- likes
- 35
- comments
- 5
Post
The best open-source image models make great pictures. We needed more: the same Character, recognizable through every style, scene, and page a story asks for. So we post-trained our own. Today we're sharing the CAI-Image family, the models behind (c.ai) Comics and image generation across the app. ✦ A multitude of art styles, 245 camera positions, and multi-Character scenes that hold one style even when every Character starts in a different look. ✦ A dedicated manga model that keeps Characters and dialogue consistent across a whole comic. Some creators are already past 100 pages. ✦ All of it runs on CAI-MM-Studio, our full-stack multimodal infrastructure. Owning every stage is why capabilities ship in weeks and inference stays affordable at consumer scale. New papers land every week, and many start with a technique and look for an application. We work in the other direction: our research starts from what storytelling asks of a model, and millions of people creating on Character.ai every day show us exactly what to train for. Read the full post: https://lnkd.in/gmNW3A6P