work: Style Reference (--sref) to lock a look and Character Reference (--cref) to hold a face across shots. Paired with the 2024 image-to-video explosion — Luma, Kling, Runway Gen-3 — V6 is what makes multi-shot AI stories, one character and one style across many frames, actually feasible.